面向变电站智能安监的行为识别与时空特征决策方法OA

To address the issues of insufficient granularity in feature extraction and limited adaptability to complex video surveillance scenarios in substations inherent in traditional power personnel behavior recognition methods,this study investigates behavior recognition technology tailored to the needs of power operation and maintenance.An end-to-end video behavior recognition framework is adopted to directly model raw surveillance videos,and a key frame extraction method based on spatiotemporal features is designed to improve inference efficiency.A behavior classification decoder is constructed to enhance the discriminative ability for multiple types of operational actions.The experimental results on the real substation operation video dataset show that the proposed method achieves an overall recognition rate of 93.7%,significantly outperforming traditional image recognition methods such as support vector machine(SVM)and multi-layer perceptron(MLP)in both recognition accuracy and processing speed.The research conclusion provides a technical reference for improving the intelligent monitoring ability of power field.