A new method for evaluating the uniformity of steel slag distribution in steel slag asphalt mixture based on deep learning

•Segmentation of steel slag in steel slag asphalt mixture based on deep learning.•Comparing the performance of three networks, U-Net segmentation works best.•Segmentation of bonded steel slag based on pixel-level watershed algorithm.•A steel slag distribution uniformity evaluation system was constru...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Construction & building materials Jg. 400; S. 132766
Hauptverfasser: Liu, Jianan, Yuan, Linjian, Wang, Zhenjun, Jing, Haosen, Shao, Tingquan, Chen, Hao
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Elsevier Ltd 12.10.2023
Schlagworte:
Asphalt mixture
Deep learning
Pixel level
Uniformity
Watershed algorithm
U-Net
Pv
MIoU
RPD
Asphalt mixture
DeepLabV3
PSPNet
Deep learning
UI
Pixel level
Watershed algorithm
Ph
Uniformity
MPA
ISSN:0950-0618, 1879-0526
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:•Segmentation of steel slag in steel slag asphalt mixture based on deep learning.•Comparing the performance of three networks, U-Net segmentation works best.•Segmentation of bonded steel slag based on pixel-level watershed algorithm.•A steel slag distribution uniformity evaluation system was constructed. A new evaluation method has been developed to assess the uniformity of steel slag asphalt mixture. The original aggregate in the method is replaced by steel slag in the range of 4.75 mm to 9.5 mm. Deep learning techniques are used in this method to select the optimal network as the segmentation network of the framework by comparing DeepLabV3+, U-Net, and PSPNet networks. Pixel level watershed algorithm is used to solve the steel slag adhesion problem for predicted images. Finally, the horizontal distribution coefficient UI and vertical distribution coefficient Pv were used to evaluate the uniformity of steel slag distribution. The results show that the U-Net network performs the best in identifying and segmenting steel slag, the training is stable without overfitting problem and the accuracy of training reaches 98.62 %, the true positive is the highist and the average accuracy is 96.98 %, UI and Pv can effectively reflect the distribution uniformity of steel slag. The end-to-end system provides a more convenient and intelligent way to evaluate the distribution uniformity of steel slag. The method can be used to quickly and accurately analyze the dispersion uniformity of aggregates in asphalt mixtures and to help in the construction inspection of road projects.
ISSN:0950-0618
1879-0526
DOI:10.1016/j.conbuildmat.2023.132766