Hyperspectral Image Analysis with Projection Pursuit and MRF Segmentation Approach – Unsupervised and Supervised

This work proposes methods for hyperspectral image analysis in both situations viz., (i) when concurrent groundtruth is unavailable and (ii) when available. The method adopts a projection pursuit (PP) procedure with entropy index to reduce the dimensionality followed by Markov Random Field (MRF) model based segmentation. Ordinal optimization approach to PP determines a set of “good enough projections” with high probability, the best among which is chosen with the help of MRF model based segmentation. When ground-truth is absent, the segmented output obtained is labeled (tuned) with the desired number of classes so that it resembles the natural scene closely. With available ground-truth, some special reflectance characteristics based on the ground-truth data of the study area in question are incorporated vis-a-vis the MRF model based segmentation stage while minimizing the energy function in the image space. In our crop study, some biophysical parameters of crops are appropriately incorporated in MRF model based segmentation stage. Subsequently segments are validated with training samples so as to yield a classified image with respect to varieties and stages of crops. Three illustrations are presented with (i) EO1 archived data for an unsupervised case, (ii) EO-1 image data with concurrent groundtruth and (iii) AVIRIS-92AV3C, Indian Pine test site image with concurrent groundtruth. The classification accuracies of some nonparametric approaches and that of proposed methodology are provided for the illustrations (ii) and (iii) along with the computational time.