XAI for Earth Observation
With the advent of the era of artificial intelligence (AI), AI has spread to various remote sensing applications, in particular building detection and segmentation. In such applications, the characteristics of remote sensing data are well learned by advanced AI algorithms, where the AI-based techniques have outperformed state-of-the-art techniques, especially, when using large labeled datasets. However, current AI-based techniques do not disclose how the data features are being selected. Therefore, current models do not provide clear physical meaning of the internal features and representations. Addressing the transparency and explainability of the employed AI models is a must, where eXplainable artificial intelligence (XAI) is widely acknowledged as a crucial step to the practical deployment of AI models in remote sensing communities. In this context, this project aims to investigate and study recent XAI methods in the litetarure and propose new methods in the context of earth observation.