An accurate SVM-based classification approach for hyperspectral image classification

One of the important tasks in analyzing hyperspectral image data is the classification process. Support Vector Machine (SVM) is the most popular and widely used classifier, and its performance is ongoing to be further improved. Recently, methods that exploit both spatial and spectral information are more sufficient, robust, useful, and accurate than those accounting for the spectral signature of pixels only. In this paper, regional texture information is extracted from the hyperspectral data by using a Spatial Pixel Association (SPA) processing to further improve the classification performance of SVM techniques. A novel approaches over SVM by exploiting SPA characteristics is proposed in order to increase the classification accuracy. Moreover, a new method that can be used to solve the misclassified-pixels problem, Control Process of Growing Classes (CPoGC), is also proposed in this manuscript. In order to demonstrate the effectiveness of the proposed scheme, experiments on AVIRIS hyperspectral data over Indian Pine Site (IPS) are conducted to compare the performance of the proposed classification approaches against some existing SVM based techniques such as SC-SVM and PSO-SVM, and some traditional methods like K-NN and K-means. Experimental results demonstrate that the proposed method clearly outperforms these well-known classification algorithms.