Fighting Poaching Through Targeted Deep Learning and Sensor Integration

Passive acoustic monitoring (PAM) has become a crucial and widespread tool for conservation monitoring, aiding the protection of species threatened by gun-based poaching through detecting calls and vocalizations. However, real-time detection of gun-based poaching activity remains an unsolved challenge despite its large ecological implications. Existing methodologies face high false positive rates and utilize computationally intensive models unsuitable for real-time field deployment. This research developed a lightweight deep neural network suitable for on-board processing and a sensor integration layer to address these limitations. The developed model achieved a 0.91 validation F1 at 935k parameters, retaining 94\% performance (F1 @ 95\% recall) of existing literature while reducing size by over 87\%. Statistical evaluation across acoustic array simulations demonstrated consistent false positive reduction through the proposed sensor integration function, presenting a promising approach for cost-effective real-time poaching detection and wildlife conservation.