Abstract:
The goal of this research is to create a reliable system for classifying local bird species using
modern deep learning architectures. The dataset forms the basis for this study and consists of 1,179
photos covering seven different species of birds. A combination of deep learning and transfer
learning algorithms, such as "InceptionV3," "VGG19," "VGG16," "MobileNetV2," "CNN01," and
"CNN02," are used in the search for precise and effective classification. The process is a
painstaking procedure that begins with web crawling to add to the dataset after data collection
from web scraping. Accurate annotations for model training are ensured via manual data labeling.
The construction and training of convolutional neural network (CNN) models is the study's central
focus. Important designs are used, including 'InceptionV3,' which is well-known for its deep and
effective design, 'VGG19' and 'VGG16' with uniform architectures, and 'MobileNetV2,' which is
especially made for devices with limited resources. Furthermore, flexibility is offered by the
generic CNN designs ('CNN01' and 'CNN02'). The experimental findings show that the
'MobileNetV2' design produced the maximum accuracy of 99.58%. This shows how well the
model can classify local bird species and make generalizations. The study highlights the
importance of transfer learning via using model training to improve productivity and faster
convergence. The accuracy obtained is proof of the effectiveness of the selected deep learning
architectures for classifying bird species .This work offers a dependable and automated method
for identifying local bird species and provides useful details about the use of modern deep learning
methods in bird study. The results have effects on protecting nature, environmental monitoring,
and the larger field of computer vision in nature research.
