HANDWRITTEN CHINESE CHARACTER CLASSIFICATION SOLUTION WITH HYPERPARAMETER OPTIMIZATION SUPPPORT

Abstract

Recently, handwritten Chinese character classification solutions based on
convolutional neural networks have become popular and achieved some
outstanding successes. However, to achieve convolutional neural networks with
high classification accuracy, the hyperparameters for these networks need to be
optimized. Based on LeNet-5 convolutional neural network architecture, this
paper presents a solution for handwritten Chinese character classification based
on convolutional neural network with the support of Hyperband hyperparameter
optimization method. Experimental results have shown that the network model
with the support of hyperparameter optimization has achieved accuracy on the
test data set up to 96%, higher than the model based on the LeNet-5 model and
the model with random hyperparameters.