Design And Implementation Of English Marking System Based On Deep Learning

The emphasis on education in contemporary society not only enriches the teaching content,but also increases the workload of teachers.On the premise of ensuring the quality of marking,it is regarded as an effective means to relieve the burden of teachers from the heavy work of marking and statistics.At present,most of the large-scale English examination papers are marked by machines with optical mark recognition technology,which adopts the form of machine-readable card,and then with the help of photoelectric marking machine.On the one hand,although this marking method is more efficient and accurate in dealing with multiple choice questions,it still needs human cooperation to fill in the blanks.On the other hand,due to the high frequency of examination practice,the consumption of a large number of machine-readable cards will bring high cost.Therefore,it is of great significance to develop a marking system for daily English test practice with lower cost.In recent years,the rapid development of deep learning makes people realize its advantages in the field of pattern recognition: the model does not need to select features manually.Based on the theory of deep learning and image processing,this dissertation designs a lightweight English marking system.The main research contents are as follows:(1)An algorithm for single character recognition based on small convolution kernel stack structure is proposed.This method considers that the convolutional neural network is too deep to be used in the recognition of small-size character images,but the semantic features of shallow features are not rich.Firstly,the Le Net-5 network is fine tuned to make it more suitable for the input characteristics of single character pictures in the marking system.Then,multiple 3×3 convolution kernels are used to replace the 5×5 convolution kernels in the original network.Experiments show that this method not only improves the recognition accuracy,but also makes the model get better convergence effect.(2)An algorithm for offline handwritten English word recognition based on the enhanced convolutional block attention module and composite convolution structure is proposed.This method takes into account the serious differences of different people's offline handwriting styles,and the feature extracted from traditional network is not strong enough.Firstly,the enhanced convolution block attention module is constructed by using the parallel connection of channel attention module and spatial attention module and the reuse of input feature map.Then,in the deep convolution layer,the dual channel convolution feature extraction structure is used to realize the composite convolution structure.Finally,the enhanced convolution block attention module and composite convolution are added to the feature extraction network to realize the recognition of handwritten English words.Experiments show that this method has some advantages compared with other mainstream offline text recognition algorithms.(3)This dissertation designs a marking system for National English test papers.The system mainly includes image processing module,character recognition module and word recognition module.Firstly,edge extraction and perspective transformation are performed on the input image to obtain the main body of the answer sheet,and then the edge of the main body of the answer sheet is extracted again to locate the student number area and the answer filling area,and then cut out the area for model recognition according to the question type,load the single character recognition model or word recognition model for end-to-end recognition,judge the right and wrong answers,and finally count the score,high frequency wrong questions and draw the histogram of wrong questions to facilitate the follow-up teaching.This system gives up the machine reading card filling and photoelectric marking machine marking,greatly reducing the cost of equipment.In the marking of the question type,further reduce the need for manual correction of the question type,make the marking system more intelligent,more convenient.