SNNS application for crop classification using HyMap data

Abstract

The goal of this paper is the presentation of a method and results for artificial neural networks crops classification based on HyMap hyperspectral data. The method that uses an ANNs does not only depend on statistical parameters of particular class and hence makes it possible to include texture information. To experiment with variable pattern size two data sets were chosen with 10 bands obtained after MNF and 5 hyperspectral vegetation indicies. Next to post classification crops maps, additional quality layers were generated to check which classes are “problematic” because of spectral similarity or errors in the training/reference data. The best accuracy was achieved using the 10 MNF bands with the 3×3 pixel sub pattern size −94,8 %.