Rapid and accurate prediction of milk yield plays an important role in the breeding of dairy goats. We improved the Mask R-CNN deep learning model based on Feature Channel Attention, anchor refinement module, and so on, making it more applicable for predicting milk yield in dairy goats. The accuracy, recall, and mIoU of the improved model for udder segmentation of dairy goats reached 92.21% ± 0.02%, 85.39% ± 0.02%, and 76.28% ± 0.01%, respectively. The predicted mean absolute error, mean squared error, and mean absolute percentage error for the milk yield in the test set were 0.149 ± 0.009, 0.042 ± 0.018, and 9.62 ± 0.014, respectively. We further validated that the udder contour features of dairy goats can serve as a basis for predicting milk yield. The method proposed in this study for predicting milk yield in dairy goats based on udder images is feasible and plays an important role in breed selection.
Ma et al. (Sun,) studied this question.