论文部分内容阅读
高分辨率SAR图像的纹理特性对于图像的解译及地物分类等具有重要的意义。根据高分辨率星载SAR图像上建筑区的纹理有别于其他地物的特点,提出了一种综合利用灰度和纹理特征的高分辨率星载SAR图像建筑区提取方法。首先对SAR图像进行斑点噪声的抑制,然后利用灰度共生矩阵计算出星载SAR图像上建筑区与非建筑区的8种纹理特征统计量,根据巴氏距离进行特征选择,并通过主成分分析去除纹理特征之间的相关性,得到了最佳纹理特征分量,将所选的特征影像与原始图像进行波段组合,利用K均值聚类算法对组合后的图像进行非监督分类;最后通过对分类图像进行后处理并提取外部轮廓,提取了建筑区。以COSMO-SkyMed SAR影像为数据源进行了实验。结果表明该方法能够有效提取高分辨率星载SAR图像中的建筑区,提取效果明显优于未利用纹理特征的方法。
The texture characteristics of high-resolution SAR images have important meanings for image interpretation and feature classification. According to the fact that the texture of building area on high-resolution space-borne SAR image is different from other features, this paper proposes a method to extract the building area of high-resolution spaceborne SAR image by comprehensively using gray-scale and texture features. Firstly, the speckle noise is suppressed in the SAR image. Then, using the gray level co-occurrence matrix, the eight texture feature statistics of the built-in area and the non-built area of the spaceborne SAR image are calculated. The feature selection is performed according to the Pahschule distance, Remove the correlation between the texture features, get the best texture feature components, the selected feature image and the original image band combination, using K-means clustering algorithm for unsupervised classification of the combined image; Finally, by classification The image is post-processed and the outer contour is extracted, extracting the building area. Experiments with COSMO-SkyMed SAR images were performed. The results show that this method can effectively extract the building area from the high-resolution spaceborne SAR image, and the extraction result is obviously better than the un-utilized texture feature.