Sow body condition is an important indicator for evaluating its production performance and guiding accurate feeding. In recent years, machine vision-based body condition scoring has received much attention. However, since most of the current studies rely on two-dimensional images and human-defined geometric features, the two-dimensional images are difficult to reflect the rich three-dimensional body features of the pig body, so the dimension of feature extraction is limited. At the same time, the limited number and dimension of human-defined features make it difficult to include detailed features related to the body condition, resulting in a large amount of information loss leading to a low scoring accuracy. A 3D point cloud dataset was constructed by deep vision technology, and a class-balanced loss function-representative surfaces (CBLoss-RepSurf) sow body condition scoring model was proposed based on the class-balanced loss function, which automatically extracted high-dimensional features related to body condition scoring, and realized end-to-end sow body condition scoring. The end-to-end sow body condition scoring was realized by automatically extracting high-dimensional features related to body condition scoring. Tested with 3,093 point clouds of 57 sows at different stages, the results showed that the accuracy of the CBLoss-RepSurf model was improved by 1.54 percentage points compared with the RepSurf model without class-balanced loss, and the scoring accuracies of sows at empty, late gestation, and weaning stages were 94.87%, 92.50%, and 85.50%, respectively, which was comparable to that of the sows scored based on the features of body mass and body size. Compared with the body mass and body size characteristics of the sow body condition scoring method, the accuracy was 6.30 percentage points, 7.00 percentage points and 0.36 percentage points higher, which can achieve more accurate scoring for different stages of sow body condition and provide feasible method support for the automatic scoring of sow body condition in large-scale aquaculture.