The Road Region Automatic Segmentation Model Combining YOLOv8 and Segment Anything Model

In order to solve the problem of low automation and accuracy in road image processing under high altitude view and complex background, We propose a new Road area Segment Anything Model (RaSAM) based on SAM (Segment Anything Model). By introducing target detection prompt generation mechanism, RaSAM realizes fully automatic segmentation of road region, which significantly improves segmentation accuracy and model generalization ability. The experimental results show that RaSAM improves mIoU and Dice indexes by about 10 percentage points compared with other mainstream segmentation models, and improves by about 5 percentage points compared with the original model. Especially in zero-sample learning, RaSAM performs more prominently, achieving 92.88% mIoU and 93.29% Dice coefficient even under complex background, which verifies its excellent segmentation effect and strong generalization ability.