Cross-Correlation Stacking for Robust Offset Tracking Using SAR Image Time-Series
Cross-Correlation Stacking for Robust Offset Tracking Using SAR Image Time-Series
复制标题
DOI：
10.1109/jstars.2021.3072240
复制
发表时间：
2021
影响因子：
5.5
通讯作者：
Shiyi Li;S. Leinss;I. Hajnsek
中科院分区：
文献类型：
--
作者：
Shiyi Li;S. Leinss;I. Hajnsek
偏移跟踪广泛应用于从遥感数据测量地表位移。位移由两个图像模板最匹配的偏移量确定。匹配可以通过归一化互相关（NCC）来评估，其中NCC峰值的高度和位置代表匹配质量和相应的偏移。获得稳健的偏移估计需要明确跟踪 NCC 本底噪声中的峰值。为了改进偏移估计，我们提出了一种互相关叠加方法，可以显着抑制 NCC 的本底噪声。我们不是从每个成对 NCC 中导出偏移量，而是堆叠一系列连续的成对 NCC，并在对 NCC 堆栈进行平均后确定偏移量。因此，跟踪受益于多个 NCC 中的冗余信息，并且对噪声更加鲁棒。我们通过使用合成孔径雷达卫星 TanDEM-X 和 Sentinel-1 A 收集的图像时间序列测量瑞士大阿莱奇冰川的流速来评估该方法。使用相对较小的 48×48$ 像素模板与 TanDEM-X 图像的七个成对 NCC 堆栈相结合，我们获得的速度场的空间覆盖范围几乎相当于在单个图像上应用 96×96$ 像素模板获得的速度场的覆盖范围对。 Sentinel-1 A 的空间覆盖范围也得到了类似的改进。结果表明，叠加方法可以极大地提高空间分辨率和所获得的速度场的覆盖范围。
Offset tracking is widely applied for measuring ground surface displacements from remote sensing data. Displacements are determined by the offset where two image templates match best. The match can be evaluated with normalized cross correlation (NCC), in which the height and location of the NCC peak represent the matching quality and the corresponding offset. Attaining robust offset estimations requires an unambiguous tracking of the peak in the NCC noise floor. To improve offset estimations, we propose a cross-correlation stacking method that can significantly suppress the noise floor of NCC. Instead of deriving offsets from each pairwise NCC, we stack a series of consecutive pairwise NCCs and determine the offset after averaging the NCC stack. Thereby, tracking benefits from the redundant information in multiple NCCs and is more robust to noise. We assessed the method by measuring the flow velocity of the Great Aletsch Glacier in Switzerland using image time series collected by the synthetic aperture radar satellites TanDEM-X and Sentinel-1 A. Using relatively small templates of $48\times 48$ pixels combined with a stack of seven pairwise NCCs of TanDEM-X images, we obtain velocity fields whose spatial coverage are almost equivalent to the coverage of velocity fields obtained with templates of $96\times 96$ pixels applied on a single image pair. Similar improvements in spatial coverage are observed for Sentinel-1 A. The results demonstrate that the stacking method can greatly enhance both the spatial resolution and the coverage of the obtained velocity fields.