JNNCE JOURNAL OF ENGINEERING & MANAGEMENT, SPECIAL EDITION 2, 2024 ISSN 2582-0079

Available online @ https://jjem.jnnce.ac.in

https: www.doi.org/10.37314/JJEM.SP0213
Indexed in International Scientific Indiexing (ISI)
Impact factor: 1.395 for 2021-22
Published on: 08 December 2024

Airborne Species Detection using YOLOv8

Pooja L1*, Sampath Kumar S2
1*
Student, 2Assistant Professor, Department of Computer Applications
JNN College of Engineering, Shivamogga

[email protected], [email protected]

Abstract
Nowadays, the majority of avian species are uncommon to find, and when they are, it might be challenging to classi-
fy them. In the natural world, birds come in a variety of sizes, shapes, and colors as viewed. Because of the variabil-
ity in bird appearance, new procedures are necessary for reliable detection and conservation, which effects identifi-
cation and monitoring operations. to provide a reliable method that facilitates accurate and quick identification of
bird species, supporting successful conservation initiatives. This paper uses the YOLOv8 object identification algo-
rithm to do a thorough analysis on avian species recognition. YOLOv8, a sophisticated object identification method
based on convolutional neural networks (CNNs), is well-known for its quick and accurate inference processes, mak-
ing it appropriate for real-time detection and classification applications. The initiative intends to greatly improve
avian monitoring efforts by utilizing the YOLOv8 algorithm, supporting environmental study and biodiversity pro-
tection.

Keywords: Deep learning, avian species detection and YOLOv8.

1. Introduction control agents, where

Pest management is one of the unique roles

that birds play since they are organic enemies
of pests. Birds also act as pollinators, seed dis-
persers, and contributors to nutrient cycling,
hence playing a significant role in ecological
balance and creating diversity. Research pro-
posals in bird species detection concentrate on
enhancing the approaches required to recog-
nize different bird species and follow up on
avian diversity to support preservation and in-
vestigation. The above differences in birds
manifest themselves in sizes, physical mor-
phology, feathers, activity, locations of habit,
movements, and songs or calls, depending on
their respective ecological requirements and
the physical climate in which they occur.
Avian species identification plays a significant
role as a basis for studying the variety of birds
and ensuring the preservation of these species.
Avi- an species identification plays a
significant role as a basis for studying the
variety of birds and ensuring the preservation
of these species. Another important function
attributed to birds is that they act as bio-
JNNCE JOURNAL OF ENGINEERING & MANAGEMENT, SPECIAL EDITION 2, 2024 ISSN 2582-0079

they help reduce the rate of pest counts and

also assist in seed dispersal and nutrient
cycling. Recording and identifying bird
species are vital in evaluating the health of a
given ecosystem and determining population
densities. Identification of endangered species
is just one aspect of the information that is
vital for decision making processes of
conserving organisms and threatened species.

Traditionally, bird detection has

involved things like field observation, bird
population census, and specimen collection.
These methods provided useful information
but were often tedious and sometimes
accompanied by errors through human
involvement. Recent advancements in IT,
especially in machine intelligence and
computer vision methods, have significantly
improved bird species detection. They also
facilitate the correction of misidentifications
when conditions are unfavorable or when dif-
ferent species resemble one another. The most
effective technology for bird detection is
known as the YOLOv8 object detection
algorithm. YOLOv8 is a fast and accurate
tool,
JNNCE JOURNAL OF ENGINEERING & MANAGEMENT, SPECIAL EDITION 2, 2024 ISSN 2582-0079

standing for "You Only Look Once, version ambient noise. They incorporate depth
8". For instance, it can monitor videos and raw convolutional network technology for
or processed images in real time to detect and classification, picture pre- processing, and
dif- ferentiate the types of birds present. This grayscale conversion. Although the work lacks
capa- bility is highly relevant for observing robustness testing across several settings and
birds in different conditions, such as thick and comparison analysis with competing
sparse forests, meadows, and city spaces, approaches, it does highlight design
where time- ly and accurate detection is friendliness and provide modules for both au-
critical. tomatic and human bird identification.

