摘要
针对规模化羊场对山羊体尺无接触式自动测量的需求,设计了一种山羊双视角图像采集装置并开发了配套的山羊体尺自动测量算法。首先,开发了山羊双视角图像自动采集装置并在养殖场完成山羊双视角图像数据集的构建;然后,采用背景减除法二值化羊体俯视图,引入简单线性迭代聚类算法(SLIC)构建侧视图超像素的纹理和颜色特征向量,训练基于支持向量机(SVM)的超像素分类器,综合利用置信度和超像素区域邻接图(RAG)获取侧视图中的羊体二值图;最后,提出了在侧视和俯视二值图像中定位关键体尺特征点的方法,自动提取山羊体高、体斜长、胸深、胸宽、管径参数,拟合得到胸围和管围参数。算法测试结果表明,羊体侧视图前景区域超像素分类正确率超过94%,算法自动提取与人工标注的侧视、俯视前景二值图的交并比分别为96.1%和97.5%。以人工使用软尺测量获得体尺参数为金标准评价算法自动提取体尺参数的精度,结果表明管围、体高、胸深、胸宽、胸围和体斜长的平均相对误差分别为5.5%、3.7%、2.6%、5.2%、4.1%和3.9%。本文开发的羊体双视角图像采集装置及相应的图像处理方法可以满足山羊体尺无接触自动测量的精度要求,为山羊体尺的高效、自动测量提供了可行的解决方案。
The body size of farm animal is an important criterion for selection and appraising of variety resources,and body size is also employed to estimate animal body weight in many existing studies.In order to meet the requirement of measuring goat body size in a non-contact manner,a double view goat body images acquisition system was designed and the corresponding algorithm for goat body size calculating was developed.The double view images acquisition system was set up by using PLC controller,photoelectric sensors,limit switches,reducer motors,and industrial cameras.Goat's top view binary image was obtained by using background subtraction.The simple linear iterative clustering algorithm(SLIC) was introduced to construct the texture and color feature vectors of the superpixel of the side view goat body image.Each superpixel of a side view image was classified as foreground and background part by using a support vector machine(SVM) based classifier.The confidence and the region adjacent graph(RAG) of each superpixel were comprehensively used to binarize the side view goat body image.A set of methods for the body size feature points locating in the side and top view binary images of goat body was proposed.These feature points were further employed to calculate the goat body size parameters such as body height,body slant length,chest depth,chest width,and pipe diameter.Then,chest girth was fitted by using chest depth and chest width,circumference of cannon bone was fitted by using pipe diameter.The developed method was tested by using the double view images of Haimen goat collected in a breeding farm.The test results indicated that SLIC,SVM and RAG can segment the goat side view foreground with an accuracy of 94%.The intersection of union(IoU) of the side and top view foreground obtained by the algorithm and the manually labeled one were respectively 96.1% and 97.5%.The goat body size parameters calculated by using the developed algorithm were compared with those measured manually.The comparison results indicated that the average errors of the circumference of cannon bone,body height,chest depth,chest width,chest girth and body slant length were 5.5%,3.7%,2.6%,5.2%,4.1% and 3.9%,respectively.Therefore,the image acquisition device developed can obtain the goat body double view images efficiently,and the proposed methods for goat image foreground segmentation and the feature points locating.Comprehensive utilization of the image acquisition device and the corresponding image processing algorithm developed provided a solution to the problem of measuring goat body size in an automatic manner.
作者
陆明洲
光二颖
陈子康
王锋
张生福
熊迎军
LU Mingzhou;GUANG Erying;CHEN Zikang;WANG Feng;ZHANG Shengfu;XIONG Yingjun(College of Artificial Intelligence,Nanjing Agricultural University,Nanjing 210031,China;Key Laboratory of Livestock Farming Equipment,Ministry of Agriculture and Rural Affairs,Nanjing 210031,China;College of Animal Science and Technology,Nanjing Agricultural University,Nanjing 210095,China;Computer College,Qinghai Nationalities University,Xining 810007,China)
出处
《农业机械学报》
EI
CAS
CSCD
北大核心
2023年第8期286-295,共10页
Transactions of the Chinese Society for Agricultural Machinery
基金
青海省科技厅基础研究计划项目(2020-ZJ-716)
中央高校基本科研业务费专项资金项目(ZJ22195003)
国家自然科学基金项目(31972615)
江苏省自然科学基金项目(BK20191315)。
关键词
山羊
体尺测量
双视角图像
图像分割
超像素
Haimen goat
body size measurement
double view images
image segmentation
superpixel
