Abstract/Summary

Change Detection (CD) in Synthetic Aperture Radar (SAR) is an essential task in the field of Earth Observation (EO). It focuses on identifying the change for the same geographical region between two SAR images acquired at different times. SAR offers several advantages over optical sensors. For instance, spaceborne SAR sensors are able to provide day/night capability to map the globe in virtually all weather conditions. Moreover, SAR’s microwave signals can pass through the cloud cover, allowing it to acquire data and generate images even in the presence of clouds, fog and dust. Despite these advantages of SAR, CD in SAR remains a highly challenging problem due to the misregistration of multi-temporal SAR images and speckle noise. Both these challenges adversely affect the performance of SAR-based CD techniques. In this thesis research, we have thoroughly discussed these challenges and have proposed novel solutions to improve the overall performance of SAR-based CD algorithms. For instance, we have proposed a deep neural network-based despeckling model (DM) that effectively suppresses speckle noise and enhances the performance of the existing CD methods. Specifically, the proposed despeckling methodology consists of two modules where the first despeckling module passes the input SAR image through a series of convolutional layers to suppress speckle noise and later feeds the resulting noise-reduced image to the subsequent change detection module. For change detection, we initiate a preclassification step employing the logarithmic ratio operator and the hierarchical FCM algorithm. Subsequently, we utilise a layer attention module that exploits correlations among multi-layer convolutions. This module produces robust cascaded feature representations learned by the network. These robust representations not only allow the proposed despeckling architecture to be resilient to multi-temporal SAR acquired from one SAR imaging process (i.e., the same number of SAR images looks before and after the change) but also enable it to deal with any combination of single or multi-look images acquired prior and after the change. In addition to this despeckling model, we have also developed a robust loss function that effectively suppresses the speckle noise, thereby improving the change detection accuracy. Both the despeckling model and the proposed noise-tolerant loss function are evaluated extensively on three public real SAR datasets, achieving superior performance compared to existing state-of-the-art SAR CD methods in all benchmark datasets.