基于纹理特征的高分辨率SAR影像居民区提取

Residential Areas Extraction In High Resolution SAR Image Based on Texture Features

利用灰度共生矩阵计算高分辨率SAR图像的纹理特征,通过统计分析选取合适的特征矢量,并基于非监督聚类分析提取居民区。对提取的居民区以一定的面积阈值剔除噪声(细小区域),并利用形态学算子对提取边界进行适当的归整,得到最终结果。在对应的光学图像上人工提取居民区范围,以此作为实验结果的评价标准。实验结果表明本方法可以得到较好的效果。

With the resolution of SAR image is improving, the inner structure of residential area shows more complex in high-resolution SAR images than low ones. An approach is proposed to extracting residential area based on SAR texture features extracted from the Gray Level Co-occurrence Matrix (GLCM). Firstly, six texture features such as energy, entropy, contrast, variance, correlation, and inverse difference moment are investigated. Secondly, the three GLCM parameters: window size, step and angle are decided. Thirdly, the feature vector is reduced from six to two. Then an unsupervised analysis is applied to the data to extract the residential area. Finally the small areas are deleted, and morphological operators are applied to adjust the sketch of the extracted area. The proposed method has been tested by using airborne SAR data at 3 m resolution.