A modified ResNet152v2 framework for bird species classification

Ornithology, the study of birds, plays a crucial role in understanding the complex interplay between avian species and their ecosystems, offering insights into biodiversity and global ecological systems. In recent years, the conservation of bird species has become increasingly critical, particularly as habitat degradation poses a significant threat to endangered species. Accurate identification and classification of bird species are essential components of conservation strategies. This research presents a modified ResNet152v2 framework designed to enhance the precision and efficiency of bird species classification. By leveraging advanced deep learning techniques, this framework aims to support conservation efforts by facilitating the accurate monitoring and protection of diverse avian populations. This study introduces a transfer learning approach for bird species identification using a modified ResNet152v2 framework, leveraging a publicly available dataset of 87,874 images spanning 515 bird species. The model achieved an impressive test accuracy of 97.67% and a categorical accuracy of 98.81%. A key contribution lies in the utilization of principal component analysis (PCA) for dimensionality reduction, optimizing the feature vector by 15%. Additionally, root mean squared propagation (RMSProp) was employed to enhance the feature vector. L2 regularization and dropout at the input and the subsequent layer were introduced to mitigate overfitting. The model underwent rigorous training, resulting in a minimum training loss of 0.0397 and a validation loss of 0.0044. The achieved average precision (98.66%), recall (97.73%), and F1-score (98.2%) values further validate the model’s effectiveness. This study presents a novel combination of techniques, emphasizing the integration of transfer learning, dimensionality reduction, and dropout for improved accuracy, resulting in a robust model with potential applications in ecological research and environmental monitoring.