Automated Wildlife Species Classification Using Deep convolutional Neural Network(CNN) : Enhancing Biodiversity Monitoring through convolutional Neural Networks

Wildlife species identification is essential for biodiversity monitoring and conservation, yet manual approaches are often slow, labor-intensive, and error-prone. This project introduces an automated recognition system based on Convolutional Neural Networks (CNNs) with ResNet-101 as the backbone for robust and efficient feature extraction. The dataset of wildlife images was preprocessed using resizing, normalization, and augmentation techniques, while oversampling was employed to address class imbalance and one-hot encoding for multi-class classification. The pretrained ResNet-101 model was fine-tuned to retain generic feature layers and adapt higher layers for species-specific learning. Bounding box localization was integrated to focus on animal regions, reducing background noise and improving detection accuracy. The model was trained using cross-entropy loss and the Adam optimizer, with dropout and batch normalization applied to minimize overfitting. The system achieved an accuracy of 69.32%, precision of 79.61%, recall of 69.38%, and an F1-score of 70.73%, demonstrating its robustness and reliability in identifying even visually similar species. The combined use of oversampling, fine-tuning, and bounding boxes significantly enhanced classification performance, making the framework scalable and adaptable for real-world conservation applications. Furthermore, this system supports automated biodiversity surveys, reduces manual effort in wildlife monitoring, and assists ecologists in rapid decision-making. Overall, the proposed model bridges the gap between artificial intelligence and ecology by providing a reliable, scalable, and sustainable solution for global biodiversity protection.