Research On Industrial Character Recognition Method Based On Convolutional Neural Network

Optical character recognition(OCR)is an important research field in machine vision.With the increasing degree of automation in production,character recognition in complex scenes,particularly in industry environment has received a significant amount of attention.Character in industry scene may have more complex background,blurry,physical damage and disturbances.These factors make the traditional character recognition method difficult to identify accurately.So automatic and accurate character recognition using machine vision technology has become an important part of the industrial production process.Based on the convolutional neural network algorithm,this thesis proposes a CNN integration model which satisfies the two main important requirements of industrial character recognition: the high recognition rate and less training time.In this dissertation,an effective data acquisition scheme is used to establish the industrial scene character dataset following an operation of data enhancement and preprocessing.Then we design the basic convolutional neural network through a series of comparative experiments.And further we establish a CNN integration model.It constructs richer feature representation for CNN feature extraction stage through multi-stage features combination,and establishes a more accurate classification decision system which integrates the structure of multiple networks referring to the thought of ensemble learning method.All these operations effectively solve the problem of losing global information in the feature extraction stage and the instability of the single network model when recognizing.At last,we compare the CNN integration model with the single network model and the CNN models proposed in other references on the same platform with industrial scene character dataset.The results in experiments demonstrate the efficiency with high recognition rate and less training time in complex industrial environment.