Матричная вероятностная нейро-фаззи система в задаче распознавания образов

Abstract

A matrix neuro-fuzzy system for quick image processing is proposed. The input signals are fed to the receptive layer of the system in the matrix form, then to all radial basis neurons of the first hidden layer. In the second hidden layer the distribution densities of the original data are determined. In the third hidden layer these estimates are corrected. In the output layer the winning class is defined. In the case of overlapping classes, both class centroids and membership levels are clarified. This system uses the matrix representation of the image, which can significantly reduce the recognition time