Land cover mapping using two neural networks and knowledge-based processing

We propose a pattern classification system for remotely sensed data which consists of three subsystems, namely, a preprocessing part by Kohonen self-organization feature mapping, a pattern classification part by a multilayered neural network classifier, and an error correcting part by a knowledge-based processor. We apply the system to analyse two kinds of data observed by two optical sensors, LANDSAT(TM) and JERS-1(OPS), and also verify the flexibility of the system empirically.