基于Mask R-CNN的牛体尺测量系统  

A bull body ruler measurement system based on Mask R-CNN

作　　者：赵建敏[1] 文博 李琦[1] ZHAO Jianmin;WEN Bo;LI Qi(School of Information Engineering,Inner Mongolia University of Science and Technology,Baotou 014010,China)

机构地区：[1]内蒙古科技大学信息工程学院,内蒙古包头014010

出　　处：《畜牧与兽医》2021年第5期42-48,共7页Animal Husbandry & Veterinary Medicine

基　　金：内蒙古自然科学基金(2019LH06006,2019MS06021);内蒙古自治区科技重大专项(2019ZD025)。

摘　　要：牛的体尺数据是衡量牛生长发育及科学育种的重要依据。为解决人工测量牛体尺时工作量大、牛应激反应剧烈等问题,提出一种基于Mask R-CNN的图像分割的牛体尺测量的方法。通过摄像头采集牛的图片,利用Mask R-CNN算法进行图像分割,提取牛体轮廓曲线并对曲线进行平滑处理,对于轮廓曲线采用分区法提取特征区域,在特征区域内利用U弦长曲率法计算曲率最大点,即为体尺测点,进而计算牛体尺数据信息。基于Ubuntu系统、Pytorch深度学习框架设计了牛体尺测量系统,在实验室验证的基础上对牧场5头牛进行现场试验,经验证,对牛体体高的实测值平均相对误差较小,其平均相对误差为4.94%;其次为体长,平均相对误差为6.84%;而对牛体体斜长检测误差较大,平均相对误差为8.36%。相比较于canny算子等传统方法提取牛体轮廓,Mask R-CNN提取目标物体轮廓更适用于复杂背景。本研究可应用于牛体无应激测量计算体尺,有利于牛精细化养殖以及建立生长电子档案。The body size data of cattle is an important basis for measuring cattle growth and development and for scientific breeding. In order to solve the problems of heavy workload of measuring cattle body size manually and severe stress response of cattle in measuring, a method for measuring cattle body size based on Mask R-CNN image segmentation was proposed. In this study, cattle pictures were collected using a camera, the Mask R-CNN algorithm was used to perform image segmentation, the contour curve of cattle body extracted and the curve was smoothed. For contour curves, the feature area was extracted using the partition method, and the U-string curvature method was used in the feature area. The point of maximum curvature was calculated, which was determined to be the body measurement point. Then, the data of the cattle body measurement was calculated. Finally, based on the Ubuntu system and the Pytorch deep learning framework, a cattle body measurement system was designed;and based on laboratory verification, the system was further verified by field experiment with five cattle in the pasture. The results showed that the average relative error of the measured body height of the cattle was small at 4.94%;and the average relative error of the measured body length was mot big, either, at 6.84%;but the detection error of the oblique length of the cattle body was a little bigger, at 8.36%. Compared with traditional methods such as suing a canny operator to extract the contours of cattle, Mask R-CNN was more effectively useful in complex backgrounds of cattle body size measurement. The Mask R-CNN-based cattle body size measurement proposed in this study might be applied to calculation of the body size of cattle without stress, which was significant for refined breeding of cattle and for creating electronic documents of their growth.

关 键 词：Mask R-CNN 图像分割 轮廓提取 体尺测量 应激反应 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] S818[自动化与计算机技术—计算机科学与技术]