Improvement in Hyperspectral Image Classification by Using Hybrid Subspace Detection Technique

Hyperspectral imaging has been an important field for researchers having numerous applications in many sectors. But this has been extensively used in ground cover classification. Due to the high dimensionality of hyperspectral imaging, it has become a daunting task for researchers and a possible solution to this is feature reduction. In this paper, we propose a hybrid technique for feature reduction by combining feature extraction and feature selection. After extracting the features in first technique, feature selection is applied. Here Principal Component Analysis (PCA), a well-known feature extraction technique has been used and normalized Mutual Information (nMI) has been chosen for selecting features. Finally classification is done using Kernel Support Vector Machine (KSVM). This proposed algorithm (PCA-nMI) offers 97.2662% classfication accuracy for AVIRIS dataset and 99.2074% classfication accuracy for HYDICE dataset.