The Reasearch On Recognition Of Steel Strip Surface Defect Based On Support Vector Machine

The cold steel strip is one of the absolutely necessary products used in electrical home appliances and petrochemical,and so on.The surface quality of steel strip has a direct effect on the product quality,therefore,many enterprises pay the most attention to control and improve the surface quality of steel strip,and it is one of the important problems concerned by enterprises,automatic surface quality inspection technology of cold steel strip has been attracted increasing attention.Pattern recognition is the key step of the whole system,and the research has been carried out on the preprocessing,feature extraction and selection and pattern recognition of surface defect images of cold steel strip.The influence of noise and optical system can make image degradation which can affect the classification directly,so the research has been carried out on the preprocessing of surface defect images of cold steel strip at first.A method which combines the adaptive median filter and homomorphic filter has been adopted to solve the problems mentioned above,adaptive median filter is an effective method for image denoise,homomorphic filter can remove nonuniform illumination of image,and finally,the method mentioned above can improve image quality effectively.Secondly,texture features and morphology features are extracted and they are used as initial features.But the mixed initial features may have some problems such as redundant features or overhigh dimension which can influence the effect of pattern recognition,therefore,a method which combines ReliefF and clustering algorithm has been adopted to select the initial features.ReliefF can select features which related to classification and clustering algorithm can remove redundant features,the basic principle of the method has been described detailly.Finally,support vector machine has been used for pattern recognition of defects,k nearest neighbor algorithm has been added to support vector machine because of the support vector machine often classify the samples nearing the super plane wrongly,k nearest neighbor algorithm also has been improved aimed at the problem that it usually classifies samples wrongly because of uneven distribution of sample category,and then a new KSVM algorithm has been put forward.The results of experiment show that the performance of the classifier improved is better than the traditional support vector machine.