A model for classifying multisource remote sensing images by Kohonen neural networks

In this work a self-organized neural network for classifying of multisource remote sensing images is proposed. The method is a modified self-organizing map based on the Kohonen model that uses training patterns which belong to the known classes in the training phase of the network. In this way we will have a supervised trained system, which results in more accurate and rapid information. Acquiring this accuracy is more guaranteed by using multisource in remote sensing such as digital terrain model data, which is highly effective for extracting the features of the collected data. The model used for the classification of multisource remote sensing images was collected from two geographical locations in Iran and then its performance in classification was compared with two other methods: maximum likelihood (MLH) statistical method, and back-propagation neural network. The applied model proved to be the best in accuracy and speed.