NEURAL NETWORK CLASSIFICATION USING REMOTE SENSING AND ANCILLARY DATA

Abstract

Digital image classification techniques are applied to generate thematic maps from remote sensing data. The conventional techniques are limited in the sense that they use only spectral information in the classification process. In visual image interpretation, besides tone (i.e. spectral characteristics), other elements such as association, pattern etc. aid in the interpretation process. Similarly the digital image classification can also be supplemented with additional information to produce accurate maps. The additional information can be obtained from ancillary data, such as topography, geophysical information, data from different sensors obtained at different times. When the classification process includes the information from such sources, it is called as multisource classification. The literature suggests that there occurs a significant improvement in the classification when data from other sources are included. In the present study, multisource classification has been performed using Artificial Neural Networks (ANN). The data from IRS 1B LISS-II, IRS 1C PAN, and topography have been utilised. The results have been evaluated with the most widely used technique namely maximum-likelihood classification. The results show that there is a significant improvement in accuracy of classification by ANN when data from other sources are incorporated. The areas located on steep slopes and covered with shadows are more appropriately classified by inclusion of topographic information in the classification process. Moreover, in general neural network approach has been found to be more accurate than maximum-likelihood for multisource classification.