基于光谱和纹理特征的ALOS影像土地利用信息提取

Study on Information Extraction of Land Use from ALOS Image Based on Spectral and Texture Characteristics

针对高分辨率遥感影像易于反映地物纹理特征的特点,综合利用地物的光谱和纹理特征进行分类,探讨适用于ALOS影像的土地利用信息提取方法。以川东丘陵地区影像为例,基于GLCM提取纹理信息,将提取的纹理特征向量采用赋权值法融合为一个综合纹理信息波段,然后采用面向对象法将其与光谱特征信息共同参与分类。与最大似然法的提取结果对比表明,考虑了纹理特征的面向对象分类方法能明显提高分类精度,Kappa精度提高了0.12;避免了椒盐现象,分割的地类边界具有更好的语义表达,更贴合地物实际分布特征;建筑用地和林地具有明显的纹理特征,而旱地纹理特征不明显。该方法不仅分出了6个基本地物类型,而且对于林地、建筑用地等类型还能进一步细分。

High resolution remote sensing images were rich in texture information. ALOS image was classified by spectral and texture characteristics of the land objects in this paper. It aims to find method suitable for extracting land use information from ALOS images. Taking the images of hills area of East Sichuan as example, the texture information extracted based on GLCM method was fused into one band represented the texture information by weighted stack. And the texture band was classified to- gether with the spectral characteristic information by objects-oriented method. The results showed that the object-oriented clas- sification considered texture information improved the classification results evidently and raised the Kappa accuracy by 0. 12 compared to the maximum likelihood classification. It avoided the phenomenon of salt and pepper noise. The boundaries of the land objects endowed with better semantic representation were more accurately accorded the spatial distribution of the reality. The building land and forest possessed obvious texture characteristic while the dry land possessed less. In addition, not only can this method classified six basis land use types, but also was able to classified more details to the types of forest and building land. It means that texture information can improve the accuracy of image classification. Object-oriented classification using spectral and textural information is suitable for ALOS images.