生猪肉产量预测的非接触实时在线机器视觉系统  

Machine vision system of nondestructive real-time prediction of live-pig meat yield

作　　者：田芳[1] 彭彦昆[1] 

机构地区：[1]中国农业大学工学院,国家农产品加工技术装备研发分中心,北京100083

出　　处：《农业工程学报》2016年第2期230-235,共6页Transactions of the Chinese Society of Agricultural Engineering

基　　金：国家科技支撑计划项目(2012BAH04B00)

摘　　要：为了准确估测生猪肉产量,该文提出了一种自动提取生猪活体特征的方法,建立了生猪肉产量预测算法。首先,基于自主研发的生猪肉产量预测的机器视觉系统装置实时采集54头活体大白猪的俯视图和侧视图,然后根据图像处理技术对所得的图像进行分析,提取生猪的体长、胸宽、臀宽、体高等体尺参数值;同时通过称量装置获得每头猪的活体质量。其次,将54头生猪样品按2∶1的比例分为校正集和验证集,基于体长、胸宽、臀宽、体高和体质量5个特征参数分别利用多元线性回归法(multiple linear regression,MLR)和偏最小二乘回归法(partial least-squares regression,PLSR)建立不同的肉产量预测模型,通过比较各预测模型的估测精度、相关系数等,确定最佳预测模型。分析结果表明,5个参数中体质量对肉产量预测的权重最大;偏最小二乘回归方法的预测结果较好,最佳预测模型的预测相关系数为0.95,预测误差为3.09 kg。试验结果证实,在检测系统中使用基于生猪的体尺参数和体质量建立的偏最小二乘回归预测模型可快速准确预测生猪肉产量,适于实际生产应用。The live pig meat yield is an important element to pig-breeding, purchase and slaughter industry. It＇s beneficial to get hold of the pig＇s growing status to predict the live pig meat yield during pig-breeding and transaction. For the traditional detection method of meat yield, the product of pig weight and meat production rate is considered as the meat yield. It depends on the experience of the stockman and is unreliable. The live pig meat yield prediction based on machine vision system is a nondestructive, real-time and precise method. In this study, a method of automatic extraction of pig body parameters was proposed, and the prediction model of pig meat yield was built. First of all, the top-view images and side-view images of 54 live white pigs were captured by the self-developed machine vision system. At the same time, the pig weight was measured with the electric animal balance. After the data collecting was finished, the pig body parameters, namely the body length, chest width, hip width and hip height, were measured using the tape by worker. Then the image processing algorithm was developed on the Microsoft VC＋＋ 2010 platform. The open source image processing library named Open CV was used to assist the processing. In the top-view binary image, the central axis of the pig back was extracted. Then the body length, chest width and hip width were calculated based on the axis. In addition, the hip height was extracted in side-view binary image after the back curve was fitted. After the image processing and weight collecting of each pig, the 54 sets of data were divided into the calibration set and validation set for modeling analysis at the ratio of 2：1. Multiple linear regression（MLR） and partial least squares regression（PLSR） were used to establish the estimation models of live pig meat yield. Results showed that, the correlation coefficient of the pig body parameters got by image processing and manual measuring all could reach 0.96. So the parameters got by image processing could be used

关 键 词：机器视觉 模型 特征提取 生猪 体质量 体尺参数 肉产量预测 

分 类 号：S818.9[农业科学—畜牧学] TP391[农业科学—畜牧兽医]