基于图像处理的大米整精米率的检测方法研究和精度分析
|
摘要
大米是我国重要粮食作物之一,整精米率是大米的重要品质因数,也是收购大米的重要指标之一。但是我国目前仍然停留在人工检测阶段,这样费时费力,同时检测结果也不够客观,不够准确。大米整精米率自动检测方法不仅可以节省时间、减少人员工作量,同时也能提高客观性和准确度。
首先,本文研究是基于数字图像处理的大米整精米率自动检测方法,利用扫描仪来获取样品米粒的图像,通过图像处理技术得到单个米粒的图像。本文通过二维Otsu阈值分割对图像进行二值化,然后根据选定区域的方法对图像进行去噪,这样可以尽量避免噪声的干扰,保持好米粒的原有信息。由于实际操作中米粒是随机摆放的,所以会有米粒粘连的情况,需要对其进行粘连分割来获得单个米粒图像。通过实验对比,采用特殊点分割算法可以得到满意的分割效果,既不会丢失米粒信息,也不会出现过分割情况。
其次,根据国家标准,整精米和碎米是根据米粒的长度来区分。鉴于稻米形状近似椭圆,本文采用最小外接矩形对米粒长度进行检测,得到的结果不但精度高,而且还可以通过顶点链码编码提高运算速度。
再次,对于整精米率的检测,国家标准中要求是整米与全部样品米的重量比值。由于基于图像处理的整精米率检测方法不能将整米和碎米分开称重,通常是采用米粒面积比来间接推算质量比。但是米粒的厚度不是完全相同,利用米粒面积比值获得质量比值会产生一定的误差。因此本文通过对米粒建立三维模型,根据米粒二维图像推测其体积,然后通过米粒的体积之比来计算整精米率,这样计算的整精米率更准确,更接近国标要求。
最后,对精度进行分析。通过多组实验,对面积计算的精度分析,对整精米和碎米面积、体积以及形态分析,对测算米粒长度的精度分析,对通过面积和体积得到的整精米率的精度分析,以及对细长米粒的整精米率的分析等。
通过图像处理的方法可以有效的测算整精米率,省时省力节省成本,可以准确检测整精米与碎米,本文算法检测的准确率可以达到98%以上。
Rice is one of the important food crops. Head rice rate is an important factor of rice quality and an important indicator of the acquisition of rice. But China still re-mains in the manual head rice rate inspection stage, so the detection is time-consuming, the test results are not objective and not accurate enough. The head rice rate automatic detection method can save time and reduce staff workload, can also improve the objec-tivity and accuracy.
Firstly, the head rice rate automatic detection method of this study is based on digital image processing. The rice image is obtained by the scanner. The single rice image is obtained using image processing technology. Image binarization is done by the two-dimensional Otsu threshold segmentation. Image denoising method based on the selected area of image is used. Therefore, it can avoid the interference of noise and keep the original rice information. Through the experimental comparison, the special point segmentation algorithm is chose to perform the separation. It can keep the rice information and avoid excessive over-segmentation situation.
Secondly, according to the national standard, the distinction between head rice and broken rice is based on rice length. Because that the rice can be approximated as ellipse, the MER method is used to calculate the length of the rice which can achieve accurate results and the computing speed can be increased by MER vertex chain code algorithm.
Thirdly, according to the national standard, the head rice rate is defined as the weight rate of the head rice and the whole sample rice. Because that the head rice rate detection method based on digital image processing can't weight the rice, the common method is to use the rate of area. Because the thickness of the rice is not completely the same, estimating head rice rate by volume of rice will be better than the area method. This thesis sets the three-dimensional model of rice, and calculates the volume accord-ing to the two-dimensional image. Then it calculates the head rice rate by the rate of volume. The method is more accurate and closer to the national standard requirements.
Finally, the precision analysis is important. It needs multiple sets of experiments. The area calculation accuracy analysis, the analysis of the whole rice and broken rice area, the rice volume and shape, the length analysis of rice, the comparison of the two method of head rice rate and the analysis of the head rice rate of the slender rice are completed.
The calculation of the head rice rate is based on image processing methods. It can detect the head rice and broken rice accurately and can save time and labor cost. The accuracy of this detection method is higher than98%.
