基于决策级数据融合的CHRIS高光谱图像分类方法研究
|
摘要
本文选取8种常用的遥感图像监督分类方法,综合利用多数投票方法与先验分类概率信息,设计了3组决策级融合分类策略,在应用地物丰富的黄河口湿地CHRIS高光谱图像数据进行分类实验的基础上,开展了不同融合策略决策分类结果的比较研究,分析了融合策略对高光谱图像总体分类精度以及不同地物分类精度的影响。结果表明:(1)3组决策级融合策略总体分类精度均大于85%,Kappa系数都优于0.77,较8种单分类器均有提高,表明决策级融合是利用已有分类器资源、提升高光谱图像分类精度的一种途径;(2)策略3的分类表现最好,即维持最小虚警率的前提下降低错分可以提高分类精度,其中芦苇、滩涂地物同时保持了很高的生产者精度和用户精度,均接近90%,水体在保持较高用户精度的情形下生产者精度有大幅提升,增幅达11%;(3)每组策略综合考虑生产者精度和用户精度的情形下分类精度最高,较仅考虑生产者精度或用户精度的情形提升了1~2个百分点,表明设计算法时应同时考虑减少错分和漏分的情况。
In this paper, three groups of strategies of decision-level data fusion classification are designed using eight classical remote sensing image supervised classification algorithms and the dint of vote method and priori classification probability information. On the base of classification experiment of CHRIS hyperspectral image locating in Yellow River Estuary wetland, the influence on the overall accuracy or the accuracy of each object with different decision-level data fusion classification algorithm for hyperspectral image are analyzed, and the main conclusions are given below:(1) All the overall accuracies and Kappa coefficients of three groups strategies of decision-level data fusion classification are greater than 85% and 0.77, respectively, and they are superior to eight classical supervised classifiers. It shows that decision-level data fusion is an effective method for improving hyperspectral image classification accuracy base on the existing resources of classifier.(2) Classification accuracy of strategy III has the best performance, and the producer accuracy and user accuracy of phragmites austrialis and tidal flat are both close to 90%, while those of water are improved significantly with an increase of 11%. It indicates that reducing commission error can improve classification accuracy in maintaining minimum false alarm rate.(3) For all strategies of decision-level data fusion classification, the classification accuracy is maximal in the case of considering producer accuracy and user accuracy simultaneously. Compared with the cases of considering producer or user accuracy only, the overall accuracies increase 1%~2%. It shows that reducing commission error and omission error should be emphasized all together when the decision-level data fusion algorithm is designed.
In this paper, three groups of strategies of decision-level data fusion classification are designed using eight classical remote sensing image supervised classification algorithms and the dint of vote method and priori classification probability information. On the base of classification experiment of CHRIS hyperspectral image locating in Yellow River Estuary wetland, the influence on the overall accuracy or the accuracy of each object with different decision-level data fusion classification algorithm for hyperspectral image are analyzed, and the main conclusions are given below:(1) All the overall accuracies and Kappa coefficients of three groups strategies of decision-level data fusion classification are greater than 85% and 0.77, respectively, and they are superior to eight classical supervised classifiers. It shows that decision-level data fusion is an effective method for improving hyperspectral image classification accuracy base on the existing resources of classifier.(2) Classification accuracy of strategy III has the best performance, and the producer accuracy and user accuracy of phragmites austrialis and tidal flat are both close to 90%, while those of water are improved significantly with an increase of 11%. It indicates that reducing commission error can improve classification accuracy in maintaining minimum false alarm rate.(3) For all strategies of decision-level data fusion classification, the classification accuracy is maximal in the case of considering producer accuracy and user accuracy simultaneously. Compared with the cases of considering producer or user accuracy only, the overall accuracies increase 1%~2%. It shows that reducing commission error and omission error should be emphasized all together when the decision-level data fusion algorithm is designed.
引文
[1]Xu L,Krzyzak A,Suen C Y.Methods of combining multiple classifiers and their applications to handwriting recognition[J].IEEE Transaction on Systems,Man,Cybernetics,1992,22(3):418-435.
[2]Freund Y.Boosting a weak learning algorithm by majority[J].Information and Computation,1995,121(2):256-285.
[3]Breiman L.Bagging predictor[J].Machine Learning,1996,24(2):123-140.
[4]Kittler J,Hatef M,Duin R,et al.On combining classifiers[J].IEEE Trans On Pattern Analysis and Machine Intelligence,1998,20(3):226-239.
[5]Lin X F,Yacou B,Sherif B J,et al.Performance analysis of pattern classifier combination by plurality voting[J].Pattern Recognition Letter,2003,24(12):1959-1969.
[6]柏延臣,王劲峰.结合多分类器的遥感数据专题分类方法研究[J].遥感学报,2005,9(5):554-563.
[7]张秀英,冯学智,刘伟.基于多分类器结合的IKONOS影像城市植被类型识别[J].东南大学学报(自然科学版),2007,37(3):399-403.
[8]张丹,杨斌,张瑞禹.多种分类器融合的遥感影像分类[J].遥感信息,2009,5:41-43.
