eCognition 8.

0: Guided Tour Level 1

Getting started Example: Simple building extraction

Classification of buildings using elevation and RGB data
Version 3.0

Legal Notices Copyright and Trademarks Copyright 2010, Trimble Navigation Limited. All rights reserved. Trimble, the Globe & Triangle logo and eCognition are trademarks of Trimble Navigation Limited, registered in the United States and in other countries. All other product names, company names, and brand names mentioned in this document may be trademark properties of their respective holders. Protected by patents US REG NO. 3,287,767; WO9741529; WO9802845; EP0858051; WO0145033; WO0205198; WO2004036337 ; EP1866849; US 6,229,920; US 7,117,131; US 6,832,002; US 7,437,004; US 7,574,053 B2; US 7,146,380; US 7,467,159 B2; US 20070122017; US 20080008349; US 12/386,380. Further patents pending. The contents of this manual are subject to change without notice. Although every effort has been made to ensure the accuracy of this manual, we can not guarantee the accuracy of its content. If you discover items which are incorrect or unclear, please contact us using this form: http://www.ecognition.com/content/training-inquiries Find more information about the product or other tutorials at http://www.ecognition.com/ http://www.ecognition.com/community Release Notice Release date: August 2010

Table of Contents
Getting started Example: _______________________________________________ 1 Simple building extraction _______________________________________________ 1 Table of Contents _______________________________________________________ 3 Introduction ___________________________________________________________ 5 Lesson 1 Introduction to Rule Set development ___________________________ 9 1.1 Get the big picture_____________________________________________ 9 1.2 Which data to use?____________________________________________ 10 1.3 Develop strategy _____________________________________________ 11 1.4 Translate the strategy into processes _____________________________ 12 1.5 Review your intermediate results ________________________________ 14 Lesson 2 Get the big picture of building extraction analysis task____________ 15 2.1 How buildings are represented in spectral image data? ______________ 15 2.2 Do buildings have significant shape characteristics? _________________ 16 2.3 Can buildings be separated using context information? ______________ 16 2.4 Is elevation information stabile characteristic for buildings?___________ 16 2.5 Conclusion: Which data to be used? Which general ideas? ____________ 17 2.6 Overview Guided Tour 1: Simple example of building extraction _______ 17 Lesson 3 Importing data ______________________________________________ 18 3.1.1 Open eCognition Developer Rule Set Mode 18 3.2 Option 1: Use workspace and import template _____________________ 19 3.2.1 Create a new workspace 19 3.2.2 Import data in the workspace with an import template 20 3.2.3 How to open a created project 21 3.2.4 How to save a project 21 3.3 Option 2 (Trial version): Creating an individual Project _______________ 22 3.3.1 Define the data to be loaded 22 3.3.2 Overview: Create Project dialog box 23 3.3.3 Define layer alias 24 3.3.4 Assign No Data values 25 3.3.5 Confirm the settings to create the project 26 Lesson 4 Developing the core strategy by evaluating the data content ______ 27 4.1 The image visualization tools ___________________________________ 27 4.1.1 Zooming into the scene 27 4.1.2 Display the DSM 28 4.1.3 Open a second viewer window to compare data 29 4.2 Evaluate the data content ______________________________________ 29 4.2.1 Evaluate the elevation data 29 4.2.2 Evaluate the RGB data 30 Lesson 5 Creating image objects _______________________________________ 32 5.1 Strategy for creating suitable image objects _______________________ 32 5.2 Translate strategy into Rule Set use algorithm multiresolution segmentation _______________________________________________ 33 5.2.1 Why multiresolution segmentation algorithm? 33 5.2.2 Insert rule for object creation 35 5.3 Review the created image objects _______________________________ 38 Lesson 6 Initial classification: classifying all elevated objects ______________ 39 6.1 Strategy to classify buildings based on elevation information _________ 39 6.2 Translate the strategy into Rule Set Object mean of DSM, algorithm assign class _______________________________________________________ 40 6.2.1 Find the feature and the threshold to represent all elevated objects (mean value of DSM) 40 6.2.2 Write rule for classifying all elevated objects 42 6.3 Review the classification result buildings and trees are classified ______ 47 Lesson 7 Refinement based on DSM: un-classifying trees __________________ 49 7.1 Strategy to separate buildings from trees: use standard deviation of DSM 49

Translate the strategy into Rule Set Standard deviation of DSM, algorithm assign class__________________________________________________51 7.2.1 Find the feature and the threshold to separate buildings from trees (Stddev. of DSM) 51 7.2.2 Write rule to separate buildings from trees 52 7.3 Review the classification result trees are de-classified; some vegetation and shadows objects are still misclassified _____________________________55 Lesson 8 Refinement based on spectral information ______________________ 56 8.1 Strategy to refine buildings based on spectral information the spectral layer ratio ____________________________________________________56 8.2 Translate the strategy into Rule Set Ratio of green, algorithm assign class ____________________________________________________________58 8.2.1 Find the feature and the threshold for spectral refinement (ratio green) 58 8.2.2 Translate strategy in a rule for refinement based on a spectral feature 60 8.3 Review the classification result the vegetated areas are de-classified; small objects still misclassified ________________________________________63 Lesson 9 Refinement based on context _________________________________ 64 9.1 Strategy to refine buildings based on context information surrounding neighbor objects ______________________________________________64 9.2 Translate Strategy in a rule for refinement based on a context feature ___66 9.2.1 The class-related feature Relative border to 66 9.2.2 Prepare the Rule Set structure 68 9.2.3 Insert process to classify 69 9.3 Review the classification result only small objects remain misclassified _71 Lesson 10 Refinement based on shape ___________________________ 72 10.1 Strategy to refine buildings based on shape information generalize objects, separate them by size ___________________________________72 10.2 Translate strategy in a rule for refinement based on a shape feature _____74 10.2.1 Merge the image objects 74 10.2.2 Find the feature and the threshold for refinement based on shape 76 10.2.3 Translate strategy in a rule for refinement based on the Area feature 77 10.3 Review the classification result ___________________________________79 Lesson 11 Exporting the result in a vector layer____________________ 80 11.1 Insert process to merge unclassified objects _______________________80 11.1.1 Prepare the Rule Set structure 81 11.1.2 Insert process to merge all unclassified objects 81 11.2 Insert process to export shape file with attributes____________________82 11.2.1 Create the feature Class name 82 11.2.2 Insert process to export vector file 83 11.3 Review the exported result ______________________________________85 Where to get additional help and information? ____________________________ 86 eCognition Community and Rule Set Exchange platform ____________________86 User Guide and Reference Book ________________________________________86 Additional Guided Tours and Tutorials ___________________________________87 eCognition Training__________________________________________________87 Consulting _________________________________________________________87 Buy Software and Services? ____________________________________________87

7.2

Introduction

Introduction
About this Guided Tour
Welcome to the Guided Tour Getting started Example: Simple building extraction. This tour is written for novice users of the eCognition software. This Guided Tour will focus on the basic steps involved in developing a Rule Set using eCognition Developer. Further information about eCognition products is available on the website http://ecognition.com.

Requirements
To perform this tutorial you will need eCognition Developer installed on a computer.

All steps of this Guided Tour, except the batch processing, can be done using eCognition Developer or its trial version (http://www.ecognition.com/products/trialsoftware ). This edition is designed for self-study.

Data included with the Guided Tour

Image data Image data can be found in the folder /Data. There are two .img files available for this Guided Tour, RGB_Level1_Simple_Example.img representing the aerial RGB data and DSM_Level1_Simple_Example.img representing the LiDAR data. Please note that the geoinformation of the data has been deleted. Data courtesy of Woolpert, a premier supplier of ADS40 imagery and LiDAR data Rule Sets Rule sets are available in the Rule Sets folder. Whenever the Guided Tour refers to a Rule Set it is to be found in this folder. Project The ready set up project can be found in the folder project. Import template Import templates can be found in the folder Data. Whenever the Guided Tour refers to an import template it is to be found in this folder.

eCognition 8.0: Guided Tour Level 1 Simple building extraction

How to use this Guided Tour

Introduction Information Symbols at the side of the document If the side is hachured and Introduction is added, this indicates that the text is giving a general introduction or methodology about the following Lesson, method or exercise. If the side is hachured and Information is added, this indicates that the text is giving information about the following exercise. If this symbol is shown, you have to follow the numbered items in the text. If you just want to work through the exercises without reading the theory part, follow only this sign.

Action!

Settings Check

If this symbol is shown, compare the settings shown in the screenshot with the settings in the according dialog box in Developer.

Rule Set Check

If this symbol is shown check the screenshot of the Process Tree with the content of the Process Tree in Developer.

Result Check

If this symbol is shown check the screenshot aside with the result Developer. It should look similar.

