陈亮,张浩舟,燕浩
Chen Liang,Zhang Haozhou,Yan Hao

• 1:安徽省科学技术研究院
• 2:合肥工业大学机械工程学院

摘要(Abstract)：

尿素泵为机动车尾气后处理系统的核心设备,泵体材料一般为铝型材,在铝型材生产过程中,受工艺等因素的影响会产生各种瑕疵,影响铝型材的质量。传统人工检测,质检的效率和准确率难以满足生产需要。本文将深度学习算法引入到缺陷检测中,结合迁移学习原理,使用小批量数据集,利用改进的YOLO模型进行训练,预测铝型材表面瑕疵。试验结果显示,尽管在小批量训练的条件下,验证集mAP值为87.43%,仍取得了98.2%的准确率,比拟人工检测的准确率,并可以快速、准确的定位缺陷部位。此技术有望革新现有质检流程,自动完成质检任务,保证产品的质量;另外,基于深度学习算法表面缺陷检测方法,鲁棒性好,具有一定的普适性,可以推广到相关的其他应用领域。
Urea pump is the core equipment of vehicle exhaust post-treatment system.Material of pump body is generally aluminum profiles.In the production process of them,due to the influence of process factors,various defects will appear,which will affect the quality of aluminum profiles.Traditional manual detection methods cannot meet the needs of production,because of its efficiency and accuracy.In recent years,deep learning algorithm.The deep learning algorithm was introduced into flaw detection.In combination with transfer learning principle and by using a small-volume dataset,the YOLO model was used for training to predict the defects of aluminum profile.The experimental results show that under the premise of small batch training,the mAP value of verification set was 87.43%,but the accuracy rate of 98.2% was still obtained,which is comparable to the accuracy of manual detection,and the defect can be quickly and accurately located.This technology can be expected to innovate existing quality inspection processes,automatically complete quality inspection tasks and guarantee product quality.In addition,the defect detection method based on deep learning is robust and has certain universality,which can be extended to other related application areas.

关键词(KeyWords)： 深度学习;迁移学习;表面瑕疵检测;YOLO模型;DCNN
deep learning;transfer learning;defect detection;YOLO model;DCNN

Abstract:

Keywords:

基金项目(Foundation):

作者(Author): 陈亮,张浩舟,燕浩
Chen Liang,Zhang Haozhou,Yan Hao

参考文献(References)：

• [1]Bordes A,Glorot X,Weston J,et al.2012,Joint learning of words and meaning representations for open-text semantic parsing[J].Artificial Intelligence and Statistics,127-135.
• [2]Ranzato M A,Susskind J,Mnih V,et al,2011,On deep generative models with applications to recognition[C]//CVPR 2011.IEEE,2857-2864.
• [3]Liu W,Anguelov D,Erhan D,et al.2016,Ssd:Single shot multibox detector[C]//European Conference on Computer Vision.Springer,Cham,21-37.
• [4]Lowe D G.Distinctive image features from scale-invariant keypoints[J].International Journal of Computer Vision,2004,60(2):91-110.
• [5]Dalal N,Triggs B.2005,Histograms of oriented gradients for human detection[C]//2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'05).IEEE,1:886-893.
• [6]徐贵力,毛罕平.利用傅里叶变换提取图像纹理特征新方法[J].光电工程,2004,31(11):55-58.
• [7]吴高洪,章毓晋,林行刚.利用小波变换和特征加权进行纹理分割[J].中国图象图形学报A辑,2001,6(4):333-337.
• [8]Lyu X.A novel defect inspection approach using image processing and support vector machines in bolts[C]//2015 Seventh International Conference on Measuring Technology and Mechatronics Automation,2015:40-43.
• [9]王宪保,李洁,姚明海,等.基于深度学习的太阳能电池片表面缺陷检测方法[J].模式识别与人工智能,2014(6):517-523.
• [10]崔云翔,蔡颖玲.上海地区太阳能-地源热泵冬季联合运行试验研究[J].流体机械,2019,47(2):65-69.
• [11]卢晨,杨洪兴,吕建,等.氨吸收式太阳能热泵-极式换热器联合供暖系统的理论与试验研究[J].流体机械,2018,46(2):60-67.
• [12]毛涛涛,潘国兵,欧阳静,等.太阳能汽车续航模型及光储优化配置方法[J].机电工程,2018,35(1):68-72.
• [13]Hinton G E,Osindero S,Teh Y W.A fast learning algorithm for deep belief nets[J].Neural Computation,2006,18(7):1527-1554.
• [14]Goh H,Thome N,Cord M.et al.Learning deep hierarchical visual feature coding[J].IEEE Transactions on Neural Networks and Learning Systems,2014,25(12):2212-2225.
• [15]Liu W,Wang Z,Liu X.et al.A survey of deep neural network architectures and their applications[J].Neurocomputing,2017,234:11-26.
• [16]Ren S,He K,Girshick R,et al.Faster r-cnn:Towards real-time object detection with region proposal networks[C]//Advances in neural information processing systems,2015,91-99.
• [17]Gur,David,Wagner,Robert F,Chan,et al.On the repeated use of databases for testing incremental improvement of computer-aided detection schemes[J].Academic Radiology,2004,11(1):103-105.
• [18]Shin H,Roth H R,Gao M.et al.Deep convolutional neural networks for computer-aided detection:CNN architectures,dataset characteristics and transfer learning[J].IEEE Transactions on Medical Imaging,2016,35(5):1285-1298.
• [19]Lin T Y,Maire M,Belongie S.et al.Microsoft COCO:Common objects in context[C]//Computer Vision ECCV 2014.Springer,Cham,2014,740-755.
• [20]Smirnov E A,Timoshenko D M,Andrianov S N.Comparison of regularization methods for imagenet classification with deep convolutional neural networks[J].Aasri Procedia,2014,6:89-94.
• [21]Pan S J,Yang Q.A survey on transfer learning[J].IEEE Transactions on Knowledge and Data Engineering,2010,22(10):1345-1359.
• [22]Hu F,Xia G S,Hu J,et al.Transferring deep convolutional neural networks for the scene classification of high-resolution remote sensing imagery[J].Remote Sensing,2015,7(11):14680-14707.