Research And Application Of Optimal Twsvm For The Identification Of Surface Defects In Steel Plates

Steel is considered to be an important standard for the development of a nation’s industrial level and is the main product of metallurgical enterprises,as well as the main raw material for the construction and decoration,machinery manufacturing and aerospace industry.In recent times,China’s aerospace sector has undergone steady advancements and progress,and has made remarkable achievements in the construction of space stations,the construction and construction of which are inseparable from steel plates,the quality control of which,especially the surface quality,is extremely strict.However,during the production of steel plates,surface defects including cracks,scratches and patches often appear due to complex factors such as raw materials,equipment and processing methods.These existing defects,if not cleaned or repaired in time,can adversely affect the appearance,corrosion resistance and fatigue strength of the product.Therefore,accurately identifying the different types of defects on the surface of steel plates becomes a current trendy topic worthy of attention.This paper provides a comprehensive analysis of the difficult problems of steel plate surface defects and constructs an all-round design from image processing to classifier,with the main research elements including the following:(1)Create a suite of image preprocessing process and feature selection method.In view of the unfavourable conditions such as noise points and blurring in the acquired images of steel plate surface defects,this paper firstly introduces a Salient Object Detection algorithm to effectively remove the influence of noise,while accurately acquiring the foreground areas of defects.Secondly,for the morphological differences of different defect types,relevant features are designed,and finally the acquired images are replaced by fused features to build the basics for the subsequently classification.(2)Based on the classification advantages of Twin Support Vector Machines(TWSVM),the Global optimized Twin Support Vector Machine(GTWSVM)algorithm is proposed.In response to the fact that the original TWSVM algorithm simply considers separating dissimilar samples without fully considering the association problem between samples,the linear discriminant analysis algorithm and the nearest neighbour algorithm are used to embed the association between samples at both global and local scales based on the original algorithm to minimise the distance between similar samples while maximising the distance between dissimilar samples to find a better decision hyperplane.Experiments show that this algorithm helps to improve the accuracy of classification compared to the original algorithm.(3)Based on the classification advantages of TWSVM,the Nearest neighbor density assisted fuzzy optimization Twin Support Vector Machine(KDFTWSVM)is proposed.In response to the TWSVM does not selectively assign contribution weights to each sample point during model training,resulting in some useless sample points interfering with the model.Therefore,based on the problems that exist,a new density fuzzy membership function is constructed that reduces the contribution of useless sample points to the model while increasing the contribution of favourable boundary sample points to the model.Finally,based on the TWSVM,the density fuzzy membership function is embedded within the model as an optimisation condition to enhance the generalisation performance of the model.Experiments show that the optimised algorithm has a significant improvement in accuracy when classifying surface defects in steel plates.(4)In this study,a steel plate surface defect identification system is designed and implemented to address the problem that traditional inspection methods require a lot of manpower and time,are less efficient and easy to error.The system adopts a new method and optimized TWSVM algorithm,which can automatically identify and classify steel plate surface defects and improve production efficiency and detection accuracy.The experiments show that the system has excellent performance and has the prospect of wide application.