Combining structural and spectral information for discrimination using pulse coupled neural networks in multispectral and hyperspectral data

The emerging field of dynamic neural networks motivated by recent biological understanding of the way in which the brain encodes discrimination information in a time signal from a large, multi-layed image suggests an approach to fusing data. Pulsed coupled neural networks (PCNNs) have shown a robust ability to segment a single spectral band image into segments for terrain categorization but have not proven to be very robust in structural identification. However, linking fields of PCNNs can easily be configured for scale and/or rotation invariance. Using a variation on the Eckhorn [1990] pulsing neuron model, a PCNN is constructed to reduce the structural information with the spectral information of a coarse resolution hyperspectral image. The configuration of the linking network is studied to try to yield meaningful pulsing signals that can be combined for enhanced segmentation or discrimination