玉米叶部病斑图像智能处理算法的研究与实现
|
摘要
利用机器视觉来解决农作物病害诊断问题,进而达到防治病害和节约成本的目的,是现代农业最显著的特点之一。目前有很多关于数字图像处理技术用于这一领域的研究,其中图像分割是最重要的一部分,它关系到特征提取和后续识别的正确率,而如何结合多种分割和预处理技术是本研究的重点。本文以大斑病、小斑病、灰斑病、弯孢叶斑病、锈病和褐斑6种玉米叶部病斑图像为研究对象,实现了玉米叶部病斑的自动识别系统。系统主要由四部分组成：图像预处理,图像分割,特征提取,分类识别,基于系统实现需要,本文主要研究了以下内容：
1.图像预处理算法的研究,主要实现了图像缩小、自动裁剪和图像增强算法。由于从大田拍摄的玉米叶片病斑图像像素过大,直接进行处理会影响效率,本文所处理的图像是由索尼a700拍摄的分辨率为3104×2064的图像,无法直接进行分割算法处理,所以要先对原图像进行缩小操作,为了裁剪出病斑所在区域,本文使用了基于人眼视觉的图像自动裁剪技术,使用分块来计算影响人眼视觉的影响因子来裁剪图像,为了更精确地裁剪出目标区域图像,本文在原有的4个影响因子中加入了颜色影响因子,裁剪效果明显要好很多。而裁剪后的图像要经过图像增强,才便于后续的分割处理,由于玉米病斑图像中有些随机杂点,通过中值滤波算法可以很好的过滤掉这些杂点。
2.图像分割算法的研究,目前,作物叶部病斑分割的方法大多是利用病叶片和病斑的灰度差异,采用固定阈值分割算法或用明确的判别标准来区分叶片部分和病斑部分。但实际上,由于病叶图像中的对象具有模糊性和不确定性,对象的边界不可能清晰明确,作物叶片和病斑本身颜色的不均匀和灰度值存在着重叠,因此很难用固定阂值或确定的准则来准确的确定病叶图像中不确定的对象。为了提高玉米叶片病斑的分割效果,本文提出了将局部阈值法与区域增长法相结合的作物病害图像自适应分割算法。
3.玉米叶片病斑特征的提取与系统验证。本文主要提取了玉米叶片病斑的几个形态特征：面积、周长、圆形度、矩形度和复杂度。通过模糊识别算法来判别病斑种类,验证各个特征对不同病斑的影响因子,通过实验得出图像缩小因子和图像裁剪分块的个数。
整个系统,基于文档视图结构,能够完成对病斑图像的预处理、病斑分割,特征提取和自动识别功能,并且所有功能都可以通过点击相应的菜单项实现,操作简单,并且处理过程可以在界面上显示,可视化好。
Using machine vision to solve the problem of crop illness diagnosis to realize the ultimate goal of crop illness prevention and saving the cost is one of the most significant characteristics of modern agriculture. Currently, lots of digital image processing techniques applied in this filed have been studied, one of the most important part is the segmentation algorithm which is vital to feature extraction and the accuracy of recognition. This paper mainly researched on how to integrate several classic segmentation algorithms together to segment 6 kinds of image ill spots on maize leaf which are Exserohilum Turcicum, Bipolaris maydis, Cercospora zeaemaydis Tehon and Daniels, Curvularia lunata Boed, Puccinia polysora, Physoderma maydis Miyabe. The system includes 4 parts which are image pre-process, image segmentation, feature extraction and classification and recognition. Based on system implementation need, the primary coverage of this paper are as follows:
1.The study on image pre-process algorithm, mainly implement the function of image-resizing, automatic image-clipping and image enhancement. Due to large pixels of images shot from the maize field with Sony a700, direct process will cause low efficiency. So firstly, we reduce the image size and then clip out the interestd area using the automatic image clipping technique based on human vision which is implemented by computing 4 influence parameters of each blocks divided. To improve the accuracy, This paper add a color influence parameter which makes a better result. Finally, by using median filtering to erase the random noise of the image.
