基于图像处理的云图分析及云层面积测量
|
摘要
进行云分类和云团分析预测的技术与方法研究,客观、准确、及时地进行云图识别和云区域估计等研究,是气象学应用中主要的研究方向。论文基于学科与方法交叉的思想,运用数字图像处理技术展开云图分析及云层面积测量的研究,在分析和测量过程中给出了几种有效的算法。
论文首先对气象云图预处理,给出一种去除下垫面的新方法——直方图阈值双峰法。该方法利用样本云与下垫面在图像中灰度差异大的特点,达到样本云与下垫面的有效分割。在去除薄雾噪声时,针对薄雾变化比较平缓,对应较低的频率分量的特点,运用了同态滤波法。实验结果表明,预处理后的云图图像提高了质量,降低了复杂度,获得了较好的图像效果。
数字图像预处理之后,论文在本章对样本云进行了进一步的分析。首先,针对传统PCA算法维数灾难问题,提出PCA的改进算法,减少压缩前后的均方误差,提高分辨力,在较低维数字图像空间得到清晰的样本云图,实现云图有效的识别。实验表明,图像SNR提高了7.64dB,识别率也大大提高。论文还对样本云的定量统计进行了讨论,给出了计数的基本思路和实现函数。算法在MATLAB中运行,仿真结果和传统的人工计数相比,不仅计数一样,而且操作很简单、省时。最后,针对样本云反射率测定现状,将图像处理技术和数学理论有机的融合,在反射率和可见光与红外亮度之间建立了数学关系,实现样本云分类的目的。
另外,卫星云图或多或少存在云遮挡的“盲区”、信息丢失,对于云图分析与解译不利。这就引出了本文要解决的第二个问题——云层面积计算。将N*M像元素阵列的图象转换成许多个互相不重叠的区域组成的块的集合。采用图像处理技术对Freeman链码算法进行改进,只要对链码边界点进行判断,实现封闭区域的迅速填充,自动的计算出不规则云层的面积。实验结果表明,运算耗时小,精确度在±1%左右。
总之,本文利用图像处理技术,为云图分析及云层面积测量提供了有效可行的解决途径,在已有理论的基础上优化和改进,为实际研究和应用提供了新思路。
The study on the technology and methods for clouds classification and clouds analysis prediction, and the research on how to identify nephogram and estimate clouds regional objectively, precisely and timely, which are main research directions in the application technology on the meteorology. Based on cross-ideologies of disciplines and methods, this paper explores on nephogram analysis and clouds area measurement by using digital image processing technology and provides effective algorithm in the process of analysis and measurement.
This paper first pre-treat nephogram and presents a new method of removing underlying surface——pad-histogram threshold value method of twin peaks. This method will make use of different gray-level between sample cloud and underlying surface in the picture to realize sample cloud and underlying surface segmentation efficiently. As for wiping off mist noise, homomorphic filtering method is used because the mist change is relatively flat, corresponding to the low frequency components. Experiment shows that pre-treat nephogram quality is improved while complexity lowered, a better image effect。
After image pre-treatment, the paper carried out further analysis of the sample cloud in this chapter. First, aiming to solve the dimension disasters problems brought by the traditional PCA, this paper gives the advanced algorithm of PCA, reduces the mean-square-error in non-compressed and compressed situations and improves the resolution in order to obtain clear sample nephogram in low dimension space of digital image. Experiment shows that the image SNR increase by7.64dB and the image recognition rate also largely improved.The paper also discussed the quantitative statistics of the sample cloud, giving the basic ideas and realization of functions of count. The algorithm is run in MATLAB, the simulation results not only count the same, but also operation is very simple, time-saving when it compares to the traditional manual. Finally, according to the current situation of sample cloud Reflectance measurement, this paper combined image processing techniques and mathematical theory organically, established mathematical relationships between the reflectivity, visual light and infrared brightness, achieving the purpose of the sample cloud classification.
In addition, satellite nephogram exists more or less "blind area" of cloud cover causing information lost which harm nephogram analysis and interpretation. This brings up a second question in this paper to solve——calculating clouds area. Making the image of N*M element in array into a set of blocks made of many non-overlapping regions. Using image processing technology will improve Freeman chain code algorithm, only judgying the boundary points of the chain code, achieving the rapid filling of the enclosed area in order to calculate the area of irregular clouds automatically. The experimental results show that the computing time-consuming small, accuracy achieves about±1%.
