Anomaly detection method of substation infrared image based on knowledge box prior

Object detection in infrared images taken by substation unmanned aerial vehicles is a vital problem in detecting thermal defects in substation equipment images. According to the characteristics of infrared images of substation unmanned aerial vehicles, an object detection method for infrared images is proposed based on the technical route of fixed-point annotation. Firstly, the Canny algorithm with adaptive thresholds is adopted, in order to ensure the effectiveness and similarity of edge feature extraction between the template image and the image to be recognized. Then, a consistency verification algorithm based on sliding windows is proposed to verify the consistency between the template image and the image to be recognized. Finally, based on the consistency verification results and the template matching algorithm with bidirectional comparison, the object detection of substation equipment in images is achieved. The experimental results show that the design in each step of the proposed method has a significant effect on improving the object detection accuracy of infrared images taken by substation unmanned aerial vehicles. According to the experiment, compared with the methods of ‘no Canny’, ‘no Otsu’ and ‘no edge rate’ in template matching, P2 increased by 2.37%, R2 increased by 6.78% and F2 increased by 4.66%. Compared with the classical target detection methods RetinaNet, ObjectBox and Yolov, P2 is increased by 10.55%, R2 is increased by 7.20%, and F2 is increased by 8.89%, which verifies the effectiveness of the proposed method.