Detecting temperature segregation in asphalt pavement construction using infrared imaging and deep learning

ABSTRACT

Temperature segregation (TS) is the uneven distribution of mixture temperature during construction. Considering segregation patterns is essential to detect TS and evaluate construction quality. This paper collected infrared images and classified three segregation categories: interlaced TS, banded TS (BTS), and clustered TS (CTS) based on temperature distribution patterns. Detecting TS using infrared image and deep learning is benefit for improving construction quality, so we labelled the BTS and CTS images to form an original training dataset (OD) for deep learning detection algorithms. We also used two image augmentation approaches to expand OD and obtained two more datasets: dataset augmented without CutMix, which is a classic image augmentation method, and dataset augmented with CutMix (CD). Based on three datasets, we trained You Only Look Once version 5 (YOLOv5) models to detect TS and compared them with Faster R-CNN and YOLOv7. CDM-YOLOv5, YOLOv5 trained on CD, outperformed others, with a mean average precision of 71.5% and an F1 score of 69.7%. We also observed performance gains using the CutMix augmentation. The findings contribute to automating construction quality evaluation and control.

KEYWORDS:

• Asphalt pavement construction
• temperature segregation
• infrared thermal image
• object detection
• YOLO

Acknowledgments

This study was sponsored by the National Natural Science Foundation of China (No. 52008311, No. 51878499, No. 52178433), the China Scholarship Council (No. 202106260113), the Fundamental Research Funds for the Central Universities (No. 22120230196), the Science and Technology Commission of Shanghai Municipality (No. 21ZR1465700), and the Shandong Province Department of Transportation (No. 2021B47). The authors are grateful for that financial support.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

Some or all data, models, or code that support the findings of this study are available from the corresponding author upon reasonable request.

Additional information

Funding

This work was supported by China Scholarship Council: [grant no 202106260113]; National Natural Science Foundation of China: [grant no 51878499, 51878499]; Science and Technology Commission of Shanghai Municipality: [grant no 21ZR1465700].