首先,本文研究是基于数字图像处理的大米整精米率自动检测方法,利用扫描仪来获取样品米粒的图像,通过图像处理技术得到单个米粒的图像。本文通过二维Otsu阈值分割对图像进行二值化,然后根据选定区域的方法对图像进行去噪,这样可以尽量避免噪声的干扰,保持好米粒的原有信息。由于实际操作中米粒是随机摆放的,所以会有米粒粘连的情况,需要对其进行粘连分割来获得单个米粒图像。通过实验对比,采用特殊点分割算法可以得到满意的分割效果,既不会丢失米粒信息,也不会出现过分割情况。
其次,根据国家标准,整精米和碎米是根据米粒的长度来区分。鉴于稻米形状近似椭圆,本文采用最小外接矩形对米粒长度进行检测,得到的结果不但精度高,而且还可以通过顶点链码编码提高运算速度。
再次,对于整精米率的检测,国家标准中要求是整米与全部样品米的重量比值。由于基于图像处理的整精米率检测方法不能将整米和碎米分开称重,通常是采用米粒面积比来间接推算质量比。但是米粒的厚度不是完全相同,利用米粒面积比值获得质量比值会产生一定的误差。因此本文通过对米粒建立三维模型,根据米粒二维图像推测其体积,然后通过米粒的体积之比来计算整精米率,这样计算的整精米率更准确,更接近国标要求。
最后,对精度进行分析。通过多组实验,对面积计算的精度分析,对整精米和碎米面积、体积以及形态分析,对测算米粒长度的精度分析,对通过面积和体积得到的整精米率的精度分析,以及对细长米粒的整精米率的分析等。
通过图像处理的方法可以有效的测算整精米率,省时省力节省成本,可以准确检测整精米与碎米,本文算法检测的准确率可以达到98%以上。
Rice is one of the important food crops. Head rice rate is an important factor of rice quality and an important indicator of the acquisition of rice. But China still re-mains in the manual head rice rate inspection stage, so the detection is time-consuming, the test results are not objective and not accurate enough. The head rice rate automatic detection method can save time and reduce staff workload, can also improve the objec-tivity and accuracy.
Firstly, the head rice rate automatic detection method of this study is based on digital image processing. The rice image is obtained by the scanner. The single rice image is obtained using image processing technology. Image binarization is done by the two-dimensional Otsu threshold segmentation. Image denoising method based on the selected area of image is used. Therefore, it can avoid the interference of noise and keep the original rice information. Through the experimental comparison, the special point segmentation algorithm is chose to perform the separation. It can keep the rice information and avoid excessive over-segmentation situation.
Secondly, according to the national standard, the distinction between head rice and broken rice is based on rice length. Because that the rice can be approximated as ellipse, the MER method is used to calculate the length of the rice which can achieve accurate results and the computing speed can be increased by MER vertex chain code algorithm.
Thirdly, according to the national standard, the head rice rate is defined as the weight rate of the head rice and the whole sample rice. Because that the head rice rate detection method based on digital image processing can't weight the rice, the common method is to use the rate of area. Because the thickness of the rice is not completely the same, estimating head rice rate by volume of rice will be better than the area method. This thesis sets the three-dimensional model of rice, and calculates the volume accord-ing to the two-dimensional image. Then it calculates the head rice rate by the rate of volume. The method is more accurate and closer to the national standard requirements.
Finally, the precision analysis is important. It needs multiple sets of experiments. The area calculation accuracy analysis, the analysis of the whole rice and broken rice area, the rice volume and shape, the length analysis of rice, the comparison of the two method of head rice rate and the analysis of the head rice rate of the slender rice are completed.
The calculation of the head rice rate is based on image processing methods. It can detect the head rice and broken rice accurately and can save time and labor cost. The accuracy of this detection method is higher than98%.
引文
[1]程映国.国际稻米贸易特点与中国稻米出口.中国稻米.2003,1：7-9.
[2]中华人民共和国国家质量监督检验检疫总局.GB/T 21719-2008.中华人民共和国国家标准.北京：中国标准出版社,2008.
[3]郭文川,朱新华.机器视觉技术在谷物识别与分级中的研究进展.粮食与饲料工业.2002,6：50-51.
