Improved segmentation and change detection of multi-spectral satellite imagery using graph cut based clustering and multiclass SVM

The Satellite image analysis automatically looks over an image to attain valuable information such as land cover classification and change detection from it. Generally, many image segmentation algorithms exploit specific spatial information between the pixel and its neighbors together with the color information to lengthen the cluster quality. Recently, a variety of clustering processes have been suggested to grasp data that is not linearly separable. The main issue in clustering algorithm is its inconsistency. To address this issue, the innovative spatial-spectral method for image segmentation and change detection based on Graph cut based clustering is proposed. In this hybrid approach, the Multispectral satellite images are preprocessed using Difference Of Offset Gaussian (DOOG) filters and then segmented by graph cut based clustering. Multi-class problems are highly expensive to solve, there is a need of a massive optimization problem. The changes between the classified images can be obtained by using image differencing method. The performance of the proposed method has been evaluated with the temporal data sets of LANDSAT images. From the experimental results, it is observed that in the proposed work, the mean value of the changed area for a particular dataset achieves a 47.2% reduction compared to the conventional system.