| With the development of science and technology,the application of pattern recognition has had a very important impact in people’s lives.Pattern recognition is the basic theory and method for the development of artificial intelligence.The main research method is the application of computer technology to recognize and classify recognized events.One of the important application areas is handwritten Chinese character recognition.The wide variety of Chinese characters,complex structures,different writing styles of different people and excessive similarity of characters in similar shapes have brought great difficulties to the recognition of handwritten Chinese characters.The continuous development of pattern recognition technology promotes the progress of AI technology and AI technology mainly includes machine learning and deep learning technology.The rise of deep learning has provided strong technical support for Chinese character recognition.Compared with traditional Chinese character recognition technology,deep learning convolutional neural network non-artificial feature extraction and end-to-end learning technology have very significant advantages.Therefore,deep learning technology is gradually replacing traditional identification methods.This paper uses deep learning technology to study the recognition of handwritten Chinese characters.The specific research work is as follows:(1)A handwritten Chinese character data acquisition device is designed.The device uses a camera with an infrared filter and uses an optical flow algorithm to collect the movement trajectory of the reflected light spot of a linear laser.The collected trajectory is binarized and finally the output is displayed on the self-developed Chinese character recognition software.(2)The experiment uses CASIA-HWDB1.1 handwritten Chinese character database from the Institute of Automation,Chinese Academy of Sciences.Because the original data image of the database is slightly blurred and in order to better meet the needs of experimental data,it is necessary to preprocess the image data of the database.Data preprocessing mainly includes:contrast enhancement,grayscale,normalization,binarization,smoothing and denoising,morphological processing.The method of enhancing image contrast is used to enhance the contrast information of image pixels.The weighted average algorithm is used for grayscale processing.The bicubic interpolation algorithm is used to normalize the image size,so that the edge features of the image are well preserved.Use the maximum between-class variance algorithm for binarization processing to achieve data foreground and background segmentation.The median filter algorithm is used for image smoothing and denoising,the characteristic information of handwritten Chinese characters is well preserved.Morphological operations are performed on the image data after median filtering,which expands the effective feature area of Chinese characters.(3)Based on the Tensor Flow deep learning framework,the Inception V3 network model is used to perform migration learning training on the data.Through multiple sets of comparative experiments,the optimal algorithm and applicable network model are determined,the data enhancement and parameter optimization strategy methods are used to obtain the optimal network model.(4)Using the desktop application development technology of Py Qt5,a set of handwritten Chinese character recognition human-computer interaction interface is designed.Through the combination of software and hardware,the handwritten Chinese character recognition mode of human-computer interaction is realized. |
