Thanks to the development of deep learning, the use of data-driven methods to detect pavement distresses has become an active research field. This research makes four contributions to address the problem of efficiently detecting cracks and sealed cracks in asphalt pavements. First, a dataset of pavement cracks and sealed cracks is created, which consists of 10,400 images obtained by a vehicle equipped with a highway condition monitor, with 202,840 labeled distress instances included in these pavement images. Second, we develop a dense and redundant crack annotation method based on the characteristics of the crack images. Compared with traditional annotation, the method we propose generates more object instances, and the localization is more accurate. Next, to achieve efficient crack detection, a semi-automatic crack annotation method is proposed, which reduces the working time by 80% compared with fully manual annotation. Finally, comparative experiments are conducted on our dataset using 13 currently prevailing object detection algorithms. The results show that dense and redundant annotation is effective; moreover, cracks and sealed cracks can be efficiently and accurately detected using the YOLOv5 series model and YOLOv5s is the most balanced model with an F1-score of 86.79% and an inference time of 14.8ms. The pavement crack and sealed crack dataset created in this study is publicly available.

The results of high-precision asphalt concrete pavement crack identification can provide help for pavement maintenance. Therefore, methods of image feature enhancement and crack identification of asphalt concrete pavement cracks are proposed. First of all, we used an industrial CCD camera mounted on a vehicle to collect an asphalt concrete pavement crack image. Then, after using the NeighShrink algorithm to denoise the acquired image, a fractional differential image enhancement algorithm was designed based on image feature blocks to enhance the image features. On this basis, crack characteristics were segmented and processed by watershed algorithm. Through crack direction identification and crack parameter extraction, crack distribution direction, crack length and width and other parameters of asphalt concrete pavement were obtained in order to achieve accurate identification of asphalt concrete pavement cracks. The experiment found that this method can effectively remove noise information from asphalt concrete crack images; after applying this method, the image entropy value of each image was improved, with a minimum improvement of 0.38 and a maximum improvement of 1.98. The time consumed by this method in identifying cracks in asphalt concrete pavement varied between 1.4 s and 2.4 s. When identifying the length of cracks in asphalt concrete pavement, the maximum deviation value was only 0.47 mm; when identifying the width of cracks in asphalt concrete pavement, the maximum deviation value was only 0.31 mm. The above results indicate that by enhancing the image features of asphalt concrete cracks, this method achieves more accurate identification results for crack distribution direction, length and width values, with high identification efficiency and good application effect.

Accurate crack segmentation must preserve thin and irregular boundaries, connect long-range structures, and suppress nuisance textures that can produce crack-like high-frequency responses. Existing convolutional, Transformer, and state-space architectures address local detail or spatial context, and recent frequency-aware designs introduce wavelet or frequency cues. However, fixed wavelet cues impose a predetermined frequency partition that is not optimized with the segmentation objective. This paper proposes CrackLIFT, a crack segmentation network built around a Lifting-based Interactive Frequency Transformation. CrackLIFT uses lifting-based decomposition to learn low- and high-frequency encoder representations, allowing the frequency partition to adapt to the segmentation objective, while specialized state-space paths and bidirectional interaction preserve complementary structural and boundary information. A wavelet-guided frequency alignment loss aligns stage-wise encoder representations with wavelet targets and adds reconstruction consistency after inverse wavelet synthesis. On CrackSeg9k, CrackLIFT achieves 82.10% mIoU and 84.48% BF-Score, while the results on Asphalt3k and Concrete3k show consistent improvements over the evaluated baselines.