Accurate identification of lamb rib area is an important content in the research of sheep carcass intelligent cutting equipment. Aiming at the problem that the color, texture and other characteristics of the carcass of the split half sheep are not obvious, and it is difficult to achieve accurate segmentation of the rib area, this paper takes the sheep rib as the research object, and proposes a sheep rib based on U-shaped convolution neural network Row image segmentation algorithm. First, collect sample images of lamb ribs, used image augmentation technology to expand the image data, and after normalization, established a lamb rib image data set. Then, the U-Net sheep segmentation image segmentation model was established, and the rib features were extracted by convolution and pooling operations, and the deep and shallow features of the rib were merged. After multiple deconvolution operations, accurate positioning of the fused features was achieved. Obtained the binary image of the rib area, so as to achieve end-to-end semantic segmentation of the image. Finally, three image semantic segmentation evaluation criteria, including accuracy (PA), average pixel accuracy (MPA), and average cross merge ratio (MIoU), were introduced to judge the segmentation performance of the network. The experimental results showed that the U-Net segmentation rib images PA, MPA, and MIoU were 92.38%, 88.52%, and 84.26%, respectively. Comparing the existing three classical image semantic segmentation methods SegNet, FCN8s, FCN32s, the U-Net average merge ratio (MIoU) were 6.47%, 15.34%, 25.86% higher than the above three methods, respectively. The image time was 48 ms shorter than the sub-optimal SegNet. For the half-sheep carcass image dataset, the MIoU of U-Net was 75.57%.