Smoking Detection in Substation Field Operation Based on Improved Faster R-CNN

Smoking in the substation may cause fire, explosion and other accidents, seriously endangering the safety of the substation. In order to realize on-site smoking behavior detection in substations, Faster R-CNN was used in this paper to detect workers and smoking behaviors and determine whether smoking behaviors existed. In order to improve the detection effect, this paper introduces multiscale representation and contextual information on the basis of the original Faster R-CNN algorithm to optimize the algorithm structure. The former can select several important feature mappings containing the most detailed and semantic information to improve the operation speed, while the latter can significantly improve the detection accuracy by utilizing the environmental information around the detected small target. According to the final experimental analysis, compared with the original Faster R-CNN algorithm, the proposed method in this paper improves Precision, Recall and AP by more than 27%, which has obvious performance advantages and can meet the requirements of substation smoking detection to a greater extent.