An Efficient Approach for Local Affinity Pattern Detection in Remotely Sensed Big Data

Mining information in Big Data requires to design a new class of algorithms and methods so that the computational complexity load is acceptable and the informativity loss is avoided. Information theory-based methodologies can represent a valid option in this sense. In this paper, we analyze a recently introduced method, called PROMODE, to efficiently detect local affinity patterns (LAPs) within Big Data sets. This processing framework operates with a computational load lower than what is required by other algorithms in literature, and is flexible enough to be applied to very heterogeneous remotely sensed datasets. Examples for spaceborne SAR and hyperspectral datasets, as well as a dataset involving Earth observations and clinical records are provided to prove this point.