A Hybrid Approach of Feature Selection and Feature Extraction for Hyperspectral Image classification

Hyperspectral imaging has gained much popularity as a field of research in recent years. For the classification of hyperspectral images the main challenge that comes across is to deal with their high dimensionality. In the cases of a low number of training samples in comparison with the high number of features of hyperspectral images, Hughes Phenomenon is encountered. To overcome this situation reducing the dimensions can put ease to it. In this purpose feature extraction and feature selection operations are performed. Basically, feature selection refers to eliminating the less informative features and feature extraction draws the concept of combining multiple features and then reducing them. In this paper, both of the feature reduction techniques i.e. Minimum Redundancy and Maximum Relevance(mRMR) and Principle Component Analysis(PCA) are applied on our dataset and corresponding results are shown. In our experiment, mRMR and PCA were applied individually on the dataset first and then combining both of them a hybrid dimension reduction approach PCA-mRMR was followed. The hybrid approach PCA-mRMR reported 87.72% accuracy which is the best among the three approaches. Finally, the Support Vector Machine (SVM) classifier was used to check the classification results.