Improved Camouflage Personnel Detection Based on YOLOv8

Aiming at the problems such as strong environmental interference, serious target occlusion and insufficient multi-scale feature fusion in camouflage target detection in complex battlefield environment, this paper proposes an improved YOLOv8 military camouflage detection algorithm. Firstly, the Triplet Attention mechanism is embedded in the C2f module of the backbone and neck network to enhance the feature characterization capability through cross-dimensional interaction. Then BiConcat is designed to replace the traditional Concat operation and realize the deep interaction of multi-level features through bidirectional cross-scale connection and adaptive weight allocation mechanism. Finally, in terms of loss function improvement, a boundary box regression strategy based on Inner-SIoU is proposed, and the auxiliary anchor frame with dynamic adjustment is used to fine-adjust the prediction frame, effectively accelerating the model convergence and improving the positioning accuracy. The experimental results show that compared with the original model, mAP0.5, R and FPS increase by 2.7%, 4.1% and 4.8% respectively. This research provides a new technical path for real-time target detection in complex battlefield environment.