Symbols for next step in Rule Set development

The following symbols show you always the next step to do. They will be prompted in the text. Get the big picture This diagram expresses that the next step to do is to develop the core strategy, containing the choice of the data to use. Get the big picture:

Figure 1: Get the big picture

Import data This diagram expresses that the next step to do is to import your data:

Introduction

Figure 2: Import data

Develop strategy This diagram expresses that the next step to do is to develop your next Rule Set development strategy step:

Figure 3: Develop strategy

Translates Strategy into Rule Set This diagram expresses that the next step to do is to translate your strategy into a Rule Set:

Figure 4: Translate into Rule Set

Review result This diagram expresses that the next step to do is to review your results:

Figure 5: Review Result

eCognition 8.0: Guided Tour Level 1 Simple building extraction

Refine/expand strategy This diagram expresses that the next step to do is to go back and develop your next strategy step, because the result did not meet your requirements:

Figure 6: Refine/expand strategy

Ready for export This diagram stands for the stage where you are satisfied with the result and you can export:

Figure 7: Result is OK, ready for export.

Symbols for stage in Rule Set development

There are also diagrams in the header of the document showing the current stage you are:

Get the big picture

Import the data

Develop strategy step

Write Rule Set

Review intermediate result

Export result

Introduction

Lesson 1 Introduction to Rule Set development

This Lesson covers the following content
Get the big picture Which data to use? Develop strategy Translate the strategy into processes Review your intermediate results

1.1

Get the big picture

The most important tool for creating a Rule Set is your expert knowledge, for example as a remote sensing professional or a geographer, and the ability to translate your recognition process into the eCognition language: the Cognition Technology Language. Behind every image analysis there is a methodology.

N Get the big picture of the general analysis task O=Choose the data P=Develop a strategy Q=Translate the strategy into a Rule Set R=Review the results S=Refine the strategy and Rule Set if necessary T=Export the results

Figure 8: The Rule Set development process.

The diagram and the symbols above will guide you through the whole tour. The individual symbols will be visible in the header of the page, with the current stage highlighted. Whenever there is a change e.g. from strategy development to Rule Set writing etc. the diagram above will show you the next phase.

eCognition 8.0: Guided Tour Level 1 Simple building extraction

1.2

Which data to use?

First step in Rule Set development is choosing the data to be used for analysis. In this Guided Tour buildings are extracted. Data requirements: The data should have a fine resolution to get correct outlines. Due to the large variability of building types and roof tops, classification only based on spectral information is quite tricky. But elevation information is a very consistent source of information to classify buildings.

eCognition allows integration of data from different sensors and with different resolutions. In this example high resolution spectral images and a lower resolution DSM (Digital Surface Model) will be loaded together without problems. The software works with objects, therefore the resolution has direct impact on the object creation. The most stabile information about buildings is their different elevation compared to their surrounding. LiDAR data contains accurate elevation information in high resolution. Spectral information will help to separate buildings from other elevated objects like trees. Conclusion: To extract buildings a combination of spectral (here RGB) and LiDAR data is an optimum.

Efficiency in development
It is important to choose a representative subset of your data to develop an image analysis routing. The first step always is to evaluate your complete data set and then choose an area you want to start with. Working with a representative subset saves development time, because the processing times are lower than testing every step on the whole data set. Nonetheless it is recommended to test the Rule Set on regular basis and not too late (!) on the complete set, to avoid surprises at the end of the calculated development time.

10

Introduction

1.3

Develop strategy

To find the best suited features and algorithms, always ask yourself, why do I recognize something as a building, tree, lake? Ask: how do I have to modify the objects so that they fulfill the criteria?

General rules
Developing a Rule Set is an iterative process Always start with the class with the most significant features. Only process in the domain you are interested in.

eCognition works with objects. The image pixels are grouped together and as a result, much more information is available. Information like the spectral signature of the whole object, the shape and size and also context information is available. All these attributes can be used and combined for classification.

Tools to develop strategies

It is essential to know how to retrieve the necessary information to describe a class or to set up rules for processing, since the raw image may not clearly show the information that is needed. eCognition Developer contains tools that support your recognition process. In eCognition Developer visualization tools and tools giving back feature values help you to display the information contained. Simply said: What you do not see/know, you can not describe. If the information is hidden in the data, use image filters provided in the software (like an edge filter) to be able to use this information as a rule. The visualization tools help to evaluate the data and/or the result

Figure 9: Different visualization possibilities of the loaded data and the classification results.

11

eCognition 8.0: Guided Tour Level 1 Simple building extraction

The Feature View helps to evaluate the feature values for the entire scene

Figure 10: Viewer displaying feature values in grey range or in a color range.

The Image Object Information window helps to evaluate values for individual objects or evaluate classifications.

Figure 11: Image Object Information window.

1.4

Translate the strategy into processes

To be able to write a Rule Set you have to translate your strategy into processes. This is done in the Process Tree window. There you can add, edit and sort the processes. In the Process Tree window you can define:

12

Introduction how shall be processed (algorithm) what and where shall be processed (domain) under which conditions (domain condition)

Figure 12: Diagram about the Process Tree and the individual processes.

A single process represents an individual operation of an image analysis routine for an image or subset. Thus, it is the main working tool for developing rule sets. A single process is the elementary unit of a rule set providing a solution to a specific image analysis problem. Every single process has to be edited to define an algorithmN to be executed on an image object domain O Combine single processes in a sequence by building a Rule Set. You can organize a process sequence with parent and child processes which are executed in a defined order. You can also load an existing Rule Set, save your rule set, and execute individual processes or the complete rule set. Developing a Rule Set does not require you to write any code, rather one selects from a set of predefined algorithms within the graphical user-interface.

13

eCognition 8.0: Guided Tour Level 1 Simple building extraction

1.5

Review your intermediate results

As already mentioned, developing a Rule Set is an iterative process. You start with a base strategy, implement the actual rules in the software, check the result, then you step back and refine/expand the strategy, modify the Rule Set and again check the result. Until you are finally satisfied with the outcome.

Figure 13: Reviewing your intermediate results is a crucial step to develop the next steps of your analysis

Figure 14:or to decide that the accuracy of the classification is ready for export

14

Get the big picture of building extraction analysis task

Lesson 2 Get the big picture of building extraction analysis task

This Lesson covers the following content
How buildings are represented in spectral image data? Do buildings have significant shape characteristics? Can buildings be separated using context information? Is elevation information stabile characteristic for buildings? Conclusion: Which data to be used? Which general ideas? Overview Guided Tour 1: Simple example of building extraction

To get the big picture you have to think about which general and consistent characteristics are contained in the data in the object-shape and whether there are context based characteristics.

2.1

How buildings are represented in spectral image data?

Roofs seen from above have a wide spectral variety, from colored roof tiles, to metal roofs. Spectral Information about buildings is inconsistent information

15

eCognition 8.0: Guided Tour Level 1 Simple building extraction

Figure 15: Example for different types of buildings.

2.2

Do buildings have significant shape characteristics?

Buildings have also a wide variety in shape and in size. They can be rectangular to circled, small family houses to bigger industry buildings. At least buildings have compared to other classes a quite large size. Partially inconsistent information

2.3

Can buildings be separated using context information?

Buildings have shadows, as they are elevated. Roofs are generally not covered by water or vegetation. consistent context information

2.4

Is elevation information stabile characteristic for buildings?

Figure 16: The DSM image layer.

Buildings have a higher elevation than their surrounding. At the edges of a building there is a very steep slope. The elevation changes suddenly. Consistent Information

16

Get the big picture of building extraction analysis task

2.5

Conclusion: Which data to be used? Which general ideas?

The most consistent and relevant characteristics of buildings is their elevation. Therefore elevation data (converted from LiDAR data) is chosen for the analysis task Building extraction. The elevation information is used for segmentation and classification. Only elevation is not sufficient to classify buildings correctly. Additional spectral information is needed, to separate e.g. buildings from trees. The elevation data used has a resolution of 2,5 feet, the aerial photo has 0,5. The spectral image layer will help to get more detailed outlines due to its higher resolution.

2.6

Overview Guided Tour 1: Simple example of building extraction

The classification will start by simply classifying all objects higher than a certain value.

In the first Guided Tour a rather flat area will be classified. This has the advantage that the initial classification of elevated objects can take place using fixed height values.

Figure 17: Classification result after first classification step.

As a result trees are classified too. To get rid of these trees, their different appearance in the DSM is used to separate them from buildings. Additionally the spectral features for vegetation and shadow will be used to clean up the classification.

Figure 18: Classification result after second classification step. After that, missing parts of the buildings will be classified using context and size information.

As a result all buildings in the subset are classified with a good accuracy according the building outline.

Figure 19: Final classification result.

17

eCognition 8.0: Guided Tour Level 1 Simple building extraction

Lesson 3 Importing data

This lesson covers the following content
Open eCognition Developer Rule Set Mode Option 1: Use workspace and import template Option 2 (Trial version): Creating an individual Project Introduction All necessary data and Rule Set components are stored in the eCognition project file (.dpr). The project files can be managed in the workspace. A workspace file (.dpj) contains image data references, projects, exported result values and references to the used Rule Sets. Furthermore, it comprises the import and export templates, result states, and metadata.

