作者：郭雪梅1, 刘桂雄2
作者单位：1. 广东省自动化研究所, 广东 广州 510070;
2. 华南理工大学机械与汽车工程学院, 广东 广州 510641
关键词：多颜色模型;k近邻算法;自学习;浪涌测试
摘要：
在浪涌测试中,由于每次识别对象不同,直接采用特征匹配每次测试前需要根据受试设备重新训练样本。先根据图像中高亮度点、白光所占比例,决策用于图像分割的颜色模型(L*a*b*、HSL、HSV),实现自适应分割;其次,提出自学习k-NN算法,以像素数n、偏心率e、密实度比r、欧拉数E为样本S特征向量X,构建数据集T0,以欧氏距离D实现样本分类;若样本置信度为k,加入预备数据集Tz中,当Tz满足条件,则扩充数据集Tz形成数据集Tz+1。结果证明:算法在9组各类样本(共21 600帧图像)识别中,准确度可达98.65%;并自学习扩充5组样本,距离矩阵变化较小,可见算法学习效率、学习准确度较高。
Multi-color space threshold segmentation and self-learning k-NN algorithm for surge test EUT status identification
GUO Xuemei1, LIU Guixiong2
1. Guangzhou Institute of Automation, Guangzhou 510070, China;
2. School of Mechanical and Automotive Engineering, South China University of Technology, Guangzhou 510641, China
Abstract: As the identification objects vary in different surge tests, new samples need to be trained for equipment under test each time when feature matching is used to identify equipment status. Therefore, multi-color space threshold segmentation and self-learning k-NN algorithm were proposed. First, color models (L*a*b*, HSL, HSV) for image segmentation were selected to realize self-adaptive division according to the proportions of high luminance points and white luminance points in the image. Second, k-NN algorithm was proposed to construct a data set T0 via a feature vector X of the sample S formed by pixel n, eccentricitye, compactness r and Euler's numbers E, and the sample S was classified through Euclidean distance D. Third, sample confidence coefficient k was added into a preliminary data set Tz'. When Tz' met the conditions, the data set Tz was expanded to form data set Tz+1. The results show that the accuracy is up to 98.65% after the k-NN algorithm is used to identify nine groups of different samples (totally 21 600 frames of images), and learns to expand to five additional samples. Moreover, the changes in distance matrix are small. It is thus evident that this algorithm is high in learning efficiency and accuracy.
Keywords: multi-color space;k-nearest neighbor algorithm(k-NN);self-learning;surge test
2016, 42(4): 107-110 收稿日期: 2015-12-21;收到修改稿日期: 2016-01-13
基金项目: 广东省前沿与关键技术创新专项(509164744030)
作者简介: 郭雪梅(1975-),女,辽宁沈阳市人,硕士,研究方向为自动化与信息工程。
参考文献
[1] WILLIAMS T, FLOOD J. Guidance note addresses surge test problems[J]. Compliance Engineering,2002(19):47-49.
[2] DASILVA D A, DACOSTA E C M, DEFRANCO J L,et al. Reliability of directly-molded polymer surge arresters:Degradation by immersion test versus electrical performance[J]. International Journal of Electrical Power and Energy Systems,2013(53):488-498.
[3] XU Y, MAENO K, NAGAHARA H, et al. Camera array calibration for light field acquisition[J]. Front Comput Sci,2015,9(5):691-702.
[4] WANG X D, WU X J, ZHANG X R. Multi-resolution texture synthesis from turntable image sequences[J]. Front Mech Eng,2012,7(1):72-80.
[5] AMATO G, FALCHI F, GENNARO C. Geometric consistency checks for kNN based image classification relying on local features[J]. 4th International Conference on SImilarity Search and APplications, SISAP 2011, June 30,2011-July 1,2011:81-88.
[6] 华才健,蹇红梅,张燕. 基于机器视觉的火花塞间隙检测技术[J]. 中国测试,2014,40(5):108-110.
[7] 陈耿新,黄坚,刘桂雄. 基于可信k-NN的面向EMC浪涌测试多状态灯模式识别[J]. 电子测量与仪器学报,2015(11):1718-1724.
[8] 顾明,郑林涛,尤政. 基于颜色空间转换的交通图像增强算法[J]. 仪器仪表学报,2015(8):1901-1907.
[9] SOLEIMANIZADEH S, MOHAMAD D, SABA T, et al.Recognition ofpartially occluded objects based on the three different color spaces(RGB,YCbCr,HSV)[J]. 3D Research,2015,6(3):1-10.
[10] CHERNOV V, ALANDER J, BOCHKO V. Integer-based accurate conversion between RGB and HSV color spaces[J]. Computers & Electrical Engineering,2015,46(8):328-337.
[11] PENG Z Y, CHANG F L, DONG W H. Color harmony system based on lab perceptual uniform color space[C]//Communications in Computer and Information Science,2015(525):45-54.
[12] KAARNA A, LIU W, KLVIINEN H. Development of color density concept with color difference formulas in respect to human vision system[J]. Front Elect Electr Eng Chin,2011,6(2):381-387.
[13] 黄坚,刘桂雄,王小辉. 滤光片表面缺陷视觉检测系统多处理器调度优化[J]. 中国测试,2015,41(10):90-93.
