Artificial intelligence (AI) and machine learning have undoubtedly brought about significant progress in image processing, including remote sensing images or, broadly speaking, Earth Observation (EO) information. Deep learning (DL), i.e. CNNs are state of-the-art but require ever-larger training sets. More recently, unsupervised learning or not fully supervised methods gain attention. In fact, scientists in the computer vision (CV) domain increasingly acknowledge the fact that animals and humans seem to learn complex tasks from a very small amount of interactions with the real world. Likewise, knowledge-based approaches, particularly (Geographic) Object based image analysis (GE)OBIA have become popular in the EO (or “remote sensing”) community. Apparently, the EO and CV communities are still somewhat neglecting each other as citation statistics reveal. The EO community strives to map, analyse and classify images and to contribute to an understanding of the status and dynamics of geographic phenomena. The raison d'être of GEOBIA is integrating multiscale EO data of increasing spatial and temporal resolution. Here, I will refer to some major achievements in the context of big EO data analytics as well as major achievements in CV. The objective of this talk is to review the role of spatial concepts in the understanding of image objects as the primary analytical units in semantic EO image analysis, and to identify opportunities to integrate deep learning and spatial concepts for multi-source remote sensing analysis.