[4]Liu W, Tao Y, Siebenmorgen T J,et al. Digital image analysis method for rapid measurement of rice degree of milling. ASAE Annual International Meeting, Florida, 1998:380-385.
[5]Wan Y N, Lin C M, Chiou J F. Adaptive classification method for an automatic grain quality inspection system using machine vision and neural network. ASAE Annual International Meeting, Milwaukee,2000:10-19.
[6]Lloyd B J, Cnossen A G, Siebenmorgen T J. Evaluation of two methods for separating head rice from brokens for head rice yield determination. ASAE Annual International Meeting, St. Joseph,2000:643-648.
[7]Yadav B. K, Jindal V. K. Monitoring milling quality of rice by image analysis. Computers and Electronics in Agriculture.2001,33(1):19-33.
[8]侯彩云,Seiichi, Oshita等.三维图像处理系统在稻米品质检测中的应用研究.农业工程学报,2001,17(3)：92-95.
[9]包晓敏,汪亚明,黄振堃.计算机视觉技术在大米轮廓检测上的应用.浙江工程学院学报.2003,20(2)：104-107.
[10]侯彩云,李慧园,尚艳芬等.稻谷品质的图像识别与快速检测.中国粮油学报.2003,18(4)：80-83.
[11]刘光蓉,周红,管庶安.基于图像处理技术的大米轮廓检测.粮食与饲料工业.2004,6：14-15.
[12]凌云.基于机器视觉的谷物外观品质检测技术研究：(博士学位论文).北京：中国农业大学.2004.
[13]尚艳芬,侯彩云,常国华等.整精米自动识别方法的研究.中国水稻科学.2004,18(5)：466-468.
[14]成芳.稻种质量的机器视觉无损检测研究：(博士学位论文).杭州：浙江大学生物系统工 程与食品科学学院,2005.
[15]凌云,王一鸣,孙鸣等.基于机器视觉的大米外观品质检测装置.农业机械学报.2005,36(9)：89-92.
[16]于润伟,金鲲鹏,朱晓慧.基于图像识别的整精米自动检测研究.中国粮油学报.2006,21(6)：147-150.
[17]陈建华.基于计算机视觉的稻米外观品质检测方法的研究：(硕十学位论文).杭州.浙江理工大学,2008：
[18]中华人民共和国国家质量监督检验检疫总局.GB1350-2009.中华人民共和国国家标准.北京：中国标准出版社,2009.
[19]顾国庆,许彦冰.数字图像区域标定的方法.上海理工大学学报,2001,23(4)：295-299.
[20]Ying Y B, Wang J P, Jiang H P. Inspecting diameter and defect area of fruit with machine vision. Transactions of the CSAE.2002,18(5):216-220.
[21]高希瑞,孟超英,籍保平.机器视觉在稻米整精米率快速计算中的应用.食品科学.2007,28(5)：268-272.
[22]毛真.大米品质动态检测算法的研究：(硕士学位论文).大连.大连理工大学,2008.
[23]谯洋.基于FPGA的光电分选机检测系统的设计：(硕士学位论文).大连.大连理工大学,2007.
[24]陈建伟,刘璎瑛.基于机器视觉技术的大米品质检测综述.粮食与食品工业.2008,15(3)：44-47.
[25]http://detail.zol.com.cn/257/256038/param.shtml
[26]http://www.cheml7.com/Product/detail/2741226. html
[27]Otsu Nobuyuki. A threshold selection method from gray-level histograms. IEEE Trans on Systems Man Cybernet.1979,9(1):62-66.
[28]刘健庄,栗文青.灰度图像的二维Otsu自动阈值分割法.自动化学报.1993,19(1)：101-105.
[29]Gonzalez R C, Woods R E. Digital Image Processing(英文版).北京：电子子业出版社,2007.
[30]郎锐.数字图像处理学Visual C++实现.北京：希望电子出版社,2003.
[31]许娟.图像去噪的非局部方法研究：(硕十学位论文).南京.南京理工大学,2009.
[32]夏德深,傅德胜,徐福培.现代图像处理技术与应用(第一版).南京：尔南大学出版社,1997.
[33]王延翔.基于均值算法的混合噪声图像滤波算法的研究与实现.北京：北京邮电大学,
2010.
