Design and Optimization of Real-time Power Scene Action Recognition System Based on Edge Deployment

With the improvement of safety and operation standardization requirements in power industry, real-time motion recognition technology has become the key to ensure the stable operation of power system. This paper presents a design and optimization scheme of real-time power scene action recognition system based on edge deployment. This research adopts deep learning technology, especially the hybrid model of convolutional neural network (CNN) and long short term memory networks (LSTM), and combines edge computing architecture to design an efficient and accurate action recognition system. By real-time monitoring and identifying the operation actions of the staff in the power scene, the system can find potential safety hazards in time and improve the safety and operation efficiency of the power system. In the aspect of algorithm research, this paper further improves the accuracy and real-time performance of motion recognition by introducing attention mechanism and lightweight network structure. In the strategy of edge deployment, model compression and pruning are used to optimize the model performance, and edge computing devices with high performance and low power consumption are selected to realize the efficient deployment and operation of deep learning models on edge devices. The experimental results show that the system has achieved high recognition accuracy on several public data sets and user-defined data sets, and is significantly better than the traditional methods. In addition, the system shows good stability and reliability in practical application, which provides strong support for the intelligent upgrade of the power industry.