Research of Low-light Image Enhancement Method for Enclosed Tank

Abstract

Due to closed environment of gas insulated switchgear (GIS) chamber, low-light images are easily produced when taking pictures under weak lighting conditions. The lost details and low contrast not only cause unpleasant subjective feelings, but also hurt the performance of many computer vision systems which are designed for normal-light images. Thus, a low-light image enhancement algorithm based on improved conditional generative adversarial network (GAN) is proposed here to solve this poor visual perception problem. Decom-Net referenced to RetinexNet used here decompose image into reflection map and illumination map. This study proposed an encode-decode convolutional neural network model as the generative model and a lightweight convolutional neural network(CNN) as the discriminative model. After joint training of generative model and discriminative model, the network outputs the final enhanced images. The experimental results demonstrate that the proposed method is more effective than existing methods in perception.