In short, this paper presents effective and feasible solution for nephogram analysis and clouds area measurement by using image processing technology. The findings is based on previous theories, optimized and improved by this paper, providing new ideas for future theory study and application.
论文首先对气象云图预处理,给出一种去除下垫面的新方法——直方图阈值双峰法。该方法利用样本云与下垫面在图像中灰度差异大的特点,达到样本云与下垫面的有效分割。在去除薄雾噪声时,针对薄雾变化比较平缓,对应较低的频率分量的特点,运用了同态滤波法。实验结果表明,预处理后的云图图像提高了质量,降低了复杂度,获得了较好的图像效果。
数字图像预处理之后,论文在本章对样本云进行了进一步的分析。首先,针对传统PCA算法维数灾难问题,提出PCA的改进算法,减少压缩前后的均方误差,提高分辨力,在较低维数字图像空间得到清晰的样本云图,实现云图有效的识别。实验表明,图像SNR提高了7.64dB,识别率也大大提高。论文还对样本云的定量统计进行了讨论,给出了计数的基本思路和实现函数。算法在MATLAB中运行,仿真结果和传统的人工计数相比,不仅计数一样,而且操作很简单、省时。最后,针对样本云反射率测定现状,将图像处理技术和数学理论有机的融合,在反射率和可见光与红外亮度之间建立了数学关系,实现样本云分类的目的。
另外,卫星云图或多或少存在云遮挡的“盲区”、信息丢失,对于云图分析与解译不利。这就引出了本文要解决的第二个问题——云层面积计算。将N*M像元素阵列的图象转换成许多个互相不重叠的区域组成的块的集合。采用图像处理技术对Freeman链码算法进行改进,只要对链码边界点进行判断,实现封闭区域的迅速填充,自动的计算出不规则云层的面积。实验结果表明,运算耗时小,精确度在±1%左右。
总之,本文利用图像处理技术,为云图分析及云层面积测量提供了有效可行的解决途径,在已有理论的基础上优化和改进,为实际研究和应用提供了新思路。
The study on the technology and methods for clouds classification and clouds analysis prediction, and the research on how to identify nephogram and estimate clouds regional objectively, precisely and timely, which are main research directions in the application technology on the meteorology. Based on cross-ideologies of disciplines and methods, this paper explores on nephogram analysis and clouds area measurement by using digital image processing technology and provides effective algorithm in the process of analysis and measurement.
This paper first pre-treat nephogram and presents a new method of removing underlying surface——pad-histogram threshold value method of twin peaks. This method will make use of different gray-level between sample cloud and underlying surface in the picture to realize sample cloud and underlying surface segmentation efficiently. As for wiping off mist noise, homomorphic filtering method is used because the mist change is relatively flat, corresponding to the low frequency components. Experiment shows that pre-treat nephogram quality is improved while complexity lowered, a better image effect。
After image pre-treatment, the paper carried out further analysis of the sample cloud in this chapter. First, aiming to solve the dimension disasters problems brought by the traditional PCA, this paper gives the advanced algorithm of PCA, reduces the mean-square-error in non-compressed and compressed situations and improves the resolution in order to obtain clear sample nephogram in low dimension space of digital image. Experiment shows that the image SNR increase by7.64dB and the image recognition rate also largely improved.The paper also discussed the quantitative statistics of the sample cloud, giving the basic ideas and realization of functions of count. The algorithm is run in MATLAB, the simulation results not only count the same, but also operation is very simple, time-saving when it compares to the traditional manual. Finally, according to the current situation of sample cloud Reflectance measurement, this paper combined image processing techniques and mathematical theory organically, established mathematical relationships between the reflectivity, visual light and infrared brightness, achieving the purpose of the sample cloud classification.
In addition, satellite nephogram exists more or less "blind area" of cloud cover causing information lost which harm nephogram analysis and interpretation. This brings up a second question in this paper to solve——calculating clouds area. Making the image of N*M element in array into a set of blocks made of many non-overlapping regions. Using image processing technology will improve Freeman chain code algorithm, only judgying the boundary points of the chain code, achieving the rapid filling of the enclosed area in order to calculate the area of irregular clouds automatically. The experimental results show that the computing time-consuming small, accuracy achieves about±1%.
In short, this paper presents effective and feasible solution for nephogram analysis and clouds area measurement by using image processing technology. The findings is based on previous theories, optimized and improved by this paper, providing new ideas for future theory study and application.
引文
[1]赖剑煌,冯国灿(译).数字图像处理疑难解释[M]北京：机械工业出版社,2005,4.
