Advanced Deep Learning Framework for Multi-Stage Bird Song Classification

Birdsong classification is vital for ecological monitoring, particularly in island wetlands. This study introduces the Multi-Stage (MS) model to address complex and mixed birdsong signals, comparing its performance against ResNet50, GoogLeNet Inception v1, and GoogLeNet Inception v4. Using a self-tested dataset of six bird species, the MS model achieved the highest recognition rate of 95.17%. The findings confirm the effectiveness of deep learning in birdsong recognition, while also highlighting the strengths and limitations of various models for specific tasks.