Ensemble of Support Vector Machines for spectral-spatial classification of hyperspectral and multispectral images

Previous studies on different satellite images have not yet introduced a single attribute with the highest accuracy for different applications. In this paper, a novel classification system with the highest strength against possible noises is offered using Support Vector Machine (SVM) and its performance is evaluated on the selected satellite images. So, an optimal high-strength classifier with the sufficient level of accuracy is proposed executing Composite Kernels and Ensemble of Classifiers. Results obtained from applying this method on IKONOS (91.65%) and AVIRIS (97.71%) satellite images (in Tehran and Indian Pine study areas) showed that the proposed method accuracy is higher than the Direct Summation of Kernels, Weighted Summation of Kernels, Cross Information Kernels and Extracted Features techniques. The main reason for this significant difference is the wide range and variety of input features.

Automated summary

Previous studies have not yet found a single attribute with the highest accuracy for different applications. This paper proposes a novel classification system using Support Vector Machine (SVM) that has the highest strength against possible noises. The performance of this system is evaluated on selected satellite images, and the results show that the proposed method outperforms other techniques in terms of accuracy.