[2]周伟,李万彪.利用GMS-5红外数据进行云的分类识别[J].北京大学学报：自然科学版,2003,39(1)：83-90.
[3]Arking. A.,1964, Latitude distribution of cloud cover from Tiros-3 photographs. Science, 143,569-572.
[4]Miller. D. B., and R. Q. Feddes,1971, Global atlas of relative cloud cover, 1967-1970. United States Air Foree. Dept. of Commerce. Washington. DC,237pp.
[5]Saunders, R. W., K. T. Kriebel,1988, An improved method for detecting clear sky and cloudy radiances from SVHRR data. Int. J. Remote Sensing,9, I,123-150.
[6]Puri, K., and N. E. Davidson,1992, The use of infared satellite cloud imagery data as proxy data for moisture and diabatic heating in data assimilation. Mon. Wea. Rev.,120, 2329-2341.
[7]Shenk, R. T. Holub and R. A. Neff:A multispectral cloud type identification method developed for tropical ocean area with Nimbus-3 MRIR measurements. Mon. Wea. Rev, 284-291.1976.
[8]Coakley, J. A., and Breatherton, F. P.,2006, Cloud coverf Rom high resolution scanner data:Decting and allowing fior partially filled fields of view. J. Geophys. Res.,87, 4917-4923.
[9]Bankert, R. L.:Cloud classification of AVHR Rimagery in maritime Regionsu sing a probabilistic neural network. J. Appl. Meteor.,33,909-918,1994.
[10]Bezdek J C. A convergence theorem for the fuzzy ISODATA clustering algorithms, IEEE Tran.,1990, PAMI-2:1-7
[11]Jonathan Lee, Ronald c.:A neural network approach to classification, IEEE Transation on geosciences and Remote Sensing. V01.28, No.5,846-855,1990.
[12]Peak, J. E, Tag P M.:Segmentation of satellite imagery using hierarchical Threshol-ding and neural networks. J. Appl. Meteor,33,605-616,1994.
[13]孙杰,李行善,左毅,等.自动测试系统集成技术[M].北京：电子工业出版社, 2004.
[14]Freeman H. Computer Processing of Line-drawing Image[J]. Computer Surveys.1974, 6(1):47-97.
[15]Freeman H. Computer Processing of Line-drawing Image[J]. Computer Surveys.
[16]孙即祥.数字图像处理[M].石家庄：河北教育出版社,1936,6.
[17]王厚大.一种计算任意封闭形状面积的方法[J].南京邮电学院学报,1997.(12).
[18]唐良瑞,马全明,景晓军等.图像处理实用技术[M].北京：化学工业出版社,2002：12-14.
[19]崔屹.数字图像处理技术与应用[M].北京：电子工业出版社,1997.
[20]张波等,基于小波变换的高分辨率快鸟遥感图像薄云去除,遥感信息,2011,3.
[21]高延增,仪表图像输入及预处理技术的研究与应用,广东工业大学,2007.
[22]赵忠明,朱重光,008,遥感图像中薄云的去除方法,环境遥感,Vol.11,No.3,pp.195-199.
[23]赖格英,刘春燕,辜晓青,2009,基于Erdas Imagine的遥测影像去云方法,江西师范大学城市与环境科学学院.
[24]王继光,2007,多光谱静止气象卫星云图的云类判别分析与短时移动预测,国防科学技术大学信息与通信工程学院.
[25]程明德.图像识别导论[M].上海：上海科学技术出版社,2003.
[26]Xue Yong. Several linear and nonlinear feature extraction and face recognition[D]. Nan Jing:Nanj ing University of Seienee,2004.
[27]齐兴敏,刘冠梅.PCA在人脸识别方法中的比较研究[J].江西蓝天学院学报2008,3(3).
[28]张翠平,苏光大.人脸识别技术综述[J].中国图像图形学报,2008(11)：7-16.
[29]WilliamK. Pratt. Digital Image Proeessing(ThirdEdition)IM]. New York:JohnWiley and Sons, Ine,2007.
[30]赵春晖,张兆礼,梅晓丹.现代图像处理技术及MATLAB实现[M].北京：人民邮电出版社,2001：269-285.
[31]舒俭,赵敏.基于K-L变换的人脸识别系统[J].华东交通大学报,2006,23(5)：70-74.
[32]郝石生等译.石油形成和分布[M].北京：石油工业出版社,1982：308-310.
[33]Milan Sonka, Vaelav Hlavae.Image Proeessing,Analysis and Machine Vision (Seeond Edition)[M].北京：人民邮电出版社,2002.