2.The study on image segmentation algorithm.currently, most of the segmentaion algorithm usually use specific criterion to differentiate between ill spots and background. But in fact,due to the fuzziness and uncertainty of the crop leaf background and the ill spots, it's hard to use specific criterion to segment the illness spot out of the image. In order to improve the effect of the segmentation. This paper propose a self-adapting algorithm combining local threshold with regional growth.
3.The study of feature extraction and system verification. This paper mainly extract several morphological characters:Area, Girth, Circularity, Rectangular degree, Complexity. This part will use Fuzzy recognition algorithm to classify illness kinds and verify the influence of different illness spot. Image reducement parameter and the number of blocks divided from image can be defined through experiment.
The system based on document-view architecture implement the function of image pre-process, image segmentation, feature extraction and automatic recognition, and all of the functions can be manipulated by clicking relevant menu. It's easy to process and have good user interface.
1.图像预处理算法的研究,主要实现了图像缩小、自动裁剪和图像增强算法。由于从大田拍摄的玉米叶片病斑图像像素过大,直接进行处理会影响效率,本文所处理的图像是由索尼a700拍摄的分辨率为3104×2064的图像,无法直接进行分割算法处理,所以要先对原图像进行缩小操作,为了裁剪出病斑所在区域,本文使用了基于人眼视觉的图像自动裁剪技术,使用分块来计算影响人眼视觉的影响因子来裁剪图像,为了更精确地裁剪出目标区域图像,本文在原有的4个影响因子中加入了颜色影响因子,裁剪效果明显要好很多。而裁剪后的图像要经过图像增强,才便于后续的分割处理,由于玉米病斑图像中有些随机杂点,通过中值滤波算法可以很好的过滤掉这些杂点。
2.图像分割算法的研究,目前,作物叶部病斑分割的方法大多是利用病叶片和病斑的灰度差异,采用固定阈值分割算法或用明确的判别标准来区分叶片部分和病斑部分。但实际上,由于病叶图像中的对象具有模糊性和不确定性,对象的边界不可能清晰明确,作物叶片和病斑本身颜色的不均匀和灰度值存在着重叠,因此很难用固定阂值或确定的准则来准确的确定病叶图像中不确定的对象。为了提高玉米叶片病斑的分割效果,本文提出了将局部阈值法与区域增长法相结合的作物病害图像自适应分割算法。
3.玉米叶片病斑特征的提取与系统验证。本文主要提取了玉米叶片病斑的几个形态特征：面积、周长、圆形度、矩形度和复杂度。通过模糊识别算法来判别病斑种类,验证各个特征对不同病斑的影响因子,通过实验得出图像缩小因子和图像裁剪分块的个数。
整个系统,基于文档视图结构,能够完成对病斑图像的预处理、病斑分割,特征提取和自动识别功能,并且所有功能都可以通过点击相应的菜单项实现,操作简单,并且处理过程可以在界面上显示,可视化好。
Using machine vision to solve the problem of crop illness diagnosis to realize the ultimate goal of crop illness prevention and saving the cost is one of the most significant characteristics of modern agriculture. Currently, lots of digital image processing techniques applied in this filed have been studied, one of the most important part is the segmentation algorithm which is vital to feature extraction and the accuracy of recognition. This paper mainly researched on how to integrate several classic segmentation algorithms together to segment 6 kinds of image ill spots on maize leaf which are Exserohilum Turcicum, Bipolaris maydis, Cercospora zeaemaydis Tehon and Daniels, Curvularia lunata Boed, Puccinia polysora, Physoderma maydis Miyabe. The system includes 4 parts which are image pre-process, image segmentation, feature extraction and classification and recognition. Based on system implementation need, the primary coverage of this paper are as follows:
1.The study on image pre-process algorithm, mainly implement the function of image-resizing, automatic image-clipping and image enhancement. Due to large pixels of images shot from the maize field with Sony a700, direct process will cause low efficiency. So firstly, we reduce the image size and then clip out the interestd area using the automatic image clipping technique based on human vision which is implemented by computing 4 influence parameters of each blocks divided. To improve the accuracy, This paper add a color influence parameter which makes a better result. Finally, by using median filtering to erase the random noise of the image.
2.The study on image segmentation algorithm.currently, most of the segmentaion algorithm usually use specific criterion to differentiate between ill spots and background. But in fact,due to the fuzziness and uncertainty of the crop leaf background and the ill spots, it's hard to use specific criterion to segment the illness spot out of the image. In order to improve the effect of the segmentation. This paper propose a self-adapting algorithm combining local threshold with regional growth.
