Abstract:
Abstract: Monitoring the growth performance is imperative to profitable sheep production. Knowledge of daily growth rates provides producers with information that can assist them in making effective management decisions. With the development of intensive sheep farming, small adjustments in production can have a large effect on overall performance and profits in growing-finishing units. The parameters of body size and body weight reflect the animal’s growth development, production performance and genetic characteristics. By using the records of the body size parameters, producers can monitor and estimate the feeding programs, herd health status, individual sheep growth characteristics, breeding, and so on. So, monitoring body size and body weight in real time is necessary. However, the present way of determining these parameters is normally by men, e.g. the sheep has to stand on a flat place with correct posture during measuring the body size with measuring stick, tape measure, and so on, and the sheep has to be tied up or hung up on scales when weighting, which has the shortcoming in causing the stress reaction of the sheep. In this work, a non-contact system with 3 high-resolution cameras was developed for automatically obtaining both the body dimension landmarks in 3 views and the body weight (BW). A software, developed in MATLAB environment, has been used to process the images and to obtain the points position in the image and the distances between the points. The measured body parameters included withers height (WH), rump height (RH), body length (BL), chest depth (CD), chest width (CW), and rump width (RW). A left camera and a right camera were used to restrain errors of WH, RH, BL and CD, and the average was performed to avoid precision reduction caused by the object deviating from the camera optical axis when using a single side camera. sixty small-tailed Han sheep (adult, females, not pregnant) with different ages (from 12 to 36 months, mean 65.48±8.58 kg) were weighed and recorded with 0.1 kg precision scale in the morning before their release for feeding in order to minimize the post-prandial variation. The measurement results in farm showed that the complementary parameters of left and right views could improve the accuracy of the measurement system, and the average of several measurements could reduce the deviation from the actual value obtained by single measurement of the multi postures. The maximum relative errors of WH, RH, BL, CD, CW and RW were 4.73%, 2.55%, 2.50%, 3.95%, 3.80% and 2.90%, respectively. In order to prove the usefulness of the monitored parameters, the body sizes of each animal were used to predict the weight by a few methods, including single variable linear regression, single variable nonlinear regression, stepwise multiple linear regression (stepwise-MLR), partial least squares regression (PLSR), radial basis function network (RBF), and support vector machine (SMV). Results showed that, the body size got by image processing and liveweight had a higher correlation. In the process of single variable analysis, only the BL was reserved to the prediction model, for it was more significant to liveweight. It was found that by using the SVM method, the standard deviation and average error in model validation were the minimum, which reached 3.82 kg and 4.32% respectively. So the parameters got by image processing can be used for monitoring the growth of sheep. Through the in-situ test, it proved that the real-time monitoring method of sheep’s growth eases the livestock measuring workload greatly and overcomes the limitations of manual measurement, and it’s worth popularizing and making more efforts to improve the precision management and welfare farming of sheep.