Publications

Abstract : Hyperspectral remote sensing has been a prime focus of geospatial technology for the past three decades. In the present work, HSI classification was done by considering open set adaptation and Generative Adversarial Networks (GAN). The test data may have additional labels than train dataset, which leads to open set domain adaptation. Sometimes, it is hard to acquire the useful information straightforwardly from HSI information because of the volume of data. Dimension reduction method such as dynamic mode decomposition (DMD) is found to be very effective for reducing the redundant features. Then as an extension to the work, also explored is a novel Chebyshev polynomial-based dimensionality reduction technique for HSI classification to check whether is it possible to reduce the dimension of each dataset further with comparable classification accuracy. The performances are analyzed in terms of classification accuracies, time for computation, and peak signal to noise ratio (PSNR).