Visible Image Recognition of Power Transformer Equipment Based on Mask R-CNN

Abstract

Visible image recognition of power substation equipment is an important part of intelligent operation and maintenance, fault location and infrared defect diagnosis. The image recognition methods used in the past have some problems, such as less details of equipment, vulnerability to environmental interference, low recognition rate and so on. In this paper, a visible image recognition model of substation equipment based on Mask R-CNN is established. The image recognition model based on Faster R-CNN is used to compare the results. The two types of image recognition models are trained by using 7,000 labeled data sets of visible photographs of power substation equipment. Using the image recognition model trained by 2,000 visible light photographs of substation equipment, this paper mainly studies the recognition effect of 11 typical substation equipment, such as transformer, current transformer and bushing. The results show that the model based on Mask R-CNN can realize the pixel-level recognition of power substation equipment and can be used for accurate identification of power substation equipment. Although it cannot achieve pixel-level recognition, the model based on Faster R-CNN is faster and more accurate than that based on Mask R-CNN in target box detection. The model based on Faster R-CNN is faster and more accurate than the model based on Mask R-CNN. Both models need to be further improved for the scene of power substation equipment.