由於河流廣布橋梁於台灣為重大之交通基礎建設，因此藉由評估橋梁劣化情況可以確保用路人之安全；再者，長期監測並分析橋梁劣化情況可以瞭解劣化區域於天災如頻繁之颱風與地震之前後變化，並可以基於此分析進行災前防範。而裂縫為一重要橋梁安全指標，傳統之橋梁裂縫以目視檢測，其需要大型機具輔助或是懸吊於橋梁下方尚可進行檢測。除了工作環境之潛在危險性，其為一勞力密集型之工作且欠缺量化檢測成果之指標。因此，本研究之目的為進行自動化橋梁裂縫區域之偵測。由於無人機之高機動性，本研究使用無人機載具搭載消費型相機進行橋樑表面影像蒐集。並利用全卷積網路(Fully convolutional networks, FCN)進行訓練後，以得到像元等級之預測成果。標記影像也為訓練模型之重要資料庫，而一般使用人工標記各像元，其需要大量人力及時間完成標記且標記不規則劣縫為一困難工作。本研究採用多重解析度影像分割技術(Multiresolution segmentation, MRS)，其將像元等級標記過程簡化為物件等級進行標記。並以影像處理技術進行變形、調整影像之伽馬値與裁切以擴增影像資料庫的數量以及多樣性。本研究將各類別誤差進行加權解決裂縫與背景像元數量比例差異甚大之問題。訓練完成之模型可以達到合理之預測成果，但FCN過度預期裂縫寬度。因此，本研究提出混合方法整合FCN預測成果與物件導向式影像分析(Object-Based Image Analysis, OBIA)，將FCN像素級偵測成果轉換為物件級別，並利用物件之特徵進行裂縫細部偵測。最後，將本研究混合方法之成果與單獨採用OBIA成果進行案例比較。單獨採用OBIA所偵測之裂縫較為貼近裂縫之邊界但由於難以制定通用各案例之指標與門檻値，導致成果完整性仍有不足之處。本研究提出之混合方法成果較為穩定，且經由加入更多訓練影像得以增強此方法的穩健度，使其能應用在更複雜裂縫偵測案例。

For the widespread rivers in Taiwan, the bridge is an essential infrastructure for the transportation. Accordingly, the evaluation of existing bridge deterioration is to ensure the safety of the public transportation. Furthermore, the long-term bridge monitoring can be analyzed to understand bridge deficient conditions under the disasters such as the frequently occurred earthquakes and typhoons, and base on the monitoring result to take related precautions. In addition, crack is the significant degradation index for the direction and width correlating with the structural damages. Conventionally, crack inspection was conducted by human in-situ visual examination. This procedure regularly requires under bridge inspection assisted with vehicle or climbing under the bridge. Despite its high-risk environment, it’s also a labor-intensive task and lacks quality assessment for the inspection result. Therefore, automatic detection of bridge crack is the goal of this research. The unmanned aerial vehicle (UAV) equipped with a commercial digital camera is adopted to acquire the image of bridge surface for its high flexibility. In addition, the fully convolutional networks (FCN) can be trained end-to-end and a pixel level prediction result can be obtained. Besides the images are essential for training the model, the high-quality labelled image is requisite. Generally, manually digitalizing each crack pixel is time consuming and it is challenging for irregular cracks. The multiresolution segmentation (MRS) is utilized to assist the labelling by selecting the objects of crack after segmentation. The augmentation is performed to increase the number and diversity of the database to make the trained model more robust by deformation, diversifying the gamma values and cropping procedure. Nevertheless, the portion of the crack is too small compared to the background in an image, which causes class imbalanced problem while training. To tackle with this problem, the weighted class loss is adopted to balance the loss between two classes. The trained result can reach reasonable accuracy, but the FCN detection results is overestimating at the detection of crack width. A hybrid method is proposed with the integration of FCN detection result and object-based image analysis (OBIA) to refine the detection result. By the hybrid method, the detection result is transformed from the pixels into the objects and the result can be refined with the features of the objects. Eventually, the hybrid result is compared to an OBIA-standalone method by the case studies. The comparisons demonstrate that the OBIA-standalone method can obtained more detail objects but lack of completeness due to universal feature indexes and a global threshold value is hard to define. The hybrid method can result in more stable and flexible for the model can fine-tuned with more crack training samples. Consequently, the proposed model can be more adaptive in the detection of crack targets.

Baatz, M., 2000. Multi resolution Segmentation: an optimum approach for high quality multi scale image segmentation, Beutrage zum AGIT-Symposium. Salzburg, Heidelberg, 2000, pp. 12-23.
