A Handwritten Digit Recognition Method Based on Improved Spiking Neural Network

Abstract

The invention discloses an unsupervised learning method for handwritten digit recognition. According to the simulated biological information processing mechanism, the pulse neural network model is firstly established, the connection mode between neurons is determined, and then the pulse time is established according to the pre- and post-synaptic trace learning rules. Relying on the plasticity model, the genetic algorithm is used to optimize the pre-synaptic initial weight and synaptic delay in the spiking neural network, train and test the MNIST data set, and realize the unsupervised learning of handwritten digit recognition. Experiments show that this method effectively improves the training speed and the recognition accuracy.

Description

technical field [0001] The invention relates to the technical field of image recognition, in particular to a handwritten digit recognition method based on an improved impulse neural network. Background technique [0002] Digital is the main medium for people to communicate and communicate, and it is the information carrier for the development of the world economy. For a long time, the processing method of manually identifying numbers and entering them has problems such as heavy workload, high cost, low efficiency, and poor timeliness. We urgently need to use the "reading" method to liberate people from heavy manual labor. Among them, the use of digital identification technology to realize automatic entry of information will undoubtedly become the fundamental method to solve this key problem. [0003] Handwritten digit recognition is an important branch of image recognition, which integrates artificial intelligence, formal language, statistical decision-making, information t...

AI Technical Summary

Problems solved by technology

The support vector machine (Support Vector Machine, SVM) based on the statistical model has greatly improved the accuracy of handwritten digit recognition, but it has too many training parameters and complex calculations, which increases the recognition process time

Convolutional neural networks (CNNs), as a current research hotspot in the field of image recognition, have a very high recognition rate, but overfitting occurs during the recognition training process, and in practice, there are few labeled data , there are many data without labels, and it is very time-consuming and boring to manually set labels for each data. At the same time, convolutional neural networks need to set and adjust a large number of parameters, which are important reasons for restricting convolutional neural networks.

Images