An Infrared Detection Scheme for Distribution Network Power Equipment Based on Deep Convolutional Auto-Encoder for Small Negative Samples

Infrared detection is frequently used to detect abnormal heating of power equipment in the distribution network. In this paper, the infrared picture of the equipment is reconstructed using the DCAE structure, and the abnormal heating area is discovered using the pixel residual between the reconstructed image and the original image, in order to keep the power grid safe and stable. The experimental results reveal that the DCAE architecture employed in this study has a recognition accuracy of 94.36%, which has greater advantages than the existing standard detection approaches. At the same time, because it does not require the massive data set required for traditional training, it is more suited to the field of infrared detection, where the number of abnormal fever samples is modest.