Research On Medical Record Text Recognition Method Based On Deep Learning

With the rapid development of deep learning,text recognition technology is gradually applied to the medical field.This paper aims to study ways to improve the recognition rate of Chinese medical records.Aiming at the problem that the identification algorithm based on convolutional neural network has low accuracy rate for Chinese medical record text recognition,a multiscale feature extraction network based on residuals is proposed,which combines the self-attention mechanism of different scales.Compared with the recognition method based on convolutional neural network,the proposed method has better feature extraction and classification ability,and effectively improves the recognition rate of Chinese medical record text.Chinese characters have a more complicated structure than other common characters.The complexity of the Chinese character vertical structure of the actual Chinese medical record text image is similar to the complexity of the horizontal structure,and the height of a single Chinese character is generally greater than the width,according to these priors of Chinese characters,this paper mainly does the following work:1.A recognition model called V-CRNN is proposed.Compared with CRNN,the model increases the vertical feature information dimension of feature maps and improves the network’s extraction of vertical fine-grained features of Chinese characters.After the vertical feature information dimension is improved,even if the number of feature map channels is reduced,the model still achieves a higher recognition rate than the original CRNN,and has better robustness to vertical incomplete characters2.A multi-scale recognition model based on residuals was designed and named Ms Net,which utilizes the characteristics of the residual network and has a deeper network structure.The model combines different scale feature information,and retains more vertical features for each scale feature map,which improves the feature extraction ability of the network.Experiments show that the feature extraction network can effectively improve the recognition rate of Chinese medical record text.3.In order to further strengthen the vertical feature extraction ability of the network,this paper proposes a two-dimensional recursive neural network model called V-Ms Net,which uses vertical bidirectional long-term memory networks to capture the vertical dependence of images in shallow networks.Experiments show that the vertical Bidirection-LSTM is effective for improving the recognition rate of Chinese characters.4.Combined with the feature extraction network proposed in this paper,a multi-scale self-attention recognition model called S-Ms Net is proposed.The model calculates the correlation of each position on the feature maps of different scales,and can capture more horizontal and vertical dependence information.Experimental results show that compared with Ms Net proposed in this paper,the multi-scale self-attention mechanism recognition model called S-Ms Net has further enhanced the feature extraction ability of Chinese medical record text images and has better recognition performance.