NOTE:
If you are working with the Trail Version, you have to jump to chapter 3.3 to create and save projects individually. The workspace functionality is not available. Or if you have access to the full licensed product you will learn how to manage projects in the workspace. Then the data in this Guided Tour will be imported in the workspace using the Customized Import tool.

3.1
Introduction

Open eCognition Developer Rule Set Mode

When starting eCognition Developer you can optionally start it in either: or Rule Set Mode Quick Map Mode

The Quick Map Mode is designed to allow a user solving simple analysis tasks without having to get involved with Rule Set development. The main steps in analyzing an image are creating objects, classifying objects and exporting results. For each of these steps, a small assortment of predefined actions is available. The Rule Set Mode provides all necessary tools to develop own Rule Sets. In all the Guided Tours, the Rule Set Mode is necessary to accomplish the exercises. 1. 2. Action! Start eCognition Developer. Select Rule Set Mode.

18

Importing data

Settings Check

Figure 20: Select Rule Set Mode to have all tools available.

3.2

Option 1: Use workspace and import template

This chapter covers the following content

Create a new workspace Import data in the workspace with an import template How to open a created project How to save a project

3.2.1 Create a new workspace

In the View Settings toolbar there are 4 predefined View Settings available. Each view setting is specific to the different phases of a Rule Set development workflow. Information

Figure 21: View Settings toolbar with the 4 predefined view setting buttons: Load and Manage Data, Configure Analysis, Review Results, Develop Rulesets.

To create, open or modify a workspace, make sure that you are in the Load and Manage Data view. 3. 4. Select the predefined view setting number 1 Load and Manage Data from the View Settings toolbar. To create a new workspace, do one of the following: Click the Create New Workspace button on the toolbar. Action!

Choose File > New Workspace from the main menu bar.

The Create New Workspace dialog box opens.

Figure 22: Create New Workspace dialog box.

19

eCognition 8.0: Guided Tour Level 1 Simple building extraction

5. 6. 7.

Enter a Name for the new workspace. The default name is New Workspace. Browse for a location to save the new workspace. Confirm with OK to create the workspace. It is displayed as the root folder in the left pane (the tree view) of the workspace window.

3.2.2 Import data in the workspace with an import template

Information There are several ways to import data into a workspace. You can create an individual project, import an existing project, use a predefined import template or the highly flexible Customized Import tool. In this Guided Tour the project used contains an RGB image file plus the LiDAR DSM. Every channel, red, green, blue and the DSM shall have an alias to make further processing more transferable. This means the loaded data will not appear with its default naming Layer1, Layer2 etc. but have meaningful names like red, green, blue and DSM. Besides that areas containing no data values have to be defined. All these settings are already defined in the customized import template file (in xml format) which is used for import. 1. Right-click in the left workspace window and choose Customized Import from the menu.

Action!

The Customized Import dialog box opens. 2. 3. Make sure that the Workspace N tab is active Click on the Load O button and browse to the GuidedTour1_eCognition8_GettingStarted_Example_SimpleBuildingExtrac tion\\Data folder. Select the CustoImport_GuidedTour_Level1.xml.

The settings defined in the .xml are loaded 4. 5. Define the Root Folder P by browsing to the folder the data is stored. Define RGB_Level1_Simple_Example.img as Masterfile Q.

Note:
Select No for the upcoming Resolve conflicts error message. The information about different number of image layers refers also to the additionally to be loaded panchromatic layer. 6. Confirm with OK R.

20

Importing data

Settings Check

Figure 23: Customized Import dialog box with settings to import the data.

Note:
Alternatively you can right-click in the left workspace window and select Import Existing Project and browse to the folder where the Guided Tour1 data is stored. There you can find the Level1_Simple_Example.dpr project file.

3.2.3 How to open a created project

There are several methods to open a project from the workspace Double-click on the project in the Workspace window. Right-click on the project and select Open from the main menu.

Note:
The currently opened project is marked in the Workspace window with an asterisk.

3.2.4 How to save a project

Save the changes of a project by any of these methods:

21

eCognition 8.0: Guided Tour Level 1 Simple building extraction

Click on the Save Project button Or press Crtl + S on your keyboard.

NOTE:
As there is no undo command, it is recommended that you save a project prior to any operation that could lead to the unwanted loss of information, such as deleting an object layer or splitting objects. To retrieve the last saved state of the project, close the project without saving and reopen.

3.3

Option 2 (Trial version): Creating an individual Project

This chapter covers the following content

Define the data to be loaded Overview: Create Project dialog box Define layer alias Assign No Data values Confirm the settings to create the project

3.3.1 Define the data to be loaded

1. Action! Click on the Create New Project button Project or go to main menu File>New

The Create Project dialog box together with the Import Image Layer dialog box opens. 2. Navigate to the folder \GuidedTour1_eCognition8_GettingStarted_Example_SimpleBuildingExtraction\ Data. Mark the following image files and click Open. RGB_Level1_Simple_Example.img04mar_multi.img DSM_Level1_Simple_Example.img

3.

The Create Project dialog box opens. 4. In the Create Project dialog box in the field Name either enter a meaningful name for the project, e.g. Building extraction. Or keep the default naming according to the first loaded image file RGB_Level1_Simple_Example.

22

Importing data

3.3.2 Overview: Create Project dialog box

The Create Project dialog box has 4 main sections. The general settings section N : the geocoding information is displayed if the Use geocoding check box is selected and the resolution is automatically detected and displayed in the Resolution field. the unit is detected automatically if auto is selected from the drop down list. The unit is automatically set to meters, but can be changed by selecting another one from the drop down list. Introduction

NOTE:
The geocoding for the data is not available with the data belonging to this Guided Tour. A subset of the loaded images can be selected by clicking the Subset Selection button. The Create Subset dialog box opens.

Figure 24: Create New Project dialog box.

23

eCognition 8.0: Guided Tour Level 1 Simple building extraction

The image layer options section O : Introduction All preloaded image layers are displayed along with their properties. To select an image layer, click it. To select multiple image layers, press Ctrl or the Shift key and click on the image layers. To edit a layer double-click or right-click an image layer and choose Edit. The Layer Properties dialog will open. Alternatively you can click the Edit button. To insert an additional image layer you can click the Insert button or right-click inside the image layer display window and choose Insert on the context menu. To remove one or more image layers, select the desired layer(s) and click Remove. To change the order of the layers select an image layer and use the up and down arrows. To set No Data values for those pixels not to be analyzed, click No Data. The Assign No Data Values dialog box opens.

The thematic layer options section P : To insert a thematic layer, you can click the Insert button or right-click inside the thematic layer display window l and choose Insert from the context menu. To edit a thematic layer works similar to editing image layers described above.

The meta data options section Q : Here you can load additional information data as an .ini file, if available.

3.3.3 Define layer alias

Information In order to generate Rule Sets that are transferable between different datasets, the loaded channels have to have aliases assigned to them. 1. Action! To assign a layer alias, select the layer in the Create Project dialog box and double-click it.

24

Importing data

NOTE:
The Layer Properties dialog opens. Important for all 2 dimensional image analysis, as it is standard for Earth Sciences, the lower part of the dialog box Multidimensional Map Parameters can be ignored.

Figure 25: The Layer Properties dialog box.

2. 3.

Assign for layer 1 blue. Confirm the alias with OK

Assign the following aliases to the other layers,:

Layer1 Layer2 Layer3

red green blue

Layer4

DSM

3.3.4 Assign No Data values

If there are areas with 0 or other artificial values not containing any relevant information, Information they can be set as No Data areas. These areas will not be processed in any way, no objects will be created for these areas. In the current example, there are areas, where the RGB contains no information but the DSM, these areas are set to be No Data. 4. 5. Click on the No Data button. In the field Global No Data Value switch on the check-box and enter the value 0. Action!

This indicates that an area is defined as No Data if any of the layers contains a 0 value pixel. 6. Confirm the settings with OK.

25

eCognition 8.0: Guided Tour Level 1 Simple building extraction

3.3.5 Confirm the settings to create the project

7. After all alias are defined, click OK at the bottom of the Create Project dialog box. The new project is created.

Lesson 3 covered the following content

Open eCognition Developer Rule Set Mode Option 1: Use workspace and import template Option 2 (Trial version): Creating an individual Project

26

Developing the core strategy by evaluating the data content

Lesson 4 Developing the core strategy by evaluating the data content

This lesson covers the following content
The image visualization tools Evaluate the data content To create an initial strategy it is essential to get familiar with the data. eCognition has layer mixing tools that help to make the important information visible. You can zoom in and out to the scale of detail you are interested in and display the data individually or as layer combinations like RGB. Two groups of tools are essential here: The zooming functions and the layer mixing tools. Additionally you have the possibility to open multiple viewers, for example to have on the one hand side the RGB displayed and on the other hand the DSM. Introduction

4.1

The image visualization tools

This chapter covers the following content

Zooming into the scene Display the DSM Open a second viewer window to compare data Layer Mixing tools are available in the View Settings toolbar. You can display individual Information image layers or combinations and you can edit the equalization of a channel.

