Deep Learning Based Infrared Thermal Image Analysis of Complex Pavement Defect Conditions Considering Seasonal Effect

Area of Science:

• Civil Engineering
• Computer Science
• Materials Science

Background:

• Deep learning excels at pavement damage detection using diverse data.
• Infra-red (IR-T) imaging offers an economical approach for detecting pavement damage, especially considering temperature variations.
• Prior research achieved 96.47% accuracy using SA-ResNet on summer pavement data.

Purpose of the Study:

• To evaluate the prediction accuracy, sensitivity, and recall of a deep learning model for pavement damage detection using winter sunny condition images.
• To compare the performance of the deep learning approach between summer and winter sunny conditions.
• To assess the viability of IR-T imaging as an economical alternative for pavement damage detection across seasons.

Main Methods:

• Utilized a deep learning architecture (SA-ResNet) for pavement damage classification.
• Compared performance metrics (accuracy, sensitivity, recall) using datasets captured during summer and winter sunny conditions.
• Analyzed prediction accuracies for different image types: Digital Camera (DC), Multispectral (MSX), and Infra-Red Thermal (IR-T).

Main Results:

• The deep learning algorithm achieved approximately 92% accuracy overall, with 95.18% in summer and 91.67% in winter.
• Digital Camera (DC) images showed the highest accuracy in both seasons (96.47% summer, 94.14% winter).
• IR-T imaging demonstrated good performance (93.83% summer, 90.173% winter), indicating its potential as a cost-effective solution.

Conclusions:

• Deep learning techniques provide reliable pavement damage categorization regardless of the season.
• Summer conditions yield slightly higher prediction accuracy compared to winter.
• Inexpensive IR-T imaging is a viable, economical option for pavement damage detection, particularly under summer sunny conditions.