综合聚类和上下文特征的高光谱影像分类

发布时间：2018-01-14 05:26

本文关键词：综合聚类和上下文特征的高光谱影像分类　出处：《武汉大学学报(信息科学版)》2017年07期 　论文类型：期刊论文

【摘要】：常规高光谱影像逐像素分类往往没有考虑空间相关性,分类结果未体现地物的空间关联和分布特征。为了在分类中充分利用空间特征,利用聚类信息并结合隐马尔可夫随机场模型讨论了高光谱遥感影像光谱-空间分类方法。首先,在不同特征提取方法(最小噪声分离、独立成分分析和主成分分析)下,使用不同聚类方法(k-均值、迭代自组织分析算法和模糊c-均值算法)借助隐马尔可夫随机场获取优化的分割图;然后,采用4连通区域标记法对分割区域标记生成图像对象,并根据支持向量机的逐像素分类结果采用多数投票法对图像对象进行分类;最后,借助凹槽窗口邻域滤波技术改进分类结果,削弱"椒盐"现象。该方法综合了监督分类和非监督分类的优势,通过聚类引入地物空间相关性信息,通过隐马尔可夫随机场引入上下文特征,较好地弥补了单纯基于光谱信息分类的不足。
[Abstract]:Conventional hyperspectral image pixel classification often does not consider spatial correlation, the classification results do not reflect the spatial correlation and distribution of features. In order to make full use of spatial features in the classification. Using the clustering information and the hidden Markov random field model, the spectral spatial classification method of hyperspectral remote sensing image is discussed. Firstly, different feature extraction methods (minimum noise separation) are used. Independent component Analysis (ICA) and Principal component Analysis (PCA), using different clustering methods, iterative self-organization analysis and fuzzy c-means algorithm, to obtain the optimized segmentation graph with the help of hidden Markov random field. Then, the 4-connected region labeling method is used to generate the image objects, and the majority voting method is used to classify the image objects according to the pixel classification results of support vector machine. Finally, we improve the classification results by using the grooved window neighborhood filtering technology, which weakens the "salt and pepper" phenomenon. This method integrates the advantages of supervised classification and unsupervised classification, and introduces spatial correlation information of ground objects through clustering. By introducing context features into hidden Markov random fields, the shortcomings of classification based on spectral information are well remedied.
【作者单位】：天津市地质调查研究院;南京大学卫星测绘技术与应用国家测绘地理信息局重点实验室;南京大学江苏省地理信息技术重点实验室;河海大学地球科学与工程学院;
【基金】：:江苏省杰出青年基金(BK2012018) 国家重大科学仪器设备开发专项(012YQ050250)~~
【分类号】：TP751
【正文快照】： 高光谱影像(hyperspectral image,HSI)具有高光谱分辨率、精细光谱特征、图像-光谱一体化表达的特性,能够提供丰富的地表信息,受到国内外学者的广泛关注[1]。高光谱影像处理内容涵盖影像融合、混合像元分解、分类、目标检测、理化参数估计等方向[2-3]。目前,国内外学者提出了

【相似文献】