Concluding Remarks and Way Forward

Methods of spectral dimensionality reduction (DR) of hyperspectral imagery (HSI) are presented in this book that includes both feature extraction and feature selection–based approaches. A fast yet efficient pooling-based feature extraction method reduces the spectral dimension in a way that the approximate signature pattern of the data is preserved. This book also presents an unsupervised ranking-based band selection method that ensures minimum correlation and maximum variance among the selected bands. The band optimisation methods using genetic algorithms and particle swarm optimisation are discussed in this book where information theory–based fitness functions are used to improve the performance. The impact of spatial dimension on DR method are analysed in this book, and it is observed that the reduction of spatial dimension could improve the efficiency of DR methods in terms of execution time. The effect of noise is also analysed for optimisation and ranking-based band selection (BS) methods. The data-driven approaches for band selection show significant advantage as they do not depend on user perception to select the required number of discriminating bands from the data. At the end, the area of future research is mentioned where advanced computation techniques may be employed to ease the overall processing of HSI.