Convolutional Neural Network for Biological Image Classification

• Designed a custom CNN architecture with sequential convolutional blocks, ReLU activations, and max-pooling layers to efficiently capture hierarchical spatial features for classification of seed shrimp appendage images
• Leveraged transfer learning by integrating pre-trained models, including EfficientNet-B0 and ResNet18, applying domain-specific image transformations to improve model performance and benchmark against custom architectures
• Performed GPU-accelerated training and inference in PyTorch with weighted random sampling to address class imbalance, achieving 81.8% accuracy and demonstrating effective feature extraction across model