Research Projects

The project aims to develop novel self-supervised learning techniques for remote sensing data using multiple modalities. While self-supervision has been successful in single-modal applications, its application in multiple modalities is limited. The aim is to develop novel SSL techniques for different types of RS data and use them for three key applications, including land-cover classification, multi-sensor change detection from bi-temporal images, and discovering new categories using information from multiple modalities. This is the first time SSL is being tested for multi-modal data under limited supervision. The goal is to improve the performance of supervised learning techniques in multi-modal remote sensing data.