Processing Landsat TM data using complex valued neural networks
Howard E. Michel, Shyamala Kunjithapatham · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 2002
Neural networks are massively parallel arrays of simple processing units that can be used for computationally complicated tasks such as image processing. This paper develops an efficient method for processing remote-sensing satellite data using complex valued artificial neurons as an approach to the problems associated with computer vision-region identification and classification-as they are applied to satellite data. Because of the amount of data to be processed and complexity of the tasks required, problems using ANNs arise, specifically, the very long training time required for large ANNs using conventional computers. These problems effectively prevent an average person from performing his own analysis. The solution presented here uses a recently developed complex valued artificial neuron model in this real-world problem. This model was then coded, run and verified on personal computers. Results show the CVN to be an accurate and computationally efficient model.