Key points are not available for this paper at this time.
Earth observation is increasingly used for mapping and monitoring processes occurring at the surface of Earth. Data acquired by satellites nowadays allow us to have a global view, consistent in time, of the state of our forests, oceans, and growing urban areas. However, such a wealth of data has little value without appropriate processing chains able to convert the pixel values to information useful for decision makers. Recently, machine learning has seen fast advances – especially thanks to the rise of deep learning methodologies – and is increasingly deployed in Earth observation image processing systems. The ever-growing models from computer vision and natural language processing have inspired developments in remote sensing, and new approaches are constantly proposed in the field. However, despite their impressive results, the ever-growing mass of approaches and solutions makes it complicated to have a holistic overview and to know the most promising approaches from the field. In this paper, we aim to fill this knowledge gap and propose to review the thriving ecosystem focusing on developing AI models for Earth observation, its recent trends, and sketch potential pathways for future advances.
Tuia et al. (Mon,) studied this question.