基于模糊三I算法的遥感影像聚类分析关键技术研究
|
摘要
随着遥感技术的发展,不断地丰富了影像信息,从而使得遥感影像的应用得到迅速地推广。尤其在影像的聚类过程中,遥感影像的高分辨率特性较好地显示了不同类别的颜色信息,在地物提取、测绘、水土流失检测、森林分类、土地覆盖情况等实际应用中充分地展示了其优越性。因而对遥感影像聚类分析的研究具有广泛的应用前景。
在目前的有监督聚类算法中,普遍认为Bayes算法的聚类效果较好,但是其结果仍然还不能满足客观的需求。主要存在的问题有:(1)传统Bayes算法具有很强的主观性。随着影像类别数的增加,由主观观察获得的类别先验概率的误差也就随着增大,需要新的方法来解决其主观性问题;(2)随着影像分辨率的提高,影像包含的数据会更加详细、信息也更加全面,因而传统的简单求和再取平均的特征提取算法难以克服样本区域中噪声点或混合像元的影响,需要新的特征提取算法与之相适应;(3)影像聚类可以视为一个模式识别问题,然而相同模式的颜色不尽相同或者不同模式的颜色存在相同部分,无疑增加了聚类的难度。如何针对遥感影像聚类分析问题,充分融入一些新的理论知识和方法进行遥感影像的聚类是非常必要的。
本文紧紧围绕以提高遥感影像聚类精度为主线,利用无人机遥感影像为主要信息数据,重点从探讨影响传统Bayes算法聚类精度因素和构造新的聚类算法两方面加以论述。在现有研究的基础上,提出了新的遥感影像聚类算法和总结了聚类精度的评价指标。
针对以上的问题,本文主要的研究工作及成果如下:
(1)在对传统Bayes聚类算法进行研究的基础上,通过构造模糊隶属度函数来修正传统先验概率的主观确定,以提高遥感影像的聚类精度。
(2)传统Bayes算法的特征提取中,各类别特征的提取均是基于样本元素权值在相同情况下求得的。通过研究灰色关联理论,并结合遥感影像高分辨率所表现出来的颜色特征,提出了一种能较好克服混合像元或者有噪声点影响的影像特征提取算法。
(3)通过利用模糊监督分类算法中多样本区域信息,研究了利用模糊三I算法的理论知识来对遥感影像进行模糊聚类分析。
总而言之,在遥感影像的聚类分析中,如何提高聚类精度是关键之处,本文着重从该方面做了一系列的研究。实验结果表明,本文提出的聚类算法较好的提高了遥感影像的聚类精度。
With the development of remote sensing technology, the image information is enriched constantly, which makes the application of remote sensing image has been rapid promotion. Especially in image clustering process, the different types of color information has been showed better by high-resolution remote sensing image characteristics. In the surface features extraction, mapping, soil erosion testing, forest classification, land covering and other practical applications fully demonstrated its superiority. Thus clustering analysis of remote sensing image has broad application prospects.
In the current supervised clustering algorithm, the results of Bayes algorithm for clustering was considered more better generally, but the results still cannot reach a satisfactory level. The main problems are: (1) Traditional Bayes algorithm has a strong subjective. With the increase in the number of image categories, the error of the category prior probability obtained by subjective observing also been increased, so it needs for new ways to solve the problem of subjectivity; (2) With the improvement of image resolution, the information of image will be more detailed, and it is also more comprehensive. So the traditional algorithms,which through simple sum and then taking the average of the feature extraction, are difficult to overcome the impact of the presence of noise point or mixed pixels in the sample area, therefore it needs new feature extraction algorithm to corresponding; (3) Image?clustering can be viewed as a pattern recognition problem, but the same pattern’s color maybe not the same or different patterns have the same part, which increasing the difficulty of clustering.?How to cluster for remote sensing image, fully integrated into some new theoretical knowledge and methods are essential.
This paper tightly around to improve the clustering accuracy of the remote sensing images as a main line, the use of low-altitude remote sensing image as the main data, focusing on discussing the two aspects: the impact factors of the traditional Bayes algorithm and constructing a new clustering algorithm. On the basis of existing research, proposed a new clustering algorithm for remote sensing images and sums up the scale to determine the best method of clustering.
To solve the above problems, the main research and results of this paper are as follows:
(1) Based on the study of the traditional Bayes clustering algorithm, modifying the traditional subjective prior probability by constructing fuzzy membership function to improve the accuracy of remote sensing image of clustering.
