An Analysis on the Subspace Detection and Classification of Hyperspectral Remote Sensing Images

Classifying hyperspectral images is a challenging task as it contains high data volume and the required spectral signatures is not available always. Therefore, the classification process faces curse of dimensionality problem. However, it is very useful in many applications after successful classification since it contains a lot of useful information about the ground objects. This complicacy can be overcome by reducing the irrelevant features before image classification. From the large input dataset, the required information can be extracted by Principal Component Analysis (PCA). Therefore, to address the aforementioned problem, PCA is applied in this thesis to reduce the input dimensionality as well as improve the classification accuracy. For appraising the efficacy of the proposed method real hyperspectral data was used and this data was also used to perform the experimental analysis. The selected subspace was evaluated using different classifiers for performing the experimental results. From the result it can be said that on real hyperspectral data, the proposed approach can achieve about 99% as training accuracy and 97% as testing or classification accuracy.