RCAF-Net: Wildlife Target Detection in Complex Forest Scenarios

This study focuses on wildlife target detection in complex forest environments in Northeast China, where monitoring images are often affected by background interference, frequent occlusion, and large target scale variation. In addition, distant animals usually appear as small targets with limited local detail information, increasing the difficulty of reliable detection. To address these challenges while maintaining deployment efficiency, an improved wildlife target detection model, RCAF-Net, is proposed based on YOLO11n. The proposed method enhances shallow feature representation, strengthens multi-scale contextual modeling, improves cross-layer feature fusion consistency, and introduces a lightweight detection head to balance detection accuracy and computational cost. Experimental results on a self-built dataset of eight typical wildlife species show that RCAF-Net achieves Precision, Recall, mAP@0.5, and mAP@0.5:0.95 values of 89.3%, 78.4%, 87.3%, and 67.3%, respectively, improving upon YOLO11n by 4.1%, 2.6%, 3.9%, and 3.4%. On the Wildlife Computer Vision Model dataset, the proposed model also achieved improved generalization performance in cross-dataset testing. In addition, the model operated at approximately 27 FPS on the Jetson TX2 NX platform. These results suggest that RCAF-Net has potential applicability for automated wildlife monitoring in complex forest environments.