基于计算机视觉技术育肥猪体重分析研究  

Study on Weight of Fattening Pigs Based on Computer Vision Technology

作　　者：张凯[1] 王春光[1] 刘涛[1] 康飞龙[1] 

机构地区：[1]内蒙古农业大学机电工程学院,呼和浩特010018

出　　处：《农机化研究》2017年第5期32-36,共5页Journal of Agricultural Mechanization Research

基　　金："十二五"国家科技支撑计划项目(2014BAD08B05)

摘　　要：为了更好地解决育肥猪的体重预估问题,本研究通过获取育肥猪在不同生长阶段的图像和质量数据,利用计算机视觉技术将猪的侧视图像进行预处理、颜色特征处理、阈值分割及图像形态学处理,经过推导计算求出猪体的侧视面积,对一维体尺参数、侧视面积与体重进行数据拟合并建立数学模型。研究结果表明:在只考虑体尺单因素的影响时,拟合出的体重与体尺的相关性较小,其平均误差也较大。通过比较逐步回归法与MLP神经网络模型发现:MLP神经网络拟合模型相关性最好,相关性R2可达到0.993,平均相对误差为1.38%,可以很好地保证估测精度,为测量猪的体重提供新的方法。In order to solve the problem of estimating the weight of the fattening pigs, the image and quality data of pigs was obtained at different growth stages. The research is based on the computer vision technology, the side image of pig will be treated with a series of measure steps which include the processing of image preprocessing, color characters, threshold segmentation and the processing of image pattern, measuring the side area of pigs after computation. Through the analysis of one dimension body size, side area and weight, it can apply the data fitting and establish mathematical model. The results show that the single factor of the body size is considered only, The correlation is lower between body weight and body size, and its average error is larger. Through the comparison of the stepwise regression method and the MLP neural network model, we find that the correlation of the MLP neural network model is the best. The correlation coefficient is 0. 993, and the average relative error is 1.38%. It can assure the precision of evaluations very well, and provide new method for measuring the weight of pigs.

关 键 词：猪体 图像处理 体重 侧视面积 数据分析 

分 类 号：S821.44[农业科学—畜牧学]