为改善面向对象的SPOT5遥感图像森林分类精度，在多分类器结合的投票法、Bayesian平均法和模糊法的基础上提出综合保守投票法和模糊法的多分类器结合新方法——投票模糊法。对最小距离、马氏距离、Bayes准则、模糊逻辑和支持向量机5个分类器结合的试验检测结果表明：采用投票模糊进行分类器结合，总体分类精度和Kappa系数均比投票法、Bayesian法和模糊法的高，也略高于分类效果最好的单个分类器——Bayes分类器，且各类型间生产者精度的差异减小。但分类器结合效果不很明显，其主要原因可能是各分类器采用同一套训练样本，分类器输出结果之间存在较高的关联性，并且所分类型较多。因此，在实际应用中，应尽可能确保各个分类器训练样本的差异性，或者尽可能地避免每个分类器都采用相同的对象特征。
Abstract
Aimed toward improve the accuracy of objectoriented classification using SPOT5 imagery, three multiclassifier combination methods, include voting rule, Bayesian mean and fuzzy fusion rule, were discussed and a new fusion approach named votingfuzzy rule was developed, which synthesized conservative voting rule and fuzzy fusion rule. Five classifiers include minimums distance, Mahalanobis distance, Bayes rule, fuzzy logic and support vector machine, were involved in the combination. The result indicated that the votingfuzzy rule had higher total accuracy and Kappa index than three other combination rules, and also the Bayes rule, the best single classifier in all classifiers; furthermore, it reduced the difference of producer accuracy between classes. However, the combination effect wasn’t as obvious as indicated in literatures. The reason might owe to the high correlativity in the outputs of five classifiers, for them shared with a sample set, and the protocol with twenty two classes. Thus classifiers shouldn’t be trained with a same training sample set, or might select difference object features in practical application.
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1]Lu D, Weng Q. A survey of image classification methods and techniques of improving classification performance[J]. International Journal of Remote Sensing, 2007, 38(5-6): 823-870.
[2]Tso B, Mather P M. Classification Methods for Remotely Sensed Data[M]. New York: Taylor and Francis Inc, 2001.
[3]Franklin J, Phinn S R, Woodcock C E, et al. Rationale and conceptual framework for classification approaches to assess forest resources and properties[G]// Wulder M, Franklin S. Methods and Applications for Remote Sensing of Forests: Concepts and Case Studies. Boston: Kluwer Academic Publishers, 2003.
[4]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.
[5]Pal M, Mather P M. An assessment of the effectiveness of decision tree methods for land cover classification[J]. Remote Sensing of Environment, 2003, 86(4): 554-565.
[6]Lu D, Weng Q. Spectral mixture analysis of the urban landscapes in Indianapolis with Landsat ETM+imagery[J]. Photogrammetric Engineering and Remote Sensing, 2007, 70(9): 1053-1062.
[7]柏延臣,王劲峰. 结合多分类器的遥感专题数据专题分类方法研究[J]. 遥感学报,2005,9(5):555-563.
[8]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.
[9]Benediktsson J A, Kanellopovlos I. Classification of multisource and hyperspectral data based on decision fusion[J]. IEEE Transactions on Geosciences and Remote Sensing, 1999, 37(31): 1367-1377.
[10]Warrender C E, Augusteijn M F. Fusion of image classification using Bayesian techniques with Markov rand infields[J]. International Journal of Remote Sensing, 1999, 20(10): 1987-2002.
[11]Steele B M. Combining multiple classifiers: an application using spatial and remotely sensed information for land cover type mapping[J]. Remote Sensing of Environment, 2000, 74(3): 545-556.
[12]Huang Z, Lees B G. Combining nonparametric models for multisource predictive for forest mapping[J]. Photogrammetric Engineering and Remote Sensing, 2004, 70(4): 415-425.
[13]Liu W, Gopal S, Woodcoock C E. Uncertainty and confidence in land cover classification using a hybrid classifier approach[J]. Photogrammetric Engineering and Remote Sensing, 2004, 70(8): 963-972.
[14]Xu L, Krzyzak A, Suenc C Y. Methods of combining multiple classifiers and their applications to handwriting recognition[J]. IEEE Transactionon Systems, Man, and Cybernetics, 1992, 22(3): 418-435.
[15]Ghosh J, Tumer K, Beck S, et al. Integration of neural classifiers for passive sonar signals[G]//Leondes C T. Control and Dynamic Systems—Advances in Theory and Applications. Londen: Academic Press, 1996.
[16]Cappelli R, Majo D, Maltoni D. Combining fingerprint classifiers[G]//Kittler Roli. Multiple Classifiers System. Berlin: Springer, 2000.
[17]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):430-434.
[18]Oussalah M. Study of some algebraical properties of adaptive combination rules[J]. Fuzzy Set and System, 2000, 114(2): 391-409.
[19]李崇贵,李春干. 森林资源调查林区GPS控制网的试验研究[J]. 林业科学,2005,41(1):19-24.
[20]杨丽英,覃征,张选平. 分类器模拟算法及其应用[J]. 西安交通大学学报,2005,39(12):1311-1314.
基金
收稿日期：2009-03-30修回日期：2009-10-26基金项目：国家自然科学基金项目(30872023)作者简介：李春干(1962—)，研究员，博士。Email: gxali@126.com。引文格式：李春干,邵国凡. 面向对象森林分类的多分类器结合方法研究[J]. 南京林业大学学报:自然科学版,2010,34(1):73-79.
