Toward Precision Biodiversity Detection: An Edge-Deployable Framework for Mitigating Data Redundancy.

Monitoring wildlife in remote areas is vital for biodiversity research, yet infrared-triggered cameras often produce large numbers of false or empty images due to environmental interference, causing data overload and power waste. To address this, this manuscript proposes WS-YOLO, a lightweight edge-optimized wildlife detection model based on improved YOLOv11n, which aims to integrate false-trigger filtering and accurate animal detection in a single model for field applications. Specifically, the model introduces three key enhancements: it adopts a Wavelet Convolution (WTConv) module to capture multi-scale and frequency-aware features, boosting robustness to complex backgrounds and small-target detection accuracy; meanwhile, a Spatial-Channel Synergistic Attention (SCSA) mechanism is incorporated to enhance spatial localization and channel-wise semantic encoding, enabling the model to focus on salient regions of wildlife; additionally, a SlideLoss function is designed to prioritize training on boundary cases, improving the handling of ambiguous detection instances. Experiments on four wildlife datasets demonstrate that WS-YOLO achieves excellent mAP@0.50 performance: 93.0% on African Wildlife, 98.0% on Amur Tiger, 98.5% on the NTLNP infrared dataset, and 79.4% on PASCAL VOC, confirming its robustness across diverse scenarios. Deployment on a Jetson Nano further validates its feasibility for real-time, energy-efficient use in the field. WS-YOLO enables accurate animal recognition and classification directly at edge devices, significantly reducing communication load and enhancing monitoring efficiency.