Abstract: This paper addresses two critical challenges in applying general image segmentation models to specific domains: high training costs and inadequate feature representation. We propose a novel ...