Research on Lightweight Semantic Segmentation Method Based on Attention Mechanism

To enhance the accuracy of lightweight semantic segmentation networks, a lightweight semantic segmentation network named TCFormer is designed. The overall architecture of the network is expounded, and each module of the network is elaborately introduced. Regarding the issue of information loss in the lightweight Transformer attention mechanism, a lightweight attention module integrating convolution and global attention mechanisms is devised to support information complementarity between global and local attention. The effectiveness of the network is verified through ablation and comparative experiments.