Research And Implementation Of Offline Handwritten Chinese Character Recognition Algorithm Based On Deep Learning

With the continuous advancement of the intelligentization of our society,artificial intelligence has been applied to more and more fields and application scenarios.At the same time,due to the increasing demand for automatic recognition,the automatic recognition of handwritten Chinese character images has become the current research focus in the field of computer vision..This thesis aims at the difficulty in detection and positioning of handwritten Chinese characters due to the difference in character density and overlapping of continuous strokes,and completes the construction of an offline handwritten Chinese character detection algorithm based on deep learning;at the same time,the number of Chinese characters is too large and the shape of handwriting is complicated and changeable.The problem of difficult recognition and classification is completed,and the construction of an offline handwritten Chinese character recognition algorithm based on deep learning is completed.Finally,an offline handwritten Chinese character recognition system based on the cascade of the above two modules is built to realize the automatic recognition of handwritten Chinese character images.The main research contents are as follows:(1)Offline handwritten Chinese character detection algorithm: This thesis first explains the shortcomings of traditional handwritten Chinese character segmentation methods,and analyzes the advantages and disadvantages of the SSD algorithm based on deep learning.Aiming at the continuous strokes and irregular arrangements between Chinese characters,combined with the characteristics of handwritten Chinese characters,the SSD algorithm model is optimized and improved by restructuring the feature extraction network,enhancing feature fusion,optimizing the preselection box,and optimizing the loss function.Network training and testing are carried out on the self-built handwritten Chinese character detection data set.The experimental results show that the m AP of the optimized network is improved by 6.04% compared with the m AP of the SSD network.(2)Offline handwritten Chinese character recognition algorithm: This thesis constructs a writing grade specification based on the writing strokes,font structure and other attributes of Chinese characters,and analyzes,decodes and evaluates the writing grade of Chinese characters on the HCL2000 data set and CASIA-HWDB1.1 data set.In order to solve the problem of difficult recognition and bloated network caused by the complicated and changeable shape of handwritten Chinese characters and the excessively large number of Chinese characters,the recognition classification accuracy and network magnitude are optimized for double grasping.Due to the large differences in the degree of freedom of handwritten Chinese characters,this thesis establishes a standard for Chinese character writing standards.For offline handwritten Chinese characters of different writing levels(Level II,Level III),an offline handwritten Chinese character recognition network integrated with Gabor and based on An offline handwritten Chinese character recognition network with residual structure,grouped convolution and channel rearrangement,and two optimized networks are trained and tested on the corresponding data sets.Two sets of experimental results show that compared with the VGG16 basic network,the optimized network achieves the optimization goal of model lightweight while maintaining or improving the classification and recognition accuracy of TOP-1.(3)For the part of the offline handwritten Chinese character recognition system: This thesis builds an offline handwritten Chinese character recognition system based on Kivy,which completes the connection from the bottom data to the high-level applications,through the functions of two modules: offline handwritten Chinese character detection and offline handwritten Chinese character recognition The cascade realizes the automatic recognition of offline handwritten Chinese character images.