The automated segmentation of sheep carcass is of great significance for improving the productivity of sheep slaughtering and processing enterprises and can contribute to a more intelligent sheep slaughtering and processing industry in China. In order to achieve accurate and efficient segmentation of the sheep carcass point cloud data into multiple splits, and provide a reference for the sheep carcass segmentation robot, a sheep carcass point cloud segmentation method was used based on surface convexity, and the Bame mutton sheep was taken as the research object. The sample point cloud data was collected on the sheep carcass segmentation production line of Meiyangyang Food Co., Ltd. in Inner Mongolia, Bayannaoer. Using the point cloud collection method, a handheld scanner was used to surround the sheep carcass. Multiple laser photosensitive films were randomly attached to the surface of the sheep carcass for three-dimensional positioning and scanning in data collection. The distance between the scanner and the sheep carcass was controlled within 200mm. The point cloud processing steps were as follows: the voxel filtering method was used to downsample the sheep carcass point cloud;the point cloud data was supervoxelized to obtain the supervoxel adjacency graph;the common edge of the adjacent point cloud in the supervoxel adjacency graph was judged by concave and convex, and the concave and convex edges were given different weights;a score function was introduced, and the relationship between the score of each point cloud and the minimum cut score according to different weights were calculated and compared;according to the comparison results, the Ransac algorithm was used to determine the segmentation plane, divide the segmentation area, and complete the segmentation of the sheep carcass point cloud. The test results showed that the average precision, average recall ratio, average F1 value and average overall accuracy of sheep carcass point cloud segmentation were 92.3%, 91.3%, 91.8% and 92.1%, respectively, and the average accuracy of each split were 92.7%, 90.7%, 92.6%, 93.2%, 92.5% and 92.2%, the average recall ratio were 86.0%, 93.2%, 92.8%, 91.6%, 90.9% and 93.4%, respectively. The average time to process a single sheep carcass point cloud was 18.82s. The applicability of this method was judged by segmenting combinations of different sheep carcass split point clouds and sheep carcass point clouds of different body weights, and the comprehensive segmentation ability of this method was verified by comparing two point cloud segmentation algorithms, namely the commonly used region grow and the Euclidean clustering. The results showed that the method can maintain high segmentation accuracy and processing speed in processing three different body types of sheep carcass point cloud samples. The segmentation effect and index results, however, showed obvious advantages: the sheep carcass point cloud can be accurately segmented into hexads, and the segmentation boundary between the splits was flat and clear. It can be used as the basis for the follow-up robots segmentation;the four indexes to evaluate the segmentation accuracy were higher than that of the region grow by 27.1%, 11.5%, 19.2% and 8.9%, and higher than that of the Euclidean clustering by 10.8%, 21.7%, 16.3% and 16.6%, respectively. Research results showed that the method had high segmentation accuracy, good real-time performance and certain applicability, and the comprehensive segmentation showed good ability.