Human Behavior Recognition for Substation Operation and Maintenance

Abstract

In the field of substation operation and maintenance, recognizing the behaviors of substation operation and maintenance personnel is crucial for maintaining the safe operation of the power grid and ensuring the safety of workers. To enhance the accuracy of human behavior recognition in substation operation and maintenance, this paper proposes an ECA-MSG3D algorithm model for human behavior recognition of work apparatus based on the Multi-Scale Pointwise Convolutional Neural Network for 3D Object Detection (MSG3D) model. The MSG3D module, which utilizes a multi-scale aggregation scheme for unified spatiotemporal graph convolution, is used for feature learning across time and space. Then, channel features are learned through the efficient channel attention (ECA) module, and the final action is obtained through global average pooling with the softmax classifier. Results show that the ECA-MSG3D model with the ECA attention module outperforms the MSG3D model with other attention modules, achieving 89.06% and 89.26% accuracy and precision on the NTU RGB+D 60 Skeleton dataset. This model provides a highly accurate human behavior recognition method for substation operation and maintenance.