Automatic crack detection is always a challenging task due to the influence of stains, shadows, complex texture, uneven illumination, blurring, and multiple scenes [2]. In the past decades, scholars have proposed a variety of image-based algorithms to automatically detect cracks on concrete surfaces and pavement. In the early studies, most of the methods are based on the combination or improvement of traditional digital image processing techniques (IPTs) [3], such as thresholding [4,5,6] and edge detection [7,8,9,10]. However, these methods are generally based on the significant assumption that the intensities of crack pixels are darker than the background and usually continuous, which makes these methods difficult to use effectively in the environment of complex background noise [11,12]. In order to improve the accuracy and integrity of crack detection, the methods based on wavelet transform [13,14] are proposed to lift the crack regions. However, due to the anisotropic characteristics of wavelets, they may not deal well with cracks with large curvatures or poor continuities [2].
In recent studies, several minimal path methods [15,16] have also been used for crack detection. Although these methods make use of crack features in a global view [3] and achieve good performance, their main limitation is that seed points for path tracking need to be set in advance [17], and the calculation cost is too high for practical application.
Activation CFD 2016 Crack
Download: https://ssurll.com/2vGoN7
Unet [32], as a typical representative of semantic segmentation algorithm, has achieved great success in medical image segmentation. There are many similarities between pavement crack detection and medical image segmentation, so it is natural to apply Unet to pavement crack segmentation.
The patchwise detection method, which divides the original pavement images into many small patches, is adopted by more researchers due to its two advantages. First, more data can be generated, and second, the localization information of cracks can be obtained. Zhang et al. [39] proposed a six-layer CNN network with four convolutional layers and two fully connected layers and used their convolutional neural network to train 99 99 3 small patches, which were split from 3264 2248 road images collected by low-cost smartphones. The output of the network was the probability of whether a small patch was a crack or not. Their study shows that deep CNNs are superior to traditional machine learning techniques, such as SVM and boosting methods, in detecting pavement cracks. Pauly et al. [40] used a self-designed CNN model to study the relationship between network depth and network accuracy and proved the effectiveness of using a deeper network to improve detection accuracy in pavement crack detection based on computer vision. In contrast with [39], which used the same number of convolution kernels in all convolution layers, Nguyen et al. [41] used a convolution neural network with an increased number of convolution kernels in each layer because the features were more generic in the early layers and more original dataset specific in later layers [42]. Eisenbach et al. [43] presented the GAPs dataset, constructed a CNN network with eight convolution layers and three full connection layers, and analyzed the effectiveness of the state-of-the-art regularization techniques. However, its network input size was 64 64 pixels, which was too small to provide enough context information. The same problem also existed in [44,45,46].
Zhang et al. put forward CrackNet [52], which is an earlier study on pixel-level crack detection based on CNN. The prominent feature of CrackNet is using a CNN model without a pooling layer to retain the spatial resolution. Fei et al. have upgraded it to Cracknet-V [53]. While CrackNet and its series versions perform well, they are primarily used for 3D road crack images, and their performances on two-dimensional (2D) road crack images have not been validated. Fan et al. [3] proposed a pixel-level structured prediction method using CNN with full connections (FC) layers, but it has the disadvantage that it requires a long inference time for testing.
The CFD dataset, published in [23], consists of 118 RGB images with a resolution of 480 320 pixels. All of the images were taken using an iPhone5 smartphone on the road in Beijing, China, and can roughly reflect the existing urban road conditions in Beijing. These crack images have uneven illumination and contain noise such as shadows, oil spots, and lane lines, and most cracks in these images are thin cracks, which make crack detection difficult. We randomly divided 70% of the dataset (82 images) for training and 30% of the dataset (36 images) for testing.
The Crack500 dataset, shared by Yang et al. in the literature [60], contains 500 original images with a resolution of 2560 1440 collected at the main campus of Temple University. Each original image was cropped into a non-overlapping image area of 640 360, resulting in 1896 training images, 348 validation images, and 1123 test images. These images are characterized by low contrast between cracks and background, as well as noise such as oil pollution and occlusions, which increase the difficulty of detection.
The DeepCrack dataset [2] contains 537 crack images, including both concrete pavement and asphalt pavement, with complex background and various crack widths, ranging from 1 pixel to 180 pixels. We kept the same data split as the original paper, with 300 images for training and 237 images for testing.
Figure 3 shows the crack detection results of six typical input images of our method and the three methods to be compared. The first column is the original input crack image, the second column is the label image corresponding to the first column image, and the next four columns are the predicted output images of the four comparison algorithms. As can be seen from Figure 3, all these algorithms could detect the rough crack profile. However, in terms of details, all three algorithms, FCN, Unet, and SegNet, had false detection and missing cracks resulting in discontinuity of cracks to a varying degree. Our algorithm was obviously better than the three algorithms, with the least false detection and missing cracks, and the closest to the ground truth.
