Automatic detection of urban areas using the Hierarchical Temporal Memory of Numenta®

Abstract

The aim of the study is to develop a new methodology for the identification of urbanized and not urbanized areas using high spatial resolution remote sensing data and adapting a new digital classification algorithm based on the functionality of the human Neocortex. We used multispectral Quickbird images for the classification of different urban areas in the Province of Cordoba (Spain). The Memory-Prediction Theory, implemented in the form of a Hierarchical Temporal Memory (HTM), was applied by means of the Nupic software, in order to obtain the classification of urban land cover types. HTM is a new computing technology that replicates the structure and function of the human neocortex. The conclusion indicates that Hierarchical Temporal Memory has great potential for extracting urban areas information from high satellite imagery and the 93.8% of the parcels has been well classified. As conclusion, this methodology can improve the level of automatization of digital classifications using high remote sensing data. Key words: Hierarchical temporal memory, urban areas, memory-prediction theory, objects based classification, Neocortex.