利用多属性剖面与双边滤波的高光谱影像分类  

Hyperspectral Image Classification Using Extended Multi-attribute Profiles and Guide Bilateral Filter

作　　者：王扣准 黄睿[1] 

机构地区：[1]上海大学通信与信息工程学院,上海200444

出　　处：《遥感信息》2016年第6期104-109,共6页Remote Sensing Information

摘　　要：随着高光谱影像空间分辨率的提高,"同谱异物"和"同谱异物"问题日益凸显,传统基于纯光谱的地物分类方法已不能满足识别精度要求。结合光谱与空间信息进行高光谱影像分类已成为领域研究热点。鉴于此,该文提出了一种基于扩展多属性剖面(Extended Multi-Attribute Profiles,EMAPs)和指引双边滤波的高光谱影像地物分类方法。首先,基于EMAPs提取高光谱影像4种属性的形态学纹理特征,通过级联纹理与光谱特征获得新特征矢量。其次,对获取的新特征进行降维,并采用指引双边滤波器滤除降维影像的噪声,同时保持影像的边缘信息。最后,采用支持向量机实现分类。高光谱影像分类实验表明,所提方法性能优于多种光谱与空间结合的分类方法。With the increase in spatial resolution of hyperspectral image, the problems of different objects with same spectrum and same objects with different spectrum have attracted more and more attentions. Since the traditional classification methods based on pure spectra cannot receive the satisfactory accuracy, integration of spectra and spatial information for classification has become a hot fieldin the context of remote sensing. A spectral-spatial method based on extended multl-attribute profiles （EMAPs） and guide bilateral filter for hyperspectral image classification is proposed. Firstly, through EMAPs. fourkinds of morphological features are extracted and concatenated into a new vector where textural and spectral information is integrated. Then,after reducing the dimensionality of EMAPs features, the guild bilateral filtering is used to filter the noise of reduced image and enhance the edge information. Finally, support vector machine is applied for classification. Classification experiments on different hyperspectral image sets demonstrate that the proposed method has a better performance than a variety of spectral and spatial combination classification methods.

关 键 词：高光谱影像 扩展多属性剖面 指引双边滤波 纹理特征提取 分类 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]