2. Related Work Mrs. D. Aruna et al., [4] The Internet of Birds

(IoB), a smartphone app, the app's objective is
Pralhad Gavali et al., [1] talk about the to help birdwatchers identify Taiwan's 27
difficulties and developments in deep learning native avian species by using deep learning
methods for bird species identification. algorithms. The algorithm used is Convolution
Because of human and environmental Neural Network. The study highlights the
variables, bird species identification was ongoing challenges with gathering features and
initially mostly dependent on human skill and the need for more research and development of
relied on visual and aural cues. This method the suggested system in a wider range of
proved labor-intensive and error prone. The environmental settings and avian species, but it
drawbacks of conventional approaches are also points out the potential benefits of
emphasized, as is the need for more automated vision in improving the accuracy of
trustworthy automated procedures. Bird avian species authorization.
species detection has substantially improved
with recent advances, especially in machine Dillon Reis et al., [5] the authors build on the
learning and computer vision. TensorFlow is advances made in the YOLO (You Only Look
used to create signatures from photos, which Once) series to improve speed as well as preci-
are then converted to grayscale. A trained sion through feature extraction and single-
dataset is then compared to the autographs to stage recognition approaches. The YOLO
determine a score. model development, which began with
YOLOv4 and YOLOv5, shows continuous at-
Haong-Tu Vo et al., [2] The study highlights tempts to enhance real-time recognition abili-
the importance of bird species identification ties in numerous fields, include robots and ob-
due to its vital function in preserving the servation. To maintain YOLOv8's position as
natural equilibrium and environmental an important step forward in real-time flying
monitoring. The study uses YOLOv5, a tracking of birds, future research may examine
trustworthy object identification method, to more model architecture improvements and
accurately identify bird species in photographs. expand the application's reach to other do-
The study also looks at the effects of mains including drone technology and aerial
optimization techniques like SGD, Adam, and surveillance.
Adamax on model performance.
EfficientNetB3 routinely beats all other opti- Nguyen et al., [6] work, that employs TINY
mizers (SGD, Adam, and Adamax) with a test YOLO and GoogLeNet to recognize and
accuracy of 98%, demonstrating remarkable identify birds in real-time, highlights important
ability to classify unknown data. developments in computer vision for wildlife
monitoring. This method offers accurate
P. Srilatha et al., [3] Using the Caltech-UCSD species identification and real-time behavioral
Birds 200 dataset, the author illustrates their analysis, which not only assists conservation
methodology and claim that image-based projects but also improves biodiversity
solutions outperform audio-based ones due to
JNNCE JOURNAL OF ENGINEERING & MANAGEMENT, SPECIAL EDITION 2, 2024 ISSN 2582-0079

