Land-cover discrimination in SPOT HRV imagery using an artificial neural network—a 20-class experiment

An artificial neural network based on the multilayer-perceptron model has been used to classify two-date multispectral SPOT High Resolution Visible (HRV) imagery on a test site in the Departement Ardeche, France. A large network consisting of 98 nodes was trained successfully to classify 20 land-cover classes. A ground dataset comprising 1881 pixels was used to verify the accuracy of the classifier. The average accuracy achieved over all classes in the verification dataset was 81 per cent, exceeding the performance of a maximum-likelihood classifier by 28 per cent.