Abstract: Traditional solo-resolution bridges detection methods turn out to be inefficient when applied to large SAR images. This paper proposes a new method based on multi-scale feature analysis. An algorithm based on dynamic programming is proposed to detect rivers, locate bridge candidates in low resolution level and obtain Regions Of Interest (ROI) in middle resolution level. Bridges are detected by analyzing the ROI. Experimental results indicate that the method can process one scene SAR image (10 000×10 000 pixels) in 15 s, with detecting rate above 90% and false alarm rate less than 5%.

Key words: SAR images, bridges detection, multi-scale

摘要： 传统的单一分辨率下的桥梁目标检测方法难以胜任整景SAR影像中桥梁目标的检测。该文提出一种基于目标多尺度特征的中分辨率SAR图像桥梁自动检测方法。在低分辨率级别下提取河流和定位潜在桥梁目标点，把定位结果映射到原始中分辨率图像数据中，完成目标检测，从而提高检测效率。实验结果表明，对幅宽10 000×10 000像素的整景SAR影像，采用该方法检测桥梁目标的速度在15 s以内，目标检测率大于90%，虚警率小于5%。

关键词: SAR图像, 桥梁检测, 多尺度

CLC Number: 

• TP391