Adhesive Handwritten Digit Recognition Algorithm Based on Improved Convolutional Neural Network

Abstract

Traditional machine learning algorithms are susceptible to many factors in the recognition of handwritten adhesion numbers, such as different people's digital writing habits, different degrees of adhesion, and low image quality, which could lead to lower digital recognition accuracy. To solve these problems, an improved convolutional neural network algorithm for handwritten adhesion digital recognition is proposed in this paper. First, an improved convolutional neural network model is provided for the large number of adhesion in handwritten digital pictures. Multilevel feature extraction is performed on the experimental images using convolution kernels of different scales, and then, the feature frame filtering algorithm is optimized to improve the recognition accuracy of handwritten adhesion numbers while enhancing the robustness of the neural network to background noise. The experimental results show that the average recognition accuracy of the improved convolution model on the experimental data set is 94%. The proposed algorithm reduces the parameter size with ensuring high recognition accuracy, and improves the recognition efficiency of the system, which is better than most state-of-the-art algorithms.