Advanced ML Methods: Bridging SAR Images and Structural Health Monitoring

The classical SAR-aided SHM directly analyzes displacement data obtained from SAR images. Although the extraction of displacement responses from SAR images via various interferometric techniques is of paramount importance to remote sensing-based SHM, the direct use and analysis of SAR-based displacement responses for monitoring and damage assessment of complex and critical civil structures is not a reliable and effective approach, especially in long-term monitoring with various uncertainties and variability in measured data. On this basis, this chapter proposes various machine learning methods in terms of unsupervised learning for SHM under environmental and operational conditions and different types of structural displacement responses retrieved from SAR images in terms of their sizes. In case of a large set of displacement data, an innovative probabilistic unsupervised learning method is proposed. For a small set of displacement data, this chapter proposes two unsupervised hybrid learning methods and mainly consists of three parts of data augmentation for expanding the size of small displacement responses, data normalization for removing the effects of environmental and operational conditions, and anomaly detection.