In the cutting process of pork loin and pork belly, the cutting path should be smooth and anatomically fit due to the fascia and fat covering of the seam surface of the pork loin and the malleability of the pork belly. Therefore, this paper proposes a cutting path generation method that combines semantic segmentation with point cloud characteristics. Firstly, the semantic segmentation of pork loin region and pork belly is realized by using DualSeg-KNet model, and then the mask edge extraction is combined with point cloud target region (ROI). Finally, the path feature points are extracted by local oriented boundary box (LOBB) feature descriptors to fit the high-precision cutting path. The experimental results show that the cutting accuracy of Dice and IOU reaches 91.2% and 83.6% respectively, which significantly improves the cutting accuracy of pork loin and pork belly.