[9]孙灏,杜培军,赵卫常.基于多分类器组合的高分辨率遥感影像目标识别[J].地理与地理信息科学,2009,25(1):33-35.
[10]贾永红,李德仁.多源遥感影像像素级融合分类与决策级分类融合法的研究[J].武汉大学学报(信息科学版),2001,26(5):431-434.
[11]Fauvel M,Chanussot J,Benediktsson J A.Decision fusion for the classification of urban remote sensing images[J].IEEE Transactions on Geoscience and Remote Sensing,2006,44(10):2828-2838.
[12]彭正林,毛先成,刘文毅,等.基于多分类器组合的遥感影像分类方法研究[J].国土资源遥感,2011,2:19-25.
[13]Briem G J,Benediktsson J A,Sveinsson J R.Boosting,bagging,and consensus based classification of multisource remote sensing data[C]//The Second International Workshop on Multiple Classifier Systems,Cambridge,2001.
[14]许凯,秦昆,杜鹢.利用决策级融合进行遥感影像分类[J].武汉大学学报(信息科学版),2009,34(7):826-829.
[15]宫大鹏,李之歆,韩建峰,杨哲海,张雅争.高光谱影像分类及组合分类器的应用[J].海洋测绘,2006,26(5):20-23.
[16]Licciardi G,Pacifici F,Tuia D,et al.Decision fusion for the classification of hyperspectral data:Outcome of the 2008 GRS-S data fusion contest[J].IEEE Transactions on Geoscience and Remote Sensing,2009,47(11):3857-3865.
[17]Hemanth R K,Lori M B.Decision-Level fusion of spectral reflectance and derivative information for robust hyperspectral land cover classification[J].IEEE Transactions on Geoscience and Remote Sensing,2010,48(11):4047-4058.
[18]He Y,Qian D,Ben M.Decision fusion on supervised and unsupervised classifiers for hyperspectral imagery[J].IEEE Transactions on Geoscience and Remote Sensing Letters,2010,7(4):875-879.
[2]Freund Y.Boosting a weak learning algorithm by majority[J].Information and Computation,1995,121(2):256-285.
[3]Breiman L.Bagging predictor[J].Machine Learning,1996,24(2):123-140.
[4]Kittler J,Hatef M,Duin R,et al.On combining classifiers[J].IEEE Trans On Pattern Analysis and Machine Intelligence,1998,20(3):226-239.
[5]Lin X F,Yacou B,Sherif B J,et al.Performance analysis of pattern classifier combination by plurality voting[J].Pattern Recognition Letter,2003,24(12):1959-1969.
[6]柏延臣,王劲峰.结合多分类器的遥感数据专题分类方法研究[J].遥感学报,2005,9(5):554-563.
[7]张秀英,冯学智,刘伟.基于多分类器结合的IKONOS影像城市植被类型识别[J].东南大学学报(自然科学版),2007,37(3):399-403.
[8]张丹,杨斌,张瑞禹.多种分类器融合的遥感影像分类[J].遥感信息,2009,5:41-43.
[9]孙灏,杜培军,赵卫常.基于多分类器组合的高分辨率遥感影像目标识别[J].地理与地理信息科学,2009,25(1):33-35.
[10]贾永红,李德仁.多源遥感影像像素级融合分类与决策级分类融合法的研究[J].武汉大学学报(信息科学版),2001,26(5):431-434.
[11]Fauvel M,Chanussot J,Benediktsson J A.Decision fusion for the classification of urban remote sensing images[J].IEEE Transactions on Geoscience and Remote Sensing,2006,44(10):2828-2838.
[12]彭正林,毛先成,刘文毅,等.基于多分类器组合的遥感影像分类方法研究[J].国土资源遥感,2011,2:19-25.
[13]Briem G J,Benediktsson J A,Sveinsson J R.Boosting,bagging,and consensus based classification of multisource remote sensing data[C]//The Second International Workshop on Multiple Classifier Systems,Cambridge,2001.
[14]许凯,秦昆,杜鹢.利用决策级融合进行遥感影像分类[J].武汉大学学报(信息科学版),2009,34(7):826-829.
[15]宫大鹏,李之歆,韩建峰,杨哲海,张雅争.高光谱影像分类及组合分类器的应用[J].海洋测绘,2006,26(5):20-23.
[16]Licciardi G,Pacifici F,Tuia D,et al.Decision fusion for the classification of hyperspectral data:Outcome of the 2008 GRS-S data fusion contest[J].IEEE Transactions on Geoscience and Remote Sensing,2009,47(11):3857-3865.
[17]Hemanth R K,Lori M B.Decision-Level fusion of spectral reflectance and derivative information for robust hyperspectral land cover classification[J].IEEE Transactions on Geoscience and Remote Sensing,2010,48(11):4047-4058.
[18]He Y,Qian D,Ben M.Decision fusion on supervised and unsupervised classifiers for hyperspectral imagery[J].IEEE Transactions on Geoscience and Remote Sensing Letters,2010,7(4):875-879.