Research And Design Of Tire Character Detection System Based On Deep Learning

Tire characters are embossed characters with identification significance composed of numbers,letters and special characters,and their identification meaning is mainly composed of tire brand,tire model,production date,production country and other information,which has an important directional role for tire dealers and tire buyers.Therefore,the imprint character test of the tire must be carried out before the tire leaves the factory to ensure that the character of the factory tire matches its model.At present,most tire manufacturing enterprises,the character detection of finished tires mainly relies on manual completion,inefficiency is also very easy to occur due to workers’ visual fatigue caused by character misdetection,so the development of tire character automatic detection system is an urgent problem for enterprises.This paper designs a set of tire character detection system for this demand of tire manufacturers,and the tire character automatic detection system designed in this paper includes five parts,namely tire image preprocessing,tire character area positioning segmentation,tire character detection,tire character logic output and tire character comparison.In the image preprocessing part,in order to enhance the foreground expression ability of tire character images,the contrast of tire character images is increased.Histogram equalization,illuminate enhancement and emphasize enhancement algorithms are applied successively to enhance the character region.The enhanced tire image is greatly improved in terms of clarity and contrast compared to the original tire image.In the Character Region Positioning Segmentation section,select the character region positioning segmentation method that is most suitable for the tires used in this article.In this paper,two sets of character area positioning schemes are designed by using the foreground extraction method and feature matching method commonly used in tire area positioning.Experiments have verified that compared with the foreground extraction method,the feature matching method is more capable of positioning the character region of the tire used in this paper.After determining the positioning method,this paper optimizes and improves the problems existing in the feature matching method,and designs a set of character region positioning method based on double-template shape matching.The improved character region positioning method splits the complete character region in all 50 test images.In the tire character detection part,this paper proposes a character detection scheme based on deep learning algorithm,which realizes the tire character positioning and classification task end-to-end through the improved Yolov5 object detection algorithm,and solves the problem that the traditional tire character detection method is difficult to separate dense characters and tire character images with high noise interference.In the tire character logic output part,this paper designs a set of post-processing character sorting scheme to solve the problem that the object detection algorithm cannot logically sort the output results.Realized the logical output of tire character branches.In the character comparison section,this paper uses Pyqt5 to design the human-computer interface.Through the man-machine interface,the image to be inspected,the detection result of the inspection image,the real character data and character comparison result of the database can be obtained in real time,and the test result and the comparison result can be recorded in a single character area,and the real-time display is recorded in the detection frame and the historical information management box to facilitate the future management of the enterprise.