monitoring efforts. These enhancements are efficiency. By leveraging advanced deep learn-
necessary to expand the potential uses of sector ing techniques and temporal analysis.
randomization enhanced deep learning models
beyond the identification of birds, such as Dharaniya et al., [10] proposed the recognition
more extensive ecological studies and of specific avian species using convolutional
conservation campaigns. neural networks (CNNs). The study's objective
is to use more sophisticated algorithms for
Mao et al., [7] present field randomization machine learning to enhance decision-making
enhanced deep learning models for bird for avian type identification. CNNs are
detection, emphasizing advancements in specifically utilized because of their ability to
computer vision tailored for wildlife immediately learn these hierarchical aspects
monitoring. Their study integrates domain from bird photos, which is essential for
randomization techniques to enhance model categorizing numerous kinds of birds based on
robustness across diverse environmental their visual traits. This method offers a
conditions, crucial for accurate avian species thorough theoretical foundation for automated
identification. By mitigating variations in bird acceptance, resolving challenges with
lighting, background, and weather, these human bird categorization and representing a
models improve generalization beyond trained significant step forward in the process of
datasets. Previous research highlights the tracking and safeguarding animals.
effectiveness of domain adaptation strategies
in enhancing model performance and Stowell and Plumbley [11] they investigated
adaptability to real-world scenarios. This the computerized categorization of avian
approach not only enhances biodiversity species using unsupervised learning of
monitoring efforts but also supports conserva- features. In order to extract valuable
tion biology by enabling precise species identi- characteristics from sound information, they
fication and behavioral analysis. suggest an unsupervised learning strategy and
draw attention to the difficulties presented by
Yang Wang et al., [8] the authors emphasize the great diversity and unpredictability of bird
the importance of birds as ecological cries. Using sophisticated machine learning
indicators, traditionally monitored through techniques, the authors show that their
labor-intensive surveys. They review approach performs better than conventional
advancements in computer vision and deep supervised learning models. Based on their
learning, noting their potential to transform research, unsupervised feature acquisition can
bird monitoring into a more efficient and low- effectively capture intricate patterns in bird
cost process. This paper addresses the lack of sounds without requiring large amounts of
large-scale bird detection datasets, which limits labeled samples, which makes it an invaluable
the application of these technologies. tool for bioacoustics studies and conservation
initiatives.
Alqaysi et al., [9] propose a novel approach for
recognizing birds around wind farms using a
Temporal Boosted YOLO-Based model, as
outlined in their study published in the Journal
of Imaging. Their model integrates temporal
information into the YOLO (You Only Look
Once) architecture, enhancing its ability to ac-
curately detects birds in real-time video feeds.
The study evaluates the model's performance
against existing methods, demonstrating prom-
ising results in terms of detection accuracy and
JNNCE JOURNAL OF ENGINEERING & MANAGEMENT, SPECIAL EDITION 2, 2024 ISSN 2582-0079

3. Proposed Methodology In the figure 3.1 Data collection involved

downloading a comprehensive bird species
dataset from Kaggle. Each image was
annotated using LabelImg, drawing precise
bounding boxes around each bird and
assigning species labels. Metadata, including
location, date, and environmental conditions,
were recorded where available. During data
preprocessing, images were cleaned to remove
low-quality ones and resized to ensure uniform
resolution. Data augmentation techniques such
as rotations, flips, cropping, and scaling were
applied to enhance dataset diversity.
For YOLOv8 training, key parameters were
set a learning rate of 0.001, a batch size of 16.
Input images were resized to 640x640 pixels.
Data augmentation was enabled to improve
model robustness. The dataset was split into
70% for training, 15% for validation, and 15%
for testing, ensuring a comprehensive
evaluation of model performance. Model
evaluation involved calculating metrics such as
precision, recall, and mean Average Precision
(mAP). Finally, the trained model was
deployed, allowing for real-time bird species
detection in new images.

4. Results and Discussion

Fig 3.1: Flow Diagram of Avian Species Fig 4.1: Bulbul Species Detection
Detection using YOLOv8
In the figure 4.1 the bird has been
recognized as a "bulbul" with a confidence
level of 0.36.
JNNCE JOURNAL OF ENGINEERING & MANAGEMENT, SPECIAL EDITION 2, 2024 ISSN 2582-0079

Fig 4.2: Bat Species Detection

In the figure 4.2 the bird has been Fig 4.4: Confusion Matrix Plotted for
recognized as a "Bat" with a confidence level YOLOv8 model
of 0.89.
The figure 4.4 gives value of
metrics/precision and metrics/recall with
respect to data taken to produce the graph.

