본문

서브메뉴

Pavement Crack Segmentation with Dense Local Geometry Features and Boundary Enhancement Loss
Pavement Crack Segmentation with Dense Local Geometry Features and Boundary Enhancement Lo...
Pavement Crack Segmentation with Dense Local Geometry Features and Boundary Enhancement Loss

Detailed Information

자료유형  
 학위논문 서양
최종처리일시  
20260202105544
ISBN  
9798265404770
DDC  
006.686
저자명  
Hsieh, Yung-An.
서명/저자  
Pavement Crack Segmentation with Dense Local Geometry Features and Boundary Enhancement Loss
발행사항  
[Sl] : Georgia Institute of Technology, 2023
발행사항  
Ann Arbor : ProQuest Dissertations & Theses, 2023
형태사항  
137 p
주기사항  
Source: Dissertations Abstracts International, Volume: 87-05, Section: A.
주기사항  
Advisor: Tsai, Yi-Chang;Yezzi, Anthony Joseph.
학위논문주기  
Thesis (Ph.D.)--Georgia Institute of Technology, 2023.
초록/해제  
요약The three-dimensional line laser scanning technology (3D laser technology) allows transportation agencies to collect high-resolution pavement range images for achieving detailed pavement condition assessment. However, the automated and accurate detection of cracks from the collected pavement data is an essential prerequisite for achieving this goal. In recent years, deep learning (DL), particularly the convolutional neural network (CNN), has emerged as the most dominant approach for automated crack detection at the pixel level (crack segmentation). However, two major bottlenecks hinder the performance of DL-based crack segmentation. First, the diverse scenarios of real-world pavement data, such as the varied geometry of cracks and diverse patterns on the pavement, challenge the robustness of CNNs for crack segmentation. Second, the CNNs for crack segmentation have poor detail-preserving capability in capturing the precise boundary, width, and connectivity of cracks, leading to incorrect estimation of crack length and width.The objective of this study is to advance automated pavement crack segmentation by addressing the two major bottlenecks that hinder the performance of CNNs on this task. To achieve this objective, we propose two methodologies. First, we propose to improve the robustness of CNNs on crack segmentation by leveraging the local geometry inherent in pavement data. To this end, we propose the Dense Local Geometry (DLG) features based on the minimal-path technique. These features are incorporated as an additional input to the CNN, providing the network with a sense of local geometry that is tailored specifically to cracks during training. To leverage the DLG features effectively, we introduce the Hybrid Fusion UNet (HFUNet), a CNN for crack segmentation based upon the hybrid fusion strategy in deep multimodal semantic segmentation. Second, we propose to enhance the detail-preserving capability of CNNs on crack segmentation by introducing the Boundary Enhancement (BE) loss. The BE loss constructs a supervisory signal on the boundaries of segmentation regions, enhancing the network's capability to perceive small deviations and disconnectivity at the boundary of cracks during training.The proposed methodologies for crack segmentation were evaluated on a realworld range image dataset with diverse conditions. The results show that the HFUNet outperforms existing CNNs in terms of Enhanced Hausdorff Distance (EHD)score (91.13), F1-score (81.82%), Intersection over Union (IoU) (69.23%), and precision (76.55%) on pavement crack segmentation. With the additional geometrical information provided by the DLG features, the major advantage of HFUNet lies in its enhanced robustness against diverse patterns on the pavement. With the BE loss, the CNNs implemented in this study achieved improved EHD, IoU, and F1-score on crack segmentation and a significant reduction in relative crack width and length errors. The results demonstrate the effectiveness and flexibility of the BE loss in enhancing the detail-preserving capability of any network. With the BE loss, the HFUNet still achieved the best EHD score (91.79), F1- score (83.47%), IoU (71.63%), and precision (78.33%) on pavement crack segmentation.The outcomes of this research will support transportation agencies in achieving detailed pavement condition assessments by providing high-granularity and precise crack segmentation from pavement range images collected with 3D laser technology.
일반주제명  
Digital cameras
일반주제명  
Deep learning
일반주제명  
Wavelet transforms
일반주제명  
Scanners
일반주제명  
Concrete
일반주제명  
Optimization techniques
일반주제명  
Boxes
일반주제명  
Cracks
일반주제명  
Lighting
일반주제명  
Lasers
일반주제명  
Neural networks
일반주제명  
Support vector machines
일반주제명  
Three dimensional imaging
일반주제명  
Geometry
일반주제명  
Semantics
일반주제명  
Information technology
일반주제명  
Mathematics
일반주제명  
Optics
기타저자  
Georgia Institute of Technology.
기본자료저록  
Dissertations Abstracts International. 87-05A.
전자적 위치 및 접속  
로그인 후 원문을 볼 수 있습니다.