[34]查金水,陈鼎才,王定成.基于机器视觉的现实叶片面积测量方法的研究[J].计算机应用,2006,26(5)：1226-1228.
[35]张田文,赵鹏,浦昭邦.基于动态轮廓线的面积测量新方法[J].中国激光,2009,32(11)：1543-1548.
[36]葛伟华,陈优广.基于边界跟踪的区域面积计算[J].计算机应用与软件,2008,25(6)：239-274.
[37]汪建根.皮革和毛皮的现代生产设备[M].西安：陕西科学技术出版社,2001,333-337.
[38]顾国庆,李国强,张薇.用顶点链编码计算区域面积的方法[J].上海理工大学学报,2008,25(3)：267-270.
[39]Shih F Y Wong WT. A One-pass Algorithm for Local Symmetry of Contours from Chain Code. Pattern Recognition[J].32(7):1203-1210.
[40]Zingaretti P, Gasparroni M, Vecci L. Fast Chain Codingo fR egion BouNdaries[J]. IEEE Transactions on Pattern Analysisand Machine Intelligence.20(4):407-414.
[41]Koplowitz. J, Plante. S. Corener Detection of Chain Code Curves. Pattern recognition [J].28(6):843-842.
[42]Yuan J, Suen C Y. An Optimal O(n)Algorithm for Identifying Line Segments from a Sequence of Chain Codes [J]. Pattern Recognition.28(4):634-644.
[43]Bfibiesca E. A Geometric Structure for Two-dimensional Shapes and Three dimensional Surfaces[J]. Pattern Recognition.24(4):483-496.
[44]Bribiesca E. A New Chain Code [J]. Pattern Recognition.32(2):234-241.
[45]Bribiesca E, Guzman A. How to Describe Pure Form and How to Measure Differences in Shapes Using Shape Numbers[J]. Pattern Recognition.12(1):101-112.
[46]J. R. Parker. Algorithmsfor image Proeessing and computer Vision, first edition. John Wiley&Sons, Inc. New York, NY, USA,2009.
[47]左东广,侯志强,韩崇昭等.灰度图像边缘检测算法综述[J].西安交通大学学报, 2008,37(4)：439-440.
[2]周伟,李万彪.利用GMS-5红外数据进行云的分类识别[J].北京大学学报：自然科学版,2003,39(1)：83-90.
[3]Arking. A.,1964, Latitude distribution of cloud cover from Tiros-3 photographs. Science, 143,569-572.
[4]Miller. D. B., and R. Q. Feddes,1971, Global atlas of relative cloud cover, 1967-1970. United States Air Foree. Dept. of Commerce. Washington. DC,237pp.
[5]Saunders, R. W., K. T. Kriebel,1988, An improved method for detecting clear sky and cloudy radiances from SVHRR data. Int. J. Remote Sensing,9, I,123-150.
[6]Puri, K., and N. E. Davidson,1992, The use of infared satellite cloud imagery data as proxy data for moisture and diabatic heating in data assimilation. Mon. Wea. Rev.,120, 2329-2341.
[7]Shenk, R. T. Holub and R. A. Neff:A multispectral cloud type identification method developed for tropical ocean area with Nimbus-3 MRIR measurements. Mon. Wea. Rev, 284-291.1976.
[8]Coakley, J. A., and Breatherton, F. P.,2006, Cloud coverf Rom high resolution scanner data:Decting and allowing fior partially filled fields of view. J. Geophys. Res.,87, 4917-4923.
[9]Bankert, R. L.:Cloud classification of AVHR Rimagery in maritime Regionsu sing a probabilistic neural network. J. Appl. Meteor.,33,909-918,1994.
[10]Bezdek J C. A convergence theorem for the fuzzy ISODATA clustering algorithms, IEEE Tran.,1990, PAMI-2:1-7
[11]Jonathan Lee, Ronald c.:A neural network approach to classification, IEEE Transation on geosciences and Remote Sensing. V01.28, No.5,846-855,1990.
[12]Peak, J. E, Tag P M.:Segmentation of satellite imagery using hierarchical Threshol-ding and neural networks. J. Appl. Meteor,33,605-616,1994.
[13]孙杰,李行善,左毅,等.自动测试系统集成技术[M].北京：电子工业出版社, 2004.
[14]Freeman H. Computer Processing of Line-drawing Image[J]. Computer Surveys.1974, 6(1):47-97.
