Research On Surface Defect Detection Method Of Rebar Based On Machine Vision

Rebar is widely used as the basic material of modern construction.During the rolling process,if defects on the surface cannot be found in time,it will result in the production of a large number of waste products.Therefore,it is necessary to quickly detect the defects in rebar to allow the staff to adjust the rolling mill Pressure or replace rolling equipment.At present,most of the rebar defects are manually detected,and there is an urgent need for an automatic and rapid intelligent detection system to solve this problem.This paper studies the detection method of rebar surface defects based on machine vision to improve the reliability,product quality and production efficiency of rebar production,and promote the transformation and upgrading of China’s rebar production industry to "smart manufacturing".This paper first analyzes the types and causes of surface defects of rebar,designs an image acquisition device for surface defects detection of rebar according to the production process requirements,and determines the workflow and software and hardware selection of the defect detection unit.Because the background area of the rebar image is complex and variable,in order to extract the area where the rebar is located,the image needs to be pre-processed first.Based on the analysis and comparison of image preprocessing algorithms commonly used in visual inspection systems,such as image enhancement,threshold segmentation,morphological processing,etc.,combining the advantages of various methods,a rebar image background removal method was designed.Aiming at the defects of bubbles and bumps on the surface of rebar,a defect detection algorithm based on Blob analysis is designed.The method of detecting straight lines by Hough transform is used to locate the edge of the longitudinal rib,and the front and side images are distinguished according to the position of the longitudinal rib.For the front image,the Blob analysis method and template matching method are used to segment the upper and lower longitudinal ribs and the transverse ribs of the rebar.For the side image,the Blob analysis method and frequency domain filtering are used to segment the side images.Then extract several types of features with large differences among various types of defects to form feature vectors,and input the feature vectors into their respective classifiers for recognition and classification.Experiments show that the designed method has good stability and practicability,and can realize the identification of two defects of bubbles and bumps on the surface of rebar.The surface defect detection algorithm based on transfer learning is studied.Use the VGG16 pre-trained network after removing the fully connected layer as a feature extractor,plus a redesigned classifier to form a new migration model,and then use the bubbles,bumps,pits,scratches,and pits in the front image A small sample defect data set for fine-tuning the network model.By changing the three parameters of the model’s learning rate,batch size and number of iterations,the accuracy change curve and the loss function change curve of the network training set and verification set are obtained.Experimental results show that the model has fast training speed and good adaptability,and its recognition effect is good compared with the Blob analysis method.