Research On Sample Recognition And Evaluation Model Of Multi-feature Fusion Of Chinese Handwriting

As one of the important skills in the learning and use of Chinese characters,the writing skills of Chinese characters have attracted the attention of the majority of Chinese character learners and users.However,the recognition model of Chinese characters has high complexity,and there are problems such as high equipment requirements and long model learning time in the actual application process.At the same time,Chinese character evaluation has some controversy in the balance between the objective factor evaluation and subjective factor evaluation of font aesthetics.Therefore,in response to these problems,combined with the previous research foundation for foreign Chinese character teaching based on force-tactile interaction devices,a multi-feature fusion recognition evaluation model of offline handwritten Chinese characters based on CNN is proposed.For the research of the model,the main research content is divided into two parts,namely the multi-feature fusion recognition model and the multi-feature fusion evaluation model,and finally completed the multi-feature fusion recognition and evaluation comprehensive model system in the form of web application.The main work completed by the study is as follows:(1)In order to verify the effectiveness of the proposed scheme,the CASIA-HWDB data set was used and some samples collected through force-tactile interaction devices were fused to form a new data set to verify the feasibility and effectiveness.(2)Aiming at the construction of multi-feature fusion recognition model,a basic multi-feature fusion recognition network model based on the Le Net-5 network’s basic structure,Goog Le Net’s Inception module and VGG’s dominant characteristics is proposed to build an adaptive network.(3)For the evaluation model of multi-feature fusion,the improved Le Net-5 model is used to extract CNN features based on the classification results of the recognition model,and the sample structural features(center of gravity feature,grid feature)are used to complete the standard intervals of various features Calculate to form a multi-feature fusion evaluation model combining neural network features and traditional structural features.At the same time,two scoring schemes of fixed proportion allocation and dynamic proportion allocation for writing aesthetic scores were proposed;(4)The integrated multi-feature fusion recognition model and evaluation model form a multi-feature fusion recognition evaluation comprehensive model,and the recognition evaluation system is implemented as a web service to complete the identification and intelligent evaluation of the sample to be tested.The research finally formed a multi-feature fusion recognition network model with a depth of 15 and the model size was reduced by about 18 MB compared to VGG16.The accuracy rate reached 96.88%,which was 0.57% higher than the improved VGG14-GAP model and 0.62% higher than Goog Le Net model.At the same time,a multi-feature fusion evaluation model was formed,and the effectiveness of the proposed intelligent allocation evaluation scheme of multi-feature fusion score ratio for writing aesthetics was verified.