Assessing band selection and image classification techniques on HYDICE hyperspectral data

Describes some preliminary results concerning the robustness and generalization capabilities of manual (parametric) methods versus machine learning methods in the band selection and subsequent classification of hyperspectral images. We specifically compare a genetic algorithm-based approach to an end-member and spectral unmixing approach in the selection of data used for classification of a HYDICE image. We then discuss and compare the classification of the image using supervised and unsupervised techniques, as well as a backpropagation neural network.