LAUSR

Frequent measurements of body weight (BW) in livestock production systems are very important because they allow the assessment of growth development of animals. However, monitoring animal growth through traditional weighing scales is laborious and stressful for animals. Thus, the objectives of this study were to: 1) assess the predictive quality of an automated computer vision system used to predict BW and average daily gain (ADG) in beef cattle; and 2) compare different predictive approaches (Multiple Linear Regression: MLR, Least Absolute Shrinkage and Selection Operator: LASSO, Partial Least Squares: PLS, and Artificial Neutral Networks: ANN). A total of 234 images of Nellore beef cattle were collected during weaning, stocker and feedlot phase. Biometric body measurements from each animal were performed using 3D images captured with the Kinect® sensor, together with their respective BW acquired using an electronic scale. The biometric measurements were used as explanatory variables for each predictive model. Prediction quality was assessed using a leave-one-out cross-validation strategy. The ANN approach resulted on higher precision and accuracy for BW prediction compared to the other methods, with Root Mean Square Error of Prediction (RMSEP) and squared predictive correlation (r2) equal to: RMSEP = 8.6 kg and r2= 0.91 for weaning; RMSEP = 11.4 kg and r2= 0.79 for stocker, and RMSEP = 7.7 kg and r2= 0.92 for beginning of feedlot. The ANN was also superior for prediction of ADG for the weaning to stocker, weaning to beginning of feedlot, weaning to end of feedlot, stocker to beginning of feedlot and beginning to end of feedlot, with RMSEP: 0.02, 0.02, 0.03, 0.10 and 0.09 kg/d, and r2: 0.67, 0.85, 0.80, 0.51 and 0.82, respectively. Overall, results indicate that an automated computer vision system is a potential tool for real-time measurement of BW and ADG in beef cattle.