Research Experiences for Undergraduates (REU), NHERI 2024: Using Street-Level Hurricane Damage Datasets to Advance AI-Supported Damage Detection and Characterization

The project aims towards reducing the time it takes to extract key features from post-disaster street-level imagery through the use of AI for damage detection and characterization, enhancing response times, and loss and resilience models. It creates a baseline for understanding and comparing the effectiveness of multiple open source VLMs for determining the presence and origin/type of damage. It examines multiple current zero-shot models for their capabilities in damage detection and characterization on street level imagery. Scientists currently working on reconnaissance and resilience for natural hazards will find this useful.

本项目旨在通过运用人工智能（AI）开展损伤检测与特征描述，缩短从灾后街道级影像中提取关键特征的时间，进而提升响应效率，并优化损失与韧性模型。它为理解和比较多款开源视觉语言模型（VLM）在判断损伤存在性、来源及类型方面的有效性建立了基准线。项目考察了当前多款零样本模型在街道级影像损伤检测与特征描述任务中的能力表现。从事自然灾害勘察与韧性研究的科研人员将发现本项目具有实用价值。