4.1.1 Zooming into the scene

Depending on the level of detail you are working on or for switching between detail and overview you need the zooming functions of eCognition Developer. They are all grouped in the Zoom toolbar. Information

Figure 26: Zoom Toolbar

27

eCognition 8.0: Guided Tour Level 1 Simple building extraction

1.

If you are working with the workspace, open the created project by doubleclicking on it in the workspace window. Or, right-click on it and select Open from the context menu. Zoom in and out of the loaded scene, use the panning function to evaluate the data.

Action!

2.

4.1.2 Display the DSM

Information You have the possibility to change the display of the loaded data using the Edit Layer Mixing dialog box. This enables you to display the individual channels of a combination. The first three bands of the loaded data are displayed by default in the Red, Green and Blue color guns, which is correct in this case. The third channel represents the elevation data. 3. To open the Edit Image Layer Mixing, do one of the following: Action! From the View menu, select Image Layer Mixing or click on the Edit Image Layer Mixing button toolbar. in the View Settings

Figure 27: Layer mixing buttons in the View Settings toolbar

Settings Check

Figure 28: Edit Layer Mixing dialog box.

4. 5.

Click on the dots for red, green and blue layers to deactivate them and click in one field of the DSM channel to activate it. At the bottom of the Edit Layer Mixing dialog box, click OK.

Action!

The image will now be displayed using the view settings you specified.

28

Developing the core strategy by evaluating the data content

Settings Check
Figure 29: Only DSM is displayed.

4.1.3 Open a second viewer window to compare data

To compare the DSM and the RGB data, open a second viewer. 1. From the Window menu, select either Split Horizontally or Split Vertically to open a second viewer window. Action!

A second window opens. 2. To link the two windows go to the menu Windows and select Side by Side View.

Now two viewer windows are open (see Figure 31) showing the same area of the loaded data. If now zooming in one window the other will follow and show the same content. Result Check

4.2

Evaluate the data content

This chapter covers the following content

Evaluate the elevation data Evaluate the RGB data

4.2.1 Evaluate the elevation data

The elevation data show, that the buildings have higher elevation than the surrounding ground. This is the most obvious information and therefore it will be used for initial classification. In the area of the subset the buildings have different heights. At the end the buildings are separated into different height classes. Introduction

Figure 30: A stabile feature to describe buildings is their difference in height.

29

eCognition 8.0: Guided Tour Level 1 Simple building extraction

Introduction

Besides the buildings, the trees are elevated objects too, but displaying the DSM you can see that leaf-off trees, in contrast to buildings, give back very heterogeneous elevation information. This heterogeneity can be used as a feature to separate the trees from buildings.

Figure 31: Left: DSM layer; Right: RGM layer mix.

4.2.2 Evaluate the RGB data

In the RGB data trees and buildings appear different too. Trees have a lower size than the buildings and in the RGB data contains the information that meadow is shining through the branches. This information separates them from the buildings too.

Figure 32: Loaded data with the workspace, displayed in two viewers. Above: the DSM layer; Below: the RGB layer mix.

30

Developing the core strategy by evaluating the data content

This example shows that height information from the LiDAR data provides key information to extract buildings. However, since trees and buildings both are elevated, only additionally using their different appearance in LiDAR and RGB layer separates buildings from all other objects with elevation. The rules in words: Buildings are higher than a certain level -> Use the height information from the DSM Trees, which have the same elevation as buildings, have a heterogenic elevation information, are smaller and have spectral vegetation features. -> Use heterogeneity and the size information and classify elevated objects with vegetation features in a separate class.

Introduction

Lesson 4 covered the following content

The image visualization tools Evaluate the data content

Next step: Develop Strategy for first analysis step!

31

eCognition 8.0: Guided Tour Level 1 Simple building extraction

Lesson 5 Creating image objects

This lesson covers the following content
Strategy for creating suitable image objects Translate strategy into Rule Set use algorithm multiresolution segmentation Review the created image objects

5.1
Information

Strategy for creating suitable image objects

The fundamental step of any eCognition image analysis is a segmentation of a scene representing an imageinto image objects. Thus, initial segmentation is the subdivision of an image into separated regions represented by basic unclassified image objects called image object primitives. To get most realistic image objects they have to represent the necessary elevation information and they should be not too big and not too small for the analysis task.

Figure 33: Diagram of relationship between Image Layer and Image Object Level.

Next step: Translate Strategy into Rule Set!

32

Creating image objects

5.2

Translate strategy into Rule Set use algorithm multiresolution segmentation

This chapter covers the following content

Why multiresolution segmentation algorithm? Creating objects using the multiresolution segmentation

5.2.1 Why multiresolution segmentation algorithm?

The first crucial decision you have to make is which algorithm to use for creating objects. Information The initially created objects are the basis for all further analysis. In the first lesson of this Guided Tour, the multiresolution segmentation is chosen to get most real-world objects in a simple and rather knowledge-free way. Simply said, the multiresolution segmentation groups areas of similar pixel values into objects. Therefore homogeneous areas result in larger objects, heterogeneous areas in smaller ones. How homogeneous/heterogeneous the objects are allowed to get is operated by the scale parameter.

Figure 34: Left RGB layer mix, Right: image object outlines. Homogeneous areas result in bigger objects, heterogeneous areas in smaller.

33

eCognition 8.0: Guided Tour Level 1 Simple building extraction

Creating objects using the multiresolution segmentation

Information Which layers to be used for creating objects? The basis for creating image objects is the input-data. According to the data and the algorithm, objects result in different object shapes. The first to evaluate is which layers contain the important information. Here in the current example there are two types of image data, the RGB and the DSM. In most segmentation algorithms it can be chosen whether to use all data available or only specific layers. This depends on where the important information is contained. Here all data is used for image object creation, the information contained in the RGB and in the DSM. Which scale parameter to be set? The scale parameter is an abstract term. It is the restricting parameter to stop the objects from getting too heterogeneous. In the current example we start with rather small objects to get the exact outlines of the buildings and to cover also the details (like windows and chimneys) as well as also the fractal structure of the leaf-off branches of the trees. For the scale parameter there is no definite rule, you have to use trial and error to find out which scale parameter results in the objects you need for further classification.

Figure 35: Process Tree with process to create image objects using the multiresolution segmentation added.

34

Creating image objects

5.2.2 Insert rule for object creation

This chapter covers the following content
Insert a parent process as wrapper for all underlying processes Insert the multiresolution segmentation process as a child process In the View Settings toolbar there are 4 predefined view settings available, specific to the different phases of a Rule Set development workflow. Information

Figure 36: View Settings toolbar with the 4 predefined view setting buttons: Load and Manage Data, Configure Analysis, Review Results, Develop Rulesets.

1. Action!

Select the predefined view setting number 4 Develop Rulesets from the View Settings toolbar.

For the Develop Rulesets view, per default one viewer window for the image data is open, as well as the Process Tree and the Image Object Information window, the Feature View and the Class Hierarchy. Information

Insert a parent process as wrapper for all underlying processes

A so called parent process is added to the Process Tree and will serve as a wrapper for all following processes. This has the advantage that you can execute the whole sequence from this process. 2. Right-click in the Process Tree window.

Action!

Result Check

Figure 37If you right-click in the Process Tree window, a context menu appears.

35

eCognition 8.0: Guided Tour Level 1 Simple building extraction

3.

Select Append New from the context menu.

The Edit Process dialog box opens. Action! 4. In the Name field enter the name Building Detection and confirm with OK.

Note:
For all parent processes which serve only as a wrapper for underlying Processes, you can edit the name. For all Processes containing an algorithm, keep the auto naming!

Insert the multiresolution segmentation process as a child process

Information According to the algorithm you choose, the parameter fields on the right side of the Edit Process dialog box change. For the multiresolution segmentation you can edit the name of the level to be created, the scale parameter and the so called homogeneity criterion. 1. 2. Select the inserted process and right-click on it. Choose Insert Child form the context menu. In the Algorithm field choose N multiresolution segmentation from the list.

Action!

TIP:
Type in the Algorithm field the first letters of the algorithm you want to use, a list of suitable algorithms is provided. 3. Keep pixel level O in the Image Object Domain.

Note:
For the first segmentation you can not change that. If you perform a second segmentation step, you can choose if you want to start from the domain pixel level again or from an already existing image object Level. 4. 5. 6. In the field Level Name O type Level 1. This will be the name of the image object level to be created. In the field Scale parameter Q enter 25. Confirm the settings with OK. The process is now added to the Process Tree.

36

Creating image objects

Settings Check

Figure 38: Edit Process dialog box with settings to create Level 1 using the multiresolution segmentation algorithm.

Rule Set Check

Figure 39: Process Tree with parent process Building Detection and child process for creating Level 1.

Execute the multiresolution segmentation process

7. Right-click on the classification process and select Execute from the context menu. Alternatively press F5 on your keyboard.

An image object level Level1 is created according to the settings in the multiresolution Action! segmentation process.

37

eCognition 8.0: Guided Tour Level 1 Simple building extraction

Next step: Review Result!

5.3
Result Check

Review the created image objects

The created objects represent the buildings even in their detailed outlines and also the roof parts.

