联合快舟一号影像纹理信息的城市土地覆盖分类

仅依靠光谱信息无法满足高分辨率遥感分类的应用需求,辅之以纹理特征信息进行分类,可提高影像分类精度。利用KZ-1卫星影像和Landsat-8卫星影像数据,基于面向对象的影像分割法和灰度共生矩阵纹理分析法对新疆石河子市局部城区进行了地表覆盖分类实验,将不同空间分辨率的全色影像纹理信息、光谱信息构成多种影像特征组合进行分类比较研究,以选择最佳的分类特征集。结果表明:KZ-1影像能为城市区域的土地覆盖分类提供丰富的纹理信息,面向对象的影像分割可较好地利用高分辨率数据的几何结构信息实现优化的影像分割,从而提高多光谱影像的分类精度,总体分类精度为90.06%,Kappa系数为87.93%,比单纯利用光谱信息分类的总体精度提高了8.02%,Kappa系数提高了9.65%,表明KZ-1数据可为光谱分类提供丰富的纹理信息,从而提高城市区域的土地覆盖分类精度。

Land Cover Classification in Shihezi City by Combing the KZ-1 Texture and Landsat-8 OLI Spectral Information

Conventional remote sensing image classification is usually based on the spectral information and can’t perform well in the classification of high spatial resolution imagery.This study presents an approach to improve the classification accuracy by combining spectral information with spatial texture features to extract from high spatial resolution bands.The KZ\|1 image and Landsat\|8 image of a portion of Shihezi City,Xinjiang were acquired and preprocessed in the study.Object\|oriented image segmentation and grey\|level co\|occurrence matrix texture analysis were used to create image objects and extract textural features for object\|oriented classification.An optimal window size and threshold values were first determined for the image segmentation operation,and the support vector machine algorithm was used to perform the classification procedure.Eight textural features such as Mean,Variance,Homogeneity,Contrast,Dissimilarity,Entropy,Angular Second Moment,and Correlation were extracted from the KZ\|1 and Lansat\|8 OLI panchromatic bands and used to create several different feature sets to conduct the SVM classifications.Classification results from these various image feature sets indicate that due to its high spatial resolution the KZ\|1 image containing abundant textural information in the study area which can achieve the classification accuracy with overall accuracy of 90.06%,and Kappa coefficient of 87.93%.Compared with conventional spectral classification,the overall accuracy of the textural classification with KZ\|1 imagery is increased by 8.02%,and Kappa coefficient increased by 9.65%.The proposed approach is valuable for combing remotely sensed imagery from different satellite platforms to extract urban expansion information quickly and accurately in Xinjiang.