Abdalqader, Ahmed F., Jin, Fei ORCID: -0003-0899-7063 and Al-Tabbaa, Abir 2016. Development of greener alkali-activated cement: utilisation of sodium carbonate for activating slag and fly ash mixtures. Journal of Cleaner Production 113 , pp. 66-75. 10.1016/j.jclepro.2015.12.010
Abeysekera, M. ORCID: -0002-1743-3254, Wu, J. ORCID: -0001-7928-3602, Jenkins, N. ORCID: -0003-3082-6260 and Rees, M. 2016. Steady state analysis of gas networks with distributed injection of alternative gas. Applied Energy 164 , pp. 991-1002. 10.1016/j.apenergy.2015.05.099
Abeysekera, Muditha ORCID: -0002-1743-3254, Jenkins, Nick and Wu, Jianzhong ORCID: -0001-7928-3602 2016. Integrated energy systems: an overview of benefits, analysis, research gaps and opportunities. HhubNet Position Paper Series, HubNet.
Abo Dhaheer, M. S., Kulasegaram, S. ORCID: -0002-9841-1339 and Karihaloo, B. L. ORCID: -0003-0657-2678 2016. Simulation of self-compacting concrete flow in the J-ring test using smoothed particle hydrodynamics (SPH). Cement and Concrete Research 89 , pp. 27-34. 10.1016/j.cemconres.2016.07.016
Adam, A., Pavlidis, D., Percival, J. R., Salinas, P., Xie, Z. ORCID: -0002-5180-8427, Fang, F., Pain, C. C., Muggeridge, A. H. and Jackson, M. D. 2016. Higher-order conservative interpolation between control-volume meshes: Application to advection and multiphase flow problems with dynamic mesh adaptivity. Journal of Computational Physics 321 , pp. 512-531. 10.1016/j.jcp.2016.05.058
Adeuyi, Oluwole Daniel and Cheah, Marc 2016. Real-time simulation experiments of DC grids. Van Hertem, Dirk, Gomis-Bellmunt, Oriol and Liang, Jun ORCID: -0001-7511-449X, eds. HVDC Grids: For Offshore and Supergrid of the Future, IEEE Press Series on Power Engineering, Wiley, pp. 371-388. (10.1002/9781119115243.ch18)
Adeuyi, Oluwole, Cheah, Marc, Liang, Jun ORCID: -0001-7511-449X and Jenkins, Nicholas ORCID: -0003-3082-6260 2016. Fast frequency response from offshore multi-terminal VSC-HVDC schemes. IEEE Transactions on Power Delivery 32 (6) , pp. 2442-2452. 10.1109/TPWRD.2016.2632860
Afzanizam Samiran, Nor, Samiran, Jo-Han, Nazri Mohd Jafar, Mohammad, Valera Medina, Agustin ORCID: -0003-1580-7133 and Chong, Cheng Tung 2016. H2-rich syngas strategy to reduce NOx and CO emissions and improve stability limits under premixed swirl combustion mode. International Journal of Hydrogen Energy 41 (42) , pp. 19243-19255. 10.1016/j.ijhydene.2016.08.095
Agathos, Konstantinos, Chatzi, Eleni and Bordas, Stephane P. A. ORCID: -0001-8634-7002 2016. Stable 3D extended finite elements with higher order enrichment for accurate non planar fracture. Computer Methods in Applied Mechanics and Engineering 306 , pp. 19-46. 10.1016/j.cma.2016.03.023
Ahmad, Muhammad Waseem ORCID: -0002-7269-4369, Mourshed, Monjur ORCID: -0001-8347-1366, Mundow, David, Sisinni, Mario and Rezgui, Yacine ORCID: -0002-5711-8400 2016. Building energy metering and environmental monitoring - A state-of-the-art review and directions for future research. Energy and Buildings 120 , pp. 85-102. 10.1016/j.enbuild.2016.03.059
Ahmad, Muhammad ORCID: -0002-7269-4369, Hippolyte, Jean-Laurent ORCID: -0002-5263-2881, Reynolds, Jonathan ORCID: -0001-9106-9246, Mourshed, Monjur ORCID: -0001-8347-1366 and Rezgui, Yacine ORCID: -0002-5711-8400 2016. Optimal scheduling strategy for enhancing IAQ, visual and thermal comfort using a genetic algorithm. Presented at: ASHRAE IAQ 2016 Defining Indoor Air Quality: Policy, Standards and Best Practices, Alexandria, VA, USA, 12-14 September 2016.
Ahmad, Muhammad ORCID: -0002-7269-4369, Mourshed, Monjur ORCID: -0001-8347-1366, Yuce, Baris ORCID: -0002-9937-1535 and Rezgui, Yacine ORCID: -0002-5711-8400 2016. Computational intelligence techniques for HVAC systems: a review. Building Simulation 9 (4) , pp. 359-398. 10.1007/s12273-016-0285-4 2ff7e9595c
Comments