Article Preview
Top1. Introduction
With the recent innovations in Earth observation techniques (sensors, satellites, and data), “urban” remote sensing or “remote sensing urban applications” have rapidly gained popularity among urban planners and, more generally, administrations in charge of territorial planning.
Indeed, remote sensing makes it possible to increase our understanding of urban areas in different ways, although the real potential of this technique is often challenged by the complexity of the urban environment itself. (Xiaojun, 2011)
Previous studies have been carried out on the urban theme, notably the work of (Antoine, 2009), he tried to develop a method able to detect and qualify changes in small areas from Very High spatial Resolution (VHR) remote sensing data acquired at different dates and from different sources by comparing the textural properties of objects of interest. First, he tried to extract objects by a region growing segmentation and then, their texture are compared.
Another work has been carried out in this direction, (Ashish, 2010) used a technique based on fuzzy clustering approach which takes care of spatial correlation between neighboring pixels of the difference image produced by comparing two images acquired on the same geographical area at different times.
Using always the VHR satellite data, (Tobias, 2017) proposed a novel object-based approach for unsupervised change detection with focus on individual buildings, he first applied a principal component analysis together with a unique procedure to determine the number of relevant principal components is performed as a predecessor for change detection; then he used k-means clustering for discrimination of changed and unchanged buildings.
(Andrew, 2017) used a different technique called Import Vector Machine (IVM) which builds upon the popular Support Vector Machine (SVM) methodology (Ribana, 2012). To obtain the optimum classification, the IVM algorithm explores all possible subsets of training data for optimal selection (termed import vectors) which are derived through successively adding training data samples until a given convergence criterion is met (Ribana, 2012). Data samples are selected according to their contribution to the classification solution.