Study On Evaluation Method Of Wear Particle Segmentation On Ferrograph Image

Ferrography is a wear particle analysis technology developed in the field of tribology, ferrograph image processing is an important content of ferrograph analysis technology. Because of the diversification of the type of the particles and each type has different characteristic in ferrograph image, which increases the difficulty of the segmentation and recognition of ferrograph image. And it makes difficult to achieve accurate segmentation. Thus the quality of the segmentation result is uneven and there is uncertainty in the extraction of wear particles. So it is necessary to evaluate the quality of the segmentation results of ferrograph image.In this paper, we evaluate the quality of the results of the segmentation of wear particle in ferrograph image, as follows:1. According to the segmentation results of background and wear particle in ferrograph image, three evaluation methods are proposed. The first one is a supervised evaluation method based on grey correlation clustering. The second one is an unsupervised evaluation method based on Deng's correlation degree, which does not require a reference sequence. In this method the region with the largest number of neighbor is the background, the rest regions are divided into wear particle, then, calculate the correlation degree between the other regions and the background region, determine whether the region is the background or wear particle according to the threshold, finally, evaluate every region. The third one is a method of unsupervised evaluation, which combining evidence theory and Deng's correlation degree. This method is an improvement of the second one by replacing Deng's correlation degree with the fusion correlation degree. Compared with the other two methods, the third one has a good evaluation result, wide application range and high accuracy.2. According to the results of edge detection of the wear particle in ferrograph image. In this paper, a quantitative evaluation method is presented, which does not require edge reference map. This method mainly contains the reconstruction similarity index fs, the edge confidence index fc, the edge shape index fp and the final evaluation index fIdx. At first the reconstructed image can be obtained by using the interpolation method based on the multiregional and anti mental distance, then calculate the reconstruction similarity index between original image and reconstructed image. In view of the characteristics of ferrograph image and in order to suppress the false edge, the edge confidence index is put forward. Considering the edge connectivity and width uniformity, the edge shape index is proposed. Then these three indicators are integrated into a final evaluation index. Finally, make an evaluation of a single edge detection algorithm and different edge detection algorithm by evaluating the results of edge detection of different background ferrograph image with the proposed index. The evaluation result is exactly the same as that of the human eye, which shows the effectiveness of the proposed method.In the end, this paper designs a quality evaluation system of wear particle segmentation of ferrograph image, which achieves the analysis of the segmentation results of background and wear particle, the results of edge detection of the wear particle in ferrograph image.