Construction And Application Of Fuzzy Ranking Min-Max Neural Network

Fuzzy neural network combines the advantages of fuzzy logic and neural network,and has achieved great success in pattern recognition,control,medicine and other fields.As a classical fuzzy neural network,fuzzy min-max neural network has strong classification performance and online learning ability,but it still has the following two limitations:(1)The classification performance of the network is affected by the input sequence of training set patterns,membership function and the processing method of overlapping areas;(2)Its extension rules may lead to the increase of overlapping areas in the network,and the extension parameters that affect the classification performance of the model need to be user-defined.In this paper,a fuzzy ranking min-max neural network model is proposed to overcome the above limitations.The main innovations are as follows:1.A fuzzy ranking min-max neural network is proposed.Compared with the known fuzzy min-max neural network,the proposed model has higher classification accuracy and lower network space complexity.In the fuzzy ranking min-max neural network model,a ranking strategy is proposed,which determines the input sequence of training set patterns into the network,makes the final network fixed,and overcomes the dependence of fuzzy min-max neural network on the input sequence of training set patterns;A membership function based on Manhattan distance is proposed,which overcomes the problem that the membership function in fuzzy min-max neural network will not decrease steadily with the increase of the distance between patterns and hyperboxes.The method based on individual contour coefficient is used to classify the patterns falling in the overlapping area,which overcomes the problem that the membership degree of the contracted area to the original hyperbox changes because the fuzzy min-max neural network uses the contracted hyperbox to eliminate the overlap.2.A fuzzy ranking min-max neural network based on particle swarm optimization is proposed.The proposed model does not need user-defined extension parameters,and has better classification performance,including two parts:(1)a new extension rule is proposed,which sets an independent extension parameter for each dimension of the hyperbox to constrain the size of the hyperbox;(2)Particle Swarm Optimization is used to optimize the extended parameters,and finally a set of extended parameters is obtained,which makes the classification accuracy of the model highest,thus improving the classification performance of the model.3.The application model of fuzzy ranking min-max neural network is proposed.Because fuzzy ranking min-max neural network has strong classification ability and online learning ability,a face recognition model and atrial fibrillation recognition model based on fuzzy ranking min-max neural network are proposed.In the face recognition model,firstly,convolution neural network is used to extract the features of face images,and then fuzzy ranking min-max neural network model is used to classify them;In the atrial fibrillation recognition model,firstly,wavelet transform is used to remove the noise in atrial fibrillation signal,and principal component analysis is used to reduce the dimension,and then fuzzy ranking min-max neural network model is used for classification.Experiments have achieved better classification accuracy than traditional models.