[34]Thomas A Nodes. Median filters:Some modifications and their properties. IEEE
Trans on Acoustics.1982,30(5):739-746.
[35]Taguchi A, Matsumoto T. Removal of impulse noise from highly corrupted images
by using noise position information and directional information of image. Electronics
and communications in Japan.2003,86(3):33-43.
[36]Zhang H. Spatially adaptive wiener filtering for image denoising using un-
decimated wavelet transform:[Ph. D dissertation]. Houston:Rice University,1999.
[37]崔屹.图象处理与分析-数学形态学方法及应用.北京：科学出版社,2000.
[38]章毓晋.图像处理和分析.北京：清华大学出版社,2000.
[39]张伟.整精米检测识别系统的研究：(硕士学位论文).大迮.大连理工大学,2009.
[40]Freeman H. Computer processing of line-drawing triages. Computing Surveys.1974,
6(1)：57-97.
[41]Freeman H. On the encoding of arbitrary geometric configuration. IRE Trans,1961,
10(2)：260-268.
[42]Freeman H. A technique for the classification and recognition of geometric
patterns. In Proceedings of the 3rd International Congress on Cybernetics, Belgium,
1961:348-368.
[43]Bribiesca E. A new chain code. Pattern Recognition.1999,32(1):235-251.
[44]Ying Yibin, Wang Jiangping, Jiang Huanyu. Inspecting diameter and defect area
of fruit with machine vision. Transactions of the CSAE.2002,18(5):216-220.
[45]方建军,张虎.板栗长短径的实时光学检测算法.农业机械学报.2006,37(5)：87-89.
[46]Andrew Fitzgibbon, Maurizio Pilu, Robert B Fisher. Direct Least Square Fitting
of Ellipses. IEEE Transaction on Pattern Analysis and Machine Intelligence.1999,
21(5):476-480.
[47]Zhengyou Zhang. Parameter Estimation Techniques:A Tutorial with Application
to Conic Fitting. Image and Vision Computing Journal.1997,15(1):59-76.
[2]中华人民共和国国家质量监督检验检疫总局.GB/T 21719-2008.中华人民共和国国家标准.北京：中国标准出版社,2008.
[3]郭文川,朱新华.机器视觉技术在谷物识别与分级中的研究进展.粮食与饲料工业.2002,6：50-51.
[4]Liu W, Tao Y, Siebenmorgen T J,et al. Digital image analysis method for rapid measurement of rice degree of milling. ASAE Annual International Meeting, Florida, 1998:380-385.
[5]Wan Y N, Lin C M, Chiou J F. Adaptive classification method for an automatic grain quality inspection system using machine vision and neural network. ASAE Annual International Meeting, Milwaukee,2000:10-19.
[6]Lloyd B J, Cnossen A G, Siebenmorgen T J. Evaluation of two methods for separating head rice from brokens for head rice yield determination. ASAE Annual International Meeting, St. Joseph,2000:643-648.
[7]Yadav B. K, Jindal V. K. Monitoring milling quality of rice by image analysis. Computers and Electronics in Agriculture.2001,33(1):19-33.
[8]侯彩云,Seiichi, Oshita等.三维图像处理系统在稻米品质检测中的应用研究.农业工程学报,2001,17(3)：92-95.
[9]包晓敏,汪亚明,黄振堃.计算机视觉技术在大米轮廓检测上的应用.浙江工程学院学报.2003,20(2)：104-107.
[10]侯彩云,李慧园,尚艳芬等.稻谷品质的图像识别与快速检测.中国粮油学报.2003,18(4)：80-83.
[11]刘光蓉,周红,管庶安.基于图像处理技术的大米轮廓检测.粮食与饲料工业.2004,6：14-15.
[12]凌云.基于机器视觉的谷物外观品质检测技术研究：(博士学位论文).北京：中国农业大学.2004.
[13]尚艳芬,侯彩云,常国华等.整精米自动识别方法的研究.中国水稻科学.2004,18(5)：466-468.
[14]成芳.稻种质量的机器视觉无损检测研究：(博士学位论文).杭州：浙江大学生物系统工 程与食品科学学院,2005.
[15]凌云,王一鸣,孙鸣等.基于机器视觉的大米外观品质检测装置.农业机械学报.2005,36(9)：89-92.
