High-accuracy handwriting recognition based on improved CNN algorithm

This paper proposes a impoved convolutional neural network model based on the Adam optimizer for network training and replacing the commonly used activation function ReLU with the advanced activation function PReLU. The algorithm will address the current problems of low accuracy and low efficiency of handwritten digit recognition based on SVM(Support Vector Machines) classifier and nearest neighbor classification techniques. In the model training, the Dropout regularization method is used to improve the generalization ability of the model and reduce overfitting; the MNIST dataset of handwritten digits is used to explore the effects of ReLU activation function and PReLU activation function on the accuracy and convergence of the model; the accuracy is compared with that of other handwritten digit recognition techniques. The experimental results show that the enhanced CNN model outperforms other recognition techniques with high algorithm convergence and an accuracy of 99.60% when trained on the MNIST dataset.