US 7,069,257 B2Granted

Pattern recognition method for reducing classification errors

AssigneeIndustrial Technology Research Institute

InventorYea-Shuan Huang

Life of this patent

Estimating the term…

Loading patent…

Abstract

A RBF pattern recognition method for reducing classification errors is provided. An optimum RBF training approach is obtained for reducing an error calculated by an error function. The invention continuously generates the updated differences of parameters in the learning process of recognizing training samples. The modified parameters are employed to stepwise adjust the RBF neural network. The invention can distinguish different degrees of importance and learning contributions among the training samples and evaluate the learning contribution of each training sample for obtaining differences of the parameters of the training samples. When the learning contribution is larger, the updated difference is larger to speed up the learning. Thus, the difference of the parameters is zero when the training samples are classified as the correct pattern type.

Claims8

Dependent claims

A pattern recognition method for reducing classification errors by using a RBF neural network including an input layer, a kernel layer having a plurality of pattern types, each having a plurality of hidden neurons, each corresponding to a kernel function, and an output layer having a plurality of output neurons, each corresponding to a pattern type of the kernel layer, the pattern recognition method comprising the steps of:

(A) inputting a feature vector formed by a plurality of training samples to the input layer, wherein the feature vector comprises a plurality of features, each forming a corresponding one of a plurality of input neurons in the input layer;
(B) calculating a measured value of the kernel layer relative to the hidden neuron by each of the input neurons based on a kernel function of each hidden neuron;
(C) for each output neuron of the output layer, multiplying the measured value of the kernel layer corresponding to the pattern type associated with the output neuron by a weight for obtaining a measured value of the output layer corresponding to the pattern type;
(D) finding a maximum one of the measured values of the output layer and a class classification error relative to a correct one of the measured values of the output layer, wherein the class classification error is obtained by calculating an error function E(x)=1−e(OT−OM)/A(n), where OM is the maximum one of the measured values of the output layer, OT is the correct measured value of the output layer, 0≦A(n+1)≦A(n), and n is an epoch iteration number;
and
(E) modifying values of parameters of the kernel function and calculating a weight of the measured values of the output layer based on the error.

3 more elements

Dependencies are inferred from the claim text.