光谱特征提取算法改进及在溢油图像中的应用
|
摘要
高光谱遥感技术作为一种新的遥感技术,受到越来越多的关注。其中主要的原因是：高光谱遥感技术能够探测到在宽波段中不可测的物质,而且高光谱遥感技术还将反映地物属性的光谱与直观的几何图像联系起来,使人们可以直观地“感受”到光谱的特征。于是,对地物的分析也转换为对光谱曲线的相似性和差异性的分析。而光谱曲线的特征提取是光谱曲线分析前提。因此,高效、准确地提取出光谱曲线的特征显得尤为重要。
本文正是基于以上因素对光谱曲线的特征提取算法进行深入地研究。将模式识别中的曲线树算法引入到光谱曲线的特征提取中,并在深入理解曲线树算法的基础上对其进行改进,使其更适用于光谱曲线的特征提取。
先通过实验室光谱仪采集了6种地物的光谱曲线。然后应用二值编码、曲线树及其改进算法对其进行特征提取和分类,验证曲线树改进算法的有效性。同时通过对特定曲线做平移、拉伸及旋转等操作,形成一组新的曲线。再应用三种算法提取这组曲线的特征,之后比较他们之间的欧氏距离。证明曲线树算法对于曲线的平移、旋转、拉伸具有良好的不变性。
最后将曲线树及其改进算法应用到实际的高光谱图像中去。先对原始的高光谱图像用对数残差和内部平均法进行反射率反演；然后应用本文中的各种曲线特征提取算法对其进行曲线的特征提取,最后对图像进行分类。通过对实验结果的分析得出改进的曲线树算法对较为“纯”物质的分类的效果较好,而对油水混合物的分类的效果有待提高的结论。
Hyperspectral remote sensing technology draws the more attention of the public as a new remote sensing technology. The main reason is:Hyperspectral remote sensing technology has a higher spectral resolution and broader spectral imaging range. That makes it possible to detect surface features characteristic that can't be detected in the ordinary remote sensing conditions. And it also makes spectral curve contact with geometry, so that people can intuitively feel the characteristics of the spectrum. Therefore, the surface features analysis also converted to the analysis of the similarities and differences in the spectral curve. The feature extraction of the spectral curve is the premise of spectroscopy. Therefore, it's important to extract the characteristics of spectral curve efficiently and accurately.
So this article in-depth studies on the spectral curve feature extraction algorithm. And the algorithm of curve tree which is in the pattern recognition is introduced to the feature extraction of the spectral curve. Then improve the curve tree algorithm by in-depth studying it, for making it more suitable for the characteristics of the spectral curve extraction.
Firstly, spectral curves of six groups of surface features were collected by laboratory spectrometer. Binary encoding algorithm, curve tree algorithm and its improved algorithm were used to classify. And some new the curve were created by processing the certain curve. So a new set of curves were formed. Then Euclidean distances between them were calculated, after the characteristics of the curves were extracted. That proves that curve tree algorithm has invariance for the translation, rotation, stretching of curve.
Finally, the curve tree algorithm and its improved algorithm were applied to the hyperspectral oil spill image. And the original hyperspectral images were inversed by the algorithm of log-residuals and internal average. Then the characteristics of the curve were extracted by the algorithms in this paper. Finally, the oil spills in the image were classified. It can be seen from result that the classification results of improved curve tree algorithm are better for the "pure" substances, and the results are bad for the situation of oil-water mixture.
