18 May 2022
Synthetic Aperture Radar (SAR) is an active radar, which can obtain high-resolution images under all-day and all-weather conditions. SAR image interpretation is to acquire key information from SAR images. It has found various applications in military and civil community. However, SAR images are strongly affected by speckle noise and geometric distortion effects. This makes the information extraction from SAR images challenging to perform.
The dissertation focuses on developing deep learning-based approaches for SAR image interpretation. Firstly, this dissertation realizes oceanic SAR eddy detection based on a novel Mask Edge Enhancement and IoU Score Region-based Convolutional Neural Network (Mask-ES-RCNN) framework. Secondly, this dissertation proposes an UNet-based semantic segmentation network with a Texture Enhancement Module (TE-UNet) for intertidal sediments and habitats classification. Finally, this dissertation presents a SAR-Optical Fusion UNet model (SOF- UNet) based on the existing largest dataset SEN12MS, which provides optical and SAR pairs to realize land cover classification. The experiment results demonstrate that these three models outperform the performances of state-of-the-art methods, showing both theoretical significance and practical value in the SAR interpretation field.
To participate in the PhD defense, please contact the Informatics Study Office via email.
The Zoom link will then be mailed to you shortly before the meeting.
Time: 16:00 (s.t.)