A U-Net Based Lightweight Semantic Segmentation Method

Image segmentation is an important research hotspot in the field of computer graphics processing, and more and more scholars pay attention to it gradually. At present, semantic segmentation models not only need to have high-precision scene parsing ability, but also must meet strict real-time requirements. Therefore, it is of great theoretical significance and practical value to deeply study the balance mechanism between accuracy and speed of semantic segmentation. Based on the U-Net model, the backbone network of feature extraction is deleted to reduce the complexity of the model. At the same time, dilated convolution is introduced to increase the receptive field, so as to better capture the context information. The model further improves the ability of feature expression through the attention mechanism. In addition, a hybrid loss function is designed to accelerate the convergence of the model and enhance the ability to capture detailed features. The proposed method is verified on the CamVid dataset. Experiments show that the model achieves $80.23 \%$ and $88.06 \%$ in mIoU and mPA, which has significant performance improvement compared with other existing algorithms, verifying the effectiveness and superiority of the proposed method in semantic segmentation.