根据高分辨力合成孔径雷达 (SAR)图像中建筑物的特性, 提出了一种基于多尺度信息融合的建筑物提取方法。以非下采样轮廓波变换 (NSCT)为多尺度分析框架, 通过融合基于 NSCT低频子带的多尺度区域分析结果提取潜在建筑物区域; 同时, 融合基于 NSCT高频信息的边缘检测结果与均值比算子结果提取边缘结构信息; 在此基础上, 结合区域与边缘结构信息对虚警进行滤除, 对漏检建筑物进行补充, 完成建筑物提取。实验结果显示: 该方法优于基于多特征融合的建筑物检测算法, 在实验所用图像上的平均查全率达到 94%, 表明文中方法的有效性。

Considering the feature of buildings in high resolution Synthetic Aperture Radar(SAR) images, an algorithm for extracting buildings from SAR image based on multi-scale information fusion is proposed. Taking the Non-Subsampled Contourlet Transform(NSCT) as multi-scale analysis framework, a multi-scale fusion segmentation method is proposed to extract the potential building regions. An edge detection method based on multi-scale data fusion is designed to extract the edge information. The results of multi-scale fusion segmentation and edge detection are combined to filter the false alarm and add the missing buildings. Experimental results show that the proposed method achieves better performance than the building detection algorithm based on feature fusion, and the average recall ratio reached 94% in the experimental images. These results prove the efficiency of the proposed approach.