Method for Identifying Dangerous Operation Actions at Power Marketing and Energy Metering Sites Using the Mask R-CNN Image Segmentation Model

Abstract

Safety issues at power marketing electricity metering sites are related to the personal safety of staff and the stable operation of the power system. Accurate identification of dangerous operating actions is crucial.Traditional CNN recognition is easily affected by background interference in recognizing dangerous operation actions at power marketing and energy metering sites and has a weak ability to express important features. This paper uses the improved Mask R-CNN image segmentation model to perform pixel-level segmentation on dangerous operation actions at the power marketing and energy metering site, and accurately locates its boundaries and actions. The study first uses the GAN model to expand the dangerous operation action images to ensure data balance, and uses the ResNeXt module to replace the ResNet in the traditional Mask R-CNN for feature extraction. Then, CBAM is embedded in the feature extraction module to enhance the extraction of spatial and temporal information of dangerous operation actions and reduce background interference. Finally, the loss function is optimized by combining Boundary loss to reduce the impact of the missing edge of the dangerous operation mask. This paper conducts experiments based on images of an actual power work site from June to December 2024. The results show that the improved Mask R-CNN performs best, with an accuracy of 96.9%, which is 2.6% higher than Mask R-CNN, and MIoU reaches 95.9%. The experimental results show that combining the Mask R-CNN image segmentation model and optimization module can effectively improve the accuracy and segmentation accuracy of dangerous operation action recognition at the power marketing and energy metering site, and ensure the safety of operators.