Huili Huang

Rapid assessment of post-earthquake infrastructure damage is essential for effective emergency response, yet current workflows remain time-consuming and labor intensive. Ground-level imagery can provide faster situational awareness than aerial or satellite observations, but its practical use is often constrained by the extensive annotation required by conventional supervised deep learning models. This study introduces SeismoMind, a decision-tree-based multimodal large language model (MLLM) framework for zero-shot post-earthquake damage understanding. Unlike conventional MLLM prompt development, which is typically guided by expert intuition, SeismoMind is informed by cognitive observations of non-experts and encodes a structured reasoning process toward expert-level damage assessment. This non-expert learning trajectory is encoded as a Decision-Tree Chain of Thought (DT-CoT) that guides the MLLM through a transparent, stepwise evaluation of infrastructure damage. SeismoMind is evaluated on two ground-view benchmarks. The Earthquake Image Database (EID) contains 13,513 online social media plateforms. The 2023 Türkiye Earthquake dataset includes 675 ground-level field images collected during post-earthquake reconnaissance by the Geotechnical Extreme Event Reconnaissance (GEER) and the Earthquake Engineering Research Institute (EERI)team. The framework achieves accuracies of 90.53% and 80.60%, respectively, which are comparable to state-of-the-art supervised computer vision baselines (91.71% and 80.59%) while requiring no task-specific training. In a size-matched limited-data setting, SeismoMind further outperforms supervised baselines by 8.83%. These findings show that encoding structured, learnable reasoning patterns observed in non-experts provide a practical path for generative AI systems to perform post-disaster damage assessment from ground-view imagery without relying on large, expensive annotated datasets.

Limited labeled imagery and tight response windows hinder the accurate damage quantification for post-disaster assessment. The objective of this study is to develop and evaluate a deep learning-based Domain-Adaptive Segmentation (DASeg) workflow to detect post-disaster damage using limited information available shortly after an event. DASeg unifies three Vision Foundation Models in an automatic workflow: fine-tuned DINOv2 supplies attention-based point prompts, fine-tuned Grounding DINO yields open-set box prompts, and a frozen Segment Anything Model (SAM) generates the final masks. In the earthquake-focused case study DASeg-Quake, the pipeline boosts mean Intersection over Union (mIoU) by 9.52% over prior work and 2.10% over state-of-the-art supervised baselines. In a zero-shot setting scenario, DASeg-Quake achieves the mIoU of 75.03% for geo-damage analysis, closely matching expert-level annotations. These results show that DASeg achieves superior workflow enhancement in infrastructure damage segmentation without needing pixel-level annotation, providing a practical solution for early-stage disaster response.

In the aftermath of earthquakes, social media images have become a crucial resource for disaster reconnaissance, providing immediate insights into the extent of damage. Traditional approaches to damage severity assessment in post-earthquake social media images often rely on classification methods, which are inherently subjective and incapable of accounting for the varying extents of damage within an image. Addressing these limitations, this study proposes a novel approach by framing damage severity assessment as a semantic segmentation problem, aiming for a more objective analysis of damage in earthquake-affected areas. The methodology involves the construction of a segmented damage severity dataset, categorizing damage into three degrees: undamaged structures, damaged structures, and debris. Utilizing this dataset, the study fine-tunes a SegFormer model to generate damage severity segmentations for post-earthquake social media images. Furthermore, a new damage severity scoring system is introduced, quantifying damage by considering the varying degrees of damage across different areas within images, adjusted for depth estimation. The application of this approach allows for the quantification of damage severity in social media images in a more objective and comprehensive manner. By providing a nuanced understanding of damage, this study enhances the ability to offer precise guidance to disaster reconnaissance teams, facilitating more effective and targeted response efforts in the aftermath of earthquakes.

The development of artificial intelligence (AI) provides an opportunity for rapid and accurate assessment of earthquake‐induced infrastructure damage using social media images. Nevertheless, data collection and labeling remain challenging due to limited expertise among annotators. This study introduces a novel four‐class Earthquake Infrastructure Damage (EID) assessment data set compiled from a combination of images from several other social media image databases but with added emphasis on data quality. Unlike the previous data sets such as Damage Assessment Dataset (DAD) and Crisis Benchmark, the EID includes comprehensive labeling guidelines and a multiclass classification system aligned with established damage scales, such as HAZUS and EMS‐98, to enhance the accuracy and utility of social media imagery for disaster response. By integrating detailed descriptions and clear labeling criteria, the labeling approach of EID reduces the subjective nature of image labeling and the inconsistencies found in existing data sets. The findings demonstrate a significant improvement in annotator agreement, reducing disagreement from 39.7% to 10.4%, thereby validating the efficacy of the refined labeling strategy. The EID, containing 13,513 high‐quality images from five significant earthquakes, is designed to support community‐level assessments and advanced computational research, paving the way for enhanced disaster response strategies through improved data utilization and analysis. The data set is available at DesignSafe: https://doi.org/10.17603/ds2‐yj8p‐hs62 .

Although large-scale labeled data are essential for deep convolutional neural networks (ConvNets) to learn high-level semantic visual representations, it is time-consuming and impractical to collect and annotate large-scale datasets. A simple and efficient unsupervised representation learning method named ScaleNet based on multi-scale images is proposed in this study to enhance the performance of ConvNets when limited information is available. The input images are first resized to a smaller size and fed to the ConvNet to recognize the rotation degree. Next, the ConvNet learns the rotation-prediction task for the original size images based on the parameters transferred from the previous model. The CIFAR-10 and ImageNet datasets are examined on different architectures such as AlexNet and ResNet50 in this study. The current study demonstrates that specific image features, such as Harris corner information, play a critical role in the efficiency of the rotation-prediction task. The ScaleNet supersedes the RotNet by ~7% in the limited CIFAR-10 dataset. The transferred parameters from a ScaleNet model with limited data improve the ImageNet Classification task by about 6% compared to the RotNet model. This study shows the capability of the ScaleNet method to improve other cutting-edge models such as SimCLR by learning effective features for classification tasks.

This research explores the utilization of the Black Marble nighttime light (NTL) product to detect and assess damage caused by hurricanes, tornadoes, and earthquakes. The study first examines average regional NTL trends before and after each disaster, demonstrating that NTL patterns for hurricanes closely align with the features of a resilience curve, unlike those for earthquakes and tornadoes. The relative NTL change ratio is computed using monthly and daily NTL data, effectively reducing variance due to daily fluctuations. Results indicate the robustness of the NTL change ratio in detecting hurricane damage, whereas its performance in earthquake and tornado assessment was inconsistent and inadequate. Furthermore, NTL demonstrates a high performance in identifying hurricane damage in well-lit areas and the potential to detect damage along tornado paths. However, a low correlation between the NTL change ratio and the degree of damage highlights the method’s limitation in quantifying damage. Overall, the study offers a promising, prompt approach for detecting damaged/undamaged areas, with specific relevance to hurricane reconnaissance, and points to avenues for further refinement and investigation.