Hybrid Technique for Classification of Hyperspectral Image Using Quadratic Mutual Information

Researchers have found profound interest in the field ‘hyperspectral imaging’ as it has numerous applications. However, the center of motivation for this task has been the immense practice of hyperspectral imaging in ground cover classification problem. But, the high dimensionality of hyperspectral images (HSI) appears to be a menace for researchers. Unprecedented feasible solution to this crux is reduction of dimensionality. Therefore, a hybrid technique has been proposed for dimensionality reduction by combining feature extraction method with feature selection method. Here, Principal Component Analysis (PCA), a renowned technique, has been utilized for feature extraction. Thenceforth, three feature selection methods named Mutual Information (MI), normalized Mutual Information (nMI) and Quadratic Mutual Information (qMI) have been chosen for selecting features from the extracted features. Subsequently, the data have been fed to Support Vector Machine (SVM). SVM is implemented using Kernel trick which we are calling Kernel SVM.