생성형 인공지능 기반 터널 콘크리트 균열 영상 데이터의 증강을 위한 파라미터 분석
Parameter analysis for augmentation of tunnel concrete crack image data based on generative AI
1
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
노후 인프라 구조물을 안전하게 유지하기 위해서는 지속적인 관리가 필요하며, 이는 터널 내부의 콘크리트 구조물에도동일하게 적용된다. 터널 구조물의 건전성 관리는 정기적인 점검과 정밀한 검사 기술을 통해 이루어진다. 기존의 검사 방법은 인력 기반으로, 작업자가 직접 현장을 방문하여 육안으로 상태를 확인하고 기록하는 방식이다. 이로 인해 구조물 상태는 경험적이고 주관적인 판단에 의해 결정될 수밖에 없는 상황이다. 이러한 문제를 개선하고 점검 결과의 객관성과 신뢰성을 높이기 위해 고해상도 카메라와 딥러닝을 활용한 방법이 활발히 연구되고 있다. 특히, 터널 구조물에 발생하는 균열을 탐지하는 신경망 모델 기반의 알고리즘은 높은 정확도를 보여주고 있다. 그러나 이러한 딥러닝 기술은 다수의 훈련영상 데이터가 확보되었다는 전제가 필요하다. 현실적으로 균열과 같은 손상 영상은 주변에서 흔히 찾아볼 수 없고, 이를확보하는 데 많은 비용과 시간이 소요된다. 이러한 문제를 해결하기 위해 본 연구에서는 생성형 AI를 이용해 균열 영상데이터를 증강하는 방법을 제안하였다. 또한, 실제와 유사한 균열 영상을 생성하기 위한 파라미터 분석을 완료하였고,그 결과 Fréchet Inception Distance가 31.73의 값을 나타내는 성능의 생성모델을 확보하였다. 이러한 방법은 향후 균열탐지 훈련 방법과 연계하여 유지관리 점검의 정확성과 신뢰성을 높이는 기술로 활용될 것으로 기대된다.
Abstract
To maintain the safety of aging infrastructure, continuous management is essential, and this applies equally to concrete structures inside tunnels. The health of tunnel structures is managed through regular inspections and precise examination techniques. Traditional inspection methods are manpower-based, where workers visit thesite in person to visually assess and record the condition. As a result, the condition of the structure is often determined based on subjective and experiential judgment. To address these issues and enhance the objectivity and reliability of inspection results, methods using high-resolution cameras and deep learning are being actively researched. Neural network model-based algorithms for detecting cracks in tunnel structures, in particular, have demonstrated high accuracy. However, such deep learning technology relies on the premise that a large amount of training image data is available. In reality, damage images such as cracks are not easily found, and collecting them involves significant costs and time. To address this challenge, this study proposes a method for augmenting crack image data using generative AI. Additionally, parameter analysis was conducted to generate crack images resembling real ones, resulting in a generative model with a performance of 31.73 Fréchet Inception Distance. This method is expected to be applied in conjunction with crack detection training methods, contributing to the improved accuracy and reliability of maintenance inspections.
참고문헌 (19)
- Bianchi, E., "Concrete crack conglomerate dataset", University Libraries, Virginia Tech, ~ (2021)
- Chen, X., "Semi-supervised semantic segmentation with cross pseudo supervision", Proceedings of the IEEE, 2021, 2613~2622
- Goodfellow, I. J., "Generative adversarial nets", Proceedings of the Advances in Neural Information Processing Systems (NIPS), pp.2672~2680 (2014)
- Heusel, M., "Gans trained by a two time-scale update rule converge to a local nash equilibrium", Proceedings of the Advances in Neural Information Processing Systems, 30, pp.6626~6637 (2017)
- Hsieh, Y. A., "Machine learning for crack detection : review and model performance comparison", Journal of Computing in Civil Engineering, 34(5), pp.04020038~ (2020)
- Huang, B., "Underwater dam crack image generation based on unsupervised image-to-image translation", Automation in Construction, 163, pp.105430~ (2024)
- Lee, S. Y., "Development of an inspection system for cracks in a concrete tunnel lining", Canadian Journal of Civil Engineering, 34(8), pp.966~975 (2007)
- Lei, M., "A novel tunnel-lining crack recognition system based on digital image technology", Tunnelling and Underground Space Technology, 108, pp.103724~ (2021)
- Li, G., "Automatic tunnel crack detection based on U-Net and a convolutional neural network with alternately updated clique", Sensors, 20(3), pp.717~ (2020)
- Li, S., "High-resolution concrete damage image synthesis using conditional generative adversarial network", Automation in Construction, 147, pp.104739~ (2023)
- Liu, J., "FDDWNet : a lightweight convolutional neural network for real-time semantic segmentation", Proceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing, pp.2373~2377 (2020)
- "Proceedings of the IEEE", 2016, 1508~1512
- Maeda, H., "Generative adversarial network for road damage detection", Computer-Aided Civil and Infrastructure Engineering, 36(1), pp.47~60 (2021)
- Munawar, H. S., "Image-based crack detection methods : a review", Infrastructures, 6(8), pp.115~ (2021)
- Phares, B. M., "Routine highway bridge inspection condition documentation accuracy and reliability", Journal of Bridge Engineering, 9(4), pp.403~413 (2004)
- Rombach, R., "High-resolution image synthesis with latent diffusion models", Proceedings of the IEEE, 2022, 10684~10695
- Shim, S., "CycleGAN based translation method between asphalt and concrete crack images for data augmentation", The Journal of The Korea Institute of Intelligent Transport Systems, 21(5), pp.171~182 (2022)
- Shim, S., "Self-training approach for crack detection using synthesized crack images based on conditional generative adversarial network", Computer-Aided Civil and Infrastructure Engineering, 39(7), pp.1019~1041 (2024)
- Zhang, K., "Self-supervised structure learning for crack detection based on cycle-consistent generative adversarial networks", Journal of Computing in Civil Engineering, 34(3), pp.04020004~ (2020)