Abstract：Concrete crack detection is a key component of structural inspection. Crack detection systems based on unmanned aerial vehicles (UAVs) have been extensively investigated in recent years, and deep learning has also been used for automated crack identification from images. However, the simple quantification of cracks remains challenging. The objective of this study was to develop a crack detection method based on deep learning and a UAV to achieve the accurate detection and effective quantification of concrete cracks without reference markers. A crack feature pyramid network (Crack-FPN), which employs a multi-scale feature fusion architecture, is proposed for the segmentation of multiple cracks. The network was trained and tested on a multi-scale and multi-scene crack dataset. An improved method was developed to calibrate the image scale field of the gimbal cameras mounted on the UAVs, and a full-field scale was established. The crack localization and quantitative measurement were completed by the obtained segmentation result and image full-field scale. Moreover, the feasibility of the proposed method was validated by a field experiment using the UAV-based system. The results reveal that the maximum crack width measurement error at various measurement distances and angles is less than 5%. The findings of this study are expected to be useful for the crack inspection of concrete structures.
丁威 俞珂 舒江鹏. 基于深度学习和无人机的混凝土结构裂缝检测方法[J]. 土木工程学报, 2021, 54(S): 1-12.
Ding Wei Yu Ke Shu Jiangpeng. Method for detecting cracks in concrete structures based on deep learning and UAV. 土木工程学报, 2021, 54(S): 1-12.
