The Application of Convolutional Neural Networks to Detect Slow, Sustained Deformation in InSAR Time Series
The Application of Convolutional Neural Networks to Detect Slow, Sustained Deformation in InSAR Time Series
复制标题
DOI：
10.1029/2019gl084993
复制
发表时间：
2019-11-07
影响因子：
5.2
通讯作者：
Bull, D.
中科院分区：
文献类型：
--
作者：
Anantrasirichai, N.;Biggs, J.;Bull, D.
探测卫星干涉合成孔径雷达(干涉合成孔径雷达)图像变形的自动化系统可用于开发全球火山和城市环境监测系统。在这里，我们探索卷积神经网络在检测包裹的干涉图中缓慢、持续的变形方面的局限性。使用合成数据，我们估计单独的形变信号的检测阈值为3.9厘米，考虑大气伪影时的检测阈值为6.3厘米。由于在不改变信噪比的情况下产生了更多的条纹，因此外包裹使这一点分别减少到1.8和5.2厘米。我们在Campi Flegrei和Dalll的累积变形时间序列上测试了该方法，在这些时间序列中，覆盖使分类性能提高了15%。我们提出了一种均值滤波方法，将不同包络参数的结果合并到旗帜变形中。在Campi Flegrei，60天后变形8.5 cm/年，在Dalll，310天后变形3.5 cm/年。这对应于3厘米和4厘米的累积位移，与基于合成数据的估计一致。
Automated systems for detecting deformation in satellite interferometric synthetic aperture radar (InSAR) imagery could be used to develop a global monitoring system for volcanic and urban environments. Here, we explore the limits of a convolutional neural networks for detecting slow, sustained deformations in wrapped interferograms. Using synthetic data, we estimate a detection threshold of 3.9 cm for deformation signals alone and 6.3 cm when atmospheric artifacts are considered. Overwrapping reduces this to 1.8 and 5.2 cm, respectively, as more fringes are generated without altering signal to noise ratio. We test the approach on time series of cumulative deformation from Campi Flegrei and Dallol, where overwrapping improves classification performance by up to 15%. We propose a mean-filtering method for combining results of different wrap parameters to flag deformation. At Campi Flegrei, deformation of 8.5 cm/year was detected after 60 days and at Dallol, deformation of 3.5 cm/year was detected after 310 days. This corresponds to cumulative displacements of 3 and 4 cm consistent with estimates based on synthetic data.