Intelligent defect detection for Power Patrol Inspection

Abstract In this paper, the insulator string images captured by Using unmanned aerial vehicles (UAVs) in power inspection are taken as research object. The image processing and deep learning methods used to label images, self-explosion fault of identified segmented strings is located. process semantic segmentation given data set enhanced, exchanged with algorithms. Based on SegNet, each pixel picture classified, then self-encoding for reference, first encoded decoded. coding layer before model uses VGG-16 network extract smaller pictures, so meet requirements segmentation, decoding after obtains classification probability value point classifies point. experimental results show that SegNet algorithm can realize recognition under complex conditions high accuracy. order identification location self-exploding position string, training be detect image. First, target detection preprocessed, transformation method rotation mirroring used; SSD selected identify locate insulator. combines advantages Faster R-CNN YOLO model. It runs faster has higher accuracy, which meets practical applications. more targeted set, improve speed accuracy positioning.