This project features a deep-CNN model with residual connections for multi-class bird species classification, achieving 90.72% accuracy. It uses compound scaling laws, data augmentation, dropout, and StepLR scheduling to improve baseline accuracy by 30%. Class Activation Maps (CAM) are used to identify key regions influencing classification.

🚀 Key Features

• ✅ 90.72% classification accuracy
• ✅ Residual connections for improved performance
• ✅ Compound scaling laws and data augmentation
• ✅ StepLR scheduling for optimized training
• ✅ Class Activation Maps (CAM) for interpretability
• ✅ 30% improvement over baseline accuracy