A transfer learning-based YOLO network for sewer defect detection in comparison to classic object detection methods
Deep learning has shown promising performance in automated sewer defect detection, however, is generally data-driven and computationally intensive. Transfer learning (TL) solves the problem of data limitations and avoids the need to build models from scratch. This study compared the performance of a...
| Main Authors: | , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2023-10-01
|
| Series: | Developments in the Built Environment |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S266616592300073X |
| _version_ | 1797689505946271744 |
|---|---|
| author | Zuxiang Situ Shuai Teng Wanen Feng Qisheng Zhong Gongfa Chen Jiongheng Su Qianqian Zhou |
| author_facet | Zuxiang Situ Shuai Teng Wanen Feng Qisheng Zhong Gongfa Chen Jiongheng Su Qianqian Zhou |
| author_sort | Zuxiang Situ |
| collection | DOAJ |
| description | Deep learning has shown promising performance in automated sewer defect detection, however, is generally data-driven and computationally intensive. Transfer learning (TL) solves the problem of data limitations and avoids the need to build models from scratch. This study compared the performance of a TL-based YOLO network (with 11 pretrained backbone CNNs) with four mainstream object detection methods (ODMs) for detecting five types of sewer defects. Results showed that the transferred YOLO methods generally outperformed the other ODMs, with improved detection precision, computation speed and intersection over union (IoU). Among the CNNs, Resnet18 achieved the best performance, while Inceptionresnetv2 was the least effective. The ODMs worked best in detecting disjoint, whereas tree root and crack were most challenging to predict. The work not only illustrated the benefits of TL, but also provided technical guidance to practitioners who lack expertise in ODMs and rely on TL for better sewer defect detection. |
| first_indexed | 2024-03-12T01:46:32Z |
| format | Article |
| id | doaj.art-7e58bfde480744e6b184bf2856b9f784 |
| institution | Directory Open Access Journal |
| issn | 2666-1659 |
| language | English |
| last_indexed | 2024-03-12T01:46:32Z |
| publishDate | 2023-10-01 |
| publisher | Elsevier |
| record_format | Article |
| series | Developments in the Built Environment |
| spelling | doaj.art-7e58bfde480744e6b184bf2856b9f7842023-09-09T04:56:21ZengElsevierDevelopments in the Built Environment2666-16592023-10-0115100191A transfer learning-based YOLO network for sewer defect detection in comparison to classic object detection methodsZuxiang Situ0Shuai Teng1Wanen Feng2Qisheng Zhong3Gongfa Chen4Jiongheng Su5Qianqian Zhou6School of Civil and Transportation Engineering, Guangdong University of Technology, No.100 Waihuan Xi Road, Guangzhou, 510006, ChinaSchool of Civil and Transportation Engineering, Guangdong University of Technology, No.100 Waihuan Xi Road, Guangzhou, 510006, China; Research Centre for Wind Engineering and Engineering Vibration, Guangzhou University, No.230 Waihuan Xi Road, Guangzhou, 510006, ChinaSchool of Civil and Transportation Engineering, Guangdong University of Technology, No.100 Waihuan Xi Road, Guangzhou, 510006, ChinaSchool of Civil and Transportation Engineering, Guangdong University of Technology, No.100 Waihuan Xi Road, Guangzhou, 510006, ChinaSchool of Civil and Transportation Engineering, Guangdong University of Technology, No.100 Waihuan Xi Road, Guangzhou, 510006, ChinaUrban Development Research Center, Guangdong Urban & Rural Planning and Design Institute, No.483 Nanzhou Road, Guangzhou, 510290, ChinaSchool of Civil and Transportation Engineering, Guangdong University of Technology, No.100 Waihuan Xi Road, Guangzhou, 510006, China; Corresponding author. School of Civil and Transportation Engineering, Guangdong University of Technology, China.Deep learning has shown promising performance in automated sewer defect detection, however, is generally