基于计算机视觉和神经网络的牛肉大理石花纹自动分级技术的研究
|
摘要
牛肉大理石花纹是牛肉质量最重要的决定因素，但大理石花纹等级是由分级员感观评定决定的，这种方法存在主观随意性，并且耗时耗力。相比之下，利用计算机视觉技术来分级是一种客观、稳定、高效和花费少的途径。本文首次在国内利用计算机视觉和神经网络技术进行牛肉大理石花纹分级的研究，所做的工作和取得的成果如下：
(1)从定量的角度明确了大理石花纹面积是大理石花纹“丰富程度”最重要的量化指标。
(2)首次在复杂背景下，应用“类差影法”结合“阈值法”有效地从背景中分离出眼肌区域，为基于计算机视觉技术的大理石花纹自动分级系统奠定实际应用的基础。
(3)探讨了在应用计算机视觉进行牛肉大理石花纹分级之前，对图像进行去噪、增强等预处理的方法，结果，中值滤波和灰度线性变换是大理石花纹图像预处理有效的方法。
(4)“区域增长”法是分离有效眼肌区域和提取图像特征参数的有效方法。
(5)沿着从大理石花纹标准图像和样本图像的特征参数两条研究路线分别建立牛肉大理石花纹等级预测模型，并比较了两条研究路线的结果。
(6)分别应用统计学方法(回归分析、判别分析)和神经网络技术(BP、LVQ、SVM)建立了牛肉大理石花纹等级预测模型，结果表明，从标准图像建立的最好模型为：Y=4.0875-0.5738*X+0.0548*X^2-0.0019*X^3(R2=0.9839)(Y-等级，X-大理石花纹面积)，对111个样本的预测准确率为72.79％，表明从标准图像建立的模型预测准确性达不到实际需要；从样本图像建立的回归模型Y=4.2858-0.4503*X+0.0261*X^2(R2=0.8302)和BP神经网路回归模型，对40个样本的预测准确率均达到了85.00％，而应用LVQ、BP、SVM神经网络分类器，预测精度分别为90.00％、87.50％和92.50％，SVM分类器是进行大理石花纹分级的有效工具。
(7)通过本文的研究结果并比较中美两国的牛肉大理石花纹等级，发现我国牛肉大理石花纹等级标准存在的一些问题并提出了修改建议。
本研究对于我国开展基于计算机视觉技术的牛肉大理石花纹自动分级系统高新
技术产品的开发，具有重要的参考价值，对推广和应用我国牛肉等级标准，提高我国
肉类质量分级水平具有重要的经济意义．
The beef marbling grade is the dominant parameter in deciding the meat quality.Traditionally, grading of beef marbling has been performed by human graders. However, the sensory inspection is subjective,time-consuming, labour-intensive and the accuracy of the tests can not be guaranteed. By contrast it has been found that computer vision inspection and grading of food products, was more consistent, efficient and cost effective.
The research of grading for beef marbling by computer vision was carried out in this paper for the first time domestically, the work that done and the accomplishment are as follows:
(1) Using quantitative method ,beef marbling area was proved to be the most important quantification index of marbling "rich degree".
(2) For the first time, the region of rib-eye was separated from the complex background using the similar algorithm of "subtracted image " and the algorithm of "threshold segmentation" .The foundation of actual application for the automatic grading system for beef marbling based on computer vision was established.
(3) The image pre-processing(noise-suppressing, image enhancement) methods were discussed ,as a result, the median filters and gray-level linear conversion were effective methods for image pretreatment.
(4) The image segmentation algorithm of "region growing" was a power tool to separate effective region of rib-eye from whole region of rib-eye and to extract the image feature values.
(5) Mathematics models were built to predict beef marbling grade from standard image and sample image respectively and this two research routes were comparied.
