A Semantic Segmentation Algorithm Based on Improved Attention Mechanism

Abstract

In the field of intelligent driving, while ensuring real-time capability, how the semantic segmentation task can more accurately grasp the boundary information to obtain accurate segmentation is an urgent problem which wants to be solved. Based on this, the article designs a semantic segmentation network GSANet (Global and Selective Attention Network) based on the visual attention mechanism. The article mainly designs an ASPP structure GASPP (Global Atrous Spatial Pyramid Pooling) with global attention information to provide long-distance detailed information for the semantic segmentation model better. Then, a new SAM (Selective Attention Module) is introduced in the decoder stage to provide different attention weight information for different spatial positions. The results show that both the ASPP with global attention and the decoder with selective attention mechanism can significantly improve the accuracy.