Mapping Deep Learning Approaches for Acoustic Bird Species Classification in the Brazilian Pantanal

Automated acoustic monitoring of bird species has strong potential for biodiversity conservation, yet key questions remain regarding effective visual representations, deep learning architectures, and preprocessing strategies, particularly in biodiversity-rich regions such as the Brazilian Pantanal. This study analyzes approaches for acoustic bird species classification. Results indicate that Mel-spectrograms and CNNs are the predominant standards, while temporal segmentation is widely adopted as a preprocessing step. Additionally, less than 6% of the reviewed studies focus on Brazil, with none addressing the Pantanal, highlighting the need to combine globally pre-trained architectures with regional datasets for automated biodiversity monitoring.