抄録
The advent of space-technology and multispectral sensors on satellites made available enormous amount of remote sensing data, the supervised classification of which is rather impossible, and is equally unrealistic to obtain a suitable training set. Application of Unsupervised Neural Network (NN) in this field is till now unexplored. In this paper we made possible the use of Unsupervised Neural Network algorithm for the analysis of NOAA satellite multispectral data. We have pointed out the difficulties for using unsupervised NN and deviced our method to overcome that by introducing proper preprocessing of data and adding an extra dimension to the input data and using special techniques for NN learning. It is elaborated with simulation results that without using the special technique we introduced, the NN fails to perform any worthwhile classification. The correctness of classification is confirmed by quantitative analysis of our simulation result.
