应对低光度下的变电站开关柜指示灯图像增强网络OA

In response to the problem in substation robot inspections where low light conditions lead to poor qual-ity capture of switch gear indicator light images,uneven illumination,and severe noise interference,affecting subsequent state detection,this paper propose a low-light indicator image enhancement network tailored for this scenario.The model connects multiple structures composed of an improved frequency-domain self-attention mechanism and a feature refinement feed-forward network to achieve precise extraction of global and local features.Combined with a brightness adjustment module,it progressively optimizes the illumination distribution and reduces feature loss through multiple re-sidual connections.Experiments show that the model performs excellently on a self-built substation indicator light data-set.Compared with advanced algorithms such as Uformer and EnlightenGAN,it achieves a PSNR of 23.89,an increase of 1.10 over the optimal comparative model LLFormer;a SSIM of 0.864,an increase of 0.061;and a LPIPS of 0.103,a decrease of 0.024.It can effectively improve the brightness of low-light images,suppress noise,and enhance illumina-tion uniformity,providing high-quality image support for inspection robots to accurately recognize indicator light states.