5.Conclusion

A reliable option is to use YOLOv8

particularly for bird species detection in
aviaries. With the use of this technology, birds
can be effectively identified and tracked in
real-time, allowing for prompt reaction to any
dangers. Because of its great precision, avian
bird protection and monitoring are effectively
carried out, minimizing the need for human
involvement and helping conservation efforts.
By accurately identifying and monitoring bird
species and enhancing wildlife management
techniques, YOLOv8 improves avian security.
Fig 4.3: Bird Object Detection This technology promotes cohabitation
between bird species and human activity,
A variety of avians have been recognized greatly aiding in the maintenance of a healthy
and assigned their proper labels: myna, stork, ecology within aviaries. All things considered,
thrush, woodpecker, ostrich, and kingfisher. YOLOv8 is an essential instrument for
Each avian is kept inside a bounding box protecting bird populations and encouraging
labeled with the species of bird that was environmentally sound aviary management.
discovered.
References

1. 1. Pralhad Gavli, Prachi Abhijeet Mhetre, Neha

Chandrakhant Patil, Nikita Suresh Bam- ane,
Harshal Dipak Buva, “Bird Species
JNNCE JOURNAL OF ENGINEERING & MANAGEMENT, SPECIAL EDITION 2, 2024 ISSN 2582-0079

Identification using Deep Learning”, Computer Vision, vol. 131, pp. 456-470, 2023.
International Journal of Engineering Research 13.Johnson, M., Thompson, R., Nguyen, L.,
& Technolo- gy (IJERT), pp. 68-72,2019. "Transfer Learning for Bird Species
2.Hoang-Tu Vo, Nhon Nguyen Thien, Kheo Recognition Using Pre-trained Convolutional
Chau Mui, "Bird Detection and Species Classi- Neural Networks," Journal of Computational
fication: Using YOLOv5 and Deep Transfer Ecology, pp. 112-120, 2023.
Learning Models", IJACSA-International 14.Zhang, Y., Wang, H., Li, X., "Multi- Sensor
Journal of Advanced Computer Science and Fusion Technique for Bird Species Detection,"
Applications, vol. 14, pp. 939-927, 2023. International Journal of Wildlife Monitoring,
3. P Srilatha, Rabin Kumar Sah, D Dinesh pp. 45-53, 2023.
Kumar, and Ramam Sripada, "Bird Species
Identification Using Deep Learning", Interna-
tional Journal of Research Publication and Re-
views, vol. 4, pp. 1962-1966, 2023.
4.Aruna D et al., "Bird Species Identification
Using Deep Learning", Dogo Rangsang Re-
search Journal, vol. 09, 592-598, 2022.
5.Dillon Reis, Jordan Kupec, Jacqueline Hong,
Ahmad Daoudi, "Real-Time Flying Ob- ject
Detection with YOLOv8", Georgia Insti- tute
of Technology, 2023.
6.Santhosh Kumar V, Anupriya K, Hari Balaji
S, Prabhavathy P, “Real Time Bird De- tection
and Recognition using TINY YOLO and
GoogLeNet”, International Journal of
Engineering Research & Technology (IJERT),
vol. 8, 2019.
7.Xin Mao et al., "Domain randomization-
enhanced deep learning models for bird
detection", pp. 1-13, 2021.
8.Yang Wang et al., “Bird Object Detection:
Dataset Construction, Model Performance
Evaluation, and Model Lightweighting”, Ani-
mals, vol. 13, 2924, 2023.
9.Alqaysi Hiba et al., “A Temporal Boosted
YOLO-Based Model for Birds Detection
around Wind Farms”, J. Imaging, vol. 7, 227.
10.Dharaniya R et.al., “Bird Species Identi-
fication Using Convolution Neural Network”,
IOS Press, 381-386, 2022.
11.Stowell D, Plumbley M D, “Automatic
Large-Scale Classification of Bird Sounds is
Strongly Improved by Unsupervised Feature
Learning”, IEEE/ACM Transactions on Audio,
Speech, and Language Processing, vol. 30, pp.
661-671, 2022.
12.Priyadarshini I, Das S K, “Hybrid Model
for Bird Species Detection Using Image and
Audio Processing”, International Journal of