Abstract:This article addresses the many problems that traditional human behavior recognition research techniques use fixed perspective action data collection as the research objective, such as lighting changes, complex backgrounds, and other factors that seriously affect the recognition effect of human behavior in videos. A neural network-based complex operator operation feature recognition method is proposed. Taking the operational characteristics of distribution network operators as the research object, the problem of identifying operational characteristics of operators is solved by collecting, preprocessing, and model learning raw data, and comparing different processing paths and modeling methods. Based on the recognition speed and accuracy, a comparison was made between the two static recognition algorithms, KNN and SVM. The experimental results showed that under the optimal parameter conditions, the recognition accuracy of the KNN model and SVM model were the same, both reaching 100%. However, in terms of recognition speed, the recognition time of the KNN model was 0.62ms, and the recognition time of SVM was 2.4ms. In comparison, the KNN algorithm took less time. Therefore, under the optimal parameters of each model, the recognition speed of the KNN model is superior to that of the SVM model.