<p>Semantic segmentation in urban street scenes is essential for autonomous driving, enabling accurate perception of complex road environments. DeepLabV3+ is a widely used segmentation framework, but its high computational complexity and limited utilization of shallow features hinder its efficiency in real-time applications. In addition, class imbalance in urban street scenes leads to suboptimal segmentation performance of the model on small target categories. To address these challenges, this paper proposes a new LightWeight Semantic Sigmentation model for urban Street Scene(LW4S). LW4S uses GhostConv instead of the standard convolution in DeepLabV3+, which reduces the model’s parameters and computational costs while maintaining feature richness. LW4S constructed the LMCF module, which integrates GhostConv with an attention mechanism to enhance the fusion of shallow and deep features, and improve the model’s recognition of fine-grained objects with minimal overhead. LW4S proposes a weighted combination loss function that combines Focal Loss with Cross-Entropy Loss (CE Loss) to address the class imbalance problem in urban street view semantic segmentation tasks. The experimental results show that compared with the baseline models, LW4S achieves a better balance between segmentation accuracy and model computational complexity, demonstrating its effectiveness in urban street scene segmentation.</p>