Blaschke, T., 2010. Object based image analysis for remote sensing. ISPRS journal of photogrammetry and remote sensing 65, 2-16.
Bottou, L., 2012. Stochastic gradient descent tricks, Neural networks: Tricks of the trade. Springer, pp. 421-436.
Canny, J., 1987. A computational approach to edge detection, Readings in Computer Vision. Elsevier, pp. 184-203.
Cha, Y.J., Choi, W., Büyüköztürk, O., 2017. Deep Learning‐Based Crack Damage Detection Using Convolutional Neural Networks. Computer‐Aided Civil and Infrastructure Engineering 32, 361-378.
Drǎguţ, L., Tiede, D., Levick, S.R., 2010. ESP: a tool to estimate scale parameter for multiresolution image segmentation of remotely sensed data. International Journal of Geographical Information Science 24, 859-871.
Fan, Z., Wu, Y., Lu, J., Li, W., 2018. Automatic Pavement Crack Detection Based on Structured Prediction with the Convolutional Neural Network. arXiv preprint arXiv:1802.02208.
Gao, Y., Guo, S., Huang, K., Chen, J., Gong, Q., Zou, Y., Bai, T., Overett, G., 2017. Scale optimization for full-image-CNN vehicle detection, Intelligent Vehicles Symposium (IV), 2017 IEEE. IEEE, pp. 785-791.
Hocenski, Z., Vasilic, S., Hocenski, V., 2006. Improved canny edge detector in ceramic tiles defect detection, IEEE Industrial Electronics, IECON 2006-32nd Annual Conference on. IEEE, pp. 3328-3331.
Hsiao, K.W., Rau, J.Y., 2017. Three-dimensional Information Extraction of Bridge Deteriorating Area through Multi-rotary UAV Imagery.
Johnson, B., Xie, Z., 2011. Unsupervised image segmentation evaluation and refinement using a multi-scale approach. ISPRS Journal of Photogrammetry and Remote Sensing 66, 473-483.
LeCun, Y., Bengio, Y., 1995. Convolutional networks for images, speech, and time series. The handbook of brain theory and neural networks 3361, 1995.
LeCun, Y., Boser, B.E., Denker, J.S., Henderson, D., Howard, R.E., Hubbard, W.E., Jackel, L.D., 1990. Handwritten digit recognition with a back-propagation network, Advances in neural information processing systems, pp. 396-404.
LeCun, Y., Bottou, L., Bengio, Y., Haffner, P., 1998. Gradient-based learning applied to document recognition. Proceedings of the IEEE 86, 2278-2324.
Long, J., Shelhamer, E., Darrell, T., 2015. Fully convolutional networks for semantic segmentation, Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 3431-3440.
Marques, A., Correia, P.L., 2012. Automatic road pavement crack detection using SVM. Lisbon, Portugal: Dissertation for the Master of Science Degree in Electrical and Computer Engineering at Instituto Superior Técnico.
Nair, V., Hinton, G.E., 2010. Rectified linear units improve restricted boltzmann machines, Proceedings of the 27th international conference on machine learning (ICML-10), pp. 807-814.
Rau, J.Y., Hsiao, K.W., Jhan, J.P., Wang, S.H., Fang, W.C., Wang, J.L., 2017. Bridge Crack Detection Using Multi-Rotary Uav and Object-Base Image Analysis. The International Archives of Photogrammetry, Remote Sensing and Spatial Information Sciences 42, 311.
Simonyan, K., Zisserman, A., 2014. Very deep convolutional networks for large-scale image recognition. arXiv preprint arXiv:1409.1556.
Sternberg, S.R., 1983. Biomedical image processing. Computer 16, 22-34.
Su, B.W., Rau, J.Y., 2017. LOD-1 Building Model Reconstruction from HRSI Derived DSM and True-orthoimage.
Talab, A.M.A., Huang, Z., Xi, F., HaiMing, L., 2016. Detection crack in image using Otsu method and multiple filtering in image processing techniques. Optik-International Journal for Light and Electron Optics 127, 1030-1033.
Zhang, A., Wang, K.C., Li, B., Yang, E., Dai, X., Peng, Y., Fei, Y., Liu, Y., Li, J.Q., Chen, C., 2017. Automated Pixel‐Level Pavement Crack Detection on 3D Asphalt Surfaces Using a Deep‐Learning Network. Computer‐Aided Civil and Infrastructure Engineering 32, 805-819.
Zhang, Y., 2014. The design of glass crack detection system based on image preprocessing technology, Information Technology and Artificial Intelligence Conference (ITAIC), 2014 IEEE 7th Joint International. IEEE, pp. 39-42.