An MSCBAM-Unet incorporating a dual-attention mechanism for cohesive stone segmentation

Stone particle size calculation faces many challenges: the large number of stones, tightly adhered and fuzzy edges make it difficult for existing methods to accurately segment individual entities, and related industry applications have put forward increasingly high requirements for accurate stone segmentation. However, the traditional U-shaped network has obvious limitations in dealing with such problems, and cannot effectively focus on the edge region and multi-scale features of stone materials. This paper proposes an enhanced MSCBAMUnet network design to solve these issues. This network is introducing a dual-attention mechanism. The multi-scale module increases stone edge segmentation accuracy by focusing on both the stone body and the edge region, reducing interference from background and other variables. The network suggested in this paper outperforms existing approaches such as U-Net [1], SegNet [2], VGG-U-Net [3], and Attention-U-Net [4] for segmentation.