Research
GetNetUPAM: Ecologically Informed Nested Cross-Validation and Noise-Robust Attention for Marine Bioacoustic Monitoring
GetNetUPAM introduces a novel hierarchical nested cross-validation framework designed for marine bioacoustic monitoring, addressing the challenges of high-noise and low-SNR conditions. It evaluates the Adaptive Resolution Pooling and Attention Network (ARPA-N), a CNN architecture that incorporates CBAM spatial attention to enhance noise suppression and generalization across diverse ecological environments. This approach significantly improves model robustness, reducing false positives by approximately 10x while maintaining a 90% recall, thereby advancing the reliability and scalability of ecological monitoring systems.
bioacousticsmodel-evaluationcross-validation