Study On Evaluation Methods Of Ferrography Image Segmentation Quality

Ferrography is a wear particle analysis and machine condition monitoring technology whichappeared in the1970s. Currently, it has become one of the most economic and effective methods ofwear detection. The processing and analysis of ferrograph image is the key of modern ferrographtechnology. Achieving the accurate segmentation of wear particle has an important impact on thesubsequent particle analysis and recognition, even the wear condition detection and fault diagnosis ofa mechanical system. Ferrograph image is complex and contains a lot of information. When it issegmented, over-segmentation or under-segmentation phenomenon will appear easily. So, an effectiveevaluation method of ferrograph image segmentation is needed which not only can improve theperformance of the algorithm, but also is the basis and foundation of subsequent abrasiveidentification and fault diagnosis.In this paper, the research on evaluation methods of ferrograph image segmentation quality hasbeen studied, as follows:(1) Aiming at evaluating the segmentation quality between background and wear particle, anevaluation method based on grey relational analysis is proposed. Firstly, this evaluation method needsdo statistics on the ratio of each color component of different background color ferrograph images inRGB space and CIEL a b space to determine the background and wear particle referenceinformation Secondly, the segmentation result of ferrograph image is obtained using Otsu method andwatershed algorithm. This step can mark each region of the segmented image. Then, based on themarked point, the sequence to be evaluated of each region is obtained. Thirdly, the grey relationalgrades between the sequence to be evaluated and reference sequence are computed using the graycorrelation model. At last, by the size of correlation, each region belongs to the background or wearparticle is figured out, and the ferrograph image segmentation quality is evaluated.(2) Aiming at evaluating the segmentation result of wear particles, an evaluation method offerrograph image segmentation results based on grey clustering is proposed. On the basis of study onevaluation indexes, this method designs "cluster number" as a new evaluation index. This study takesWatershed algorithm segmentation results under different gradient threshold as evaluation objects,the area factor, shape factor, Hamming distance, centroid distance and cluster number as evaluationindexes, superior, normal and poor as ash classes. At last, clustering coefficient vector matrix is got.Through the data of its rows and ranks, all segmentation results are ranked and sorted. In this paper, Visual C++6.0is used as a software platform and OpenCV library used to fulfilledthe algorithms. The experimental results demonstrate that the evaluation methods based on greysystem theory can be used to evaluate ferrograph image segmentation quality and they are have goodfeasibility and effectiveness.