To address the problem of decreasing disease detection accuracy due to shadow interference in complex road scenes, this study proposed a road shadow removal algorithm based on a weakly supervised generative adversarial network. By constructing a dynamic adaptive normalization strategy and an improved residual module, the algorithm optimized image feature representation under the condition of limited shadow pairing samples. The specific design was as follows: First, an auxiliary classifier with an attention mechanism was introduced to enhance the semantic perception ability of the network for road scenes. Second, the residual structure was improved by adopting cross-layer skip connections to effectively alleviate the problem of gradient vanishing in deep layers of the network. Finally, a hybrid loss function that combined the perceptual loss and the adversarial loss was designed to enhance the structure preservation ability of shadow removal. The experimental results show that the image quality evaluation metrics ERMSE, RPSNR, and SSSIM of the shadow removal algorithm are 1.083 4, 27.568 2, and 0.794 9. The images after shadow removal using the proposed algorithm were fed into the detection model. The recognition accuracy of six typical road diseases such as cracks and potholes are improved by 3.3%‒10.2%, which verifies the effectiveness of the method.
截至2023年底,中国公路通车总里程数达到540万km,其中,高速公路通车总里程数已经达到18.4万km,公路养护里程占公路总里程的99.9%,规模庞大、结构复杂的道路交通网络,给以公路为核心的道路基础设施可持续养护工作带来了诸多挑战。传统道路病害一般通过人工巡检检测,可靠性强,但效率低,每日人均检测里程不足10 km[1];面对大规模、常态化的持续检测作业,传统检测手段落地实施难度较大。随着深度学习、计算机视觉、建筑信息模型(Building Information Modeling, BIM)等技术应用于道路病害检测后,大幅提升了检测效率[2-5]。然而,在实际道路场景中,受自然光照变化影响,图像中阴影区域亮度常显著低于周围区域。这种阴影效应严重干扰图像质量,影响目标检测、跟踪及配准等关键计算机视觉任务的鲁棒性与精度。因此,有效削弱阴影干扰,不仅有助于提升图像整体质量,还对提高病害检测算法的稳定性与准确性具有重要意义。
OliveiraH, CorreiaP L. Supervised strategies for cracks detection in images of road pavement flexible surfaces[C]//2008 16th European Signal Processing Conference. Lausanne, Switzerland. IEEE, 2008: 1-5.
[2]
FeiY, WangK C P, ZhangA, et al. Pixel-Level cracking detection on 3D asphalt pavement images through deep-learning-based CrackNet-V[J]. IEEE Transactions on Intelligent Transportation Systems, 2020, 21(1): 273-284.
[3]
LiuY H, YaoJ, LuX H, et al. DeepCrack: A deep hierarchical feature learning architecture for crack segmentation[J]. Neurocomputing, 2019, 338: 139-153.
[4]
LeCunY, BengioY, HintonG. Deep learning[J]. Nature, 2015, 521(7553): 436-444.
YanLupeng, WuYifei, ChenBo, et al. Research on information technology in highway defect inspection based on BIM and reality capture technology[J]. Journal of China & Foreign Highway, 2020, 40(2): 300-306.
GaoFan, HeChunguang, WangPengzhang, et al. Application of digital image processing in cracks of concrete pavement[J]. Journal of China & Foreign Highway, 2023, 43(1): 44-51.
LiuKun, MoHongliu, LiuYanfeng, et al. Research on optimization of pavement distress identification model in complex environment[J]. Highway Engineering, 2024, 49(5): 106-115, 166.
[13]
KhanS H, BennamounM, SohelF, et al. Automatic shadow detection and removal from a single image[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2016, 38(3): 431-446.
[14]
HosseinzadehS, ShakeriM, ZhangH. Fast shadow detection from a single image using a patched convolutional neural network[C]//2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). Madrid, Spain. IEEE, 2018: 3124-3129.
[15]
ZouQ, CaoY, LiQ Q, et al. CrackTree: Automatic crack detection from pavement images[J]. Pattern Recognition Letters, 2012, 33(3): 227-238.
PanGuangzhen, SunYanqing, WangFeng. Removal of vehicle shadow based on Fast RCNN model[J]. Computer Engineering and Design, 2018, 39(3): 819-823.
[18]
CarlsonA, VasudevanR, Johnson-RobersonM. Shadow transfer: Single image relighting for urban road scenes[PP/OL]. V2. arXiv (2019-09-26)[2025-9-30].
[19]
WangJ F, LiX, YangJ. Stacked conditional generative adversarial networks for jointly learning shadow detection and shadow removal[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City, UT, USA. IEEE, 2018: 1788-1797.
[20]
HuX W, JiangY T, FuC W, et al. Mask-ShadowGAN: Learning to remove shadows from unpaired data[C]//2019 IEEE/CVF International Conference on Computer Vision (ICCV). Seoul, Korea. IEEE, 2019: 2472-2481.
[21]
LiuZ H, YinH, MiY, et al. Shadow removal by a lightness-guided network with training on unpaired data[J]. IEEE Transactions on Image Processing, 2021, 30: 1853-1865.
[22]
CunX D, PunC M, ShiC. Towards ghost-free shadow removal via dual hierarchical aggregation network and shadow matting GAN[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2020, 34(7): 10680-10687.
[23]
LiuZ H, YinH, WuX Y, et al. From shadow generation to shadow removal[C]//2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Nashville, TN, USA. IEEE, 2021: 4925-4934.
[24]
JinY Y, SharmaA, TanR T. DC-ShadowNet: Single-image hard and soft shadow removal using unsupervised domain-classifier guided network[C]//2021 IEEE/CVF International Conference on Computer Vision (ICCV). Montreal, QC, Canada. IEEE, 2021: 5007-5016.
[25]
KubiakN, MustafaA, PhillipsonG, et al. S3R-net: A single-stage approach to self-supervised shadow removal[C]//2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). Seattle, WA, USA. IEEE, 2024: 5898-5908.
[26]
CreswellA, WhiteT, DumoulinV, et al. Generative adversarial networks: An overview[J]. IEEE Signal Processing Magazine, 2018, 35(1): 53-65.
[27]
RadfordA, MetzL, ChintalaS. Unsupervised representation learning with deep convolutional generative adversarial networks[PP/OL]. V2. arXiv (2016-01-07)[2025-09-30].
[28]
KimJ, KimM, KangH, et al. U-GAT-IT: Unsupervised generative attentional networks with adaptive layer-instance normalization for image-to-image translation[PP/OL]. V4. arXiv (2020-04-08)[2025-09-30].
UlyanovD, VedaldiA, LempitskyV. Instance normalization: The missing ingredient for fast stylization[PP/OL]. V3. arXiv (2017-11-06)[2025-09-30].
[31]
ZhouB L, KhoslaA, LapedrizaA, et al. Learning deep features for discriminative localization[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas, NV, USA. IEEE, 2016: 2921-2929.