(2) In the traditional algorithms on feature extraction, the feature extraction of each category is based on the same weight of the sample element. By studying the theory of gray relation, and combined with the color characteristic of high-resolution remote sensing image, an algorithm on feature extraction of image is proposed which can overcome the mixed pixel or noise points influence better.?
(3) By using variety of information in the region on fuzzy surveillance cluster algorithm, studied the triple I algorithm of fuzzy theory to take cluster analysis on remote sensing images. In conclusion, how to improve the clustering accuracy is a key point in clustering analysis of remote sensing images. This article has done a series of studies focuses on the aspects, experiments show that the proposed algorithm in this article is better to improve the accuracy of clustering on remote sensing images.
在目前的有监督聚类算法中,普遍认为Bayes算法的聚类效果较好,但是其结果仍然还不能满足客观的需求。主要存在的问题有:(1)传统Bayes算法具有很强的主观性。随着影像类别数的增加,由主观观察获得的类别先验概率的误差也就随着增大,需要新的方法来解决其主观性问题;(2)随着影像分辨率的提高,影像包含的数据会更加详细、信息也更加全面,因而传统的简单求和再取平均的特征提取算法难以克服样本区域中噪声点或混合像元的影响,需要新的特征提取算法与之相适应;(3)影像聚类可以视为一个模式识别问题,然而相同模式的颜色不尽相同或者不同模式的颜色存在相同部分,无疑增加了聚类的难度。如何针对遥感影像聚类分析问题,充分融入一些新的理论知识和方法进行遥感影像的聚类是非常必要的。
本文紧紧围绕以提高遥感影像聚类精度为主线,利用无人机遥感影像为主要信息数据,重点从探讨影响传统Bayes算法聚类精度因素和构造新的聚类算法两方面加以论述。在现有研究的基础上,提出了新的遥感影像聚类算法和总结了聚类精度的评价指标。
针对以上的问题,本文主要的研究工作及成果如下:
(1)在对传统Bayes聚类算法进行研究的基础上,通过构造模糊隶属度函数来修正传统先验概率的主观确定,以提高遥感影像的聚类精度。
(2)传统Bayes算法的特征提取中,各类别特征的提取均是基于样本元素权值在相同情况下求得的。通过研究灰色关联理论,并结合遥感影像高分辨率所表现出来的颜色特征,提出了一种能较好克服混合像元或者有噪声点影响的影像特征提取算法。
(3)通过利用模糊监督分类算法中多样本区域信息,研究了利用模糊三I算法的理论知识来对遥感影像进行模糊聚类分析。
总而言之,在遥感影像的聚类分析中,如何提高聚类精度是关键之处,本文着重从该方面做了一系列的研究。实验结果表明,本文提出的聚类算法较好的提高了遥感影像的聚类精度。
With the development of remote sensing technology, the image information is enriched constantly, which makes the application of remote sensing image has been rapid promotion. Especially in image clustering process, the different types of color information has been showed better by high-resolution remote sensing image characteristics. In the surface features extraction, mapping, soil erosion testing, forest classification, land covering and other practical applications fully demonstrated its superiority. Thus clustering analysis of remote sensing image has broad application prospects.
In the current supervised clustering algorithm, the results of Bayes algorithm for clustering was considered more better generally, but the results still cannot reach a satisfactory level. The main problems are: (1) Traditional Bayes algorithm has a strong subjective. With the increase in the number of image categories, the error of the category prior probability obtained by subjective observing also been increased, so it needs for new ways to solve the problem of subjectivity; (2) With the improvement of image resolution, the information of image will be more detailed, and it is also more comprehensive. So the traditional algorithms,which through simple sum and then taking the average of the feature extraction, are difficult to overcome the impact of the presence of noise point or mixed pixels in the sample area, therefore it needs new feature extraction algorithm to corresponding; (3) Image?clustering can be viewed as a pattern recognition problem, but the same pattern’s color maybe not the same or different patterns have the same part, which increasing the difficulty of clustering.?How to cluster for remote sensing image, fully integrated into some new theoretical knowledge and methods are essential.
This paper tightly around to improve the clustering accuracy of the remote sensing images as a main line, the use of low-altitude remote sensing image as the main data, focusing on discussing the two aspects: the impact factors of the traditional Bayes algorithm and constructing a new clustering algorithm. On the basis of existing research, proposed a new clustering algorithm for remote sensing images and sums up the scale to determine the best method of clustering.
To solve the above problems, the main research and results of this paper are as follows:
(1) Based on the study of the traditional Bayes clustering algorithm, modifying the traditional subjective prior probability by constructing fuzzy membership function to improve the accuracy of remote sensing image of clustering.