本文正是基于以上因素对光谱曲线的特征提取算法进行深入地研究。将模式识别中的曲线树算法引入到光谱曲线的特征提取中,并在深入理解曲线树算法的基础上对其进行改进,使其更适用于光谱曲线的特征提取。
先通过实验室光谱仪采集了6种地物的光谱曲线。然后应用二值编码、曲线树及其改进算法对其进行特征提取和分类,验证曲线树改进算法的有效性。同时通过对特定曲线做平移、拉伸及旋转等操作,形成一组新的曲线。再应用三种算法提取这组曲线的特征,之后比较他们之间的欧氏距离。证明曲线树算法对于曲线的平移、旋转、拉伸具有良好的不变性。
最后将曲线树及其改进算法应用到实际的高光谱图像中去。先对原始的高光谱图像用对数残差和内部平均法进行反射率反演；然后应用本文中的各种曲线特征提取算法对其进行曲线的特征提取,最后对图像进行分类。通过对实验结果的分析得出改进的曲线树算法对较为“纯”物质的分类的效果较好,而对油水混合物的分类的效果有待提高的结论。
Hyperspectral remote sensing technology draws the more attention of the public as a new remote sensing technology. The main reason is:Hyperspectral remote sensing technology has a higher spectral resolution and broader spectral imaging range. That makes it possible to detect surface features characteristic that can't be detected in the ordinary remote sensing conditions. And it also makes spectral curve contact with geometry, so that people can intuitively feel the characteristics of the spectrum. Therefore, the surface features analysis also converted to the analysis of the similarities and differences in the spectral curve. The feature extraction of the spectral curve is the premise of spectroscopy. Therefore, it's important to extract the characteristics of spectral curve efficiently and accurately.
So this article in-depth studies on the spectral curve feature extraction algorithm. And the algorithm of curve tree which is in the pattern recognition is introduced to the feature extraction of the spectral curve. Then improve the curve tree algorithm by in-depth studying it, for making it more suitable for the characteristics of the spectral curve extraction.
Firstly, spectral curves of six groups of surface features were collected by laboratory spectrometer. Binary encoding algorithm, curve tree algorithm and its improved algorithm were used to classify. And some new the curve were created by processing the certain curve. So a new set of curves were formed. Then Euclidean distances between them were calculated, after the characteristics of the curves were extracted. That proves that curve tree algorithm has invariance for the translation, rotation, stretching of curve.
Finally, the curve tree algorithm and its improved algorithm were applied to the hyperspectral oil spill image. And the original hyperspectral images were inversed by the algorithm of log-residuals and internal average. Then the characteristics of the curve were extracted by the algorithms in this paper. Finally, the oil spills in the image were classified. It can be seen from result that the classification results of improved curve tree algorithm are better for the "pure" substances, and the results are bad for the situation of oil-water mixture.
引文
[1]Jia X P, Richards J A. Efficient transmission and classification of hyperspectral image data[J]. IEEE Transactions on Geoscience and Remote Sensing.2003, 41(5):1129-1131.
[2]童庆禧,张兵,郑兰芬.高光谱遥感的多学科应用[M].北京：电子工业出版社2006.5.
[3]浦瑞良,宫鹏.高光谱遥感技术及其应用[M].北京：高等教育出版社,2000.8.
[4]Landgrebe D A. Information Processing for Remote Sensing[M], [s.1.]:the World Scientific Publishing,2000.
[5]杨倩倩.高光谱溢油图像特征提取在油种识别中的应用[D]：(硕士学位论文).辽宁：大连海事大学,2009.
[6]臧影.高光谱溢油图像波段选择在油膜厚度估算中的应用[D]：(硕士学位论文).辽宁：大连海事大学,2009.
[7]阮建武,邢立新.遥感数字图像的大气辐射校正应用研究[J].遥感技术与应用.2004,19(3)：206-208.
[8]王天星,阎广建,任华忠等.高光谱传感器光谱性能参数反演与反射率恢复[J].光谱学与光谱分析.2010,30(10)：2714-2718.
[9]张永宁,丁倩,李栖筠.海上溢油污染遥感监测的研究[J].大连海事大学学报.1999,25(3)：1-5.
[10]张永宁,丁倩,高超等.油膜波谱特征分析与遥感监测溢油[J].海洋环境科学.2000,19(3)：5-10.
[11]赵冬至,丛丕福.海面溢油的可见光波段地物光谱特征研究[J].遥感技术与应用.2000,15(3)：160-164.
[12]陆应诚,田庆久,王晶晶等.海面油膜光谱响应实验研究[J].科学通报.2008,53(9)：1086-1088.
[13]陆应诚,田庆久,齐小平等.海面油膜高光谱遥感信息提取[J].光谱学与光谱分析2008.
[14]陆应诚,田庆久,齐小平等.海面甚薄油膜光谱响应研究与分析[J].光谱学与光谱分析.2009,29(4)：986-989.
[15]朱绍攀,陈宇.大气辐射校正方法分析[J].地理空间信息.2010,8(1)：113-116.
