Insulator Defect Detection and Recognition Utilizing Stochastic Configuration Network and Convolutional Neural Network

Power outages on transmission lines are often caused by insulator failure. However, low insulator detection efficiency, poor generalization of insulator defect feature extraction methods, and a scarcity of defective insulator samples have made it challenging to detect and recognize insulator defects. To address these issues, a new method based on stochastic configuration networks (SCNs) and convolutional neural networks (CNNs) has been proposed. This method employs deep transfer learning and is capable of detecting and recognizing insulator defects. The YOLOv7 CNN is used to extract the insulator region, while the StyleGanv3 convolutional adversarial generative network generates defect insulator images. Next, the ResNet CNN extracts semantic information from insulator images as feature vectors, which are classified by the Stochastic Configuration Networks classifier. Experimental results indicate that this method is effective in detecting the insulator area and identifying insulator defects.