Abstract: Mining of target class/es from voluminous hyperspectral data has received greater attention in the recent past. Various dimensionality reduction approaches have been proposed in this regard. However, mining of data by targeting regions in hyperspectral data that are a collection of classes is yet under-addressed. The proposal here is to design a dimensionality reduction technique based on Fisher’s Linear Discriminant Analysis (LDA) to derive the optimal feature subspace based on the training set of the Region of Interest (RoI) so that the accurate mapping of all the classes present in the RoI in the input could be possible. Also, a compression technique that enables the lossless decompression of the RoI is proposed. Experiments are carried on standard benchmark datasets such as the ROSIS Pavia University dataset. The results are satisfactory and are reported.
Keywords: Hyperspectral data, RoI, LDA, Dimensionality reduction, Compression.
Keywords: Hyperspectral data, RoI, LDA, Dimensionality reduction, Compression.
