Research On Defect Detection System Of Rebar Head With Thread Based On Machine Vision

As the key part of rebar mechanical connection,the quality of the rebar head with thread is an important factor affecting the overall quality of the building.The commonly used inspection methods for rebar head is low in efficiency and poor in accuracy and no longer meet the needs of modern inspection.In this paper,a machine vision based defect detection system for rebar head with thread is studied.(1)The design of the rebar head with thread defect detection system is completed and a detection platform is built.The detection system consists of three parts: image acquisition module,image processing module,and software interface module.The image of the rebar head with thread is collected by the testing platform and transmitted to the PC in real time and detected and recognized and the test result is displayed through the software interface.(1)A dimension detection method of rebar head defect is proposed.Aftrer camera calibration is performed,the image is pre-processed and through ROI extracted to segment the measurement area,after that,Harris Corner detection and Canny Edge Detection are used to achieve the extraction of rebar head measurement points,and finally through the dimensional parameter measurement algorithm to obtain the measurement parameters to complete the dimensional detection.(3)An improved surface defect detection method for rebar head with thread was proposed,and a data set for rebar head with thread was established.FPN feature pyramid network,DCN,and ROI Align were used to replace ROI Pooling in the network structure.OHEM,Soft-NMS and multi-scale training are used in the testing stage to improve the detection accuracy of surface defects of rebar head with thread.(4)The design of the host computer software for detecting the defect of the threaded rebar head with thread is completed.The software interface is written by QT and HIK SDK.The software implements the detection of dimensional defects and surface defects of rebar head with thread.The test results show that the accuracy of the dimension detection algorithm studied in this paper can reach 0.05 mm and the accuracy of surface defect detection can reach 82.82%,which meets the requirements of the system.