By improving the standard U-Net architecture, this paper proposes a novel semantic segmentation model, which incorporates multiple attention mechanisms to enhance the model’s capacity to capture multi-scale features. Specifically, we introduce the Efficient Multi-Scale Attention Module with CrossSpatial Learning (EMA), Spatial and Channel Squeeze and Excitation (SCSE), and Squeeze-andExcitation (SE) mechanisms into the standard U-Net network. These modules assist the network in learning significant information from feature maps at multiple scales while suppressing interference from irrelevant background. Experimental results demonstrate that incorporating attention mechanisms effectively enhances the prediction accuracy of the standard U-Net network for lane line semantic segmentation. The new model outperforms the standard U-Net model on our custom dataset, with particularly significant improvements in lane detection accuracy in scenarios with certain interference.