Enhancing Bird Species Classification with PaliGemma and Custom EfficientNet Architectures

Bird species classification is a critical task in biodiversity studies, ecological research, and conservation efforts. Our research presents a comprehensive approach to bird species classification using a custom-modified EfficientNet-B0 model, enhanced with a Squeeze-and-Excitation (SE) block to improve feature sensitivity. The dataset used comprises a diverse collection of bird images, which are preprocessed and split into training, validation, and test sets. The custom model incorporates the SE block into the EfficientNet-B0 structure, with a dropout layer and a connected layer added for classification. Using an additional Squeeze-and-Excitation (SE) block to improve feature sensitivity and a specifically tailored EfficientNet-B0 model, our work provides a comprehensive approach to bird species classification. The dataset consists ofseveral hundred preprocessed images of birds, split into test, validation, and training sets. For classification, the dropout layer and connected layer are added, and the custom model integrates the SE block into the EfficientNet-B0 framework. In our model design, the learning rate is adjusted as needed using a learning rate scheduler, and SGD is employed for model training. The findings show that the custom EfficientNet-B0 model boosts classification accuracy highlighting how well the SE block works to improve model performance.The paper shows deep learning methods for animal tracking and conservation. Index Terms- Convolutional Neural Networks (CNNs), Squeeze-and-Excitation block, Deep learning, EfficientNet-B0,Fine-grained Image Classification Bird Species Classification neural networks wildlife monitoring pali jewel