Research of System for Correcting Exam Papers Based on Convolution Neural Network

Abstract

In order to solve the problem of consuming too much time and energy in correcting exam papers, a system for correcting papers, which is based on convolutional neural network, is studied. Taking primary school mathematics papers for example, after taking a photo of a paper through a mobile phone, and then uploading to the system, the system will identify the digit answer and compare it with the standard answer by using digit recognition method based on convolution neural network, so as to automatically get a score for the paper. Because digit recognition is a classification problem, the thesis firstly compares the classification results of several algorithms in machine learning on MNIST database, and then selects the convolution neural network with the highest recognition accuracy rate for the system implementation. Finally, the system for correcting papers is achieved through image acquisition, image uploading, image transformation, digit preprocessing, convolution neural network classification, answer comparison scoring and so on. The experimental result shows that the accuracy rate of the system for correcting papers has reached 99.9%, which can be applied in practice.