Identification of body size characteristic points based on the Mask R-CNN and correlation with body weight in Ujumqin sheep.

Abstract

The measurements of body size data not only reflect the physical fitness, carcass structure, excellent growth condition, and developmental relationship among tissues and organs of animals but are also critical indicators to measure the growth and development of sheep. Computer vision-based body size identification is a non-contact and stress-free method. In this study, we analyzed different body size traits (height at wither, body slanting length, chest depth, chest circumference, shank circumference, hip height, shoulder width, and rump width) and the body weight of 332 Ujumqin sheep and significant correlations (P < 0.05) were obtained among all traits in Ujumqin sheep. Except for shoulder width, rump width, and shank circumference, all were positively correlated, and the effect of sex on Ujumqin sheep was highly significant. The main body size indexes affecting the body weight of rams and ewes were obtained through stepwise regression analysis of body size on body weight, in order of chest circumference, body slanting length, rump width, hip height, height at wither, and shoulder width for rams and body slanting length, chest circumference, rump width, hip height, height at wither and shoulder width for ewes. The body slanting length, chest circumference, and hip height of ewes were used to construct prediction equations for the body weight of Ujumqin sheep of different sexes. The model's prediction accuracy was 83.9% for the rams and 79.4% for ewes. Combined with a Mask R-CNN and machine vision methods, recognition models of important body size parameters of Ujumqin sheep were constructed. The prediction errors of body slanting length, height at wither, hip height, and chest circumference were ~5%, chest depth error was 9.63%, and shoulder width, rump width, and shank circumference errors were 14.95, 12.05, and 19.71%, respectively. The results show that the proposed method is effective and has great potential in precision management.