Research on Behavior Recognition of Transmission Line Live Working Based on Temporal-Channel-Spatial Attention Mechanism

Safety distance is the primary factor to ensure the personal safety of live working, and numerical limits have been implemented for various types of safety distances in the production system. However, it is still a challenge to determine the potentials of the operators according to the different behavioral categories of live working in the monitoring process, so as to realize the automatic triggering of the distance restrictions. In this paper, we propose a temporal-channel-spatial attention mechanism-based model (TCS-3DRN) for efficiently recognizing behavior features of live working operators. Temporal attention module (TAM), channel attention module (CAM), and spatial attention module (SAM) are able to compute the attention weight matrices along the temporal, channel, and spatial dimensions, respectively, to obtain local features, which are jointly fused into a 3D residual network, to enhance the capability of action recognition for live working. Experimental results show that the classification accuracy of TCS-3DRN is at least 9.5% higher than that of other common models, which can provide a reference for the intelligentization of the safety monitoring of live working.