Advancing Bird Classification: Harnessing PSA-DenseNet for Call-Based Recognition

This study focuses on call-based bird classification utilizing the PSA-DenseNet model. A dataset comprising 14,311 natural call audio clips from 20 bird species was collected. Results demonstrate that our model achieves impressive accuracy metrics: \(95.8\%\) accuracy, \(94.7\%\) macro \(F_{1}\) score, and \(94.8\%\) recall, surpassing existing models such as DenseNet, MobileNet V2, ResNet, and EPSANet. Moreover, the probability of correct classification for each bird species consistently exceeds \(85.29\%\) . A normalized confusion matrix is provided to illustrate performance. Our findings underscore the outstanding capability of PSA-DenseNet in call-based bird classification tasks.