Digital Recognition Based on Improved LENET Convolution Neural Network
Lishan Jia, Yi Gang Sun · 2018
To promote the performance of LeNet-5 convolutional neural network, an improved LeNet-5 convolutional neural network was proposed. The improved neural network model was trained using MNIST character library. The effects on the performance of final recognition of the parameters of LeNet-5, including the different number of iterations, amounts of samples of each batch and the network learning rate were analyzed. The comparison of performances between Le-Net-5 and improved LeNet-5 was made. The results showed that the structure of improved network is simpler, the training parameters are fewer, the training time is shorter, the recognition rate is higher, the improved network can overcome the fitting phenomenon. The performance of the improved network is better that of traditional methods.