Intelligent diagnostic method for composite insulator surface temperature distribution realized by visible images

Efficient and accurate diagnosis of thermal defects in composite insulators is essential for maintaining the safe and stable operation of transmission lines. Traditional thermal defect diagnostic methods rely on infrared thermal imagers. However, the high cost limits their large-scale application in power inspections. Therefore, this research proposes a method to directly diagnose the surface temperature distribution of composite insulators from their visible images using a deep learning algorithm. The You Only Look Once version 7 is used for the precise localization of composite insulators in visible images, and their backgrounds are cropped. These cropped images are fed into the improved U-Net for accurate diagnosis and distortion-free visualization of insulator surface temperature distribution. Then, the model is further fine-tuned using physics-improved U-Net to enhance its accuracy. The results show that the model can diagnose the surface temperature distribution of composite insulators at different angles and brightness. The temperature diagnostic range is from 15 to 125 °C, achieving a mean absolute error of 0.780 °C and a root mean squared error of 1.130 °C, and the diagnosis time for a single image is 30 ms, which realizes a wide temperature range, high accuracy, and real-time temperature diagnosis of composite insulators based on visible images.