3.The study of feature extraction and system verification. This paper mainly extract several morphological characters:Area, Girth, Circularity, Rectangular degree, Complexity. This part will use Fuzzy recognition algorithm to classify illness kinds and verify the influence of different illness spot. Image reducement parameter and the number of blocks divided from image can be defined through experiment.
The system based on document-view architecture implement the function of image pre-process, image segmentation, feature extraction and automatic recognition, and all of the functions can be manipulated by clicking relevant menu. It's easy to process and have good user interface.
引文
[1]冀荣华,祁力钧,傅泽田.机器视觉技术在精细农业中的研究进展.农业机械化研究,2007.11,11：1-5.
[2]Seginer I, Elster R T, Goodrum J W, et al Plant wilt detection by computer-vision tracking of leaf tips Trans of the A SAE,1992,35(5):1563-1567
[3]Ling P P, Cheng Z, Musacchio D J. Quantificastion of xomatic coffee embryo growth using machine vision. Trans of the ASAE,1995,38(6):1911-1917
[4]E.J.Van Henten, J.Hemming, B.A.J.Van Tuijl, et al.An autonomous robot for harvesting cucumbers [J].Autonomous Robots,2002,(13):241-258.
[5]H. Aghayeghazvini, A. Afzal, M. Heidarisoltanabadi, S. Malek, L.Mollabashi. Determining Percentage Of Broken Rice By Using Image Analysis.2008.
[6]纪寿文,王荣本,陈佳娟等.应用计算机图像处理技术识别玉米苗期田间杂草的研究[J].农业工程学报,2001,17(2)：154-156.
[7]田有文,李成华.基于统计模式识别的植物病害彩色图像分割方法.吉林大学学报,2004.4,34(2)：291-293.
[8]马骏,王建华.一种基于数学形态学的植物病虫识别方法.深圳大学学报,2004,21(1)：72-75.
[9]Wenhua Mao, Baoping Jia, Xiaochao Zhang, Xiaoan Hub. Detection and position method of apple tree image.2008
[10]刘立波,周国民.人眼视觉特征在植物叶片图像提取中的应用.计算机工程与应用,2009(19)
[11]肖超云,朱伟兴.基于Otsu准则及图像熵的阈值分割算法.计算机工程,2007.7,33(14)：188-209
[12]王永波,陈继荣.二维Otsu阈值分割算法的改进及应用.计算机仿真,2008.4,25(4)：263-266
[13]雷小锋,谢昆青,林帆,夏征义.一种基于K-Means局部最优性的高效聚类算法.软件学报,2008.7,19(7)：1683-1692
[14]Tapas Kanungo, Nathan S.Netanyahu, Angela Y.Wu. An Efficient k-Means Clustering Algorithm:Analysis and Implementation. IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL 24, NO.7, JULY 2002
[15]K.U.Leuven. Automatic detection of'yellow rust' in wheat using reflectance measurements and neural networks. Computers and Electronics in Agriculture,2004,44:173-188
[16]汤秀娟.玉米叶部病斑自动识别系统研究与实现,北京邮电大学硕士生论文,2009(3)
[17]S.A. Hojjatoleslami, J. Kiltler. Region Growing:A New Approach. IEEE Transactions on Image Processing,1998.7,7(7):1079-1084
