Semantic Segmentation of Road Landscape Based on Improved Deeplabv3+

Image semantic segmentation is a critical task in the field of computer vision. However, traditional Deeplabv3+ faces challenges such as a large number of parameters and insufficient accuracy when handling small objects and object edges. Therefore, we propose an improved lightweight road semantic segmentation algorithm, WDeeplabv3+, based on the Deeplabv3+ network. (1) We introduce a Three-Parallel Down-sampling Feature Extraction Module (TPDFE), which enriches low-level features and enhances the model's ability to capture details. (2) We replace Bilinear interpolation up-sampling with CARAFE up-sampling and propose a Grouped Spatial Feature Fusion Module (GSFFM). After feature interaction, we use attention guidance for feature fusion, effectively alleviating the semantic gap in feature integration while achieving network lightweighting. (3) Shuffle attention is added to improve the model's feature expression ability and its capacity to capture long-range dependencies. At the same time, it retains important information while reducing computational complexity. Experimental results show that on the CamVid dataset, mIoU and mPA increased by 3.16% and 2.10%, respectively, while the number of parameters and computation load decreased by 88% and 96%.