TY - GEN
T1 - Semantic segmentation of sewer pipe defects using deep dilated convolutional neural network
AU - Wang, M. Z.
AU - Cheng, J. C.P.
N1 - Publication details (e.g. title, author(s), publication statuses and dates) are captured on an “AS IS” and “AS AVAILABLE” basis at the time of record harvesting from the data source. Suggestions for further amendments or supplementary information can be sent to [email protected].
PY - 2019
Y1 - 2019
N2 - Semantic segmentation of closed-circuit television (CCTV) images can facilitate automatic severity assessment of sewer pipe defects by assigning defect labels to each pixel in the image, from which defect types, locations and geometric information can be obtained. In this study, a deep convolutional neural network (CNN), namely DilaSeg, is developed based on dilated convolution for improving the segmentation of sewer pipe defects including cracks, tree root intrusion and deposit. Sewer pipe CCTV images are extracted from inspection videos and are annotated to be used as the ground truth labels for training the model. DilaSeg is constructed with dilated convolution for producing feature maps with high resolution. Both DilaSeg and the state-of-the-art model, fully convolutional network (FCN), are trained and evaluated on the annotated dataset using the same hyper-parameters. The results of the experiments indicate that the proposed DilaSeg improved the segmentation accuracy significantly compared with FCN, with 18% of increase in mean pixel accuracy (mPA) and 22% of increase in mean intersection over union (IoU) with a fast detection speed. © 2019 International Association for Automation and Robotics in Construction I.A.A.R.C. All rights reserved.
AB - Semantic segmentation of closed-circuit television (CCTV) images can facilitate automatic severity assessment of sewer pipe defects by assigning defect labels to each pixel in the image, from which defect types, locations and geometric information can be obtained. In this study, a deep convolutional neural network (CNN), namely DilaSeg, is developed based on dilated convolution for improving the segmentation of sewer pipe defects including cracks, tree root intrusion and deposit. Sewer pipe CCTV images are extracted from inspection videos and are annotated to be used as the ground truth labels for training the model. DilaSeg is constructed with dilated convolution for producing feature maps with high resolution. Both DilaSeg and the state-of-the-art model, fully convolutional network (FCN), are trained and evaluated on the annotated dataset using the same hyper-parameters. The results of the experiments indicate that the proposed DilaSeg improved the segmentation accuracy significantly compared with FCN, with 18% of increase in mean pixel accuracy (mPA) and 22% of increase in mean intersection over union (IoU) with a fast detection speed. © 2019 International Association for Automation and Robotics in Construction I.A.A.R.C. All rights reserved.
KW - Convolutional neural network (CNN)
KW - Defect segmentation
KW - Dilated convolution
KW - Semantic segmentation
KW - Sewer pipe defect
KW - Visual inspection
UR - http://www.scopus.com/inward/record.url?scp=85071434510&partnerID=8YFLogxK
UR - https://www.scopus.com/record/pubmetrics.uri?eid=2-s2.0-85071434510&origin=recordpage
U2 - 10.22260/isarc2019/0078
DO - 10.22260/isarc2019/0078
M3 - RGC 32 - Refereed conference paper (with host publication)
T3 - Proceedings of the 36th International Symposium on Automation and Robotics in Construction, ISARC 2019
SP - 586
EP - 594
BT - Proceedings of the 36th International Symposium on Automation and Robotics in Construction, ISARC 2019
PB - International Association for Automation and Robotics in Construction I.A.A.R.C)
T2 - 36th International Symposium on Automation and Robotics in Construction, ISARC 2019
Y2 - 21 May 2019 through 24 May 2019
ER -