A PSO-SVM-Based Change Detection Algorithm for Remote Sensing Optical Images

Change detection is considered as one of the challenging issues in the field of remote sensing. The multi-temporal images used to detect the changes generally have significant illumination variations which affect the performance of a change detection method. To address this issue, a machine learning (ML)-based change detection algorithm is proposed for the remote sensing optical images. The proposed method is a combination of Support Vector Machines (SVM) and Particle Swarm Optimization (PSO). In this method, the PSO is utilized in a novel way to provide the optimized feature vectors. These feature vectors are used in SVM to accurately determine the changed and unchanged pixels. The proposed method is very effective in identifying the changes in remote sensing optical images having significant illumination variations. It can give comparatively higher correct classification (PCC) values, and lower false alarm (PFA) as well as total error (PTE) values than the state-of-the-art methods. Experiments on six pairs of Landsat images demonstrate the effectiveness of the proposed method.