Research On Character Recognition Algorithm Based On Deep Convolutional Neural Network

Chinese character recognition has a very wide application prospect and practical value in various image recognition,for example,it can be applied to mail sorting,ticket identification,license plate recognition,street sign recognition,ID card recognition,blind assisted reading,etc.with the convenience of automated identification,reduced manual operations,saving time and labor costs,and facilitating people's lives.Blind people cannot obtain such visual information due to congenital or acquired reasons.Therefore,they are an urgently needed group,and their need for acquiring external information through text reading is more urgent.In order to solve the problem of dyslexia of the blind,domestic and foreign experts and scholars have studied the blind reader,but its products have problems such as narrow application scope,difficult operation and low efficiency.Therefore,it is particularly important to design a more concise,convenient and efficient reader for the blind and assist the blind to read books and other paper documents.In this paper,the blind reader is taken as the research background to study the efficient image text recognition algorithm of Chinese character.1.Design a depth of character recognition based on deep learning convolution neural network(Han-DCNN),relative to the traditional Chinese character recognition algorithm based on structural pattern recognition,support vector machine,artificial neural network has the disadvantages of low accuracy and poor robustness,it not only improves Chinese character recognition accuracy,but also for Chinese characters that contain noise has higher recognition rate.After testing,the accuracy rate on the training dataset and test data is as high as 0.99,indicating that the neural network model can perfectly fit the Chinese character recognition training set.2.A new intelligent blind reader for application testing and hardware functional verification was designed and built: in the embedded platform of the corte-A9 architecture equipped with Linux real-time operating system,the page image of the books will be captured and sent to the cloud server by the 4G communication module.In the cloud server,the character recognition will be completed using digital image preprocessing and deep learning algorithm.The recognition result will be sent back through 4G module to tell the user through voice broadcast module of the front-end.However,this blind reader relies on the network communication module to transmit images and recognition results,the real-time performance is greatly reduced,and the detection and recognition results are often highly delayed.3.In order to improve the real-time performance of the reader,a lightweight deep convolutional neural network character recognition model Han-Net for embedded platform was proposed,which discards the use of network communication module to upload images to the server for detection and realizes text detection and recognition directly on the embedded platform.Compared with other neural networks,the algorithm of Han-Net has lower time and space complexity with high accuracy.Experimental comparison and analysis show that,Han-Net has a very fast running speed on any computing platform,and can fully utilize the computing power of the embedded computing platform.The memory consumption and calculation amount are also very small,which is suitable for the lightweight operation requirements of embedded platform.