Share
Export Citation
Automatic Crack Detection on Bridge Damage in Low Lighting Conditions Using YOLOv11-SegBased on Drone Imagery
Azzahra A.A.M.
2026 IEEE International Conference on Artificial Intelligence and Mechatronics Systems Aims 2026
Abstract
Detection of bridge damage in low-light conditions is a significant challenge in conventional infrastructure inspections. The study proposes an automated crack detection system for bridge structures that uses drone imagery captured in low-light conditions and is enhanced by an image pre-processing pipeline integrated with the YOLOv11-Seg instance segmentation model. The pre-processing pipeline consisted of Contrast Limited Adaptive Histogram Equalization (CLAHE), Gamma Correction (γ 0.7), Grayscale Conversion, and Denoising applied to a dataset of 200 drone images, with the training subset augmented from 140 to 560 images, resulting in a total of 620 images used for training, validation, and testing. The dataset is divided into training (70%), validation (20%), and testing (10%) data. The results showed that the combination of CLAHE, Gamma Correction, Grayscale, and Denoising or abbreviated (CGGD) pre-processing produced the best performance, with a precision of 0.77, recall of 0.58, mAP50 0.61, and mAP50-95 of 0.44 in the training phase, as well as an accuracy of 0.69, precision of 0.78, recall of 0.65, F1-Score of 0.68, mAP50 0.62, and mAP50-95 of 0.38 in the off-baseline testing phase and other partial pre-processing combinations. PSNR analysis confirmed that CGGD combinations were most effective at maintaining image fidelity for crack detection. These findings prove that adaptive pre-processing combined with YOLOv11-Seg provides a reliable and efficient solution for automated bridge inspection under low-light conditions.