The effectiveness of non-parametric classifiers in a limited training set

This paper presents a comparative analysis of the effectiveness of the method of support vector machine and artificial neural networks for classification of satellite images medium spatial resolution as an example of a high degree of heterogeneity and limited training data. The results of field-based researches have been used for test cases generation. Neural network approach showed the best result for classification accuracy (89,9 % vs. 86,2 % support vector), but was significantly less speed.