(2) In the traditional algorithms on feature extraction, the feature extraction of each category is based on the same weight of the sample element. By studying the theory of gray relation, and combined with the color characteristic of high-resolution remote sensing image, an algorithm on feature extraction of image is proposed which can overcome the mixed pixel or noise points influence better.?
(3) By using variety of information in the region on fuzzy surveillance cluster algorithm, studied the triple I algorithm of fuzzy theory to take cluster analysis on remote sensing images. In conclusion, how to improve the clustering accuracy is a key point in clustering analysis of remote sensing images. This article has done a series of studies focuses on the aspects, experiments show that the proposed algorithm in this article is better to improve the accuracy of clustering on remote sensing images.
引文
[1] Ian D.Howarth and Werner Schmutz.Near-Infrared Spectroscopy of Galactic Wolf-Rayet Stars[M].Lecture Notes in Physics,1992.
[2]任学申.遥感影像在土地利用中的应用研究[J].测绘,2009,32(4):152-154.
[3] Langford M,Unwin D J. Generating and mapping population density surfaces within a geographical information system[J].The Cartographic Journal,1994,31:21-26.
[4]薛利红,曹卫星,罗卫红.基于冠层反射光谱的水稻产量预测模型[J].遥感学报,2005,9(1):100-105.
[5]姜晓剑,刘小军,田永超,等.基于遥感影像应用到作物生长的监测系统的设计与实现[M].农业工程学报,2010,26(3):256-162.
[6]薛利红,曹卫星,罗卫红,等.基于冠层反射光谱的水稻群体叶片氮素状况监测[J].中国农业科学,2003,36(7):807-812.
[7]石峰,何宏林,张英.青海玉树Ms7.1级地震地表破裂带的遥感影像解译[J].震灾防御技术,2010,5(2):220-227.
[8]王卫安,竺幼定.高分辨率卫星遥感图像及其应用[J].测绘通报,2000,(6):20-23.
[9]龙艳,张永红,马晶.DMC卫星影像在海啸灾情土地覆盖类型变化分析中的应用[J].测绘科学,2006,31(1):64-66.
[10]柳晶辉,邵芸,黄初冬.基于遥感影像的城市森林分类提取及生态价值估算研究[J].地理与地理信息科学,2007,23(4):33-36.
[11]陈建裕,毛志华,吴均平,等.高分辨率遥感影像中提取特定地物的分区多阶段混合分类方法[J].地球信息科学,2005,7(3):91-95.
[12]颜军,陈水利,吴云东.基于模糊Bayes-Gauss判别法的遥感影像的聚类[J].集美大学学报(自然科学版),2011,16(2):154-158.
[13] Sanghamitra Bandyopadhyay and Sankar Pal.Theoretical Analysis of the GA-classifier[M]. Natural Computing Series,2007.
[14] Arit Thammano and Uraiwan Kakulphimp. Genetic Algorithm-Based Clustering and Its New Mutation Operator. Lecture Notes in Computer Science,2006,Volume 4113, Intelligent Computing,Pages 703-708.
[15] Duo Chen, Du-Wu Cui and Chao-Xue Wang.Weighted Fuzzy C-Means Clustering Based on Double Coding Genetic Algorithm. Lecture Notes in Computer Science , 2006 , Volume 4113, Intelligent Computing, Pages 622-633;
[16]李海民.遗传算法性能及其在聚类分析中应用的研究[D].西安:西安电子科技大学,1999.
[17]冯征,阎敏,张智峰.一种基于PSO的模糊聚类算法[J].计算机工程与应用.2006.27:150-151.
[18]陈水利,李敬功,王向公.模糊集理论及其应用[M].北京:科学出版社,2005.
[19] Bezdek. Pattern Recognition with Fuzzy Objective Function Algorithms[M].New York:Plenum Press,1981.
[20] Wang F J. Improving remote sensing image analysis through fuzzy information representation[J].Photogram Eng Remote Sensing,1990,56(8):1163-1169.
[21]王文宇,李博.基于eCognition的高分辨率遥感图像的自动识别分类技术[J].北京建筑工程学院学报,2006,22(4):26-29.
[22]杜凤兰,田庆久,夏学齐,等.面向对象的地物分类法分析与评价[J].遥感技术与应用,2004,19(1):20-24.
[23]陈丹峰,林培,汲长远.自组织网络与模糊规则结合在遥感土地覆盖分类中的应用[J].中国土地科学,1998,12(5):42-44.
