Research On Industrial Character Recognition Method Based On CNN Integration Model

As the frontier field of machine vision,character recognition has been widely used in many aspects of realistic life especially in industrial production,character recognition technology is playing increasingly more and more important role in promoting the intelligence of industrial production.Due to such a complicated production environment,such as motion blur,position change,and even a series of uncertain interference factors such as obstructions,lighting condition change,physical damage,cracks,etc.,traditional character recognition methods are difficult to ensure the accuracy and efficiency of recognition.It has seriously restricted the development of industrial intelligent production in China.With the continuous development of deep learning algorithms,the character recognition based on convolutional neural network has received increasing attention in machine vision.Based on the convolutional neural network,this paper constructs an integrated network model for industrial character recognition,which achieves good recognition results under character recognition,especially in the industrial environment.The main contents of this article are as follows:(1)A character data set for model training is constructed.In view of the insufficient number of character image samples by collected,the data set is appropriately processed with data enhancement processing,finally,the established data set includes partial data samples in the natural scene of ICDAR2003 and industrial character images after data enhancement processing.(2)Aiming at the influence of network parameters on the recognition effect in the convolutional neural network,a specific contrast experiment is designed to investigate the influence of the network parameters of the convolutional neural network on the recognition rate,such as the depth of the neural network and the size of the convolution kernel.And its number,etc.,it provides relatively specific reference information for constructing a suitable convolutional neural network model.(3)The influence of convolutional layer size on feature extraction of convolutional neural network model is analyzed,and the feature maps extracted by the convolutional layer and the pooled layer are visualized.Aiming at the insufficiency of single feature extraction in traditional model,a multi-level feature extraction structure is proposed,the validity of the method was verified by using the established data,and the recognition rate was significantly improved compared with the traditional single-level feature extraction model.(4)Combining the integrated learning Bagging algorithm with the proposed multi-level feature extraction structure,three basic classifier structures for model integration are designed.A character recognition algorithm based on convolutional neural network integration model is proposed.The comparison experiments on the character dataset verify the effectiveness of the proposed integrated model algorithm.Compared with the traditional convolutional neural network model,the robustness and generalization ability are improved.