[16]于润伟,金鲲鹏,朱晓慧.基于图像识别的整精米自动检测研究.中国粮油学报.2006,21(6)：147-150.
[17]陈建华.基于计算机视觉的稻米外观品质检测方法的研究：(硕十学位论文).杭州.浙江理工大学,2008：
[18]中华人民共和国国家质量监督检验检疫总局.GB1350-2009.中华人民共和国国家标准.北京：中国标准出版社,2009.
[19]顾国庆,许彦冰.数字图像区域标定的方法.上海理工大学学报,2001,23(4)：295-299.
[20]Ying Y B, Wang J P, Jiang H P. Inspecting diameter and defect area of fruit with machine vision. Transactions of the CSAE.2002,18(5):216-220.
[21]高希瑞,孟超英,籍保平.机器视觉在稻米整精米率快速计算中的应用.食品科学.2007,28(5)：268-272.
[22]毛真.大米品质动态检测算法的研究：(硕士学位论文).大连.大连理工大学,2008.
[23]谯洋.基于FPGA的光电分选机检测系统的设计：(硕士学位论文).大连.大连理工大学,2007.
[24]陈建伟,刘璎瑛.基于机器视觉技术的大米品质检测综述.粮食与食品工业.2008,15(3)：44-47.
[25]http://detail.zol.com.cn/257/256038/param.shtml
[26]http://www.cheml7.com/Product/detail/2741226. html
[27]Otsu Nobuyuki. A threshold selection method from gray-level histograms. IEEE Trans on Systems Man Cybernet.1979,9(1):62-66.
[28]刘健庄,栗文青.灰度图像的二维Otsu自动阈值分割法.自动化学报.1993,19(1)：101-105.
[29]Gonzalez R C, Woods R E. Digital Image Processing(英文版).北京：电子子业出版社,2007.
[30]郎锐.数字图像处理学Visual C++实现.北京：希望电子出版社,2003.
[31]许娟.图像去噪的非局部方法研究：(硕十学位论文).南京.南京理工大学,2009.
[32]夏德深,傅德胜,徐福培.现代图像处理技术与应用(第一版).南京：尔南大学出版社,1997.
[33]王延翔.基于均值算法的混合噪声图像滤波算法的研究与实现.北京：北京邮电大学,
2010.
[34]Thomas A Nodes. Median filters:Some modifications and their properties. IEEE
Trans on Acoustics.1982,30(5):739-746.
[35]Taguchi A, Matsumoto T. Removal of impulse noise from highly corrupted images
by using noise position information and directional information of image. Electronics
and communications in Japan.2003,86(3):33-43.
[36]Zhang H. Spatially adaptive wiener filtering for image denoising using un-
decimated wavelet transform:[Ph. D dissertation]. Houston:Rice University,1999.
[37]崔屹.图象处理与分析-数学形态学方法及应用.北京：科学出版社,2000.
[38]章毓晋.图像处理和分析.北京：清华大学出版社,2000.
[39]张伟.整精米检测识别系统的研究：(硕士学位论文).大迮.大连理工大学,2009.
[40]Freeman H. Computer processing of line-drawing triages. Computing Surveys.1974,
6(1)：57-97.
[41]Freeman H. On the encoding of arbitrary geometric configuration. IRE Trans,1961,
10(2)：260-268.
[42]Freeman H. A technique for the classification and recognition of geometric
patterns. In Proceedings of the 3rd International Congress on Cybernetics, Belgium,
1961:348-368.
[43]Bribiesca E. A new chain code. Pattern Recognition.1999,32(1):235-251.
[44]Ying Yibin, Wang Jiangping, Jiang Huanyu. Inspecting diameter and defect area
of fruit with machine vision. Transactions of the CSAE.2002,18(5):216-220.
[45]方建军,张虎.板栗长短径的实时光学检测算法.农业机械学报.2006,37(5)：87-89.
[46]Andrew Fitzgibbon, Maurizio Pilu, Robert B Fisher. Direct Least Square Fitting
of Ellipses. IEEE Transaction on Pattern Analysis and Machine Intelligence.1999,
21(5):476-480.
[47]Zhengyou Zhang. Parameter Estimation Techniques:A Tutorial with Application
to Conic Fitting. Image and Vision Computing Journal.1997,15(1):59-76.