摘要
[目的]嫩度是肉品质量的首要指标,其影响牛肉的消费和商业价值;寻找合适的嫩度指标,快速、无损、客观地预测牛肉嫩度一直是肉品学研究的热点之一。[方法]本文基于机器视觉技术和图像处理方法,分割牛肉图像的肌间结缔组织区域,提取肌间结缔组织的特征参数,运用统计学方法关联该特征参数和熟肉剪切力值,结合经过专门训练的评级小组的分级,采用Stepwise多元线性回归(Stepwise-MLR)建模,对牛肉嫩度进行预测和分级。[结果]70个样本图像的结缔组织特征数据全部用于训练模型,采用留一法交叉验证(Leave-one-out cross validation)测试模型,验证模型的牛肉嫩度判别系数(R^2)为0.857,剩余标准误差(residual standard error,RSEC)为6.453;将牛肉分为嫩、中等、老3个等级,全部预测集的总体等级预测正确率为88.57%。[结论]肌间结缔组织特征是预测牛肉嫩度的重要指标,本文所用的软硬件方法对牛肉嫩度的快速、无损、客观预测和分级具有一定的实用价值及指导意义。
[Objectives]Tenderness is the primary indicator of the meat quality. It influences the consumption and commercial value of the beef. Looking for suitable indicators of tenderness and predicting the tenderness with a rapid,non-destructive,and objective method has always been one of research focuses. [Methods]In this paper,the area of connective tissue between the muscles was segmented based on computer vision technology and image processing methods to extract features. Then statistical methods were used to find the relationship between characteristic parameters and cooked-beef shear force value. And combining with rating by a trained panel,the beef tenderness model was established by Stepwise-multiple linear regressions to predict cooked-beef tenderness and grading. [Results]The connective tissue feature data for 70 sample images were used to train and test sample tenderness model in a rotational leave-one-out scheme. Beef tenderness discrimination coefficient of the model R^2 was 0.857,and RSEC was 6.453. Through cross validation,the beef was classified into tender,medium and tough groups with 88. 57% classification accuracy. [Conclusions]Experimental results showed that image features of connective tissue between muscles were important indicators of beef tenderness.The hardware and software which was able to predict beef tenderness levels quickly and non-destructively had good practical value and guiding significance.
出处
《南京农业大学学报》
CAS
CSCD
北大核心
2016年第5期865-871,共7页
Journal of Nanjing Agricultural University
基金
国家现代农业(肉牛)产业技术体系项目(nycytx-38)
农业科技成果转化资金项目(SQ2011ECC100043)
国家自然科学基金项目(61503187)
关键词
牛肉嫩度
结缔组织
机器视觉
图像处理
Stepwise多元线性回归
留一法交叉验证
beef tenderness
connective tissue
computer vision
image processing
Stepwise-multiple linear regressions（Stepwise-MLR）
Leave-one-out cross validation
