Research On Machine Vision Based Image Recognition Method For Steel Plate Surface Defects

The development of the country’s economy is inseparable from the support of steel materials,and the current development of the modern economy has increasingly high requirements for the quality of steel products.Due to the influence of the environment and equipment in the manufacturing of steel plates,the surface will produce various and complex defects such as inclusions,scratches and pitting,which will seriously affect the quality and performance of steel plates,so it is necessary to effectively detect the defects of steel plates.In this paper,the surface defects of hot-rolled strip steel are the object of study,and the detection method of machine vision is used to study and design the image classification recognition and target localization detection method of steel plate surface defects,mainly carrying out the following work.(1)Steel plate surface defect image pre-processing technology research.In order to improve the quality of the defect image,three aspects of image pre-processing are required before the image feature extraction:① for the problem of noise in the original image,four kinds of noise removal algorithms are compared using the peak signal-to-noise ratio,and finally the Wiener filter noise removal algorithm is selected;② in order to increase the contrast between the defect and the background in the image,three classical enhancement algorithms are compared using the average gradient value,and finally the Laplace enhancement algorithm;③to address the shortcomings of OTSU segmentation techniques,a 2-dimensional OTSU image segmentation algorithm combined with the improved whale algorithm is proposed,i.e.,the improved whale algorithm is used to find the optimal threshold of the OTSU method.From the experimental results of defect segmentation,the segmentation algorithm proposed in this paper has more satisfactory performance in steel plate surface defect images,with shorter running time and reduced 0.55 s.(2)Research on feature extraction and selection methods for steel plate surface defect images.In order to make full use of the image information,44 grayscale features,geometric shape features and texture features are extracted from the pre-processed steel plate defect image,and 27 features with more contribution and more recognition information are selected with a threshold value greater than 0.2 using Fisher’s criterion.(3)Research on the classification and recognition algorithm of steel plate surface defects.For the problem of complex and diverse defect types,the BP neural network,support vector machine(Support Vector Machine,SVM)and improved SVM models were used for classification and recognition experiments.The results show that the improved SVM model has the best classification recognition effect,with the highest classification accuracy(95.54%)and the lowest accuracy standard deviation(0.59),which proves that the improved SVM model has stronger generalization ability and stability.(4)Research on surface defect localization detection algorithm for steel plates.To address the shortcomings of small defect images on the steel plate surface and the shortcomings of ResNet-50 residual network feature extraction,two parts of the Faster R-CNN network,the large convolution kernel and the residual structure of feature extraction,are optimized to reduce the feature parameters,deepen the network depth,and improve the defect location recognition rate.From the experimental results,it is concluded that both the two individual optimization algorithms and the combined optimization algorithm improve the comprehensive detection effect of the original model,in which the average accuracy of the combined optimization algorithm for six defects reaches 82.47%.