(6) Mathematics models were developed to predict beef marbling grade using statistics method(regression analysis,discriminant analysis) and neural networks technology(BP,LVQ,SVM) respectively.As a result, from the standard image,the best model is: Y=4.0875-0.5738 X+0.0548 X 2-0.0019 X 3 (R2=0.9839) (Y-grade of beef marbling,X-aera of beef marbling), the forecast accuracy rate for 111 samples was so low (72.79%) that this model did not satisfy the actual needs; from sample image,the model Y=4.2858-0.4503 X+ 0.0261 X 2 (R2=0.8302) and BP neural network regression model had reached 85.00% for the forecast accuracy rate of 40 samples. Applying LVQ, BP and SVM neural network classfication, the forecast precision was 90.00%, 87.50% and 92.50% respectively, SVM was an effective tool for predicting beef marbling grade.
(7) Some problems existing in the beef marbling standard were found and some suggestions were made based on the results from this paper and the comparison of the beef marbling standard in China and U.S.A.
This research has important reference value for the developing new and high technical products based on computer vision and neural networks to grade for beef marbling of our country.It has important economic meaning for the extending and application beef grade standard, raising the grading technology of meat quality in China.
(1)从定量的角度明确了大理石花纹面积是大理石花纹“丰富程度”最重要的量化指标。
(2)首次在复杂背景下，应用“类差影法”结合“阈值法”有效地从背景中分离出眼肌区域，为基于计算机视觉技术的大理石花纹自动分级系统奠定实际应用的基础。
(3)探讨了在应用计算机视觉进行牛肉大理石花纹分级之前，对图像进行去噪、增强等预处理的方法，结果，中值滤波和灰度线性变换是大理石花纹图像预处理有效的方法。
(4)“区域增长”法是分离有效眼肌区域和提取图像特征参数的有效方法。
(5)沿着从大理石花纹标准图像和样本图像的特征参数两条研究路线分别建立牛肉大理石花纹等级预测模型，并比较了两条研究路线的结果。
(6)分别应用统计学方法(回归分析、判别分析)和神经网络技术(BP、LVQ、SVM)建立了牛肉大理石花纹等级预测模型，结果表明，从标准图像建立的最好模型为：Y=4.0875-0.5738*X+0.0548*X^2-0.0019*X^3(R2=0.9839)(Y-等级，X-大理石花纹面积)，对111个样本的预测准确率为72.79％，表明从标准图像建立的模型预测准确性达不到实际需要；从样本图像建立的回归模型Y=4.2858-0.4503*X+0.0261*X^2(R2=0.8302)和BP神经网路回归模型，对40个样本的预测准确率均达到了85.00％，而应用LVQ、BP、SVM神经网络分类器，预测精度分别为90.00％、87.50％和92.50％，SVM分类器是进行大理石花纹分级的有效工具。
(7)通过本文的研究结果并比较中美两国的牛肉大理石花纹等级，发现我国牛肉大理石花纹等级标准存在的一些问题并提出了修改建议。
本研究对于我国开展基于计算机视觉技术的牛肉大理石花纹自动分级系统高新
技术产品的开发，具有重要的参考价值，对推广和应用我国牛肉等级标准，提高我国
肉类质量分级水平具有重要的经济意义．
The beef marbling grade is the dominant parameter in deciding the meat quality.Traditionally, grading of beef marbling has been performed by human graders. However, the sensory inspection is subjective,time-consuming, labour-intensive and the accuracy of the tests can not be guaranteed. By contrast it has been found that computer vision inspection and grading of food products, was more consistent, efficient and cost effective.
The research of grading for beef marbling by computer vision was carried out in this paper for the first time domestically, the work that done and the accomplishment are as follows:
(1) Using quantitative method ,beef marbling area was proved to be the most important quantification index of marbling "rich degree".
(2) For the first time, the region of rib-eye was separated from the complex background using the similar algorithm of "subtracted image " and the algorithm of "threshold segmentation" .The foundation of actual application for the automatic grading system for beef marbling based on computer vision was established.
(3) The image pre-processing(noise-suppressing, image enhancement) methods were discussed ,as a result, the median filters and gray-level linear conversion were effective methods for image pretreatment.
(4) The image segmentation algorithm of "region growing" was a power tool to separate effective region of rib-eye from whole region of rib-eye and to extract the image feature values.
(5) Mathematics models were built to predict beef marbling grade from standard image and sample image respectively and this two research routes were comparied.
