Image-based damage conditional assessment of large-scale infrastructure systems using remote sensing and deep learning approaches

H. Pan, Z. Zhang, X. Wang, Z. Lin
North Dakota State University,
United States

Keywords: Deep learning, AI, remoste sensing, damage conditonal assessment, structural health monitoring


Assessment of health state of large-scale infrastructure systems are crucial to ensure their operational safety and take timely emergency response if necessary. However, the complexity and long distance of these systems post a great challenge. While complex systems successfully incorporate the satellite imagery through remote sensing for conditional assessment, these conventional methods are often case-by-case base, and thus may not make full use of relevant information from sensitive features. In this study, we propose the image-based damage and conditional assessment of large-scale systems using deep learning approaches. The support vector machine as a shallow learning algorithm is selected as a comparison. The deep convolutional neural networks are optimally designed by layers and architecture for the satellite images to extract the sensitive features for assessment. The findings show that the machine learning methods exhibit great potential for infrastructure assessment, such as high bridges, and oil/gas pipeline assessment at both spatial and temporary scales over conventional methods, and particularly the deep learning could quantitatively provide valuable information for infrastructure conditional assessment.