[16]徐长健.基于小波变换的高光谱溢油图像压缩方法的研究[D]：(硕士学位论文).辽宁：大 连海事大学,2010.
[16]Ouimet M, Bengio Y. Sparse Greedy Spectral Clustering and Kernel PCA[R]. Montreal:IRIS Machine Learning Workshop,2004.
[17]Bachmann C M, Ainsworth T L, Fusina R A. Exploiting manifold geometry in hyperspectral imagery[J]. IEEE Transactions on Geoscience and Remote Sensing.2005, 43(3):441-454.
[18]辛宪会,赵泳,翟辉琴等.面向分类的高光谱遥感影像数据特性的研究[J].海洋测绘.2005,25(6)：22-25.
[19]范冬娟,张韶华,高光谱影像反射率反演方法的研究[J].海洋测绘.2006,26(3)：28-30.
[20]Shah C A, Arora M K, Robila S A etal. ICA Mixture Model Based Unsupervised Classification of Hyperspectral Imagery[C]. Proceedings of 31st Applied Imagery Pattern Recognition Workshop,2002:29-35.
[21]杜培军,陈云浩,方涛等.高光谱遥感数据光谱特征的提取与应用[J].中国矿业大学学报.2003,32(5)：500-504.
[22]陈文霞,陈安升,蔡之华.基于高光谱吸收特征参数的分类研究[J].计算机工程与应用.2008,44(28)：230-232.
[23]李飞,周成虎,陈荣国.基于光谱曲线形态的高光谱影像检索方法研究[J].光谱学与光谱分析.2008,28(11)：2482-2486.
[24]Lee C, Landgebe D A. Feature Extraction Based on Decision Boundarise[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence,1993,15(4):388-400.
[25]闻兵工,冯伍法,刘伟.基于光谱曲线整体相似性测度的匹配分类[J].测绘科学技术学报.2009,26(2)：128-131.
[26]邵涛.基于光谱信息的高光谱图像目标识别方法的研究[D]：(硕士学位论文).黑龙江：哈尔滨工业大学,2010.
[27]Dundar M M, Landgrebe D A. Toward an optimal supervised classifier for the analysis of hyperspectral data[J]. IEEE Transactions on Geoscience and Remote Sensing.2004,42(1):271-277.
[28]胡倩.基于知识的高光谱图像波普匹配技术研究与应用[D]：(硕士学位论文).北京：中国地质大学,2010.
[29]王志平,杨建峰,薛彬.利用高光谱数据进行地物识别分类研究[J].光子学报.2008,37(3)：561-565.
[30]Hoffbeck J P, Landgrebe D A. Classification of high dimensional multispectral data[D]:(PHD). West Lafayette:Purdue University.1995
[31]李素敏,万燕,曾培峰.N链码对异形纤维特征参数的作用[J].东华大学学报(自然科学版).2008,34(5)：608-613.
[32]吴昊.高光谱遥感图像数据分类技术研究[D].(博士学位论文).安徽：国防科技大学,2004.
[33]谢秋昌.基于高光谱数据的分类技术研究[D].(硕士学位论文).陕西：长安大学,2008
[34]杨国鹏,余旭初,刘伟等.面向高光谱遥感影像的分类方法研究[J].测绘通报.2007,10：17-20.
[35]Dai X L, Khorram S. A feature-based image registration algorithm using improved chain-code representation combined with invariant moment[J]. IEEE Trans. on Geoscience and remote Sensing.1999,37(5):2351-2362.
[36]喜文飞,方源敏,隋玉成等.一种改进的矢量曲线特征点提取方法[J].,江西科学.2011,29(2)：282-284.
[37]Park H J, Ji M S. Shape matching using the modified histogram based on chain code[J]. Proceeding of SPIE.2002,52(83):114-121.
[38]陈孝春,叶懋冬,倪臣敏.基于二叉树的曲线描述方法[J].中国图象图形学报.2007,12(1)：116-120.
[39]陈孝春.二维形状的描述和识别的研究：(硕士学位论文).浙江：浙江大学,2006.
[40]Saghri J, Freeman H. Analysis of the precision of the generalized chain codes for the representation of planar curves[J]. IEEE Trans, on Pattern Analysis and Machine Intelligence.1981,3(5):533-539.
