Texture Extraction And Identification Of Wear Particles Based On Co-occurrence Matrix And Hough Transform

During operation of the mechanical equipment,different kinds of wear particles will be produced due to mechanical wear.Abnormally large wear appearing often indicates that the running state of mechanical equipment is in some problem,and need to stop in time to check.Wear particles analysis and identification is the most important part of ferrographic image analysis technology.Getting the feature of wear particles such as texture feature and shape feature and determining the type of wear particles play an important role in ferrographic image analysis technology which is moving toward to the derection of intelligence and automation.The main research content is as follows:Traditional Otsu threshold method in removing background often make some error segmentation,so in the HSV color space Otsu threshold background deduction algorithm is proposed.The algorithm obtains a good effect.Region marking method based on neighborhood is studied,and a wear particles segmentation algorithm based on region marking is proposed.Two texture extraction algorithms based on gray level co-occurrence matrix and gray gradient co-occurrence matrix are studied,and the two kinds of algorithm is applied to extracte texture feature of the fatigue and severe sliding wear particles.Using SVM based on extracted texture features classify the wear particles.Finally,the advantages and disadvantages of the two algorithms are compared through experiments.The principle of Hough Transform is studied,and the algorithm of particle texture extraction based on the Hough Transform is proposed.Firstly,the algorithm uses Canny operator to deal with gray image,and get the texture seed image.The image contains pixels with significant changes in gray.Secondly,the linear or circular texture primitives are detected by Hough Transform.In the end,the identification of the abnormal wear particles,such as severe sliding and fatigue wear particles,is carried out based on detection of texture primitives.The experimental results show that the proposed algorithm based on HoughTtransform gets a good effect.Compared with the texture extraction method based on co-occurrence matrix,the algorithm does not require training samples,and is consistent with the visual perception?The algorithm is an effective method of texture extraction and identification of wear particles.