Research On Cold-rolled Aluminum Surface Defects Classification Based On Support Vector Machines

With the rapid development of social modernization and the increasing market competition,people have more and more demands for manufacturing line,which not only meet the needs of high efficiency but also guarantee the quality of products.Metal aluminum in transportation,tool manufacture,the application of space science and technology,people's daily life plays an important role.However,in the aluminum production process,due to the outdated production equipment,production batch or quality of raw materials and different production processes and other reasons the aluminum surface have a number of defects.These defects reduce the product appearance,corrosion resistance and wear resistance in a large extent and has brought huge losses to the cold-rolled sheet production enterprises.Therefore,it is very important to detect and classify the surface defect of cold rolled aluminum sheet.In this paper,the process of classifying the surface defects of cold rolled aluminum sheet,using the support vector machine classification algorithm and the BP neural network classification algorithm,by analyzing the experimental results showed that BP neural network classification method of surface defects of cold rolled aluminum plate overall recognition accuracy is not high,the classification process needs a long time,the poor generalization ability and the classification model training requires a lot of defect sample,but the correct classification of oil stain rate is higher;Support vector machine(SVM)classification algorithm for the defects of the overall classification accuracy rate is higher,but for the oil stains and other complex defects recognition rate but not up to the satisfactory effect.In order to make the recognition rate of the surface defects of cold rolled aluminum sheet and the recognition rate of single category defects all reach a high level,a classification algorithm based on support vector machine and BP neural network is proposed in this paper.Firstly,the method of reducing background and median filter is used to pre process the surface defect image of cold rolling aluminum plate,and then the defect region is separated from the surface of cold rolled aluminum sheet by adaptive threshold method;In feature extraction,the gray feature,shape feature and geometric feature are extracted by combining the target image and the segmented image;Training support vector machine(SVM)and BP neural network combined classification model,BP neural network hidden layer uses a layer structure,according to certain rules by comparing the results of several experiments to set the number of hidden layer nodes;Radial basis function is used as kernel function of SVM,the penalty factor and kernel parameter are determined by cross validation method and using one to one classification strategy to carry on the classification of many kinds of defects;Using a combination of support vector machine(SVM)and BP neural network classification model to classify the test sample,the BP model of artificial neural network to classify judgment sample tests whether the oil stain,if considered to be the oil stain and output classification results,if that is the second type of defects and put the sample sent to support vector machine classification model to classify,which is divided into bubbles,broken skin,scratches,holes,black,the final output of the classification results.In this paper,the classification of the six types of cold rolled aluminum plate surface defects,mainly through the OpenCV and VC++ programming to complete the relevant experiments.Finally,the experimental results show that the classification algorithm based on support vector machine and BP neural network has high classification accuracy and can meet the requirements of real-time.