| With the rapid development of information processing technology,replacing people with computers to complete some of the work is an important development direction in the future,and handwritten numeral recognition is one of the typical representative.How to use computer to classify and recognize handwritten numerals intelligently and improve the recognition accuracy and efficiency has become an important goal in this field.Artificial neural network(ANN)has been widely used in forecasting,classification and recognition because of its powerful approximation effect and nonlinear processing ability.However,the traditional neural network generally uses the number of pixels in the image as the input layer neurons in handwritten numeral classification and recognition,resulting in the excessive number of neurons in the whole network,which leads to long learning time and low efficiency.According to the existing problems of traditional neural network,PCA neural network is used to classify and recognize handwritten numerals.The original sample data are reduced from 784 to 10,40,50 and 58 dimensions respectively,and the hidden layer neural network is used to classify and recognize the handwritten numerals.The number of meta-data is 30 and 100,and then compared with the original sample from three aspects: learning time,learning efficiency and classification accuracy.Finally,the input of neural network is reduced from 784 dimensions to about 50 dimensions,with the same accuracy as 784 dimensions,and the learning speed of network is greatly improved.When the number of hidden layer neurons was 30,the learning time was reduced by 87 seconds,the efficiency was increased by 50%,and the highest accuracy was increased by 0.06%.When the number of hidden layer neurons is 100,the learning time is reduced by 556 seconds,the efficiency is increased by 83.2%,and the highest accuracy is increased by-0.41%.The effectiveness of the method is proved. |
PDF Full Text Request