Infrared Image Recognition Method of High Voltage Porcelain Insulator Based on Self-Constructed CNN

To achieve online automatic monitoring of high voltage porcelain insulator strings by infrared thermography, it is essential to obtain the temperature information of the iron cap and disc. In order to accurately extract the temperature, this paper proposes a self-constructed convolutional neural network (CNN) for automatic identification of the iron cap and disk area in an infrared image. The sample set of the algorithm consists of insulator images from multiple substations in various regions, without loss of generality. After training, the network finally outputs four classifiers of iron caps, discs, aluminum fittings and cables. Then we use these classifiers to identify the corrected insulator string region image. Finally, we use different colors to identify the target area and extract the temperature of the area with in-house code. By evaluating the relative temperature difference between an individual insulator and its adjacent insulators, we can discriminate whether there is deterioration in the insulator string. The experimental results show that the self-constructed CNN can achieve excellent recognition results for insulator string iron caps and disc surfaces of different voltage levels.