[18]Jianping Fan, David K.Y.Yau. Automatic Image Segmentation By Integrating Color-Edge Extraction And Seaded Region Growing. IEEE on Image Processing, 2001.10,10(10):1454-1466
[19]Jie Wu, Skip Poehlman, Michael D. Noseworthy, Markad V. Kamath. Texture feature based automated seeded region growing in abdominal MRI segmentation. Biomedical Science and Engineering,2009,2,1-8
[20]N.A.Mat-Isa, M.Y.Mashor, N.H.Othman. Seeded Region Growing Features Extraction Algorithm;Its Potential Use in Improving Screening for Cervical Cancer. International Journal of The Computer,the Internet and Management,2005.1,13(1):61-70
[21]尹建军,王新忠,毛罕平.RGB与HSI颜色空间下番茄图像分割的对比研究.农机化研究.2006.11,11(1)：171-174
[22]赵玉霞,王克如,白中英,李少昆谢瑞芝,.基于图像识别的玉米叶部病害诊断研究.中国农业科学,2007,40(4)：698-703
[23]毛罕平,张艳诚,胡波.基于模糊C均值聚类的作物病害叶片图像分割方法研究.农业工程学报,2008.9,24(9)
[24]马晓丹,祁广云.基于神经网络的大豆叶片病斑的识别与研究.黑龙江八一农垦大学学报.2006.4,18(2)：84-87
[25]林晓燕,刘文耀,陈晓冬,曹茂永.杨树病害孢子的图像识别技术研究.仪器仪表学报,2003.8,24(4)：364-366
[26]钱玮,陈卫,白石磊,韦穗,丁燕.基于形态学及SVM思想的病变图像识别方法.中国图象图形学报,2003,10,8(10)：1201-1204
[27]张梅,张祖勋.基于微分不变量和区域增长法的深度图分割.计算机工程,2008.10,34(19)
[28]章毓晋.图像分割评价技术分类和比较.中国图像图形学报,1996.6,1(2)：151-157
[29]刘立波,周国民,基于图像评价的灯箱最佳成像环境因素确定方法.计算机应用研究,2009(12)
[30]毛罕平,徐贵力.基于遗传算法的蔬菜缺素叶片图像特征选择研究.江苏大学学报,2003,24
[31]胡春华,李萍萍.基于图像处理的黄瓜缺氮与缺镁判别的研究.江苏大学学报,2004,25
[32]田有文,王滨,唐晓明.基于纹理特征和支持向量机的玉米病害的识别.沈阳农业大学学报,2005,36(6)：730-732.
[33]田有文,李天来,李成华,朴在林,孙国凯,王滨.基于支持向量机的葡萄病害图像识别方法. 农业工程学报,2007,23(6)：175-180.
[34]张红涛.储粮害虫图像识别中的特征抽取研究.郑州大学硕士学位论文.2002,28-30.
[35]战国科,夏哲雷.基于支持向量机的飞机图像识别算法.现代电子技术,2007,21：127-129.
[36]魏邦龙.用模糊模型识别的隶属度方法判别作物类型.甘肃省农科院测试室.
[37]刘素华.基于模糊理论的仓储物昆虫的分类识别研究.郑州大学硕士学位论文.2003
[38]唐强.基于图像模式识别技术的昆虫识别研究.昆明理工大学硕士学位论文.2006.
[2]Seginer I, Elster R T, Goodrum J W, et al Plant wilt detection by computer-vision tracking of leaf tips Trans of the A SAE,1992,35(5):1563-1567
[3]Ling P P, Cheng Z, Musacchio D J. Quantificastion of xomatic coffee embryo growth using machine vision. Trans of the ASAE,1995,38(6):1911-1917
[4]E.J.Van Henten, J.Hemming, B.A.J.Van Tuijl, et al.An autonomous robot for harvesting cucumbers [J].Autonomous Robots,2002,(13):241-258.
[5]H. Aghayeghazvini, A. Afzal, M. Heidarisoltanabadi, S. Malek, L.Mollabashi. Determining Percentage Of Broken Rice By Using Image Analysis.2008.
[6]纪寿文,王荣本,陈佳娟等.应用计算机图像处理技术识别玉米苗期田间杂草的研究[J].农业工程学报,2001,17(2)：154-156.
[7]田有文,李成华.基于统计模式识别的植物病害彩色图像分割方法.吉林大学学报,2004.4,34(2)：291-293.
[8]马骏,王建华.一种基于数学形态学的植物病虫识别方法.深圳大学学报,2004,21(1)：72-75.
