Recognition And Classification Of Cast Aluminum Alloy Defects Based On Fractal Theory Research

Aluminum alloy material is largely used among metal material. With the rapiddevelopment of materials engineering, higher requirement is needed for metalmaterials and demand is acceleratingly growing. However, major workplace accidentsfrequently occur due to material defects, bringing huge losses and threatening thesafety of civilians seriously. Metallographic analysis is an important means ofstudying and function test. With the aid of computer image processing technology,metallurgical defect can be identified and classified, avoiding the inefficiency ofmanual analysis, labor intensity and poor reproducibility.Research object of this essay is metallographic image of casting aluminum alloy,and metallographic image features are extracted and classified by the means of aseries of digital image processing methods. In order to improve the accuracy andcalculation speed of metallographic analysis, this essay takes on several researches toimprove related image processing method according to the characteristics of thecasting aluminum alloy defect images, realizing correct classification of imagedefects.This article is organized as follows:(1) Preprocessed by anisotropic diffusion filter technology, noise is removed andcharacteristic boundary is preserved within metallographic image. Not only enhancingthe original image, but also correcting uneven brightness phenomenon, consequentlyproviding good preparation for image segmentation.(2) An improved image segmentation method is proposed. Renyi entropy isintroduced into Otsu segmentation method, the objective function of the segmentationmethod is maximized within between-class variance and Renyi entropy of the image isalso maximized. After experimental comparison, it is proved that the effectiveness isimproved by modified Otsu segmentation.(3) Characteristic parameters of segmentation image are extracted by applyingfractal theory into metallurgical defect image feature extraction. This essay mainlyadopts box dimension, widely used in fractal theory. First, we extract the segmentedimage, after that describe defect characteristics of area, compactness, irregularity,invariant moment and fractal parameters for the purpose of image recognition andclassification.(4) PCA method is harnessed for dimension reduction of original multi- dimensional data firstly, and then FCM is taken for fuzzy clustering analysis to obtainfuzzy membership matrix. After fuzzy clustering, processed data is trained andclassified through SVM classifier, and the classification result of defects is obtainedthrough experiments.Finally, the work of this paper is summarized, and the future research direction isalso discussed.