[24] DOMN,VETTERLIM.The contour let transform: An efficient directional multi resolution image representation[J].IEEE Transaction Image Processing,2005,14(12):2091-2106.
[25]洛建承,周成虎,杨艳.具有部分监督的遥感影像模糊聚类方法研究及应用[J].遥感技术与应用,1999,14(4):37-43.
[26]鱼海涛.基于模糊集理论的图像增强算法研究[D].西安:西安科技大学,2005.
[27]尹军梅,杨明.一种面向单个正例的Fisher线性判别分类方法[J].南京师范大学学报(工程技术版),2008,8(3):61-65.
[28]付丽,孙红帆,杨勇,等.基于贝叶斯分类器的图像分类技术[J].长春理工大学学报(自然科学版),2009,32(1):132-134.
[29]边肇棋,张学工.模式识别[M].北京:清华大学出版社,2001.
[30] Guojun Wang.Non-classical Mathematical Logic and Approximate Reasoning (second edition)[M]. Beijing:Science Press,2008.
[31]周保魁,王国俊.不同蕴涵算子下的三I算法[J].陕西师范大学学报(自然科学版),1998,26(4):1-5.
[32]党安荣,王晓栋,陈晓峰,等.ERDAS IMAGINE遥感图像处理方法[M].北京:清华大学出版社,2003.
[33] Laviolette M,Seaman J W, Barrett J D,et al.A probabilistic and statistical view of fuzzy methods[J].Technometrics,1995,37(3):249-261.
[34]苑飞.基于模糊熵三I算法构造的模糊系统的概率表示及其逼近性能[J].海南师范大学学报(自然科学版),2008,21(4):472-477.
[35] Kiran R.Bhutani,Abdella Battou.An application of fuzzy relations to image enhancement[J]. Pattern Recognition Letters,1995,16:901-909.
[36] Di Y C,Deng Y P. Performance evaluation based on the grey relational grade of image segmentation[J].Infrared and Laser Engineering,1999,28(3):20-22.
[37]马苗,等.灰色理论在图像工程中的应用研究进展[J].中国图象图形学报,2007,12(11): 1943-1951.
[38]颜军,陈水利,吴云东.基于模糊灰色关联度的影像特征提取算法[C]//李茂清.计算机与系统工程教学研究论文集.美国:科学与教育出版社,2010:100-105.
[39] Guo Fang-fang,Chen Tu-Yun,Xia Zun-Quan.Triple I methods for Fuzzy Reasoning Based on Maximum Fuzzy Entropy Principle[J].Fuzzy Systems and Mathematics 2003,17(4):55-59.
[40]陈忠.高分辨率遥感图像分类技术研究[D].北京:中国科学院遥感应用研究所,2006.
[2]任学申.遥感影像在土地利用中的应用研究[J].测绘,2009,32(4):152-154.
[3] Langford M,Unwin D J. Generating and mapping population density surfaces within a geographical information system[J].The Cartographic Journal,1994,31:21-26.
[4]薛利红,曹卫星,罗卫红.基于冠层反射光谱的水稻产量预测模型[J].遥感学报,2005,9(1):100-105.
[5]姜晓剑,刘小军,田永超,等.基于遥感影像应用到作物生长的监测系统的设计与实现[M].农业工程学报,2010,26(3):256-162.
[6]薛利红,曹卫星,罗卫红,等.基于冠层反射光谱的水稻群体叶片氮素状况监测[J].中国农业科学,2003,36(7):807-812.
[7]石峰,何宏林,张英.青海玉树Ms7.1级地震地表破裂带的遥感影像解译[J].震灾防御技术,2010,5(2):220-227.
[8]王卫安,竺幼定.高分辨率卫星遥感图像及其应用[J].测绘通报,2000,(6):20-23.
[9]龙艳,张永红,马晶.DMC卫星影像在海啸灾情土地覆盖类型变化分析中的应用[J].测绘科学,2006,31(1):64-66.
[10]柳晶辉,邵芸,黄初冬.基于遥感影像的城市森林分类提取及生态价值估算研究[J].地理与地理信息科学,2007,23(4):33-36.
[11]陈建裕,毛志华,吴均平,等.高分辨率遥感影像中提取特定地物的分区多阶段混合分类方法[J].地球信息科学,2005,7(3):91-95.
[12]颜军,陈水利,吴云东.基于模糊Bayes-Gauss判别法的遥感影像的聚类[J].集美大学学报(自然科学版),2011,16(2):154-158.