Figure 40: Outlines of objects created with multiresolution segmentation algorithm.

These objects will be the basis to classify in a first step all elevated objects in the scene.

Next step: Develop next analysis step!

Lesson 5 covered the following content

Strategy for creating suitable image objects Translate strategy into Rule Set use algorithm multiresolution segmentation Review the created image objects

38

Initial classification: classifying all elevated objects

Lesson 6 Initial classification: classifying all elevated objects

This lesson covers the following content
Strategy to classify buildings based on elevation information Translate the strategy into Rule Set Object mean of DSM, algorithm assign class Review the classification result buildings and trees are classified

6.1

Strategy to classify buildings based on elevation information

Introduction

As described in the chapter Evaluate the loaded data the assumption is made that buildings are always elevated. The subset used for this Guided Tour has not much change in terrain elevation, so simply the definite elevation above sea level is used as a threshold to classify elevated objects.

Figure 41: First step is to classify all elevated objects, buildings and trees.

Now this knowledge has to be transfered into the Rule Set!

Next step: Translate Strategy into Rule Set!

39

eCognition 8.0: Guided Tour Level 1 Simple building extraction

6.2

Translate the strategy into Rule Set Object mean of DSM, algorithm assign class

This chapter covers the following content

Find the feature and the threshold to represent all elevated objects (mean value of DSM) Write rule for classifying all elevated objects Review the classification result buildings and trees are classified

6.2.1 Find the feature and the threshold to represent all elevated objects (mean value of DSM)
Information Every object contains lots of feature values. There are: Object features like layer features, shape features and position features Class-related features: Context information about neighbors, sub objects below and super objects above Scene features: Features relative to scene, like the overall brightness, etc. Process-related features: Features with which you can expand the domain concept to deal only with individual objects, not the whole class or level Metadata: Deal with external, additional information, if available Feature variables : Used for defining individual names for features

For the initial classification of elevated objects, the mean value of the object concerning the DSM is used.

Use the Feature View tool

Information The most crucial part in Rule Set development is to find the optimal features and values for classifying image objects in the one or the other class. The Feature View is a tool that helps finding the optimal features and to determine threshold values for classification. With the Feature View the values for all objects are displayed in the viewer in grey values. But there is also the possibilities to show value ranges in color.

Note:
If not already open, there are several possibilities to open the Feature View tool: Choose in the menu Tools>Feature View, alternatively select the Feature View button from the Tool toolbar.

40

Initial classification: classifying all elevated objects

Select the feature 1. 2. 3. In the Feature View window browse to Object features>Layer Values>Mean. Double-click on DSM. Move your cursor over the objects in the viewer and the exact feature values for the object appears. Action!

As you can see the elevated objects appear quite bright, this means they have high values for the feature Mean DSM.

Result Check

Figure 42: All objects appear now in grey values representing the respective DSM value.

But how to get the correct threshold to separate elevated and not elevated objects?

Update the feature range

Besides moving the mouse over the objects and guessing a threshold value, you can use Information the color displayed range and so find the threshold value range. First the whole range for the feature Mean DSM has to be calculated internally (from minimum to maximum value). 1. 2. Select the feature Mean DSM in the Feature View and right-click on it. From the menu select Update Range. Action!

Visualize a certain area of the whole feature range

1. Click the check box at the bottom of the Feature View window. This activates the display of the feature range and the minimum value (here 737.92) is displayed in the left box and the maximum value (here 810.87) is displayed in the right box. Result Check

41

eCognition 8.0: Guided Tour Level 1 Simple building extraction

Figure 43: All image objects are colored in a smooth transition from blue (low values) to green (high values).

Information

With the arrows beside the value boxes you can increase or decrease the start and end of the range. All objects which are not within this range will be displayed in grey again. 2. To isolate the high values (areas with high elevation), click the up arrow Nto the right of the minimum value.

Action!

This will increase the low end of the range. Only objects within this new range are now displayed in color. 3. Continue until you reach the value 765 or type it in.

Result Check

Figure 44: All image objects within a value range from 765 to 810 are colored, all objects outside this range (lower than 765) are displayed in grey values.

The rule derived from the feature and the values can be formulated like this: all objects with a mean elevation of more than 765 are possible buildings.

NOTE:
Be sure to update the range of feature values each time you select a different feature. Otherwise, the range of the recent feature is used.

6.2.2 Write rule for classifying all elevated objects

Information Formulate the rule in words: All objects with a mean elevation of more than 765 are buildings.

Prepare the Rule Set structure

1. 2. In the Process Tree right-click on the last process and select Append New from the context menu. Insert Classification of Buildings in the Name field and confirm with OK.

Action!

42

Initial classification: classifying all elevated objects

3. 4.

In the Process Tree right-click on the Classification of Buildings process and select Insert Child from the context menu. Insert Classification based on elevation in the Name field and confirm with OK.

Rule Set Check

Figure 45: Process Tree with the two parent processes Classification of Buildings and Classification based on elevation.

Create a class in the Class Hierarchy window

1. Right-click in the Class Hierarchy window and select Insert Class from the context menu. Action!

The Class Description dialog box opens. 2. 3. Enter Building in the Name field. Keep the default color. Confirm with OK.

Figure 46: The Class Description dialog box. The name is defined as Building. Red is chosen as color.

The class Building is added to the Class Hierarchy window.

43

Result eCognition 8.0: Guided Tour Level 1 Simple building extraction Check

Figure 47: Class Hierarchy with class Building inserted.

Insert process to classify

1. Right-click on the process Classification based on elevation and select Insert Child from the context menu.

Action!

Choose algorithm and object domain 2. 3. 4. Choose assign class N from the algorithm drop-down list. Check if Level 1O is set as level domain In the field Class Filter keep none P.

Define condition 5. Click in Threshold condition field and again click on the Q=next to it.

The Select Single Feature dialog box opens.

Figure 48: The Select Single Feature dialog box.

6. 7. 8.

Browse to Object features>Layer Values>Mean>DSM and double-click on it. Choose the larger or equal than (>=) as operator and enter the value 765. Confirm with OK.

The Edit threshold condition dialog box opens. The condition is added to the process. Settings Check

44

Initial classification: classifying all elevated objects

Figure 49: The Edit threshold condition dialog box.

Define the target class 9. In the Algorithm Parameter section in the right pane, select Building Rfrom the drop-down list. Action!

10. Confirm the process settings with OK. The process is now added to the Process Tree.

Settings Check

Figure 50: Edit Process dialog box with settings to classify objects with Mean DSM higher than 765 to the class Building.

Rule Set Check

Figure 51: Process Tree with process for classification added.

45

eCognition 8.0: Guided Tour Level 1 Simple building extraction

Execute the classification process

11. Right-click on the classification process and select Execute from the context menu. Alternatively press F5 on your keyboard. Action! All objects fitting in the rule defined in the process are assigned to the class Building.

Next step: Review Result!

46

Initial classification: classifying all elevated objects

6.3
1. 2.

Review the classification result buildings and trees are classified

and make Action!

In the View Settings toolbar select the View Classification button sure that the Show or Hide Outlines button is deselected.

Move the cursor over the classification and the assigned class will appear as a tool tip next to it. Select the Pixel View or Objects Mean View button transparency view. Select the Show or Hide Outlines button the classification colors. to switch on and off the

3.

4.

and the outlines will be displayed in

Result Check

Figure 52: Classification view not transparent, transparent and with outlines view switched on.

All elevated objects are elevated. Besides the buildings there are also trees classified. This means add additional rules to refine the result must be added. You have to develop a new strategy to find additional rules to minimize the misclassifications.

Figure 53: Classification result.

47

eCognition 8.0: Guided Tour Level 1 Simple building extraction

Next step: Develop next analysis step!

Lesson 6 covered the following content

Strategy to classify buildings based on elevation information Translate the strategy into Rule Set Object mean of DSM, algorithm assign class Review the classification result buildings and trees are classified

48

Refinement based on DSM: un-classifying trees

Lesson 7 Refinement based on DSM: un-classifying trees

This lesson covers the following content
Strategy to separate buildings from trees: use standard deviation of DSM Translate the strategy into Rule Set Standard deviation of DSM, algorithm assign class Review the classification result trees are de-classified; some vegetation and shadows objects are still misclassified

7.1

Strategy to separate buildings from trees: use standard deviation of DSM

Introduction

In a second processing step buildings have to be separated from trees, as both are elevated objects. Here again the DSM information will be used. Evaluate object features for DSM to find a separating feature

Figure 54: Image layer DSM. Trees appear more heterogeneous than buildings.

If you evaluate the DSM image layer you can see that trees have very high elevation values close to very low elevation values. This is due to the leaf-off tree branches. The laser hits a branch and gives back a high elevation value, between the branches the laser goes through onto the ground, this results in a low elevation value. So it is characteristic that there are significant changes in elevation close to each other.

49

eCognition 8.0: Guided Tour Level 1 Simple building extraction

Introduction

Figure 55: The laser hits one time the branch, the other time it goes onto the ground. Due to this effect, the leaf-off trees appear heterogeneous.