[41]Cosmin G. Distance set for shape filters and shape recognition[J]. IEEE Trans. on Image Processing,2003.12(10):1274-1285.
[42]刘淑娟,周恩辉,张有会等.变长夹角链码及其生成算法研究[J].河北师范大学学报(自然科学版).2010,34(6)：652-655.
[43]Merlin P M, D J Farber. A parallel mechanism for detecting curves in pictures[J]. IEEE Trans. Comput.24(1):96-98.
[44]Zhang J, Zhang X, Krim H. Object representation and recognition in shape spaces[J]. Pattern Recognition.2003,36(5):1143-1154.
[45]Hannu K,Tapio S, Matti P. An Experimental Comparison of Autoregressive and Fourier-Based Descriptors in 2D Shape Classification[J]. IEEE Trans.on Pattern Analysis and Machine Intelligence.1995,17(2):201-207.
[46]Ashok S,Jodi E. Generalized Hough transform for natural shapes[J]. Pattern Recognition.1997,18:473-480.
[47]黄国华.基于相对高度的曲线特征提取算法[J].邵阳学院学报(自然科学版).2011,8(1)：29-33.
[2]童庆禧,张兵,郑兰芬.高光谱遥感的多学科应用[M].北京：电子工业出版社2006.5.
[3]浦瑞良,宫鹏.高光谱遥感技术及其应用[M].北京：高等教育出版社,2000.8.
[4]Landgrebe D A. Information Processing for Remote Sensing[M], [s.1.]:the World Scientific Publishing,2000.
[5]杨倩倩.高光谱溢油图像特征提取在油种识别中的应用[D]：(硕士学位论文).辽宁：大连海事大学,2009.
[6]臧影.高光谱溢油图像波段选择在油膜厚度估算中的应用[D]：(硕士学位论文).辽宁：大连海事大学,2009.
[7]阮建武,邢立新.遥感数字图像的大气辐射校正应用研究[J].遥感技术与应用.2004,19(3)：206-208.
[8]王天星,阎广建,任华忠等.高光谱传感器光谱性能参数反演与反射率恢复[J].光谱学与光谱分析.2010,30(10)：2714-2718.
[9]张永宁,丁倩,李栖筠.海上溢油污染遥感监测的研究[J].大连海事大学学报.1999,25(3)：1-5.
[10]张永宁,丁倩,高超等.油膜波谱特征分析与遥感监测溢油[J].海洋环境科学.2000,19(3)：5-10.
[11]赵冬至,丛丕福.海面溢油的可见光波段地物光谱特征研究[J].遥感技术与应用.2000,15(3)：160-164.
[12]陆应诚,田庆久,王晶晶等.海面油膜光谱响应实验研究[J].科学通报.2008,53(9)：1086-1088.
[13]陆应诚,田庆久,齐小平等.海面油膜高光谱遥感信息提取[J].光谱学与光谱分析2008.
[14]陆应诚,田庆久,齐小平等.海面甚薄油膜光谱响应研究与分析[J].光谱学与光谱分析.2009,29(4)：986-989.
[15]朱绍攀,陈宇.大气辐射校正方法分析[J].地理空间信息.2010,8(1)：113-116.
[16]徐长健.基于小波变换的高光谱溢油图像压缩方法的研究[D]：(硕士学位论文).辽宁：大 连海事大学,2010.
[16]Ouimet M, Bengio Y. Sparse Greedy Spectral Clustering and Kernel PCA[R]. Montreal:IRIS Machine Learning Workshop,2004.
[17]Bachmann C M, Ainsworth T L, Fusina R A. Exploiting manifold geometry in hyperspectral imagery[J]. IEEE Transactions on Geoscience and Remote Sensing.2005, 43(3):441-454.
[18]辛宪会,赵泳,翟辉琴等.面向分类的高光谱遥感影像数据特性的研究[J].海洋测绘.2005,25(6)：22-25.
[19]范冬娟,张韶华,高光谱影像反射率反演方法的研究[J].海洋测绘.2006,26(3)：28-30.
[20]Shah C A, Arora M K, Robila S A etal. ICA Mixture Model Based Unsupervised Classification of Hyperspectral Imagery[C]. Proceedings of 31st Applied Imagery Pattern Recognition Workshop,2002:29-35.
