A Shadow-Robust Pavement Damage Detection Framework Based on RACycle-GAN and DDE-YOLOv8

Pavement construction worldwide has expanded significantly, making maintenance of pavement infrastructure a highly challenging task. Traditional manual inspection methods are inefficient and costly, while deep learning-based object detection algorithms often underperform in complex scenarios such as shadowed pavement surfaces. To address these issues, by combining a shadow removal model (Residual Attention Cycle GAN, RACycle-GAN) and Damage Detection Enhanced YOLOv8 (DDE-YOLOv8), this paper proposes a shadow-robust pavement damage detection framework. By integrating the Residual Encoder-Decoder Block, two-branch spatial attention, and cycle consistency into the original DCGAN architecture, the proposed RACycle-GAN significantly enhances shadow removal performance, fundamentally solving the issue of detection under shadow interference. Additionally, DDE-YOLOv8 is introduced to further enhance the damage detection performance. Experimental results demonstrate that RACycle-GAN greatly improves detection performance, and DDE-YOLOv8 outperforms the original version. The combination of RACycle-GAN and DDE-YOLOv8 achieves a leading performance in pavement damage detection, demonstrating the effectiveness of the shadow-robust damage detection method in this paper.