Which feature represents the characteristics of leaf-of trees? The features of the category Mean will not help here. A very suitable feature to describe heterogeneity or homogeneity is the standard deviation.

Next step: Translate Strategy into Rule Set!

50

Refinement based on DSM: un-classifying trees

7.2

Translate the strategy into Rule Set Standard deviation of DSM, algorithm assign class

This chapter covers the following content

Find the feature and the threshold to separate buildings from trees (Stddev. of DSM) Write rule to separate buildings from trees Review the classification result trees are de-classified; some vegetation and shadows objects are still misclassified

7.2.1 Find the feature and the threshold to separate buildings from trees (Stddev. of DSM)
Information

The feature standard deviation of DSM

The standard deviation of pixel values of an object gives back information whether the pixel values of an object are all quite similar or if there are significant differences.

For the leaf-off trees high differences in elevation is expected, the standard deviation of these areas should be high. In opposite, the building surfaces should give back low values for standard deviation.

51

eCognition 8.0: Guided Tour Level 1 Simple building extraction

Information

The pixels of building objects are close around the mean value of the object, this means that the dispersion is low. In opposite the pixels of a tree object have a wide dispersion around the mean of the object.

Find threshold using the Feature View tool

1. 2. Action! 3. 4. Browse to the feature Object features>Layer Values>Standard deviation>DSM. Right-click on it and select Update Range from the context menu. Switch on the check box at the bottom of the Feature View window. Increase the lower range to the value 6. Most of the objects which are actually trees have values above 6, most of the building objects have values below.

Rule Set Check

Figure 56: Left: classification view; Middle: whole range of the standard deviation of DSM; Right: high standard deviation values are highlighted in color.

The trees appear in green; this indicates high values for the standard deviation of DSM. If you increase the low values, only the trees remain colored.

7.2.2 Write rule to separate buildings from trees

Introduction Formulate the rule in words: Unclassify all Building objects with standard deviation above 6. Only one condition is used, so again use the algorithm assign class. 1. Select the last process in the Process Tree, right-click and select Append New from the context menu.

Action!

Choose algorithm and object domain

2. Choose assign class from the algorithm list.

52

Refinement based on DSM: un-classifying trees

3.

Click in the field Class Filter and again click on the button next to it. Select Building as domain by switching on the check-box next to it.

Figure 57: The class Building is selected.

Tip:
If you want to change to the list view in the Edit Classification Filter dialog box, simply select the last button. Then you can select the classes by double-clicking

Define condition
4. 5. 6. 7. Click in Threshold condition field and again click on the =next to it. Browse to Object features>Layer Values>Standard deviation>DSM and double-click on it. In the Edit threshold condition choose the larger or equal than (>=) as operator and enter the value 6. Confirm with OK.

Define the target class

8. 9. In the Parameter section in the left pane, keep unclassified. Confirm the process settings with OK.

The process is now added to the Process Tree.

Settings Check

Figure 58: Edit Process dialog box with settings to un-classify all Building objects with a standard deviation of DSM larger than 6.

53

eCognition 8.0: Guided Tour Level 1 Simple building extraction

Rule Set Check

Figure 59: Process Tree with second process for classification added.

Execute the classification process

10. Right-click on the process and select Execute from the context menu. Alternatively press F5 on your keyboard. Action!

Next step: Review Result!

54

Refinement based on DSM: un-classifying trees

7.3

Review the classification result trees are de-classified; some vegetation and shadows objects are still misclassified
Result Check

All Building objects with a value higher than 6 for the feature standard deviation of DSM are unclassified again. But still there are too many objects classified as Building. The classification is not ready for export!

Figure 60: Left: Classification before refinement; Right: Classification after refinement.

This means you have to go back to the Develop Strategy stage to find additional rules to minimize the misclassifications.

Lesson 7 covered the following content

Strategy to separate buildings from trees: use standard deviation of DSM Translate the strategy into Rule Set Standard deviation of DSM, algorithm assign class Review the classification result trees are de-classified; some vegetation and shadows objects are still misclassified

Next step: Translate Strategy into Rule Set!

55

eCognition 8.0: Guided Tour Level 1 Simple building extraction

Lesson 8 Refinement based on spectral information

This lesson covers the following content
Strategy to refine buildings based on spectral information the spectral layer ratio Translate the strategy into Rule Set Ratio of green, algorithm assign class Review the classification result the vegetated areas are de-classified; small objects still misclassified

8.1

Strategy to refine buildings based on spectral information the spectral layer ratio

Introduction

After refining the classification using the standard deviation of the DSM layer still some trees are classified as Building. Some of them are coniferous trees, so they are not leafoff and therefore give back a quite homogeneous elevation area. But these coniferous trees have a significant spectral difference to buildings. Roofs are usually not covered by vegetation; therefore a feature must be found that represents vegetation in a stabile way.

Figure 61: The black arrows are pointing to some trees not de-classified with the previous step.

Figure 62: RGB layer mixing

It is obvious that the green layer contains significant information about vegetation, but only in comparison with the other 2 layers.

56

Refinement based on spectral information

Introduction

Figure 63: The individual layers, red, green, blue

One way of comparing image layers is to create a ratio. The relevant ratio would be green/(red+green+blue). This knowledge must be transfered into the Rule Set now. Outlook on Process Tree:

Figure 64: Process Tree processes added to refine classification based on spectral feature.

Next step: Translate Strategy into Rule Set!

57

eCognition 8.0: Guided Tour Level 1 Simple building extraction

8.2

Translate the strategy into Rule Set Ratio of green, algorithm assign class

This chapter covers the following content

Find the feature and the threshold for spectral refinement (ratio green) Translate strategy in a rule for refinement based on a spectral feature Review the classification result the vegetated areas are de-classified; small objects still misclassified

8.2.1 Find the feature and the threshold for spectral refinement (ratio green)
Information Within eCognition Developer there are standard features, always listed in the feature view, but there is also the possibility to customize features for your specific needs. You can create, save and load such Customized Features. Here in this example a feature is needed which expresses the ratio: green/(red+green+blue).

Note:
The software provides a standard ratio calculation. These ratio features are calculated on basis of all inserted layers. This means in our case that the DSM will be included in the calculation.

Load a customized feature

1. 2. Action! 3. Right-click in the Feature View and select Load. Browse to the folder RuleSets in your Guided Tour directory and select CustRatioGreen.duf.

The customized feature is added to the Feature View Browse to Object Features>Customized.

Figure 65: The customized feature listed in the Feature View.

4.

Right-click on the feature CustRatioGreen and select Edit.

The Edit Customized Feature dialog box opens.

58

Refinement based on spectral information

Check the calculation for the customized feature

5. Check the correct calculation: [Mean green]/([Mean blue]+[Mean green]+[Mean red]).

The mean value for the layer green of the object is compared to the overall brightness of the object.

Result Check

Figure 66: The loaded customized feature in the Edit Customized Feature dialog box.

6.

Close the dialog by clicking OK.

Find threshold to separate the class Vegetation

1. Browse to the feature Object features>Customized>CustRatioGreen and double-click on it.

Action!

Result Check

Figure 67: Feature View of the customized feature Cust Ratio Green

The vegetated areas appear very bright, this means the objects have high values. The amount of green in comparison to the other image layers is here dominant. 2. 3. 4. Right-click on it and select Update Range from the context menu. Switch on the check box at the bottom of the Feature View window. Increase the lower range to the value 0.36. Most of the vegetated areas have values above 0.36. Action!

59

eCognition 8.0: Guided Tour Level 1 Simple building extraction

Result Check

Figure 68: RGB layer mixing.

Figure 69: Feature View range (above 0.36) of the customized feature Cust Ratio Green

Figure 70: Vegetation is represented by values higher than 0.36 for the feature CustRatioGreen.

As a rule from the feature and the range it can be formulated that: Unclassify all Building objects with a value for customized ratio green of more than 0.36.

8.2.2 Translate strategy in a rule for refinement based on a spectral feature

Prepare the Rule Set structure
1. 2. In the Process Tree right-click on the last process and select Append New from the context menu. Insert Refinement based on spectral information in the Name field and confirm with OK.

Action!

60

Refinement based on spectral information

Insert process to classify

1. In the Process Tree right-click on the Refinement based on spectral information and select Insert Child from the context menu.

Choose algorithm and object domain Again only one condition for classification is used, so the algorithm assign class is appropriate. 2. 3. Choose assign class from the algorithm list. As Class Filter select Building.

Define condition 4. 5. 6. 7. Click in Threshold condition field and again click on the =next to it. Browse to Object features>Customized>Cust Ratio Green and double-click on it. In the Edit threshold condition choose the larger or equal than (>=) as operator and enter the value 0.36. In the field Active class keep unclassified.

Confirm the settings 8. Confirm with OK.

Settings Check

Figure 71: Edit Process dialog box with settings to un-classify Building objects with a value for the customized feature higher than 0.36.

