Application of Fast Feature Pyramids and Soft-Cascade in Image Fault Detection for Angle Cock
-
摘要: 为了提升列车折角塞门的故障检测效率,提出了一种基于快速特征金字塔和Soft-Cascade的故障图像检测算法。首先,构建快速特征金字塔模型来提取图像多尺度聚合通道特征;其次,利用向量化后的多尺度聚合通道特征来训练Soft-Cascade故障分类器;最后,利用训练好的分类器来判断待检折角塞门是否含有故障。实验结果表明:该算法的故障检测正确率为97.33%,离线检测速度高达43 fps(每张图像仅需23 ms),检测效率高于其他算法。该算法训练时间短,检测速度快,硬件要求低,能满足列车折角塞门的故障检测要求。
-
关键词:
- 机器视觉 /
- 折角塞门 /
- 快速特征金字塔 /
- Soft-Cascade算法
Abstract: The fault detection algorithm based on the fast feature pyramids and Soft-Cascade was proposed in order to improve the efficiency of the fault detection for the angled cock. Firstly, the fast feature pyramids model was constructed to extract the image multi-scale aggregate channel features. Secondly, the vectorized multi-scale aggregate channel features was used to train the Soft-Cascade fault classifier. Finally, the trained classifier was used to detect whether the angle cock contains a fault. The experimental results show that the fault detection accuracy of the proposed algorithm is of 97.33%, the offline detection speed is up to 43 fps (only 23 ms per image), and the detection efficiency is higher than that by using other algorithms. The present algorithm has the short training time, fast detection speed and low hardware requirements, which can meet the requirements of fault detection for the angle cock.-
Key words:
- machine vision /
- angle cock /
- fast feature pyramids /
- Soft-Cascade algorithm
-
表 1 实验所用的数据库
名称 正样本 负样本 总数 训练样本 800 800 1600 测试样本 400 400 800 表 2 不同特征对故障检测的影响
特征组合 CDR/% MDR/% FDR/% Gray 40.03 59.97 0 LBP 63.18 18.17 18.65 GradM 53.7 11.41 34.89 HOG 93.25 4.66 2.09 Gray+LBP 44.21 55.15 0.64 GradM+LBP 46.95 53.05 0 Gray+GradM 57.55 0.81 41.64 Gray+HOG 92.45 4.82 2.73 GradM+HOG 94.25 4.50 1.25 LBP+HOG 90.04 2.25 7.71 Gray+GradM+HOG 92.77 3.05 4.18 GradM+HOG+LBP 97.33 1.28 1.39 表 3 不同置信度阈值对故障检测的影响
置信度阈值 CDR/% MDR/% FDR/% 0 89.24 5.78 4.98 5 89.22 5.47 5.31 10 89.87 4.50 5.63 15 90.03 3.54 6.43 20 95 2.41 2.59 25 97.33 1.28 1.39 30 94.88 0.80 4.32 35 89.87 0.48 9.65 40 88.75 0.48 10.77 45 87.79 0.16 12.05 50 87.79 0 12.21 表 4 不同算法性能对比结果
算法 CDR/% MDR/% FDR/% 训练时间/s 测试时间/s LBP+Cascade 48.87 51.13 0 76 0.048 HOG+Cascade 82.96 16.56 0.48 151 0.115 本文算法 97.33 1.28 1.39 84 0.023 -
[1] 孙国栋, 杨林杰, 梅术正, 等.一种基于几何特征的列车集尘器形状匹配算法[J].中国机械工程, 2016, 27(2):230-234, 257 doi: 10.3969/j.issn.1004-132X.2016.02.014Sun G D, Yang L J, Mei S Z, et al. A shape matching method for train dust collector based on geometrical features[J]. China Mechanical Engineering, 2016, 27(2):230-234, 257(in Chinese) doi: 10.3969/j.issn.1004-132X.2016.02.014 [2] 赵大兴, 王博, 周唯倜, 等.基于有向线段重构轮廓与成对几何直方图的缓解阀盖图像匹配算法[J].中国机械工程, 2017, 28(6):721-727 doi: 10.3969/j.issn.1004-132X.2017.06.014Zhao D X, Wang B, Zhou W T, et al. Image matching algorithm of release valve bonnet based on contours reconstructed by oriented line segments and pairwise geometrical histogram[J]. China Mechanical Engineering, 2017, 28(6):721-727(in Chinese) doi: 10.3969/j.issn.1004-132X.2017.06.014 [3] Zhang J M, Wang S M, Yang T X. An automatic method for detecting side frame key using histogram of oriented gradient and local binary pattern[J]. International Journal of Control and Automation, 2017, 10(7):163-178 doi: 10.14257/ijca [4] Liu L, Zhou F Q, He Y Z. Automated visual inspection system for bogie block key under complex freight train environment[J]. IEEE Transactions on Instrumentation and Measurement, 2016, 65(1):2-14 doi: 10.1109/TIM.2015.2479101 [5] 孙国栋, 汤汉兵, 林凯, 等.