(6) Mathematics models were developed to predict beef marbling grade using statistics method(regression analysis,discriminant analysis) and neural networks technology(BP,LVQ,SVM) respectively.As a result, from the standard image,the best model is: Y=4.0875-0.5738 X+0.0548 X 2-0.0019 X 3 (R2=0.9839) (Y-grade of beef marbling,X-aera of beef marbling), the forecast accuracy rate for 111 samples was so low (72.79%) that this model did not satisfy the actual needs; from sample image,the model Y=4.2858-0.4503 X+ 0.0261 X 2 (R2=0.8302) and BP neural network regression model had reached 85.00% for the forecast accuracy rate of 40 samples. Applying LVQ, BP and SVM neural network classfication, the forecast precision was 90.00%, 87.50% and 92.50% respectively, SVM was an effective tool for predicting beef marbling grade.
(7) Some problems existing in the beef marbling standard were found and some suggestions were made based on the results from this paper and the comparison of the beef marbling standard in China and U.S.A.
This research has important reference value for the developing new and high technical products based on computer vision and neural networks to grade for beef marbling of our country.It has important economic meaning for the extending and application beef grade standard, raising the grading technology of meat quality in China.
引文
1．周光宏．美国牛肉的分级制．中国牧业通讯，1999(3)：32-33
2．中国农业年鉴．农业出版社，2001．
3．周光宏，刘丽．国外牛肉分级标准介绍兼论建立中国牛肉分级制度的意义．中国畜产与食品，1997，4(5)：1-2
4．林祥金．我国肉牛生产存在的问题与发展对策．当代畜牧，2001(5)：1-2
5．周光宏．我国肉牛业现状及对策．畜牧与兽医，2000，32(1)：16-17
6．高峰，周光宏．我国肉牛业发展现状及趋势．畜禽业，2002(4)：54-55
7．高峰，周光宏．建立和完善我国优质牛肉生产全程质量控制体系．黄牛杂志，2002，28(1)：43-45
8. McDonald, T. and Chen, Y. R. A geometric model of marbling in beef longissimus dorsi. Trans. ASAE, 1992, 35 (3) : 1057-1062
9. Gerrard, D. E., Gao, X., Tan, J. Beef marbling and colour score determination by image processing. Journal of Food Science. 1996, 61 (1), 145-148
10. Li, J., Tan, J., Martz, F.A.. Predicting beef tenderness from image texture features. In: 1997 ASAE Annual International Meeting Technical Papers, Paper No. 973124, ASAE, 2950 Niles Road, St. Joseph, MI 49085-9659, USA
11. Lu, J., Tan, J., Gerrard, D.E.. Pork quality evaluation by image processing. In: 1997 ASAE Annual International Meeting Technical Papers, Paper No. 973125, ASAE, 2950 Niles Road, St. Joseph, MI 49085-9659, USA
12 方如明．计算机图像处理与米的品质检测．农业工程学报，1992，8(3)：104～11216
13．张建平等．烟叶颜色测量与分析．江苏工学院学报，1993，14(4)：7～1317
14．刘禾．计算机视觉在水果自动分级中的应用：[博士论文]．北京：中国农业大学，199418
15．陈顺三．奶牛体型线性评定的图像处理技术研究：[博士论文]．北京：中国农业大学，199619
16．王丰元等．种子形状参数检测的计算机图像处理技术．农业机械学报，1995，26(2)：52～5720
17．李庆中．苹果自动分级中计算机视觉信息快速获取与处理技术的研究．[博士论文]．北京：中国农业大学，200005
18．王树文．计算机视觉技术在农产品自动检测与分级的研究—番茄的表面缺陷自动检测与分类．[硕士论文]．东北农业大学，200205．
19. Brandon JR, Howarth MS, Searcy SW. A neural network for carrot tip classification. ASAE Paper, 90754922
20. TaoY, Deck SH, etal. Rule-supervised automated machine vision inspection of produce. ASAE Paper, 91700323