[21]杜培军,陈云浩,方涛等.高光谱遥感数据光谱特征的提取与应用[J].中国矿业大学学报.2003,32(5)：500-504.
[22]陈文霞,陈安升,蔡之华.基于高光谱吸收特征参数的分类研究[J].计算机工程与应用.2008,44(28)：230-232.
[23]李飞,周成虎,陈荣国.基于光谱曲线形态的高光谱影像检索方法研究[J].光谱学与光谱分析.2008,28(11)：2482-2486.
[24]Lee C, Landgebe D A. Feature Extraction Based on Decision Boundarise[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence,1993,15(4):388-400.
[25]闻兵工,冯伍法,刘伟.基于光谱曲线整体相似性测度的匹配分类[J].测绘科学技术学报.2009,26(2)：128-131.
[26]邵涛.基于光谱信息的高光谱图像目标识别方法的研究[D]：(硕士学位论文).黑龙江：哈尔滨工业大学,2010.
[27]Dundar M M, Landgrebe D A. Toward an optimal supervised classifier for the analysis of hyperspectral data[J]. IEEE Transactions on Geoscience and Remote Sensing.2004,42(1):271-277.
[28]胡倩.基于知识的高光谱图像波普匹配技术研究与应用[D]：(硕士学位论文).北京：中国地质大学,2010.
[29]王志平,杨建峰,薛彬.利用高光谱数据进行地物识别分类研究[J].光子学报.2008,37(3)：561-565.
[30]Hoffbeck J P, Landgrebe D A. Classification of high dimensional multispectral data[D]:(PHD). West Lafayette:Purdue University.1995
[31]李素敏,万燕,曾培峰.N链码对异形纤维特征参数的作用[J].东华大学学报(自然科学版).2008,34(5)：608-613.
[32]吴昊.高光谱遥感图像数据分类技术研究[D].(博士学位论文).安徽：国防科技大学,2004.
[33]谢秋昌.基于高光谱数据的分类技术研究[D].(硕士学位论文).陕西：长安大学,2008
[34]杨国鹏,余旭初,刘伟等.面向高光谱遥感影像的分类方法研究[J].测绘通报.2007,10：17-20.
[35]Dai X L, Khorram S. A feature-based image registration algorithm using improved chain-code representation combined with invariant moment[J]. IEEE Trans. on Geoscience and remote Sensing.1999,37(5):2351-2362.
[36]喜文飞,方源敏,隋玉成等.一种改进的矢量曲线特征点提取方法[J].,江西科学.2011,29(2)：282-284.
[37]Park H J, Ji M S. Shape matching using the modified histogram based on chain code[J]. Proceeding of SPIE.2002,52(83):114-121.
[38]陈孝春,叶懋冬,倪臣敏.基于二叉树的曲线描述方法[J].中国图象图形学报.2007,12(1)：116-120.
[39]陈孝春.二维形状的描述和识别的研究：(硕士学位论文).浙江：浙江大学,2006.
[40]Saghri J, Freeman H. Analysis of the precision of the generalized chain codes for the representation of planar curves[J]. IEEE Trans, on Pattern Analysis and Machine Intelligence.1981,3(5):533-539.
[41]Cosmin G. Distance set for shape filters and shape recognition[J]. IEEE Trans. on Image Processing,2003.12(10):1274-1285.
[42]刘淑娟,周恩辉,张有会等.变长夹角链码及其生成算法研究[J].河北师范大学学报(自然科学版).2010,34(6)：652-655.
[43]Merlin P M, D J Farber. A parallel mechanism for detecting curves in pictures[J]. IEEE Trans. Comput.24(1):96-98.
[44]Zhang J, Zhang X, Krim H. Object representation and recognition in shape spaces[J]. Pattern Recognition.2003,36(5):1143-1154.
[45]Hannu K,Tapio S, Matti P. An Experimental Comparison of Autoregressive and Fourier-Based Descriptors in 2D Shape Classification[J]. IEEE Trans.on Pattern Analysis and Machine Intelligence.1995,17(2):201-207.
[46]Ashok S,Jodi E. Generalized Hough transform for natural shapes[J]. Pattern Recognition.1997,18:473-480.
[47]黄国华.基于相对高度的曲线特征提取算法[J].邵阳学院学报(自然科学版).2011,8(1)：29-33.