Research On Recognition Method Of Steel Coil Label Based On Computer Vision

The steel coil label is equivalent to the "ID card" of the steel coil,which contains the information of the specifications,models,manufacturer and batch numbers of the steel coil,which plays a very important role in the intelligent warehousing and logistics management of the steel coil.For the processing of steel coil label,manual method is still used to identify and record,and there are many shortcomings such as error-prone,poor real-time performance and safety hidden dangers.It cannot satisfy the demands of the growth of intelligent manufacturing in the metallurgical business.Therefore,it is of great significance to study the automatic and high-speed proof of steel coil label for realizing intelligent warehouse management and intelligent logistics process management of steel coil.In this thesis,the automatic recognition of steel coil label is realized by machine learning and deep learning.The main work includes three parts: preprocessing of steel coil image,label character segmentation and label recognition.The specific research contents are as follows:(1)Research on steel coil image preprocessing method,the gray scale transformation,filtering de-noising,image enhancement,binarization and character contour edge detection are used to preprocess steel coil image.The main goal of preprocessing is to simplify the image information of label,remove noise interference and highlight character features,so as to lay a good foundation for the segmentation and recognition of label characters.(2)Research on the segmentation method of steel coil label characters,the horizontal and vertical scanning segmentation technology of image pixels is used to realize the segmentation of steel coil label characters.And the segmented labeled characters are normalized to a single character image of uniform size,which is ready for recognition.(3)Research on coil label character recognition method.First,TWSVM algorithm is proposed to recognize coil label characters.After analyzing and studying the characteristics of twin support vector machine,TWSVM of binary classification is converted into TWSVM of multi-classification,which effectively improves the coil label recognition and classification effect.Simultaneously,an advanced Lenet-5 neural network is proposed based on the study of the traditional Lenet-5 network construction.The C5 layer in the traditional Lenet-5 network is deleted,the S4 pooling layer is directly connected with the full connection level,the Dropout level is added to the complete affiliation layer,and a large amount of neurons in the complete affiliation layer and output layer is ameliorated.By practicing and testing the improved Lenet-5network model,the network model parameters are determined.Finally,the analysis and research consequence evidence that the advanced Lenet-5 network pattern has distinct superiorities in the accuracy and real-time of steel coil label recognition.