Handwritten Character Recognition Based on Improved Convolutional Neural Network

Because of the characteristics high redundancy, parallelism and nonlinearity in handwritten character recognition model, convolutional neural networks (CNNs) are becoming first choice to solve these complex problems. The complexity, types characters, similarity dataset, optimizers all have a great impact on network resulting low accuracy, loss, other In view existence problems, an improved LeNet-5 model is proposed. Through increasing its layers fully connected layers, higher quality features can be extracted. Secondly, more dataset called EMNIST selected many experiments carried out. After experiments, Adam optimization algorithm finally chosen optimize model. Then, for processing problems pre-processed divided into different parts processed. A better-divided result after comparative experiments. Finally, accuracy characters achieved. experimental results show that reached at 88% test set, loss low.