The analysis of remotely sensed data with neural networks

This paper describes the application of artificial neural networks (ANN) towards the supervised classification of multispectral satellite data. A direct comparison with the maximum likelihood (ML) classifier is performed. The ANN is implemented as a multilayer perceptron and trained by the backpropagation algorithm. Two Landsat Thematic Mapper scenes covering the district of Niederbayern in Germany and extensive ground truth form the basis for this approach. Confusion matrices are used to compare the results of the classification process. In addition the generalization capabilities of ANN are demonstrated.