[13] Sanghamitra Bandyopadhyay and Sankar Pal.Theoretical Analysis of the GA-classifier[M]. Natural Computing Series,2007.
[14] Arit Thammano and Uraiwan Kakulphimp. Genetic Algorithm-Based Clustering and Its New Mutation Operator. Lecture Notes in Computer Science,2006,Volume 4113, Intelligent Computing,Pages 703-708.
[15] Duo Chen, Du-Wu Cui and Chao-Xue Wang.Weighted Fuzzy C-Means Clustering Based on Double Coding Genetic Algorithm. Lecture Notes in Computer Science , 2006 , Volume 4113, Intelligent Computing, Pages 622-633;
[16]李海民.遗传算法性能及其在聚类分析中应用的研究[D].西安:西安电子科技大学,1999.
[17]冯征,阎敏,张智峰.一种基于PSO的模糊聚类算法[J].计算机工程与应用.2006.27:150-151.
[18]陈水利,李敬功,王向公.模糊集理论及其应用[M].北京:科学出版社,2005.
[19] Bezdek. Pattern Recognition with Fuzzy Objective Function Algorithms[M].New York:Plenum Press,1981.
[20] Wang F J. Improving remote sensing image analysis through fuzzy information representation[J].Photogram Eng Remote Sensing,1990,56(8):1163-1169.
[21]王文宇,李博.基于eCognition的高分辨率遥感图像的自动识别分类技术[J].北京建筑工程学院学报,2006,22(4):26-29.
[22]杜凤兰,田庆久,夏学齐,等.面向对象的地物分类法分析与评价[J].遥感技术与应用,2004,19(1):20-24.
[23]陈丹峰,林培,汲长远.自组织网络与模糊规则结合在遥感土地覆盖分类中的应用[J].中国土地科学,1998,12(5):42-44.
[24] DOMN,VETTERLIM.The contour let transform: An efficient directional multi resolution image representation[J].IEEE Transaction Image Processing,2005,14(12):2091-2106.
[25]洛建承,周成虎,杨艳.具有部分监督的遥感影像模糊聚类方法研究及应用[J].遥感技术与应用,1999,14(4):37-43.
[26]鱼海涛.基于模糊集理论的图像增强算法研究[D].西安:西安科技大学,2005.
[27]尹军梅,杨明.一种面向单个正例的Fisher线性判别分类方法[J].南京师范大学学报(工程技术版),2008,8(3):61-65.
[28]付丽,孙红帆,杨勇,等.基于贝叶斯分类器的图像分类技术[J].长春理工大学学报(自然科学版),2009,32(1):132-134.
[29]边肇棋,张学工.模式识别[M].北京:清华大学出版社,2001.
[30] Guojun Wang.Non-classical Mathematical Logic and Approximate Reasoning (second edition)[M]. Beijing:Science Press,2008.
[31]周保魁,王国俊.不同蕴涵算子下的三I算法[J].陕西师范大学学报(自然科学版),1998,26(4):1-5.
[32]党安荣,王晓栋,陈晓峰,等.ERDAS IMAGINE遥感图像处理方法[M].北京:清华大学出版社,2003.
[33] Laviolette M,Seaman J W, Barrett J D,et al.A probabilistic and statistical view of fuzzy methods[J].Technometrics,1995,37(3):249-261.
[34]苑飞.基于模糊熵三I算法构造的模糊系统的概率表示及其逼近性能[J].海南师范大学学报(自然科学版),2008,21(4):472-477.
[35] Kiran R.Bhutani,Abdella Battou.An application of fuzzy relations to image enhancement[J]. Pattern Recognition Letters,1995,16:901-909.
[36] Di Y C,Deng Y P. Performance evaluation based on the grey relational grade of image segmentation[J].Infrared and Laser Engineering,1999,28(3):20-22.
[37]马苗,等.灰色理论在图像工程中的应用研究进展[J].中国图象图形学报,2007,12(11): 1943-1951.
[38]颜军,陈水利,吴云东.基于模糊灰色关联度的影像特征提取算法[C]//李茂清.计算机与系统工程教学研究论文集.美国:科学与教育出版社,2010:100-105.
[39] Guo Fang-fang,Chen Tu-Yun,Xia Zun-Quan.Triple I methods for Fuzzy Reasoning Based on Maximum Fuzzy Entropy Principle[J].Fuzzy Systems and Mathematics 2003,17(4):55-59.
[40]陈忠.高分辨率遥感图像分类技术研究[D].北京:中国科学院遥感应用研究所,2006.