Semantic Segmentation Algorithm of Nighttime Road Surface Water Based on Improved Low-Light Image Enhancement

At present, when semantic segmentation algorithms with high segmentation performance in daytime scenes are directly applied to nighttime scenes, it is difficult to annotate data due to the influence of light and low visibility, especially in the segmentation task of road surface water logging in urban waterlogging monitoring, the effect of water reflection is enhanced, resulting in poor segmentation results. Therefore, this paper proposes a semantic segmentation algorithm for nighttime road flooding. The algorithm introduces SCI low-light image enhancement network to enhance the nighttime road surface water images to reduce the influence of light, and adds a feature enhancement module to solve the problem of image detail loss. At the same time, the spatial attention mechanism is introduced to solve the problem of excessive enhancement and distortion in local areas with good light. Finally, the fusion effect is verified by the Upernet-Swein-T semantic segmentation model. The results show that the mIoU of the fused algorithm reaches 90.03%, which is 1.7% higher than that of the original model algorithm, and the ability of predicting night road surface water is improved.