Research On Recognition Method Of Steel Pipe Marking Based On Convolutional Neural Network

With the development of the metallurgical manufacturing industry,the steel pipe identification as the "identity card" of the steel pipe plays a very important role in the manufacturer’s inspection of the quality of the steel pipe and the later data traceability of the steel pipe.However,the traditional method for operators to visually identify and manually record the steel pipe marks is time-consuming and laborious,and is limited by the influence of the workers’ labor enthusiasm and mental state,and errors are prone to occur.In recent years,with the rapid development of computer vision technology,industrial production has gradually evolved to automation.The use of machines to automatically recognize inkjet characters not only improves production efficiency,but also reduces errors during manual inspection.This paper combines image processing and deep learning knowledge to propose a steel pipe identification scheme based on convolutional neural network to realize automatic identification of steel pipe identification.The main contents are as follows:1.Aiming at the problems of corrosion,breakage,abrasion of inkjet characters,noise and light effects caused by the shooting environment,a set of preprocessing schemes suitable for steel pipe marking images are proposed.And for the small amount of noise and other interference after preprocessing,the image was processed twice,including repairing and filling gaps,removing small area noise,corrosion operation to remove "burrs" and classification expansion processing,and finally obtained a more ideal binary value Character image.2.In view of the adhesion,tilt and large-area noise interference of the steel pipe logo image,a set of character segmentation schemes are studied.First,the least square method is used to correct the image tilt,and the interference noise points that affect the character segmentation are removed through character positioning;then use The two segmentation methods of connected domain method and upper and lower contour analysis method realize character segmentation by layer by layer;finally,the segmentation algorithm is compared and verified.Experiments show that this algorithm combines the advantages of the two segmentation methods and improves the segmentation.The accuracy rate is significantly better than other algorithms.3.Aiming at the problem of low data volume of steel pipe character images,a character recognition algorithm based on improved Res Net-18 network model is proposed.The algorithm adjusts the number of network layers and parameters of the Res Net-18 model,introduces migration learning and regularization methods to optimize the training process;through comparative experiments,it proves the advantages of the character recognition algorithm in this paper with high accuracy and short reasoning time;finally summarized The character segmentation and recognition algorithm in this article designed a set of Py Qt5-based steel pipe identification recognition system,and deployed the system on an industrial flat panel.Through the interface display,the recognition results are more intuitive and more conducive to the operation of workers.