Rapid and accurate detection of internal defects in bridges has always been a major concern of the management and maintenance departments. In the present study, an intelligent method for the detection of structural internal defects is proposed based on infrared thermography and deep learning. Through theoretical analysis, numerical simulations and laboratory experiments, the classification, localization and quantification of internal defects of concrete structures were achieved with the infrared thermography and deep learning method. The mean average precision for classification and localization of internal defects is 96.59%, the mIoU for pixel-level segmentation is 95.19%, and the average relative error for damage quantification is 0.70%. The feasibility of the trained model is verified with new images, and the results show that the trained model can capture infrared thermal features of internal defects with different sizes and depths. This method has the advantages of low cost, high accuracy, easy operation, and large area scanning of concrete structures, which can provide a good reference for the detection of internal defects of concrete structures.

Reinforced Concrete (RC) bridges are prone to deterioration throughout their service life, which can compromise their structural integrity. Among the visible indicators of deterioration, cracks are particularly significant, underscoring the need for their early detection. Traditional inspections, primarily based on visual inspection, face challenges such as restricted access to certain bridge areas, complex logistics, and excessive operational costs and time. To address these challenges, the engineering community is working toward implementing more efficient technologies for bridge inspection. This paper proposes an automatic deep learning-based system for the segmentation and quantification of surface cracks in RC bridges using images captured by unmanned aerial vehicles (UAVs). The proposed framework combines patch-based and pixel-level segmentation using binary image classification and crack segmentation models. Additionally, an alternative method based on a laser module with a diffractive optical element (DOE) was implemented to obtain an automatic scale factor, enabling the conversion of pixel-level information into metric measurements for damage quantification. Experimental results show that the binary classification model achieved an F1-score of 0.93 on the test set, while the segmentation model achieved an F1-score of 0.83. Overall system performance yielded a coefficient of determination (R²) of 0.70 and an average inference time of 2.32 seconds per high-resolution image. Through integrating UAVs, lasers, and deep learning, the proposed approach offers an alternative for bridge inspections, reducing cost and time while supporting the rapid generation of preliminary reports.