Research On Non-contact Yak Body Size Measurement And Body Weight Estimation Method

In order to solve the problems of high cost,inconvenience and possible harm to yaks and people caused by the traditional method of measuring yak body size and weight during breeding.This paper proposes a non-contact measurement by introducing semantic segmentation and binocular vision.In the process of research,this paper explores the corresponding solutions aiming at specific problems in practice by analyzing specific implementation of semantic segmentation technology and binocular vision technology,collecting 3410 yak images on 200 yaks.This paper studies noncontact measurement of body size and body weight of yak.This paper has the following contributions:(1)Yak image foreground extraction based on U-net.Traditional image processing methods are unable to fulfill foreground extraction task in an image with complex background.Deep learning methods can solve this problem effectively,and separate the yak from the image accurately.This paper obtain a yak foreground extraction model based on U-net by yak image collecting,annotation with labelme,and training with U-net.Experiments show that the accuracy is above 97%.(2)Research on labeling algorithm of yak body size measurement points.Measurement points labeling is the key to get accurate yak body size.To get a accurate yak body size from yak foreground needs an accurate measurement points labeling of body height,body oblique length and chest depth.This paper designs and realize the algorithm of measuring point labeling and identification,yak contour extraction and yak center point calculation.The algorithm can accurately and efficiently find yak body size measurement points and fulfill contour extraction.Experiments show that the algorithm has high stability and accuracy.(3)Measurement of yak body size based on binocular vision.We design matching algorithm of yak measurement point on binocular image with a combination of yak foreground,body measurement points and yak binocular image.We calculate the yak body height,body oblique length and chest depth by calibrating binocular camera to get external parameter information of the left and right cameras,and getting the depth image of the yak image.Experiments show that the average error of yak body height is around 2.59 cm,oblique length is around 2.66 cm,and chest depth is around 5.94 cm.The results prove that our method based on binocular vision can effectively reduce the error and accurately calculate the yak body size.(4)Research on yak body size estimation.There is a high correlation between yak body weight and body size.We get four yak body size estimation models with linear regression and support vector machine regression.The average errors of the four models are 10.78%,13.01%,10.26% and 12.44% separately,proving that this method has certain advantages in estimating yak body weight.According to the above research and experiments,our non-contact measurement method of yak body size and weight can effectively save manpower and material resources in actual breeding of yak.Further experiments prove that our method can be very useful in practice.To achieve intelligent yak breeding to provide certain technical support.