Research On The Key Technology Of Steel Plate Surface Detect Detection System Based On Machine Vision

As the basic industry of national economy,steel industry plays an important role in economic construction and social development.Steel plate is the basic material in the fields of automobile manufacture,aerospace,national defense equipment and construction.The quality of steel plate directly affects the quality and performance of the industry product.So the detection of the surface quality of steel plate is also one of the important aspects of steel plate quality detection.With the development of image processing technology,artificial intelligence and neural network theory,the technology of steel plate surface defect detection based on machine vision has become the focus and direction of domestic and foreign scholars and iron and steel enterprises.In this paper it is to focused on the pre-processing algorithm,defect segmentation algorithm,detect feature selection and extraction method,defect classification algorithm of steel plate surface defect image.Main works in this paper are as follows:(1)Under the image pre-processing stage,it is focused on the common uniform illumination algorithms and it is determined the best solutions to uneven illumination by comparing experiment data.The method is studied to improve homomorphic filtering algorithm based on Retinex uniform illumination defect image enhancement algorithm.In order to reduce dust and noise disturbance,common image denoising methods are studied combining experimental analysis,and the final choice is bilateral filtering method.Experiments show that this method not only eliminates the noise,but also retains the details to the greatest extent,which is convenient for subsequent processing.(2)On the basis of obtaining high-definition steel plate images,different image edge detection algorithm and image segmentation techniques are analyzed.Experiments show that the Canny edge detection algorithm which has relatively strong detection ability is the best choice,and an efficient threshold segmentation algorithm of variable coefficient is proposed based on iterative threshold segmentation and adaptive threshold segmentation,which can accurately achieve the target defect region with isolated background.(3)There is different information included in unsegmented and segmented images,so invariant moment feature,texture feature and gray level feature are selected as classification features.Because the singular samples in the extracted feature data will make the subsequent classifier converge very slowly,a variety of data processing methods is studied in this paper,and finally the standard transformation method is selected to standardize the extracted feature data.By using principal component analysis(PCA)to reduce the dimensionality of the extracted features,the feature selection is realized,and the complexity of the subsequent classifier design is reduced,which lays a solid foundation for the classification and recognition of defects.(4)The defect classification algorithm of BP neural network and Support vector machine are studied,including selection of kernel function and the determination of parameters.The experiments show that the classification effect of support vector machine is better than BP neural network.