Abstract: Abstract: The live pig meat yield is an important element to pig-breeding, purchase and slaughter industry. It’s beneficial to get hold of the pig’s growing status to predict the live pig meat yield during pig-breeding and transaction. For the traditional detection method of meat yield, the product of pig weight and meat production rate is considered as the meat yield. It depends on the experience of the stockman and is unreliable. The live pig meat yield prediction based on machine vision system is a nondestructive, real-time and precise method. In this study, a method of automatic extraction of pig body parameters was proposed, and the prediction model of pig meat yield was built. First of all, the top-view images and side-view images of 54 live white pigs were captured by the self-developed machine vision system. At the same time, the pig weight was measured with the electric animal balance. After the data collecting was finished, the pig body parameters, namely the body length, chest width, hip width and hip height, were measured using the tape by worker. Then the image processing algorithm was developed on the Microsoft VC++ 2010 platform. The open source image processing library named OpenCV was used to assist the processing. In the top-view binary image, the central axis of the pig back was extracted. Then the body length, chest width and hip width were calculated based on the axis. In addition, the hip height was extracted in side-view binary image after the back curve was fitted. After the image processing and weight collecting of each pig, the 54 sets of data were divided into the calibration set and validation set for modeling analysis at the ratio of 2:1. Multiple linear regression (MLR) and partial least squares regression (PLSR) were used to establish the estimation models of live pig meat yield. Results showed that, the correlation coefficient of the pig body parameters got by image processing and manual measuring all could reach 0.96. So the parameters got by image processing could be used to modeling. The correlation between the 5 parameters, namely the body length, chest width, hip width, hip height and weight, was not significant. Pig weight had a higher correlation with meat yield and the correlation coefficient was 0.92. In the process of Stepwise-MLR analysis, only pig weight and chest width were reserved to the prediction model (P<0.05), for they were more significant to meat yield. The correlation coefficients in model prediction and model validation were 0.94 and 0.88 respectively. The method of Enter-MLR and PLSR got higher correlation coefficients, which were all above 0.95. Because PLSR had the smaller standard deviation in model validation which was 3.09 kg, and its average relative error was the minimum which was 3.21%, it was confirmed to be the best model to predict the live pig meat yield. This study showed that, the PLSR model built based on pig’s weight and body parameters could predict live pig meat yield effectively during the pig-breeding. Through the field experiment, we know that the system of live pig meat yield prediction is labor-saving and resource-saving. It’s worth popularizing and more efforts have to be applied to improve the prediction precision in the future.