验证图像识别在螺栓松动中的有效性.docx

1、研究生周工作总结表（每周）姓名时间本周实验工作进展：实验方案设计：实验目的：验证图像识别在螺栓松动中的有效性实验装置：M6螺栓（外六角，银白）M6螺栓（外六角，黑）M6螺栓（内六角，黑），木板（灰，黑，红，蓝，原色）每木板预计：20螺栓实验流程设计：1 .验证RCNN在螺栓定位中的精确度，并与霍夫变换定位作比较原色木板上，订10个黑色外六角螺栓与IO个银白色外六角螺栓。RCNN训练集：单独黑色外六角、单独银白色外六角、混用分别观察其识别精确度同色木板上：灰色木板对银白色螺栓，黑色木板对黑色螺栓再次观察两种方法识别精度2 .混合测试：本周具体进展情况分别对5种不同色木板，各放置七个外六角银白、外

2、六角黑与内六角黑螺栓，测试方法有效性。RCNN方法预用：网络选取：alexnet （预训练后网络）利用论文中图像数据，简单应用RCNN方法进行目标检测:手动标注训练集：轮II送代II经过的时间I小枇里揆失II小此里准确度II小批里RMSEIRPNMtni-batAccuracy(hh：m：ss)I1I1I00:00:04I1.9627I40.94%0.16I67.8I5OI00:01:17I0.7126I97.04%I0.13I100.15I1OOI00:02:30I0.4789I96.25%I0.11I100.2OI140I00:03:24I0.1655I98.24%0.11I100.精度一

3、般，预计为训练量太小(仅7张图片)参考文献1. Image-basedBolt-IooseningDetectionTechniqueofBoltJointinSteelBridgesJ.H.Parkl,T.H.Kim2,J.T.Kim32. Bolt-LooseningMonitoringFrameworkUsinganImage-BasedDeepLearningandGraphicalModelHaiChienPham1,Quoc-BaoTa23. Quasi-autonomousbolt-looseningdetectionmethodusingvision-baseddeeplear

4、ningandimageprocessingThanh-CanhHuynha,Jae-HyungParkb4. Vision-basedtechniqueforbolt-looseningdetectioninwindturbinetowerJae-HyungParka,Thanh-CanhHuynhb,5. ImageRegistration-BasedBoltLooseningDetectionofSteelJoints,byXiangxiongKong*0rcIDandjianLiOrcID6. 1.atinAmericanJournalofSolidsandStructures.Pri

5、ntversionISSN1679-78170n-lineversionISSN1679-7825.Lat.Am.j.solidsstruct.vol.14no.12RiodeJaneiroDec.20177. Fullyautomatedvision-basedloosenedboltdetectionusingtheViola-JonesalgorithmLovedeepRamana,WooramChoi,Young-JinCha8. Boltlooseningangledetectiontechnologyusingdeeplearning.XuefengZhaoYangZhangNiannianWang测试集结果：（非重复训练集图像）名目D&a会9. MonitoringofCorrodedandLoosenedBoltsinSteelStructuresviaDeepLearningandHoughTransforms,byQuoc-BaoTaOrcIDandJeong-TaeKim*