基于多特征融合和BP-AdaBoost算法的列车关键零件故障自动识别[J].中国机械工程, 2017, 28(21):2588-2594 doi: 10.3969/j.issn.1004-132X.2017.21.013Sun G D, Tang H B, Lin K, et al. Automatic fault recognition for key parts of train based on multi-feature fusion and BP-AdaBoost algorithm[J]. China Mechanical Engineering, 2017, 28(21):2588-2594(in Chinese) doi: 10.3969/j.issn.1004-132X.2017.21.013 [6] 张杨.基于改进马尔科夫随机场与精确高度函数的列车故障图像层次特征匹配[D].武汉: 湖北工业大学, 2017 http://cdmd.cnki.com.cn/Article/CDMD-10500-1017815200.htmZhang Y. Hierarchical feature matching of fault images in TFDS based on improved Markov random field and exact height function[D]. Wuhan: Hubei University of Technology, 2017(in Chinese) http://cdmd.cnki.com.cn/Article/CDMD-10500-1017815200.htm [7] Kim H C, Kim W Y. Automated inspection system for rolling stock brake shoes[J]. IEEE Transactions on Instrumentation and Measurement, 2011, 60(8):2835-2847 doi: 10.1109/TIM.2011.2119110 [8] Zhou F Q, Zou R, Gao H. Dust collector localization in trouble of moving freight car detection system[J]. Journal of Zhejiang University SCIENCE C, 2013, 14(2):98-106 doi: 10.1631/jzus.C1200223 [9] Zou R, Xu Z Y, Li J Y, et al. Real-time monitoring of brake shoe keys in freight cars[J]. Frontiers of Information Technology & Electronic Engineering, 2015, 16(3):191-204 http://cn.bing.com/academic/profile?id=83ba9d028c64749e85242b1723298249&encoded=0&v=paper_preview&mkt=zh-cn [10] Liu L, Zhou F Q, He Y Z. Automated status inspection of fastening bolts on freight trains using a machine vision approach[J]. Proceedings of the Institution of Mechanical Engineers, Part F:Journal of Rail and Rapid Transit, 2016, 230(7):1629-1641 doi: 10.1177/0954409715619603 [11] Nan G, Yao J E. A real-time visual inspection method of fastening bolts in freight car operation[C]//Proceedings of SPIE 9675, AOPC 2015: Image Processing and Analysis. Beijing, China: SPIE, 96752G http://adsabs.harvard.edu/abs/2015SPIE.9675E..2GN [12] Dollár P, Appel R, Belongie S, et al. Fast feature pyramids for object detection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2014, 36(8):1532-1545 doi: 10.1109/TPAMI.2014.2300479 [13] Bourdev L, Brandt J. Robust object detection via soft cascade[C]//Proceedings of 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. San Diego: IEEE, 2005: 236-243 https://www.researchgate.net/publication/4156416_Robust_object_detection_via_soft_cascade [14] Liu L, Zhou F Q, He Y Z. Automated visual inspection system for bogie block key under complex freight train environment[J]. IEEE Transactions on Instrumentation and Measurement, 2015, 65(1):2-14 http://www.wanfangdata.com.cn/details/detail.do?_type=perio&id=45edc2ad9ff31ca33242fbd6f79c758c [15] Everingham M, Eslami S M A, Van Gool L, et al. The pascal visual object classes challenge:a retrospective[J]. International Journal of Computer Vision, 2015, 111(1):98-136 http://cn.bing.com/academic/profile?id=3e662660e695a9372b17f5b7d0000495&encoded=0&v=paper_preview&mkt=zh-cn