Multispectral image classification: a supervised neural computation approach based on rough-fuzzy membership function and weak fuzzy similarity relation

A supervised neural network classification model based on rough-fuzzy membership function, weak fuzzy similarity relation, multilayer perceptron, and back-propagation algorithm is proposed. The described model is capable of dealing with rough uncertainty as well as fuzzy uncertainty associated with the classification of multispectral images. The concept of weak fuzzy similarity relation is used for generation of fuzzy equivalence classes during the calculation of rough-fuzzy membership function. The model allows efficient modelling of indiscernibility and fuzziness between patterns by appropriate weights being assigned using the back-propagated errors depending upon the rough-fuzzy membership values at the corresponding outputs. The effectiveness of the proposed model is demonstrated on classification problem of IRS-P6 LISS IV image of Allahabad area. The results are compared with statistical (minimum distance to means), conventional Multi-Layer Perceptron (MLP) and Fuzzy Multi-Layer Perceptron (FMLP) models. The better overall accuracy, user's and producer's accuracies and kappa coefficient of the proposed classifier in comparison to other considered models demonstrate the effectiveness of this model in multispectral image classification.