21. Liao K, Paulsen MR, Reid JF, etal. Corn kernel breakage classification by machine vision
using a neural network classifier. Trans of the ASAE, 36(6):1949～1953
22. YangQ. Finding stalk and calyxby using structured lighting. Computers and Electronics on Agriculture, 1993(3):33～3711
23. Guedalia D, Ddan Y. A dynamic artificial neural network for coding and classification of multi-sensor quality information. ASAE Paper, 94305312
24. Ozer N, Engel BA, Simon JE. Fusion classification techniques for fruit quality. Trans of the ASAE, 1995,38(6):1927～193413
25. Elater RT, Goodrum JW.Detection of crack sin eggs using machine vision. Trans of the ASAE, 1991,34(1):307～312
2．中国农业年鉴．农业出版社，2001．
3．周光宏，刘丽．国外牛肉分级标准介绍兼论建立中国牛肉分级制度的意义．中国畜产与食品，1997，4(5)：1-2
4．林祥金．我国肉牛生产存在的问题与发展对策．当代畜牧，2001(5)：1-2
5．周光宏．我国肉牛业现状及对策．畜牧与兽医，2000，32(1)：16-17
6．高峰，周光宏．我国肉牛业发展现状及趋势．畜禽业，2002(4)：54-55
7．高峰，周光宏．建立和完善我国优质牛肉生产全程质量控制体系．黄牛杂志，2002，28(1)：43-45
8. McDonald, T. and Chen, Y. R. A geometric model of marbling in beef longissimus dorsi. Trans. ASAE, 1992, 35 (3) : 1057-1062
9. Gerrard, D. E., Gao, X., Tan, J. Beef marbling and colour score determination by image processing. Journal of Food Science. 1996, 61 (1), 145-148
10. Li, J., Tan, J., Martz, F.A.. Predicting beef tenderness from image texture features. In: 1997 ASAE Annual International Meeting Technical Papers, Paper No. 973124, ASAE, 2950 Niles Road, St. Joseph, MI 49085-9659, USA
11. Lu, J., Tan, J., Gerrard, D.E.. Pork quality evaluation by image processing. In: 1997 ASAE Annual International Meeting Technical Papers, Paper No. 973125, ASAE, 2950 Niles Road, St. Joseph, MI 49085-9659, USA
12 方如明．计算机图像处理与米的品质检测．农业工程学报，1992，8(3)：104～11216
13．张建平等．烟叶颜色测量与分析．江苏工学院学报，1993，14(4)：7～1317
14．刘禾．计算机视觉在水果自动分级中的应用：[博士论文]．北京：中国农业大学，199418
15．陈顺三．奶牛体型线性评定的图像处理技术研究：[博士论文]．北京：中国农业大学，199619
16．王丰元等．种子形状参数检测的计算机图像处理技术．农业机械学报，1995，26(2)：52～5720
17．李庆中．苹果自动分级中计算机视觉信息快速获取与处理技术的研究．[博士论文]．北京：中国农业大学，200005
18．王树文．计算机视觉技术在农产品自动检测与分级的研究—番茄的表面缺陷自动检测与分类．[硕士论文]．东北农业大学，200205．
19. Brandon JR, Howarth MS, Searcy SW. A neural network for carrot tip classification. ASAE Paper, 90754922
20. TaoY, Deck SH, etal. Rule-supervised automated machine vision inspection of produce. ASAE Paper, 91700323
21. Liao K, Paulsen MR, Reid JF, etal. Corn kernel breakage classification by machine vision
using a neural network classifier. Trans of the ASAE, 36(6):1949～1953
22. YangQ. Finding stalk and calyxby using structured lighting. Computers and Electronics on Agriculture, 1993(3):33～3711
23. Guedalia D, Ddan Y. A dynamic artificial neural network for coding and classification of multi-sensor quality information. ASAE Paper, 94305312
24. Ozer N, Engel BA, Simon JE. Fusion classification techniques for fruit quality. Trans of the ASAE, 1995,38(6):1927～193413
25. Elater RT, Goodrum JW.Detection of crack sin eggs using machine vision. Trans of the ASAE, 1991,34(1):307～312
© 2004-2018 中国地质图书馆版权所有 京ICP备05064691号 京公网安备11010802017129号 地址：北京市海淀区学院路29号 邮编：100083 电话：办公室：(+86 10)66554848；文献借阅、咨询服务、科技查新：66554700 |