Development of an algorithm for detecting defects in glass insulators based on computer vision using a neural network approach

The article presents an algorithm for detecting defects in glass insulators using computer vision. Insulators, which are key elements of electrical networks, are subject to various defects, such as bubbles, chips and deformations. Such damage can significantly reduce the service life of insulators. In traditional production conditions, these defects are detected manually, which reduces productivity and increases the likelihood of human factor-based errors.
To solve the problem related to manual control restraints, the authors developed an algorithm based on the use of a neural network. The main task of the algorithm is to automatically identify defects that have a significant impact on the mechanical and electrical insulation properties of products.
The authors collected a data set for training the neural network and supplemented it with generated images to increase the sample of the location and shape of the considered defects. The paper describes in detail the steps of data preprocessing, including augmenting the contrast to increase the detectability of defects and reducing noise. Fragmenting is described to process defects of various sizes and shapes. Such fragmenting allows detecting defects of different sizes relative to the insulator size.