Executive Summary : | Hyperspectral image (HSI) classification is a crucial method for analyzing land-use and land-cover information in remotely sensed hyperspectral data. However, traditional analytic methods struggle to extract relevant information due to the unique statistical and geometrical aspects of hyperspectral images. Deep learning-based methods have evolved to extract abstract features with joint spectral-spatial properties. This research aims to investigate different deep learning techniques and integrate complementary methods to develop robust classification frameworks for hyperspectral imagery. The objectives of this research project include developing robust deep spectral-spatial HSI classification schemes that preserve useful information, addressing the problem of training deep neural network models effectively with limited training data, integrating different complementary deep learning techniques for effective classification, and developing a GUI-based tool for classification based on deep learning techniques. The expected outcomes include robust and efficient classification methods for hyperspectral imagery, quality research publication in international repute, trained students and research scholars in HSI classification, and a GUI-based software tool beneficial to researchers and professionals. The new methods developed in this work will address key challenges in the HSI classification domain and can be utilized for various applications, such as land use land cover mapping. The proposed research will culminate in the development of robust methods based on deep learning frameworks in a GUI-based software tool, which will find significant applications in academia and higher education for teaching deep learning in a simplified manner compared to commercial tools. |