Research On Surface Defect Recognition Algorithm Of Steel Workpiece Based On Machine Vision

Iron and steel industry has a pivotal position in the economy in our country.Although China's iron and steel output is in the forefront of the world,there are still many problems.For steel workpieces,in the process of production and use,due to uneven force,unstable optical elements and other factors,the surface defects of workpieces are often caused.These defects not only affect the accuracy of products,but also easily breed safety risks.At present,many enterprises use artificial method to detect the defects of steel workpiece.The results of manual inspection are not only affected by subjective emotions,but also have low precision and efficiency,and the labor cost of manufacturers is also relatively high.In recent years,with the rapid development of artificial intelligence technology,the application of machine vision in defect detection is more and more.For some manufacturers,using machine vision detection instead of the initial artificial visual inspection is the only way to improve the production level and quality.Aiming at the actual needs of steel workpiece surface defect detection,this paper proposes a set of workpiece surface defect detection algorithm based on machine vision:Firstly,the camera calibration principle and the transformation relationship between image coordinate system and other coordinate system are studied.The distortion of camera is explained,the calibration of camera is completed,and the camera internal reference matrix is obtained.This paper expounds the image preprocessing technology involved in the detection process,introduces the image filtering methods,compares the results of each filtering method,and finally chooses wavelet median filtering as the preprocessing method.Then,several commonly used edge detection operators are introduced,and according to the comparison of experimental results,Canny edge detection operator with the best detection effect is adopted.Aiming at the problems that the traditional Canny operator is sensitive to noise and has low adaptability,an improved algorithm based on Ostu and morphology is proposed.By comparing the experimental results of the improved algorithm with other algorithms,it is found that the improved algorithm has obvious advantages in edge detection.Secondly,in order to classify the defects of workpiece,the characteristics of two image feature extraction methods,local binary pattern and local phase quantization,are studied,and a feature extraction method combining the two methods is proposed,which makes the extracted image details more abundant and directly affects the accuracy of later classification.Support vector machine(SVM)model is introduced to classify defects.The penalty factor and kernel function in the algorithm determine the classification performance.In order to get the best classification effect,the support vector machine model is modified by using the optimization ability of particle swarm optimization to select the best two parameters.The recognition accuracy of the improved SVM algorithm is improved by 18.33%.Once more,a defect detection model based on yolov4 algorithm is proposed.Aiming at the problem that the original algorithm model is not effective for small target defect detection on steel workpiece,K-means + + algorithm is selected to optimize the anchor frame,and the network structure is improved by adding a feature scale,which makes the detection layer divide the image grid more finely.And,the data set is expanded by image enhancement.On the surface of experimental results,the total map value of the improved algorithm is improved by 12.1.Finally,the platform of workpiece surface defect online detection is designed,and the specific structure and content of motion system,vision system and upper computer system of the whole platform are introduced in detail.Combined with the structure and function requirements of the experimental platform,the selection of PLC main control equipment and its supporting servo motor and controller is completed.According to the accuracy and installation requirements of vision system,the selection of light source,industrial camera and lens is completed.The whole set of visual inspection system can realize image acquisition,real-time defect warning and other functions.