data-driven and computationally intensive. Transfer learning (TL) solves the problem of data limitations and avoids the need to build models from scratch. This study compared the performance of a TL-based YOLO network (with 11 pretrained backbone CNNs) with four mainstream object detection methods (ODMs) for detecting five types of sewer defects. Results showed that the transferred YOLO methods generally outperformed the other ODMs, with improved detection precision, computation speed and intersection over union (IoU). Among the CNNs, Resnet18 achieved the best performance, while Inceptionresnetv2 was the least effective. The ODMs worked best in detecting disjoint, whereas tree root and crack were most challenging to predict. The work not only illustrated the benefits of TL, but also provided technical guidance to practitioners who lack expertise in ODMs and rely on TL for better sewer defect detection.http://www.sciencedirect.com/science/article/pii/S266616592300073XObject detectionTransfer learningPretrained CNNsYOLO networkSewer defects |
| spellingShingle | Zuxiang Situ Shuai Teng Wanen Feng Qisheng Zhong Gongfa Chen Jiongheng Su Qianqian Zhou A transfer learning-based YOLO network for sewer defect detection in comparison to classic object detection methods Developments in the Built Environment Object detection Transfer learning Pretrained CNNs YOLO network Sewer defects |
| title | A transfer learning-based YOLO network for sewer defect detection in comparison to classic object detection methods |
| title_full | A transfer learning-based YOLO network for sewer defect detection in comparison to classic object detection methods |
| title_fullStr | A transfer learning-based YOLO network for sewer defect detection in comparison to classic object detection methods |
| title_full_unstemmed | A transfer learning-based YOLO network for sewer defect detection in comparison to classic object detection methods |
| title_short | A transfer learning-based YOLO network for sewer defect detection in comparison to classic object detection methods |
| title_sort | transfer learning based yolo network for sewer defect detection in comparison to classic object detection methods |
| topic | Object detection Transfer learning Pretrained CNNs YOLO network Sewer defects |
| url | http://www.sciencedirect.com/science/article/pii/S266616592300073X |
| work_keys_str_mv | AT zuxiangsitu atransferlearningbasedyolonetworkforsewerdefectdetectionincomparisontoclassicobjectdetectionmethods AT shuaiteng atransferlearningbasedyolonetworkforsewerdefectdetectionincomparisontoclassicobjectdetectionmethods AT wanenfeng atransferlearningbasedyolonetworkforsewerdefectdetectionincomparisontoclassicobjectdetectionmethods AT qishengzhong atransferlearningbasedyolonetworkforsewerdefectdetectionincomparisontoclassicobjectdetectionmethods AT gongfachen atransferlearningbasedyolonetworkforsewerdefectdetectionincomparisontoclassicobjectdetectionmethods AT jionghengsu atransferlearningbasedyolonetworkforsewerdefectdetectionincomparisontoclassicobjectdetectionmethods AT qianqianzhou atransferlearningbasedyolonetworkforsewerdefectdetectionincomparisontoclassicobjectdetectionmethods AT zuxiangsitu transferlearningbasedyolonetworkforsewerdefectdetectionincomparisontoclassicobjectdetectionmethods AT shuaiteng transferlearningbasedyolonetworkforsewerdefectdetectionincomparisontoclassicobjectdetectionmethods AT wanenfeng transferlearningbasedyolonetworkforsewerdefectdetectionincomparisontoclassicobjectdetectionmethods AT qishengzhong transferlearningbasedyolonetworkforsewerdefectdetectionincomparisontoclassicobjectdetectionmethods AT gongfachen transferlearningbasedyolonetworkforsewerdefectdetectionincomparisontoclassicobjectdetectionmethods AT jionghengsu transferlearningbasedyolonetworkforsewerdefectdetectionincomparisontoclassicobjectdetectionmethods AT qianqianzhou transferlearningbasedyolonetworkforsewerdefectdetectionincomparisontoclassicobjectdetectionmethods |