SAFENET: Image-Based Infrastructure Classification of Bridges and Viaducts for Seismic Risk Assessment

Controlled Benchmarking and Component-Aware Ablation for Railway Viaduct Structural and Damage Segmentation

Automated damage inspection of railway viaducts requires pixel-level identification of structural components and surface damage such as cracking and rebar exposure. A common assumption in bridge inspection is that damage segmentation improves when component information is provided alongside the image. This study tests that assumption on the Tokaido synthetic viaduct dataset using controlled comparisons between segmentation models with and without component information. Both damage and structural component segmentation are evaluated across multiple architectures, and the trained component model is assessed on real viaduct photographs against a baseline model requiring no task-specific training. Under the original random split, explicit component conditioning does not produce a measurable improvement in damage segmentation: all tested strategies remain within 0.008 mean Intersection-over-Union (mIoU) of a baseline without component input, and this null result persists even when component predictions are reliable. Under a leakage-controlled scene-disjoint split, however, the same component-aware variants show a small positive trend (up to +0.019 mIoU over three seeds), so the effect of component conditioning depends on the evaluation protocol. The best unconditioned model reaches 0.569 mIoU for damage segmentation; for real-photo component segmentation, the trained model reaches 0.424 mIoU compared with 0.250 mIoU for the training-free baseline. These results show that multitask benefits reported in bridge inspection do not automatically translate into gains from explicit use of component information on synthetic viaduct data, where damage occurs almost exclusively on columns yet is too sparse for structural element identity to yield more than a marginal localisation gain. The multi-architecture benchmark and the measured real-photo structural transfer gap provide reference baselines for subsequent work on component-aware and transfer-robust inspection.

AI-Based Image and Data Analysis for Automated Assessment of Residential Damage in Seismic Regions

Earthquakes remain a critical threat to global infrastructure. Recent catastrophic events, such as the 2023 Kahramanmaraş earthquakes in Türkiye and Syria, underscore the vital necessity of rapid, accurate post-disaster building damage evaluations. Structural collapse under seismic loading leads to substantial loss of life and severe economic disruption, particularly in regions dominated by aging building stocks that predate modern seismic design codes. To address the limitations of conventional manual inspections, this study introduces a comprehensive artificial intelligence (AI) framework designed to automate and enhance post-earthquake structural assessments. Leveraging a heterogeneous dataset from the 2021 Haiti earthquake, which includes both categorical building attributes and post-disaster imagery, the proposed approach employs rigorous data preprocessing and exploratory analysis to identify key vulnerability indicators and resolve data inconsistencies. Independent predictive pipelines were developed utilizing state-of-the-art machine learning algorithms for tabular data and deep learning architectures for image analysis. Subsequently, a novel hybrid meta-classifier was implemented to fuse these distinct modalities. By integrating spatial and structural context with direct visual evidence of damage, the hybrid model is successful in estimating structural damage severity. Among all evaluated approaches, this multimodal framework significantly improved predictive reliability. The hybrid model achieved a classification accuracy of 89%, consistently outperforming isolated tabular and image-based models. These findings highlight the efficacy of multimodal data fusion in disaster analytics and suggest that AI-driven hybrid architectures can serve as robust, scalable decision support tools for structural engineers and emergency response agencies.

Rapid post-earthquake damage assessment using patch-level CNNs and VLM

Rapid and accurate post-earthquake damage assessment is critical for effective disaster response and enabling sustainable community recovery. This study addresses the urgent need for timely and precise quantification of earthquake-induced building damage, moving beyond simple detection, to inform actionable engineering decisions. It explores AI-driven image classification to expedite damage evaluation by identifying damaged buildings from post-disaster photos much faster than conventional methods. A total of 154,380 indoor image patches (224 × 224 pixels) from photos of buildings affected by the 2020 M6.4 Petrinja, Croatia earthquake, were compiled. These patches include both close-up cracks and wider building context. Two convolutional neural networks (ResNet-18 and ConvNeXt-Tiny) were trained on the patch dataset for binary damage classification. Additionally, large vision-language model (VLM) Gemma 3 27B was evaluated in a zero-shot binary classification setting by providing it with a descriptive prompt to classify each patch as damaged or undamaged. ConvNeXt-Tiny achieved the highest performance, with a macro F1-score of 0.849 and a damaged-class F1-score of 0.731, slightly outperforming ResNet-18. Gemma 3 27B achieved a macro F1-score of approximately 0.6 but a damaged-class F1-score of 0.366, with many false positives. Integrating such a granular approach into rapid loss assessment frameworks significantly enhances decision-making capabilities and post-disaster recovery efforts, providing a more detailed understanding of structural integrity across affected areas.

MDeepLab: a real-time lightweight network for accurate road crack segmentation

Damage Classification in Historical Buildings Through Transfer Learning Approaches

Historical buildings are important cultural assets that reflect the identity of cities and preserve the collective memory of societies. However, these structures are increasingly exposed to environmental degradation and human-induced impacts, making their systematic documentation and condition assessment essential for effective conservation strategies. Recent advances in artificial intelligence have provided powerful tools for image-based analysis in the field of heritage preservation. In particular, transfer learning enables the adaptation of pre-trained deep learning models to domain-specific tasks with limited labeled data. In this study, a deep transfer learning-based framework is proposed for automatic damage detection and classification in historical buildings. A new near-balanced dataset of 20,000 images spanning six deterioration categories was developed and made publicly available. Ten convolutional neural network and transformer architectures pre-trained on ImageNet were systematically compared under a unified Bayesian optimization protocol. Experimental results on a held-out test set show that EfficientNetB3 achieves the highest classification accuracy (97.65%), while AlexNet obtains the lowest performance (83.89%); the validation set was used exclusively for hyperparameter tuning. The results demonstrate that transfer learning-based models can effectively identify visually observable deterioration patterns and provide reliable support for automated documentation processes. The proposed framework contributes to the development of data-driven decision-support tools for digital documentation and condition assessment in heritage conservation.

Rapid Building Damage Detection from Remote Sensing Images: a Novel Lightweight Network with Contrastive Learning

Abstract. Accurate and timely building damage detection (BDD) is crucial for disaster emergency response. Although deep learning-based change detection methods have made significant progress in remote sensing, their practical application in disasters still faces two major challenges: (1) Existing high‑accuracy models are typically computationally complex and difficult to deploy for real‑time inference on edge devices. (2) Model performance heavily relies on large amounts of annotated data, but disaster data are extremely scarce. To address these challenges, this paper proposes a novel lightweight Local-Global Interaction Network (LGINet) for efficient BDD. The core of LGINet is the proposed Local-Global Interaction Unit (LGIU), which achieves efficient fusion of detailed and contextual features through a dual‑path architecture and channel‑wise cross‑attention mechanism. Furthermore, a Frequency Difference Enhancement Unit (FDEU) is proposed to generate more accurate damage features, and contrastive learning is employed to reduce the model’s sensitivity to weather conditions and its reliance on annotated data. Experimental results on the xBD and WBD datasets show that LGINet achieves F1-scores of 81.76% and 80.91%, respectively, with an inference speed of 47.83 FPS. It achieves the best balance between accuracy and efficiency, outperforming existing methods.