Research And Realization Of Automatic Marking System Based On Deep Learning

Written examination is a very necessary approach in various tests to determine knowledge and skills,however,manual marking of paper examination papers is tedious and easy to make errors.Compared with it automatic scoring is more fair,accurate,and efficient.With the development of artificial intelligence and image processing technology,automatic marking through image recognition has gradually become a achievable and very promising technology today.Automatic scoring can also be divided into subjective scoring and objective scoring,and objective scoring can be divided into optical mark symbol recognition and text recognition.The research purpose of this thesis is to design an automatic test paper scoring system for objective items of character recognition.Text recognition applies the deep learning method and technology,which can recognize the handwritten numerals and character symbols.Compared with the optical line mark,the method of marking the text recognition has the advantages of no need for a separate answer card,no restriction of the answer pen,and no need to draw a mark separately after answering the question.In addition,because the semantic recognition system is complicated,the accuracy is poor,and the versatility is poor,the subjective questions are avoided from being processed,but the artificial scoring of the subjective items can be identified and counted.The system mainly recognizes handwritten characters,uses the convolutional neural network VGG16 in deep learning to recognize handwritten characters,and optimizes and improves the model accordingly.The model is implemented using Python+Tensor Flow for training and implementation,and the model effect Compared with Let Net-5,the entire system is implemented using Java programming.The content of this thesis also includes image preprocessing and correction,image content segmentation and extraction,text image standardization,and automatic recognition of handwritten text.The selected algorithm used in this paper is efficient and reliable.As a result,the whole test paper processing is integrated into a recognition module plus a score statistics module,data management module,etc.,and integrates it into an automatic marking system based on C / S architecture.