61

eCognition 8.0: Guided Tour Level 1 Simple building extraction

Rule Set Check

Figure 72: Process Tree with parent process Refinement based on spectral information added and with child processes for classification.

Execute the classification process

9. Right-click on the process and select Execute from the context menu. Alternatively press F5 on your keyboard.

Action!

Next step: Review Result!

62

Refinement based on spectral information

8.3

Review the classification result the vegetated areas are de-classified; small objects still misclassified

All Building objects with a value higher than 0.36 for the feature Cust Ratio Green are now unclassified again.

Figure 73: Left: Classification view before refinement; Middle: Feature View for Ratio Green; Right: Classification after refinement.

But still there are objects misclassified as Building and some building areas are not yet classified. The classification is not ready for export yet! This means the strategy has to be refined, additional rules must be added to minimize the misclassifications.

Lesson 8 covered the following content

Strategy to refine buildings based on spectral information the spectral layer ratio Translate the strategy into Rule Set Ratio of green, algorithm assign class Review the classification result the vegetated areas are de-classified; small objects still misclassified

Next step: Develop next analysis step!

63

eCognition 8.0: Guided Tour Level 1 Simple building extraction

Lesson 9 Refinement based on context

This lesson covers the following content
Strategy to refine buildings based on context information surrounding neighbor objects Translate Strategy in a rule for refinement based on a context feature Review the classification result only small objects remain misclassified

9.1

Strategy to refine buildings based on context information surrounding neighbor objects

Introduction

Some areas of the buildings have not been classified yet or were de-classified again because they fulfilled one of the conditions before. Some of the not classified objects are highly surrounded by Building objects. If an unclassified object has a high common border to Building objects it should also belong to the class Buildings.

Figure 74: Some unclassified objects are highly surrounded by Building objects.

This knowledge must be transferred into the Rule Set now. Within eCognition Developer you can express neighborhood relationships of objects using the Class-Related (context) features. This is possible within the image object level, for super-objects above or sub-objects below.

64

Refinement based on context

Introduction

Figure 75: Diagram of Image Object Hierarchy and network.

In the current example neighborhood within the same level will be analyzed. The rule must describe the situation that if an unclassified object has a high common border to Building objects it belongs to the class Building too.

Figure 76: All objects have neighborhood relationships, which can be used as information for classification.

This knowledge must be transferred into the Rule Set now. Outlook on Process Tree:

Figure 77: Process Tree with processes added for classification based on context information.

Next step: Translate Strategy into Rule Set!

65

eCognition 8.0: Guided Tour Level 1 Simple building extraction

9.2

Translate Strategy in a rule for refinement based on a context feature

This chapter covers the following content

The class-related feature Relative border to Prepare the Rule Set structure Insert process to classify Review the classification result only small objects remain misclassified

9.2.1 The class-related feature Relative border to

Information The class-related feature Relative border to describes the amount of the overall border of an object compared to the border to a specific class.

Figure 78: The selected object (blue outlines) has a border to Buildings about 0.6 and a border to unclasssified about 0.4

The range of the feature reaches from 0 to 1. Objects with no border to Building objects have the value 0 Objects with a low common border to Building objects have low values Objects with a high common border to Building objects have high values Objects which are completely surrounded by Building objects have the value 1

Create the feature Relative border to Buildings

1. In the Feature View window browse to Class-Related features>Relations to neighbor objects> Relative Border to.

Action!

66

Refinement based on context

Note:
Per default the list of class-related features is empty. You first have to create the class you want to evaluate. If you had a lot of classes and all relationships to all classes would be displayed always, this would be very confusing. This is why you have to create the individual relationships on demand. 2. Double-click on Create new Rel. border to.

The Create Rel. border to dialog box opens. 3. 4. From the Value drop-down list select the class Building, as the border to Building objects must be evaluated. Confirm the settings with OK.

Settings Check

Figure 79: The Create Rel. border to dialog box with the class Building selected.

The feature is added to the Feature View list.

Result Check

Figure 80: Feature View with new feature Rel. border to Building created.

Find the threshold

1. 2. 3. 4. Double-click the feature Class-Related features>Relations to neighbor objects> Relative Border to Building. Right-click on it and select Update Range from the context menu. Switch on the check box at the bottom of the Feature View window. Increase the lower range to the value 0.5. Action!

67

eCognition 8.0: Guided Tour Level 1 Simple building extraction

Result Check

Objects appearing in a color range form blue to green, have at least half of its overall border common with Building objects. Objects with a lower common border to Building objects appear now in a grey range. If the value is too low, too many neighboring objects would be included.

Figure 81: The selected object has a Rel. border to Buildings of 0.3.

If the value is too high, too few neighboring objects would be included.

Figure 82: The selected object has a relative border to Buildings of 0.6.

Figure 83: The selected object has a Rel. border to Buildings of 0.5.

As a rule from the feature and the range it can be formulated that: Unclassified objects with a relative border to Building objects more than 0.5 shall be classified as Building too.

9.2.2 Prepare the Rule Set structure

1. 2. In the Process Tree right-click on Refinement based on spectral information and select Append New from the context menu. Insert Refinement based on context information in the Name field and confirm with OK.

Action!

68

Refinement based on context

9.2.3 Insert process to classify

1. In the Process Tree right-click on the Refinement based on context information and select Insert Child from the context menu.

Choose algorithm and object domain

2. 3. Choose assign class from the algorithm list. Define unclassified as Class filter.

Define condition
4. Click in the Threshold condition field and browse to Class-Related features>Relations to neighbor objects> Rel. border to Building and doubleclick on it. In the Edit threshold condition choose the larger or equal than (>=) as operator and enter the value 0.5. Confirm with OK.

5. 6.

Settings Check

Figure 84: Edit Process dialog box with settings to classify all unclassified objects with a high common border to Buildings as Buildings too.

Rule Set Check

Figure 85: Process Tree with parent process Refinement based on context and shape added and with child processes for classification.

69

eCognition 8.0: Guided Tour Level 1 Simple building extraction

Execute the classification process

7. Right-click on the process and select Execute from the context menu. Alternatively press F5 on your keyboard.

Action!

Next step: Review Result!

70

Review the classification result only small objects remain misclassified

9.3

Review the classification result only small objects remain misclassified

Unclassified objects with a relative border to Building objects more than 0.5 are classified as Building too.

Figure 86: Classification view before and after the refinement.

But still there are too many misclassifications. The classification is not ready for export! This means the development strategy has to be refined to find additional rules to minimize the misclassifications.

Next step: Develop Strategy for fist analysis step!

Lesson 9 covered the following content

Strategy to refine buildings based on context information surrounding neighbor objects Translate Strategy in a rule for refinement based on a context feature Review the classification result only small objects remain misclassified

71

eCognition 8.0: Guided Tour Level 1 Simple building extraction

Lesson 10 Refinement based on shape

This lesson covers the following content
Strategy to refine buildings based on shape information generalize objects, separate them by size Translate strategy in a rule for refinement based on a shape feature Review the classification result

10.1 Strategy to refine buildings based on shape information generalize objects, separate them by size
Introduction There are still some misclassified objects, but they are very small. The size of the object of interest is the separating condition here, based on the assumption, that buildings have a certain size.

Figure 87: Outlines of unmerged objects.

As all objects are now still as small as they were created by the initial multiresolution segmentation, they have to be merged first according to their classification. After that the area feature can be used to separate the big buildings from the small misclassified objects.

Figure 88: Outlines of merged Building objects.

72

Refinement based on shape

This means you have to add two rules: Rule 1) Merge all Building objects. Rule 2) Unclassify all too small Building objects. This knowledge must be transferred into the Rule Set now. Outlook on Process Tree:

Introduction

Figure 89: Process Tree with process added to merge all Building objects and to un-classify too small Building objects.

Next step: Translate Strategy into Rule Set!

73

eCognition 8.0: Guided Tour Level 1 Simple building extraction

10.2 Translate strategy in a rule for refinement based on a shape feature

This chapter covers the following content
Merge the image objects Find the feature and the threshold for refinement based on shape Translate strategy in a rule for refinement based on the Area feature Review the classification result

10.2.1 Merge the image objects

Information eCognition Developer provides several algorithms to merge image objects. The most simple one is the algorithm merge region. With this algorithm you can merge neighboring objects according to their class.

Prepare the Rule Set structure

1. 2. In the Process Tree right-click on the last process and select Append New from the context menu. Insert clean up too small Building objects in the Name field and confirm with OK.

Action!

Insert process to merge Building objects

1. In the Process Tree right-click on the Clean up too small Building objects and select Insert Child from the context menu.

Choose algorithm and object domain 2. 3. Action! 4. 5. Choose merge region from the algorithm list. Select Building as Class filter. Keep Threshold condition empty. Confirm with OK.

74

Refinement based on shape

Settings Check

Figure 90: Edit Process dialog box with settings to merge all Building objects.

Rule Set Check

Figure 91: Process Tree with parent process clean up too small Building objects added and with child processes to merge Building objects.

Execute the classification process 6. Right-click on the process and select Execute from the context menu. Alternatively press F5 on your keyboard. Action!

