VHR Satellite Image Time Series Analysis for Illegal Building Monitoring Using Multi-Dimensional Histogram Earth Mover's Distance

In remote sensing, temporal sequence of images called Satellite Image Time Series (SITS) covering the same scene allows land cover observation, understanding, analysis and monitoring. Besides, several conventional techniques for controlling and monitoring anarchic urban expansion have been initiated but they remain not sufficient to overcome this issue. This paper proposes an automatic method of detection and monitoring of anarchic urban expansions starting from multi-sources and multi-temporal data (VHR satellite images and geographic information data). First, the illegal urban areas are extracted using an original SVM based technique integrating expert knowledge and auxiliary data by means of ontology construction. This leads to the formalization of the expert semantic information and the urban construction rules (often in sentences form) and their confrontation with the classification results. Secondly, a spatio-temporal regions' similarity framework is proposed using a novel matrix based on Multi-dimensional histograms Earth Mover's Distance (EMD). To this end, an explicit definition of a Spatio-Temporal Region (STR) is given in order to build its characteristic matrix called Multi-Temporal Region Matrix (MTRM). Afterwards, using this matrix, a Cross-STR Similarity Matrix (CSTRSM) is computed between STR of in order to reveal regions with similar temporal fingerprint. The resulted spatio-temporal map describes urban areas types and their temporal changes.