Multi-Spectral Image Classification Using Improved Backpropagation Neural Networks

2.0 Conventional Backpropagation Model This paper deals with the application of neural network approach for pattern classification of remotely-sensed multispectral image data. The ability to classify multispectal data correctly and quickly is very important to the remote sensing community. Previously, the statistical pattern recognition method or the multivariate approach is widely used. However, not all data can be modeled by a convenient multivariate statistical model. The neural network classifier presents a convenient and distribution-free approach to multi-spectral classification. We have used an improved version of the conventional backpropagation model by initializing certain weights using self-organized approach. As a result, the network training time is reduced substantially. Both the methodology of this improved approach and results obtained using multispectral data are presented here.