Convolutional Neural Network Algorithm And It's Application In The Printing Character Recognition

The Printing character is widely used in the manufacturing industry.Such as bottled liquids,including mineral water bottles,edible oil bottle packaging products.All have the printing character to record the product serial number,production date,origin place.This has an important role in product traceability,quality control,inventory management and so on.Manual read and recognition the printing characters,the way is not accurate and not anti-fatigue,especially when to deal with the industry in large quantities,high speed of production.Convolutional neural network algorithm,as the first to achieve a breakthrough in deep learning,in the direction of twodimensional image recognition,is showing a higher accuracy than other traditional algorithms.For this,especially for the complex scenes,multi class character font,convolutional neural network algorithm is applied to the printing character recognition,has important theoretical significance and practical application value.This paper analyzes the deficiency in the past to the Printing character recognition method,expounds the advantage of convolutional neural network in image recognition and classification.Then,from the shallower to the deeper basic theoretical formula of artificial neural network and convolutional neural network is derived.The classical Lenet-5 convolution neural network model is studied in detail,and the number,connection and training parameters of each layer of network are detailed.And thus based do a lot of convolutional neural network experimental parameters,to optimize the parameters of convolutional neural network based on the character to spurt the code,including convolution kernel number,fully connected layer node number,learning step,batch size and improve the printing character recognition accuracy.The dissertation also proposes the recognition system solutions of the printing character,including system hardware composition,and character segmentation method is proposed,including line segmentation,partitioning column.And compared the recognition algorithm of convolutional neural network and other traditional algorithms in the printing character set accuracy.According to the different origin of printing characters,selecting the appropriate training set,different pretreatment methods of character test.Finally,using the C++ language to rewrite the forward calculation of the convolution neural network,using the parameters of Python training,reducing the hardware requirements of industrial control computer in the production environment.The main innovations of this paper are as follows:1 for the first time convolution neural network algorithm is applied to the printing character recognition;2.Lots of experiments,optimize the convolutional neural network model,can improve the accuracy of the recognition.The total accuracy rate reached 99.80%;3.As the printing characters of the different areas have different brightness and different font,we proposed appropriate pre-processing method to improve the model generalization recognition ability to characters;4.the paper uses C++language to rebuild the forward calculation of the network,reducing the time of prediction recognition.