Comparison of advanced neural network architectures for hyperspectral data classification

We investigate the performance of two advanced neural network architectures proposed earlier for hyperspectral data classification. While the first architecture uses feature reduction based on the samples of the classes, the second architecture uses a completely unsupervised approach for feature reduction using auto-associative neural networks. The aim of this study is to identify the pros and cons of such multi-level neural network architectures while classifying hyperspectral data.