Research and Implementation of Handwritten Numbers Recognition System Based on Neural Network and Tensorflow Framework
Jing Bai, Jie Fei · Journal of Physics Conference Series · 2020
Abstract It is difficult for a computer to recognize handwritten numbers with a high accuracy because of many factors such as shape, size and others. This paper applies a neural network based on Tensorflow framework to implement the handwritten numbers recognition system. The system is compiled and run in vim under Mac operating system with an 8g RAM. This system first builds a neural network, then trains the model on the MNIST dataset to output the accuracy of handwritten numbers recognition. We can input the handwritten picture and get the prediction result through this system. Then we optimize this system on neural network, activation function, loss function, learning rate and generalization to improve the accuracy. Finally, we prove that this system has a higher accuracy closed to 98% for recognizing.