The use of Back Propagating Artificial Neural Networks in Rare Vegetation Communities Classification from High-Resolution Satellite Imagery
要旨This paper has presented artificial neural networks (ANNs) for rare vegetation communities’ classification using remotely sensed data. Three variants of training of the Multi Layer Perceptron (MLP) based on three different classification schemes are used. At first 12 types of rare vegetation communities were defined and the main classification scheme was designed on that basis. After preliminary statistical tests for training samples, two modification algorithms of the classification scheme were defined: the first one led to creating a scheme, which consisted of 7 classes, and the second one led us to creating of 5 classes’ scheme. Testing results show that the use of ANNs of 5 classes’ scheme can produce higher classification accuracies than other alternative. The training procedures of these classifiers are described in details along with analysis and post processing products using Geoinformation Technologies. Ancillary geospatial data: DTM and its derivable (DEM, Slope, Aspect), as well as topographical, hydrological data and land use maps were created in order to support post classification operations. This result demonstrates that a level of classification accuracy achieved by artificial neural networks is higher than those generated by the statistical classifiers.