MARC

 008260126s2023        us                              c    eng  d
■001000017360542
■00520260202105544
■006m          o    d                
■007cr#unu||||||||
■020    ▼a9798265404770
■035    ▼a(MiAaPQ)AAI32315487
■035    ▼a(MiAaPQ)GeorgiaTech75539
■040    ▼aMiAaPQ▼cMiAaPQ
■0820  ▼a006.686
■1001  ▼aHsieh,  Yung-An.
■24510▼aPavement  Crack  Segmentation  with  Dense  Local  Geometry  Features  and  Boundary  Enhancement  Loss
■260    ▼a[Sl]▼bGeorgia  Institute  of  Technology▼c2023
■260  1▼aAnn  Arbor▼bProQuest  Dissertations  &  Theses▼c2023
■300    ▼a137  p
■500    ▼aSource:  Dissertations  Abstracts  International,  Volume:  87-05,  Section:  A.
■500    ▼aAdvisor:  Tsai,  Yi-Chang;Yezzi,  Anthony  Joseph.
■5021  ▼aThesis  (Ph.D.)--Georgia  Institute  of  Technology,  2023.
■520    ▼aThe  three-dimensional  line  laser  scanning  technology  (3D  laser  technology)  allows  transportation  agencies  to  collect  high-resolution  pavement  range  images  for  achieving  detailed  pavement  condition  assessment.  However,  the  automated  and  accurate  detection  of  cracks  from  the  collected  pavement  data  is  an  essential  prerequisite  for  achieving  this  goal.  In  recent  years,  deep  learning  (DL),  particularly  the  convolutional  neural  network  (CNN),  has  emerged  as  the  most  dominant  approach  for  automated  crack  detection  at  the  pixel  level  (crack  segmentation).  However,  two  major  bottlenecks  hinder  the  performance  of  DL-based  crack  segmentation.  First,  the  diverse  scenarios  of  real-world  pavement  data,  such  as  the  varied  geometry  of  cracks  and  diverse  patterns  on  the  pavement,  challenge  the  robustness  of  CNNs  for  crack  segmentation.  Second,  the  CNNs  for  crack  segmentation  have  poor  detail-preserving  capability  in  capturing  the  precise  boundary,  width,  and  connectivity  of  cracks,  leading  to  incorrect  estimation  of  crack  length  and  width.The  objective  of  this  study  is  to  advance  automated  pavement  crack  segmentation  by  addressing  the  two  major  bottlenecks  that  hinder  the  performance  of  CNNs  on  this  task.  To  achieve  this  objective,  we  propose  two  methodologies.  First,  we  propose  to  improve  the  robustness  of  CNNs  on  crack  segmentation  by  leveraging  the  local  geometry  inherent  in  pavement  data.  To  this  end,  we  propose  the  Dense  Local  Geometry  (DLG)  features  based  on  the  minimal-path  technique.  These  features  are  incorporated  as  an  additional  input  to  the  CNN,  providing  the  network  with  a  sense  of  local  geometry  that  is  tailored  specifically  to  cracks  during  training.  To  leverage  the  DLG  features  effectively,  we  introduce  the  Hybrid  Fusion  UNet  (HFUNet),  a  CNN  for  crack  segmentation  based  upon  the  hybrid  fusion  strategy  in  deep  multimodal  semantic  segmentation.  Second,  we  propose  to  enhance  the  detail-preserving  capability  of  CNNs  on  crack  segmentation  by  introducing  the  Boundary  Enhancement  (BE)  loss.  