CNN-Based Bird Sound Detection: A Comparative Performance Study

Abstract

The use of automated systems for biodiversity monitoring has positioned bird sound detection as a crucial   technology for modern ecological research. This study introduces a deep learning framework for bird sound detection, demonstrating strong performance in challenging acoustic environments. By leveraging feature extraction techniques such as Mel spectrograms and comparing multiple convolutional neural network architectures, this approach achieves competitive results. In our experiments, the best-performing model ResNet50 achieved an accuracy of 90.75%, with a recall of 0.89, a precision of 0.91, and an F1-score of 0.90 on the test set. These results highlight the method’s potential for real-time biodiversity monitoring, providing a reliable tool for ecological research, citizen science initiatives, and conservation efforts.