ECBirdNet: A Lightweight Convolutional Neural Network with Attention Mechanism for Bird Audio Classification

Birdsong classification plays a crucial role in biodiversity monitoring and ecological research, offering a non-invasive approach to assess species distribution and behavior. However, the complexity and variability of bird vocalizations pose significant challenges to achieving highaccuracy classification. In this study, we propose a novel convolutional neural network model, ECBirdNet, which is based on EfficientNetB0 and incorporates the Convolutional Block Attention Module (CBAM) to enhance feature extraction capabilities and improve classification performance. Using Mel spectrograms as input features, the model was trained and evaluated on a dataset comprising 30 bird species. Experimental results demonstrate that ECBirdNet outperforms several baseline models, including ResNet34, DenseNet121, MobileNetV2, and the original EfficientNetB0. Specifically, ECBirdNet achieved an accuracy of $97.49 \%$, a precision of $97.50 \%$, a recall of $97.47 \%$, and an F 1 -score of $97.45 \%$ on the validation set. These findings confirm that integrating the lightweight architecture of EfficientNet with attention mechanisms significantly improves the accuracy of bird audio classification. Overall, ECBirdNet provides an efficient and accurate solution for birdsong recognition tasks, with strong potential for application in large-scale wildlife monitoring systems.