Support Vector Machines for Automatic Multi-class Change Detection in Algerian Capital Using Landsat TM Imagery

Recently, Support Vector Machines (SVMs) have shown a practical relevance in various image processing applications. This paper investigates their applicability for land cover and land use change detection using multi-sensor images of remote sensing. Then, the most widely used approaches for multi-class SVMs, which are the One-Against-All and the One-Against-One with both Max-Win and DDAG decision rules are implemented to perform multi-class change detection. SVMs are evaluated in comparison with artificial neural networks using different accuracy indicators. The results obtained showed that SVMs are much more efficient than artificial neural networks and highlighted their suitability for land cover change detection.