Research On Surface Defect Recognition Of Steel Strips Based On AdaBoost Classifier

With the rapid development of domestic and international economy,strip steel is widely used in production and life,and its surface quality has been paid more and more attention.Through the analysis of the development status of strip steel in the world,it is found that the samples have many disadvantages such as low contrast and uneven gray scale,which seriously depress the product quality and restrict the development of enterprises.Therefore,it is an urgent task to effectively identify the defects of the strip surface and improve the quality of the strip surface.Firstly,based on image noise characteristics,adaptive median filtering method in the process of image denoising is proposed in this thesis.This method can effectively remove image noise and preserve image details and obviously reduce the image blurring that caused by filtering.In edge detection algorithm,the Canny operator is used to extract the defect edge,strengthening the edge feature and improving the image segmentation quality obviously.From the experimental results,the segmentation effect is satisfactory.Secondly,in the aspect of feature extraction,the texture features and shape geometric features,which can characterize the feature of the defect image,a total of 40 dimensional feature as the original data set,are extracted.Considering the recognition speed and reliability,a improved ReliefF feature selection algorithm is putted forward in this thesis.This algorithm can remove the correlation among the characteristics and reduce the redundancy among features.Experimental results show that using this algorithm to extract features can be more effective in identifying strip defects.Finally,based on analyzing the advantages and disadvantages of several classifiers and the scope of application,the AdaBoost classifier is proposed in the thesis.The classifier needs fewer samples in the training process,reduces the training time and improves the accuracy.The simulation results show that the proposed classifier significantly improve the accuracy of classification of strip defects.