DyFusion-PCT: An improved point cloud transformer for automatic and non-contact morphometric measurement of sheep

With the development of smart animal husbandry, automated and non-contact animal body size measurement technology is of great significance for improving breeding efficiency and animal welfare. However, the complex geometric shape, hair coverage and dynamic posture changes of sheep pose challenges to traditional point cloud measurement methods. To this end, this paper proposes an improved Point Cloud segmentation model, DyFusion-PCT. This model introduces a multi-scale feature fusion mechanism and a dynamic context-aware module based on the Point Cloud Transformer to enhance the recognition ability and dynamic adaptability of the complex geometric structure of sheep bodies. We have built a complete automated body size measurement system, including multi-view point cloud collection, 3D reconstruction, semantic segmentation and body size extraction pipeline. The experiment was based on 176 sets of point cloud data from 30 small-tailed Han sheep. The results showed that Dyffusion PCT achieved an IoU of approximately 88.7% and a Dice coefficient of 94.9% on the test set, and the recall rate in the key measurement area reached 91.2%. In terms of core body size indicators such as body height, body length, chest depth and chest circumference, the average absolute error is significantly lower than that of traditional methods, and the relative error is controlled between 1.57% and 4.52%, approaching the measurement accuracy of manual measurement. In addition, the model still maintains stable segmentation and measurement performance under dynamic postures. This research provides an efficient, accurate and practical technical solution for the automatic measurement of sheep body rulers, which has a clear engineering application prospect.