[15]Freeman H. Computer Processing of Line-drawing Image[J]. Computer Surveys.
[16]孙即祥.数字图像处理[M].石家庄：河北教育出版社,1936,6.
[17]王厚大.一种计算任意封闭形状面积的方法[J].南京邮电学院学报,1997.(12).
[18]唐良瑞,马全明,景晓军等.图像处理实用技术[M].北京：化学工业出版社,2002：12-14.
[19]崔屹.数字图像处理技术与应用[M].北京：电子工业出版社,1997.
[20]张波等,基于小波变换的高分辨率快鸟遥感图像薄云去除,遥感信息,2011,3.
[21]高延增,仪表图像输入及预处理技术的研究与应用,广东工业大学,2007.
[22]赵忠明,朱重光,008,遥感图像中薄云的去除方法,环境遥感,Vol.11,No.3,pp.195-199.
[23]赖格英,刘春燕,辜晓青,2009,基于Erdas Imagine的遥测影像去云方法,江西师范大学城市与环境科学学院.
[24]王继光,2007,多光谱静止气象卫星云图的云类判别分析与短时移动预测,国防科学技术大学信息与通信工程学院.
[25]程明德.图像识别导论[M].上海：上海科学技术出版社,2003.
[26]Xue Yong. Several linear and nonlinear feature extraction and face recognition[D]. Nan Jing:Nanj ing University of Seienee,2004.
[27]齐兴敏,刘冠梅.PCA在人脸识别方法中的比较研究[J].江西蓝天学院学报2008,3(3).
[28]张翠平,苏光大.人脸识别技术综述[J].中国图像图形学报,2008(11)：7-16.
[29]WilliamK. Pratt. Digital Image Proeessing(ThirdEdition)IM]. New York:JohnWiley and Sons, Ine,2007.
[30]赵春晖,张兆礼,梅晓丹.现代图像处理技术及MATLAB实现[M].北京：人民邮电出版社,2001：269-285.
[31]舒俭,赵敏.基于K-L变换的人脸识别系统[J].华东交通大学报,2006,23(5)：70-74.
[32]郝石生等译.石油形成和分布[M].北京：石油工业出版社,1982：308-310.
[33]Milan Sonka, Vaelav Hlavae.Image Proeessing,Analysis and Machine Vision (Seeond Edition)[M].北京：人民邮电出版社,2002.
[34]查金水,陈鼎才,王定成.基于机器视觉的现实叶片面积测量方法的研究[J].计算机应用,2006,26(5)：1226-1228.
[35]张田文,赵鹏,浦昭邦.基于动态轮廓线的面积测量新方法[J].中国激光,2009,32(11)：1543-1548.
[36]葛伟华,陈优广.基于边界跟踪的区域面积计算[J].计算机应用与软件,2008,25(6)：239-274.
[37]汪建根.皮革和毛皮的现代生产设备[M].西安：陕西科学技术出版社,2001,333-337.
[38]顾国庆,李国强,张薇.用顶点链编码计算区域面积的方法[J].上海理工大学学报,2008,25(3)：267-270.
[39]Shih F Y Wong WT. A One-pass Algorithm for Local Symmetry of Contours from Chain Code. Pattern Recognition[J].32(7):1203-1210.
[40]Zingaretti P, Gasparroni M, Vecci L. Fast Chain Codingo fR egion BouNdaries[J]. IEEE Transactions on Pattern Analysisand Machine Intelligence.20(4):407-414.
[41]Koplowitz. J, Plante. S. Corener Detection of Chain Code Curves. Pattern recognition [J].28(6):843-842.
[42]Yuan J, Suen C Y. An Optimal O(n)Algorithm for Identifying Line Segments from a Sequence of Chain Codes [J]. Pattern Recognition.28(4):634-644.
[43]Bfibiesca E. A Geometric Structure for Two-dimensional Shapes and Three dimensional Surfaces[J]. Pattern Recognition.24(4):483-496.
[44]Bribiesca E. A New Chain Code [J]. Pattern Recognition.32(2):234-241.
[45]Bribiesca E, Guzman A. How to Describe Pure Form and How to Measure Differences in Shapes Using Shape Numbers[J]. Pattern Recognition.12(1):101-112.
[46]J. R. Parker. Algorithmsfor image Proeessing and computer Vision, first edition. John Wiley&Sons, Inc. New York, NY, USA,2009.
[47]左东广,侯志强,韩崇昭等.灰度图像边缘检测算法综述[J].西安交通大学学报, 2008,37(4)：439-440.