SAR图像分块CFAR迭代的极地冰山检测

通过迭代CFAR算法，本文发展了一种针对分块SAR图像的冰山检测方法。考虑到滑动窗口运算负担大、计算效率低，首先对SAR图像进行分块，以提取分块内的亮目标冰山。利用高斯模型表征后向散射系数的统计分布，冰山检测阈值可简单地表达为均值和方差的线性组合，将分块内像素逐个比对阈值以检测冰山。考虑到同一场景中尺寸变化大的冰山影响检测精度，以识别的冰山像素做种子执行区域生长，从而提取大尺寸的冰山。针对单个分块迭代上述处理，以降低高斯模型表征分块统计分布的误差，提高冰山检测精度。利用2013年11月22日和29日获取自极地海域的两景RADARSAT-2图像开展验证试验。结果表明，数量多、尺寸变化大并嵌入在海冰等极地常见情形下的冰山，能被文中方法有效识别，选取区域内正确率高达85%以上，且具有良好的运行效率。

Iceberg detection of polar regions using CFAR iteration for SAR image blocks

In this paper, an iceberg detection method based on image blocks is proposed using iterative CFAR algorithm. Considering the large computational burden and low computational efficiency of sliding windows, the SAR image is blocked first to detect iceberg shown as bright objects in blocks. The Gauss model is used to characterize the statistical distribution of backscatter coefficient. The iceberg detection threshold can be simply expressed as a linear combination of mean and variance (μ+n_σ), compared with which the iceberg pixels are detected. Considering the impact of large size icebergs in the same scene, the identified iceberg pixels as seeds are grown to detect large size icebergs. In order to reduce the error of Gauss model to characterize the block statistical distribution and improve the accuracy of iceberg detection, iterative processing is done for a single block. The method is validated by two RADARSAT-2 images acquired on November 22 and 29, 2013 in polar sea area. The results show that these icebergs with large number, size change and embedded into, which is common on the poles, can be effectively detected by the method in this paper, the accuracy rate is more than 85%, and it has high operation efficiency.