This paper investigates intelligent enhancement techniques for low-light images based on deep learning. An image illuminance classification network was designed using an improved VGG network, with network width and depth pruning and the introduction of dilated convolutions to achieve lightweight network structure. By setting a probability threshold, low-light images are input into the subsequent image enhancement network for processing. Then, an image fault recognition method based on the VGG network + Retinex-Net and the improved YOLOv5 network is proposed. The analysis results show that the image fault recognition method proposed in this paper, based on Retinex-Net and the improved YOLOv5 network, can quickly and accurately detect and identify faults in power equipment.
