Automatic pavement crack detection based on single stage salient-instance segmentation and concatenated feature pyramid network

Regular inspection of pavement cracks is an important task to ensure the safety of the transportation system. At present, many pavement crack detection methods still rely on the manual way. These methods are usually time-consuming and subjective. Moreover, although the automatic crack detection meth...

Full description

Saved in:
Bibliographic Details
Published in:The international journal of pavement engineering 2022-10, Vol.23 (12), p.4206-4222
Main Authors: Li, Gang, Lan, Dongchao, Zheng, Xuan, Li, Xue, Zhou, Jian
Format: Article
Language:English
Subjects:
Algorithms
crack analysis
crack parameter calculation
Cracks
Datasets
Error analysis
Flaw detection
Inspection
Instance segmentation
Mathematical analysis
Pavement crack detection
Pavements
Recall
S4Net-CFPN
single stage salient-instance segmentation
Transportation networks
Transportation systems
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Regular inspection of pavement cracks is an important task to ensure the safety of the transportation system. At present, many pavement crack detection methods still rely on the manual way. These methods are usually time-consuming and subjective. Moreover, although the automatic crack detection method has made great progress recently, there are still difficulties such as poor anti-interference ability and low detection efficiency. Therefore, this paper proposes a pavement crack detection algorithm, which can solve the above problems well. This algorithm combines single stage salient-instance segmentation (S4Net) and concatenated feature pyramid network (CFPN), which greatly improves the ability to acquire feature information. Experiments show that on the noise-free dataset, the average precision, average recall, and F1-score are 0.9331, 0.9358, and 0.9344, respectively. On the complex noise dataset, the average precision, average recall, and F1-score are 0.8244, 0.8653, and 0.8443, respectively. Compared with other methods, our method has the advantages of strong anti-noise ability, high detection accuracy and fast detection speed. In addition, we propose a method for calculating the physical size of cracks. Through error analysis, the relative errors of calculating the length and width of the cracks are 0.056 and 0.084 respectively, which can meet the needs of engineering inspection.
ISSN:1029-8436
1477-268X
DOI:10.1080/10298436.2021.1938045