Handwritten number recognition method based on fractional calculus and generalized inverse neural network

Abstract

The invention discloses a handwritten number recognition method based on fractional calculus and a generalized inverse neural network. The method comprises the following steps: obtaining a handwritten number image set, and selecting a single hidden layer neural network model according to the handwritten number image set; utilizing fractional calculus and a weight matrix and a weight vector in a generalized inverse calculation single hidden layer neural network; and according to the weight matrix and the weight vector, obtaining neural network parameters, and calculating the sample precision. The method is applied to handwritten number recognition in pattern recognition and is high in recognition precision. The method uses fractional derivatives to optimize the input weight matrix because the fractional derivatives can reflect the performance of a neural network model better than integer derivatives. A regularization term is introduced into the method, and thus the method has better generalization performance and is enhanced in system controllability. Meanwhile, the regularization term can reduce noise existing in a dataset to achieve noise reduction.

Description

technical field [0001] The invention relates to the technical field of pattern recognition, in particular to a handwritten number recognition method based on fractional calculus and generalized inverse neural network. Background technique [0002] With the development of technologies such as artificial intelligence, information retrieval and massive data processing, pattern recognition has become a research hotspot. In pattern recognition, the recognition of handwritten digits has become an urgent problem to be solved because of its complicated calculation and recognition rate. The use of knowledge and experience is essential in the recognition of handwritten digits. However, the acquisition of knowledge is indeed the current bottleneck problem. In particular, the knowledge acquisition process of the human brain is almost a potential process. Neural network as a Methods for extracting knowledge and experience that are not easily articulated are playing an increasingly impor...

AI Technical Summary

Problems solved by technology

[0006] Therefore, in the existing recognition of handwritten digits, the traditional neural network based on the integer order gradient descent method is slow in training and has low accuracy; although the ELM algorithm is fast, it requires too many nodes in the hidden layer of the network ; Although the ELM algorithm based on the steepest descent method has been improved, the accuracy needs to be improved due to the use of integer order derivatives

Images