The  BE  loss  constructs  a  supervisory  signal  on  the  boundaries  of  segmentation  regions,  enhancing  the  network's  capability  to  perceive  small  deviations  and  disconnectivity  at  the  boundary  of  cracks  during  training.The  proposed  methodologies  for  crack  segmentation  were  evaluated  on  a  realworld  range  image  dataset  with  diverse  conditions.  The  results  show  that  the  HFUNet  outperforms  existing  CNNs  in  terms  of  Enhanced  Hausdorff  Distance  (EHD)score  (91.13),  F1-score  (81.82%),  Intersection  over  Union  (IoU)  (69.23%),  and  precision  (76.55%)  on  pavement  crack  segmentation.  With  the  additional  geometrical  information  provided  by  the  DLG  features,  the  major  advantage  of  HFUNet  lies  in  its  enhanced  robustness  against  diverse  patterns  on  the  pavement.  With  the  BE  loss,  the  CNNs  implemented  in  this  study  achieved  improved  EHD,  IoU,  and  F1-score  on  crack  segmentation  and  a  significant  reduction  in  relative  crack  width  and  length  errors.  The  results  demonstrate  the  effectiveness  and  flexibility  of  the  BE  loss  in  enhancing  the  detail-preserving  capability  of  any  network.  With  the  BE  loss,  the  HFUNet  still  achieved  the  best  EHD  score  (91.79),  F1-  score  (83.47%),  IoU  (71.63%),  and  precision  (78.33%)  on  pavement  crack  segmentation.The  outcomes  of  this  research  will  support  transportation  agencies  in  achieving  detailed  pavement  condition  assessments  by  providing  high-granularity  and  precise  crack  segmentation  from  pavement  range  images  collected  with  3D  laser  technology.
■590    ▼aSchool  code:  0078.
■650  4▼aDigital  cameras
■650  4▼aDeep  learning
■650  4▼aWavelet  transforms
■650  4▼aScanners
■650  4▼aConcrete
■650  4▼aOptimization  techniques
■650  4▼aBoxes
■650  4▼aCracks
■650  4▼aLighting
■650  4▼aLasers
■650  4▼aNeural  networks
■650  4▼aSupport  vector  machines
■650  4▼aThree  dimensional  imaging
■650  4▼aGeometry
■650  4▼aSemantics
■650  4▼aInformation  technology
■650  4▼aMathematics
■650  4▼aOptics
■690    ▼a0800
■690    ▼a0501
■690    ▼a0489
■690    ▼a0405
■690    ▼a0752
■71020▼aGeorgia  Institute  of  Technology.
■7730  ▼tDissertations  Abstracts  International▼g87-05A.
■790    ▼a0078
■791    ▼aPh.D.
■792    ▼a2023
■793    ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17360542▼nKERIS▼z이  자료의  원문은  한국교육학술정보원에서  제공합니다.

Preview

Export

ChatGPT Discussion

AI Recommended Related Books


    New Books MORE
    Statistics for the past 3 years. Go to brief

    Подробнее информация.

    • Бронирование
    • не существует
    • моя папка
    • Первый запрос зрения
    • Non-Book Loan Application
    • Nighttime Book Loan Application
    материал
    Reg No. Количество платежных Местоположение статус Ленд информации
    TF14797 전자도서 대출가능 My Folder 부재도서신고 비도서대출신청 야간 도서대출신청

    * Бронирование доступны в заимствований книги. Чтобы сделать предварительный заказ, пожалуйста, нажмите кнопку бронирование

    Books borrowed together with this book

    Related Popular Books

    Available after logging in.