Application of a hybrid neural network model for multispectral remotely sensed image classification in the Belopa area South Sulawesi of Indonesia

The main motivation for the research presented in this paper is to design and to apply a framework based on a hybrid neural network model for classification of remote sensing data. The approach based on the combination of the Self-Organizing Map (SOM) and a Learning Vector Quantization (LVQ) method. Specifically, The SOM acting as a preprocessor and provides an approximate method for computing the feature map vectors of land cover over study area in an unsupervised manner. The supervised learning technique using the LVQ method with training the neural network based on a competitive learning rule, that uses class information to move the voronoi vectors slightly, so as to improve the learning time process and the quality of the classifier decision regions. Furthermore, the window size of the feature map topology was selected based on the spatial distance between the winning neuron and its neighbors automatically. The approach has been tested and verified on a Landsat-TM multispectral imagery of land cover over Belopa area, South-Sulawesi, Indonesia. Based on the experimental results, it is shown that the proposed hybrid model outperforms the original LVQ method in average classification performance.