Stacked Res2Net-CBAM with Grouped Channel Attention for Multi-Label Bird Species Classification

Identification of bird species through automatic analysis of their vocalizations holds great promise for various fields, such as ecology, conservation monitoring, and vocal behavioural studies. In recent years this has become a research-active area, and many studies have used deep-learning models to classify bird calls. However, small and imbalanced datasets often limit the performance of these models. In this paper, We explore the effectiveness of Res2Net and Convolutional block attention module (CBAM) with Spatial attention (SA) and Grouped Channel attention (GCA) using a sequential aggregation strategy (Se) for multi-label bird classification. The proposed framework which uses fewer parameters than other residual attention frameworks and can be used directly for audio and image classification tasks. Our findings show that our proposed framework is superior to state-of-the-art models with an F1 score of 72.20% using Mel-spectrogram features.