CNN based Image Processing for Crack Detection on HT Insulator’s Surface

The proposed work provides an innovative method for employing real-time image processing to find flaws or cracks in electrical equipment using real-time image processing with Raspberry Pi in Proteus software. There are two distinct phases to this strategy. To find any cracks in the electrical equipment, an image of the apparatus is first taken, and then the image is analyzed using a convolutional neural network (CNN). Manual inspection of the equipment is a traditional approach to finding cracks in electrical equipment. Manual inspection, however, takes a lot of time and effort. A number of challenges need to be addressed to develop an effective insulator fault detection system, acquire a large and diverse dataset of insulator images, develop a robust and accurate CNN model, and deploy the insulator fault detection system in a real-world setting. This study proposed a new method for segmentation of the insulator image by using the idea of the U-Net segmentation technique with VGG classification. While the part of segmentation result is still significantly influenced by the threshold of segmentation; this strategy increases the effect of segmentation of insulator pictures with many ups and downs. As a result, insulator fault identification using morphological characteristics and classifiers was accomplished. Compared with other unsupervised learning methods, this approach had strong observations of self-learning, for maintaining multiple classifiers which are weak into classifiers that are strong.