[9]Wenhua Mao, Baoping Jia, Xiaochao Zhang, Xiaoan Hub. Detection and position method of apple tree image.2008
[10]刘立波,周国民.人眼视觉特征在植物叶片图像提取中的应用.计算机工程与应用,2009(19)
[11]肖超云,朱伟兴.基于Otsu准则及图像熵的阈值分割算法.计算机工程,2007.7,33(14)：188-209
[12]王永波,陈继荣.二维Otsu阈值分割算法的改进及应用.计算机仿真,2008.4,25(4)：263-266
[13]雷小锋,谢昆青,林帆,夏征义.一种基于K-Means局部最优性的高效聚类算法.软件学报,2008.7,19(7)：1683-1692
[14]Tapas Kanungo, Nathan S.Netanyahu, Angela Y.Wu. An Efficient k-Means Clustering Algorithm:Analysis and Implementation. IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL 24, NO.7, JULY 2002
[15]K.U.Leuven. Automatic detection of'yellow rust' in wheat using reflectance measurements and neural networks. Computers and Electronics in Agriculture,2004,44:173-188
[16]汤秀娟.玉米叶部病斑自动识别系统研究与实现,北京邮电大学硕士生论文,2009(3)
[17]S.A. Hojjatoleslami, J. Kiltler. Region Growing:A New Approach. IEEE Transactions on Image Processing,1998.7,7(7):1079-1084
[18]Jianping Fan, David K.Y.Yau. Automatic Image Segmentation By Integrating Color-Edge Extraction And Seaded Region Growing. IEEE on Image Processing, 2001.10,10(10):1454-1466
[19]Jie Wu, Skip Poehlman, Michael D. Noseworthy, Markad V. Kamath. Texture feature based automated seeded region growing in abdominal MRI segmentation. Biomedical Science and Engineering,2009,2,1-8
[20]N.A.Mat-Isa, M.Y.Mashor, N.H.Othman. Seeded Region Growing Features Extraction Algorithm;Its Potential Use in Improving Screening for Cervical Cancer. International Journal of The Computer,the Internet and Management,2005.1,13(1):61-70
[21]尹建军,王新忠,毛罕平.RGB与HSI颜色空间下番茄图像分割的对比研究.农机化研究.2006.11,11(1)：171-174
[22]赵玉霞,王克如,白中英,李少昆谢瑞芝,.基于图像识别的玉米叶部病害诊断研究.中国农业科学,2007,40(4)：698-703
[23]毛罕平,张艳诚,胡波.基于模糊C均值聚类的作物病害叶片图像分割方法研究.农业工程学报,2008.9,24(9)
[24]马晓丹,祁广云.基于神经网络的大豆叶片病斑的识别与研究.黑龙江八一农垦大学学报.2006.4,18(2)：84-87
[25]林晓燕,刘文耀,陈晓冬,曹茂永.杨树病害孢子的图像识别技术研究.仪器仪表学报,2003.8,24(4)：364-366
[26]钱玮,陈卫,白石磊,韦穗,丁燕.基于形态学及SVM思想的病变图像识别方法.中国图象图形学报,2003,10,8(10)：1201-1204
[27]张梅,张祖勋.基于微分不变量和区域增长法的深度图分割.计算机工程,2008.10,34(19)
[28]章毓晋.图像分割评价技术分类和比较.中国图像图形学报,1996.6,1(2)：151-157
[29]刘立波,周国民,基于图像评价的灯箱最佳成像环境因素确定方法.计算机应用研究,2009(12)
[30]毛罕平,徐贵力.基于遗传算法的蔬菜缺素叶片图像特征选择研究.江苏大学学报,2003,24
[31]胡春华,李萍萍.基于图像处理的黄瓜缺氮与缺镁判别的研究.江苏大学学报,2004,25
[32]田有文,王滨,唐晓明.基于纹理特征和支持向量机的玉米病害的识别.沈阳农业大学学报,2005,36(6)：730-732.
[33]田有文,李天来,李成华,朴在林,孙国凯,王滨.基于支持向量机的葡萄病害图像识别方法. 农业工程学报,2007,23(6)：175-180.
[34]张红涛.储粮害虫图像识别中的特征抽取研究.郑州大学硕士学位论文.2002,28-30.
[35]战国科,夏哲雷.基于支持向量机的飞机图像识别算法.现代电子技术,2007,21：127-129.
[36]魏邦龙.用模糊模型识别的隶属度方法判别作物类型.甘肃省农科院测试室.
[37]刘素华.基于模糊理论的仓储物昆虫的分类识别研究.郑州大学硕士学位论文.2003
[38]唐强.基于图像模式识别技术的昆虫识别研究.昆明理工大学硕士学位论文.2006.