PSI - Issue 64

Shirley J. Dyke et al. / Procedia Structural Integrity 64 (2024) 21–28 Author name / Structural Integrity Procedia 00 (2019) 000–000

25

5

Fig. 2. Image localization: red – the path of the inspector, blue – the point cloud model.

Inspired by Choi et al.’s research, Wogen et al. developed a method to apply the S-CNN to support bridge inspections. They develop a comprehensive methodology that integrates similarity techniques with GPS data to fuse both sets of information to provide a tool for human inspectors (2024). Similarity search is use to directly retrieve historical inspection images for highway bridges. Thus, the inspector can rapidly pull up an image of a specific bridge component collected in one or more previous years simply by taking an image in the current day, and using that as a query image to the tool. A dataset of over 1000 real-world bridge deck images is used, and a metric is proposed that leverages both the image similarity and the GPS. In concert, these AI-based methods add contextual information, particularly spatial data, to the data sets, and thereby reduce the time that the human engineer requires for such tasks. 4. I nformation Fusion The advances in automation mentioned in the prior section will empower engineers to perform intricate tasks with great efficiency. Images are no longer just pictures. Their content is readily extracted and organized, while their context is harnessed for data reuse. But beyond mere organization of these data lies the potential for deeper synthesis. Lenjani et al. worked toward automating the post-event data collection and analysis process, developing a method for the fusion of pre-event and post-event information (2020). Relevant building characteristics are automatically extracted using physical attributes such as first-floor elevation, number of stories, and construction material, properties extracted from existing street view image databases. Post-event condition of residential buildings is assessed using reconnaissance images, coupled with a probabilistic approach to merge the findings from multiple images. This approach significantly boosts the efficiency of post-disaster building assessments and ensures a more reliable outcome by considering visible damage from different perspectives. Building on this concept, Liu et al. developed an automated technique for classifying the damage level of buildings after a seismic event (2022). The methodology integrates computer vision with a Naïve Bayes fusion algorithm, processing images to assess visible damage to reinforced concrete and masonry components. By using the confidence in the damage classification of each image, the technique delivers an overall damage state of the building based on the set of images collected. This approach is validated against a dataset of thousands of images from buildings affected by various natural disasters. The fusion of data is also an application for condition state assessment of bridge decks. Through a number of methodologies that automate the detection and measurement of cracking in bridge decks exists via image classification and semantic segmentation, Zhang et al. developed a method to turn those quantitative values into a bridge deck condition rating based on current accepted practices. Central to their exploration is a cost-benefit analysis, weighing the direct costs of inspections against the potential consequences of AI misclassifications. More recently, Iturburu et al. rapidly and automatically classify the vulnerability of reinforced concrete buildings,