A SVM-Based Change Detection Method from Bi-Temporal Remote Sensing Images in Forest Area

The reliability of support vector machines for classifying multi-spectral images of remote sensing has been proven in various studies. In this paper, we investigate their applicability for land cover change detection in forest regions. Firstly, multidate remote sensing images are co-registered and we have stacked the NDVI index layers of two dates in red, green, blue bands composite to perform a supervised classification. Secondly, sample pixels were manually selected from changed and unchanged area to be used in the training stage. Thirdly, for each pixel SVM produces a single output through its decision function, high detection overall accuracy (>96%) and overall Kappa coefficient (>0.89) were achieved using two landsat images covering an 8-years period in study area. Lastly, SVM-based change detection with different kernel functions was compared using statistical evaluations.