75

eCognition 8.0: Guided Tour Level 1 Simple building extraction

10.2.2 Find the feature and the threshold for refinement based on shape
Information Within eCognition Developer objects can also be evaluated according to their shape characteristics, like their area. 1. 2. Action! In the Feature View window browse to Object features>Geometry>Extent. Double-click on the feature Area.

Result Check

Figure 92: Feature View of the feature Area.

The buildings appear very bright, this means the objects have high values. The area of buildings is significantly higher than the area of not building objects. 3. 4. Action! 5. Right-click on it and select Update Range from the context menu. Switch on the check box at the bottom of the Feature View window. Increase the lower range to the value 7000. All of the not Building objects have values below 7000.

Result Check

Figure 93: Feature View range of the feature Area.

As a rule from the feature and the range it can be formulated that: Unclassify all Building objects with a value for the feature Area of less than 7000.

76

Refinement based on shape

10.2.3 Translate strategy in a rule for refinement based on the Area feature
Insert process to classify
1. In the Process Tree right-click on the last process and select Append New from the context menu. Action!

Choose algorithm and object domain Again only one condition is used, so use the algorithm assign class. 2. 3. Choose assign class from the algorithm list. Select Building as Class filter.

Define condition 4. 5. 6. In the Threshold condition field browse to Object Features> Geometry>Extent>Area and double-click on it. In the Edit threshold condition choose the smaller than (<) as operator and enter the value 7000. Confirm with OK.

Settings Check

Figure 94: Edit Process dialog box with settings to un-classify all Building objects smaller than 7000 pixel.

77

eCognition 8.0: Guided Tour Level 1 Simple building extraction

Rule Set Check

Figure 95: Process Tree with process to un-classify Building objects.

Execute the classification process 7. Right-click on the process and select Execute from the context menu. Alternatively press F5 on your keyboard.

Action!

Next step: Review Result!

78

Review the classification result

10.3 Review the classification result

No misclassified Building object remains. The classification is ready for export!

Figure 96: Classification before and after the refinement.

This means the development is ready for the next step, the Export.

Next step: Develop next analysis step!

Lesson 10 covered the following content

Strategy to refine buildings based on shape information generalize objects, separate them by size Translate strategy in a rule for refinement based on a shape feature Review the classification result

79

eCognition 8.0: Guided Tour Level 1 Simple building extraction

Lesson 11 Exporting the result in a vector layer

This lesson covers the following content
Insert process to merge unclassified objects Insert process to export shape file with attributes Review the exported result Introduction The buildings are classified correctly now and there are no misclassifications. The classification is ready for export. In this Guided Tour the classification will be exported as a vector .shp file. The buildings should be exported together with their size information. Outlook on Process Tree:

Figure 97: Process Tree with process added to export a vector layer of the result.

11.1 Insert process to merge unclassified objects

This chapter covers the following content
Prepare the Rule Set structure Insert process to merge all unclassified

Information

To clean up and reduce the number of objects in the scene first all unclassified objects must be merged.

80

Exporting the result in a vector layer

11.1.1 Prepare the Rule Set structure

1. 2. In the Process Tree collapse the process Classification of Buildings and right-click on it. Select Append New from the context menu. Insert Export in the Name field and confirm with ok. Action!

11.1.2 Insert process to merge all unclassified objects

1. 2. 3. 4. 5. 6. In the Process Tree right-click on the Export and select Insert Child from the context menu. Choose merge region from the algorithm list. Select unclassified as Class filter. Keep no condition. Confirm with OK. Execute the process.

Settings Check

Figure 98: Edit Process dialog box with settings to merge all unclassified objects.

Rule Set Check

Figure 99: Process Tree with parent process Export added and with child processes to merge unclassified objects.

81

eCognition 8.0: Guided Tour Level 1 Simple building extraction

11.2 Insert process to export shape file with attributes

This chapter covers the following content
Create the feature Class name Insert process to export vector file eCognition offers several algorithms to export e.g. current views, statistics and also vector layers.

11.2.1 Create the feature Class name

Information When exporting a file, per default no feature is defined to be additionally exported. To export now the class name to a vector file or any other object related file (object statistic or thematic raster file), the new feature Class name has to be added to the feature list (like area feature of other features). The Class Name feature returns the name of the class of an image object or the name of the super-class of an image object or the name of the class of the super-object of an image object

This is controlled by the Distance value. 1. 2. In the Feature View window browse to Class-Related features>Relations to Classification. Double-click on Create new Class name.

Action!

The Class name dialog box opens. 3. 4. Keep the default settings. Confirm the settings with OK.

The feature is added to the Feature View list.

Result Check

Figure 100: The feature Class name in the Feature View tree.

82

Exporting the result in a vector layer

11.2.2 Insert process to export vector file

1. In the Process Tree right-click on the last process and select Append New from the context menu. Action!

Choose algorithm and object domain 2. 3. Choose export vector layers from the algorithm list. Keep the Class filter and Threshold condition as default.

Define the parameter 4. 5. 6. In the field Export mode keep Use export item. In the field Export item name insert Building. This will be the name of the exported file. Click in the field Attribute table.

The Select Multiple Features dialog box opens. 7. 8. Browse to the Class-Related features>Relations to Classification>Class name and double-click on it to move it to the Selected window. Do the same for the Area feature (Object Features>Geometry>Extent>Area). Confirm with OK.

The features are added to the export specifications.

Result Check

Figure 101: The Select Multiple Features dialog box.

9.

In the algorithm parameter field Shape Type change from Points to Polygons.

10. Keep all other default settings. 11. After clicking OK, the Edit Attribute Table Columns window appears, where other settings are possible. 12. Execute the process. The shape file together with its attributes is exported to the location where the image data is stored. Action!

83

eCognition 8.0: Guided Tour Level 1 Simple building extraction

Settings Check

Figure 102: Edit Process dialog box with settings to export a vector layer.

Rule Set Check

Figure 103: Process Tree with process added to export a vector layer of the result.

Lesson 11covered the following content

Insert process to merge unclassified objects Insert process to export

84

Review the exported result

11.3 Review the exported result

A shape file has been exported. Review it with ArcGIS or simply check the exported .dbf file.

Figure 104: Final classification of the buildings

Figure 105: Exported .dbf table containing the area and the classification of every object.

Lesson 11 covers the following content

Insert process to merge unclassified objects Insert process to export shape file with attributes Review the exported result

85

eCognition 8.0: Guided Tour Level 1 Simple building extraction

Where to get additional help and information?

eCognition Community and Rule Set Exchange platform

The eCognition community helps to share knowledge and information within the user, partner, academia and developer community to benefit from each other's experience. The community contains next to other content: Wiki: collection of eCognition related articles (e.g. Rule Set tips and tricks, strategies, algorithm documentation...). Discussions: ask questions and get answers. File exchange: share any type of eCognition related code such as Rule Sets, Action Libraries, plug-ins... Blogs: read and write insights about whats happening around our industry..

Share your knowledge and questions with other users interested in using and developing image intelligence applications for Earth Sciences at http://community.ecognition.com/.

User Guide and Reference Book

Together with the software a User Guide and a Reference book is installed. You can access them in the Developer interface in the main menu Help>eCognition Developer User Guide or Reference Book. The Reference Book lists detailed information about algorithms and features, and provides general reference information. Both are also available in the Wiki section of the community http://community.ecognition.com/home/app_wiki

86

Where to get additional help and information?

Additional Guided Tours and Tutorials

If you are interested in further auto learning material you can work through the other Guided Tours you can download them on the eCognition community (http://community.ecognition.com/home/app_wiki ). Overview over the current Guided Tours: Level1_GuidedTour_LiDAR_SimpleExample Level2_GuidedTour_MappingImperviousSurface Level3_GuidedTour_LiDAR_AdvancedExample

Tutorials can be found in the Wiki section of the community http://community.ecognition.com/home/training-material

eCognition Training
eCognition Training Services offer a carefully planned curriculum that provides handson, real-world training. We are dedicated to enhancing customers image analysis skills, helping these organizations to accomplish their goals. Tailored courses are available to meet the needs of customers engaged in Life and Earth Sciences. Our courses are held in our classrooms around the world and at customer's sites. We offer regular trainings as Open Classes, where anyone can register and as In-Company Training. We also offer Customized Courses to satisfy customer's unique image analysis needs, thereby maximizing the training effect. For more information please see our website at http://www.ecognition.com/learn/trainings . eSeminars Join one of the live, online eSeminars: http://www.ecognition.com/learn/trainings/web-based-self-study-training

Consulting
eCognition Consulting Services experts have deep insight into products, best practices and professional project management skills. As a result, we are able to deliver rapid implementations that maximize the return on investment and minimize total cost of ownership. Consulting can also provide full-service, turnkey solutions, including project management, off-site rule development, QA and implementation reviews. Customers can count on our consultants considerable expertise in the application of technology.

Buy Software and Services?

If you want to purchase software or if you want to know more about pricing please visit http://www.ecognition.com/buy and get in contact with our sales team.

87