Design And Application Of Steel Plate Number Detection And Recognition System Based On Machine Vision

Under the background of the rapid development of national infrastructure construction and the rapid development of society and economy,the demand for steel plates in various production activities continues to rise.The steel plate number(hereinafter referred to as the steel plate number)is the only code marked in the production process of the steel plate,and it is a string of numerical characters and texts marked by the enterprise for each plate.This number reflects the production process,steel material,manufacturing process,production use and other information of the steel plate,and is of great significance in the quality tracking,quality traceability and inventory control of the steel plate.When there is an error in the plate number record,the production process is cumbersome,and there will be problems such as incorrect tracking information,loss of plate,and disordered materials on the huge steel plate production line.In order to meet the needs of plate number detection and identification of the plate production department of a large domestic steel mill,this thesis builds a plate number detection and identification system based on machine vision technology based on its existing production line and automated equipment system.The system includes the research content and the result is as follows:1)In the text detection module,this thesis designs a text detection algorithm that uses DBNet coarse positioning and traditional visual algorithm to correct the text detection algorithm,which achieves 99.9% and 99.3% accuracy in the detection of printed and handwritten plate numbers,respectively.2)In the recognition module of this thesis,an end-to-end text recognition algorithm is designed in this thesis.The main body adopts the Bi LSTM-Attention structure,and the accuracy rates of 99.5% and 86.15% are respectively achieved in the recognition of printed and handwritten plate numbers.3)In terms of data enhancement,this thesis develops an image restoration algorithm for jet-printed text in this industrial scenario,which effectively solves the problem of image damage caused by the interference of some data in the database by strong production light sources,enriches database samples,and improves data utilization.The research of this thesis realizes the rapid detection and identification of steel plate numbers on the production line,meets the requirements of system accuracy,runs stably,has strong practicability,and promotes intelligent information processing and information-based material tracking in the steel plate production process.