摘要
为了解决传统牛体尺参数测量中存在测量难度大、牛应激反应强等问题,试验采用Mask RCNN算法结合牛体尺测点识别的方法,基于双目视觉原理测定牛体尺参数,并利用实验室8个黄牛模型和内蒙古乌兰察布市察哈尔右翼中旗某牧场7头西门达尔牛对建立的牛体尺参数无接触测量系统进行验证,并与人工测量值进行比较分析。结果表明:利用Mask RCNN算法对牛图像进行分割并提取牛轮廓曲线,极大的缩短了轮廓提取的时间并且更适用于复杂背景;划分牛体轮廓曲线区域,利用最大曲率等方法可以准确的识别牛体尺测点;利用双目视觉原理将牛体尺测点像素坐标转化为空间3维坐标,利用欧氏距离计算牛体尺参数。实验室系统验证结果与人工测量值比较,其体长、体高、体斜长的平均相对误差分别为1.41%、0.92%和1.37%;某牧场实地系统验证结果与人工测量值比较,其体长、体高、体斜长的平均相对误差分别为6.09%、5.78%、6.85%,均满足测量需求。说明Mask RCNN算法能够较好地在复杂背景中提取牛轮廓,牛体尺测点识别准确,能够在饲养过程中实现牛体尺参数的无接触测量。
In order to solve the problems of traditional measurement of cattle body size parameters, such as the difficulty in measurement and strong stress response of cattle, the method of identifying the measurement points of cattle body size, based on the principle of binocular vision was used in the experiment. The results showed that using Mask RCNN algorithm to segment the cattle image and extract the cattle contour curve can greatly shorten the extraction time and wasmore suitable for the complex background. By dividing the contour curve region of the bovine body, the measuring points of the bovine body measurement can be identified accurately by means of the maximum curvature.Based on the principle of binocular vision,the pixel coordinates of the measuring pointscan be transformed into 3 d coordinates, and the Euclidean distance can beused to calculate the cattle body size parameters. The results of laboratory system verification are compared with those of manual measurement, the average relative errors of length, height and slant length are 1.41%, 0.92% and 1.37%, the result of the pasture system verification are compared with those of manual measurement, the average relative errors of lenght, height and slant length are 6.09%, 5.78% and 6.85%, all meets the needs of sequencing. It indicated that the Mask RCNN algorithm can better extract the contour of the cattle in the complex background, identifythe measurement point accurately, and realize thenon-contact measurement of the cattle body sizeparameters in the breeding process.
作者
李琦
刘伟
赵建敏
LI Qi;LIU Wei;ZHAO Jianmin(College of Information Engineering,Inner Mongolia University of Science and Technology,Baotou 014010,China)
出处
《黑龙江畜牧兽医》
CAS
北大核心
2020年第12期46-50,159,160,共7页
Heilongjiang Animal Science And veterinary Medicine
基金
内蒙古自然科学基金项目“基于注意力机制的牛个体身份识别方法研究”(2019LH06006)
内蒙古自然科学基金项目“基于畜牧大数据的牲畜放牧轨迹挖掘与优化生产决策研究”(2019MS06021)
内蒙古自治区科技成果转化项目“基于物联网的现代畜牧业生产监管及产品溯源服务平台推广应用”(CGZH2018041)。