基于异构双流图卷积网络的人体行为识别OA

Aiming at the urgent demand for real-time and high-precision abnormal behavior recognition in military security scenarios,a heterogeneous dual-stream graph convolutional network is proposed.By decoupling skeletal features into geometric features and coordinate features,a dual-stream network is constructed to process heterogeneous information.An early fusion mechanism is adopted to dynamically integrate features,and a channel bottleneck constraint module is designed to optimize computational efficiency,so as to achieve the optimal balance between accuracy and efficiency and meet the deployment requirements on edge devices.Experimental results demonstrate that the proposed model achieves better performance compared with other mainstream models,and can provide effective solutions for military security applications such as abnormal activity recognition and sentry behavior monitoring in military sites.