基于差商的油液监测铁谱图像自适应分割  

Self-adaptive segmentation of oil monitoring ferrographic image based on difference quotient

作　　者：温广瑞[1,2] 徐斌[1] 张志芬[1] 陈峰[1] 

机构地区：[1]西安交通大学智能仪器与监测诊断研究所,陕西西安710049 [2]新疆大学机械工程学院,新疆乌鲁木齐830047

出　　处：《光学精密工程》2017年第5期1322-1330,共9页Optics and Precision Engineering

基　　金：国家自然科学基金资助项目(No.51421004;No.51365051);教育部新世纪优秀人才支持项目(No.NCET-13-0461)

摘　　要：针对油液监测中铁谱磨粒图像分割阈值难以选取的问题,本文提出一种基于差商的自适应铁谱图像分割算法。首先,将铁谱磨粒灰度图像转换成三维灰度直方图,并对其进行切片分析;然后,引入Newton插值多项式,将不同切片所得的频数作为切片灰度-频数曲线的插值点,基于差商构造第一类可接受函数和第二类可接受函数,结合实验数据确定两类误差,选取同时满足两类误差的最小灰度值作为分割阈值;最后,用本文方法对不同类型的磨粒图像以及添加高斯噪声和椒盐噪声后图像分别进行分割实验,并与经典的迭代阈值法、Otsu算法、最大熵法进行了比较。实验结果表明,本文方法受噪声干扰较小,误检率和漏检率整体优于其他3种算法。对分割所得的磨粒图像进行特征提取,并利用支持向量机进行识别,本文方法对3种故障磨粒识别准确率最高,达到82.86%,虽在运行时间上无明显优势,但综合性能最优,能满足油液监测过程中铁谱图像自适应分割的需求。Aiming at problem that segmentation threshold value of a ferrographic image is difficult to select in oil monitoring, a self-adaptive ferrographic image segmentation algorithm based on difference quotient was introduced. Firstly, the ferrographic abrasive particle image was converted into three- dimensional grey histogram and then a slice analysis was made on it; then, by introducing Newton interpolation polynomial, the pixel number obtained from different slices was took as interpolating point of slice grayscale-frequency curve; the first kind of acceptable function and the second kind of acceptable function were established based on difference quotient, and two kinds of errors were identified by combination of experimental data. The minimum gray value which simultaneously satisfied the two kinds of errors was selected as segmentation threshold value. Finally, segmentation experiments on different types of ferrographic images and ferrographic images with Gaussian noise and salt ~ pepper noise were conducted to compare the performance of proposed algorithm and three classical algorithms including iterated thresholding method, Otsu algorithm and maximum entropy. The experimental result indicates that the proposed algorithm is rarely interfered by noise and its average false positive rate and average omission rate is overall superior to other three algorithms. Through conducting feature extraction on ferrographic image and identification by support vector machine, it can be found that the proposed method has the highest identification accuracy rate on three faulted abrasive particles, which reaches 82. 86%. Although there are no obvious advantages on operation time, but the method has optimal comprehensive property and can meet the requirement for making a self-adaptive segmentation on ferrographic image in the process of oil monitoring.

关 键 词：油液监测 铁谱图像 差商 图像分割 

分 类 号：TP394.1[自动化与计算机技术—计算机应用技术]