Search arXivSearch

arXiv subjects

Zhangding Liu

Publications and source records attributed to Zhangding Liu.

2 recordsLinked to original sources

MCANet: A Multi-Scale Class-Specific Attention Network for Multi-Label Post-Hurricane Damage Assessment Using UAV Imagery

Hurricanes cause widespread damage to buildings, roads, and other infrastructure, making timely post-disaster damage assessment critical for emergency response and recovery planning. Unmanned aerial vehicle (UAV) imagery provides high-resolution observations of affected areas, but post-hurricane scenes are difficult to classify because multiple damage categories often co-occur within the same image, appear at different spatial scales, and include visually similar severity levels as well as rare but operationally important classes. To address these challenges, this study presents MCANet, a multi-label classification framework for post-hurricane UAV damage assessment. MCANet integrates a Res2Net-based backbone for multi-scale representation with class-specific residual attention to aggregate spatial evidence separately for each damage category. Evaluation on the RescueNet dataset, which includes 4,494 UAV images collected after Hurricane Michael and annotated with 10 damage categories, shows that MCANet achieves the highest mean average precision (mAP) among the evaluated models, with an mAP of 91.37%. Compared with Vision Transformer (ViT-B/16), MCANet improves mAP by 1.90 percentage points while using approximately 50% fewer parameters and 52% fewer giga floating-point operations (GFLOPs). Class-specific residual attention contributed an mAP increase of 1.52 percentage points over class-agnostic attention (CA); the Res2Net backbone alone produced concentrated gains on scale-heterogeneous or underrepresented classes such as Road Blocked and Pool. The single-head configuration achieved the highest overall mAP, whereas the two-head configuration provided targeted gains for Road Blocked and Building Major Damage. MCANet supports rapid post-disaster image screening and provides structured outputs for downstream emergency-management workflows.

cs.CV

Knowledge-Guided Vision-Language Inference for Image-Based Urban Flood Depth Estimation

Timely floodwater depth estimates support road accessibility assessment and emergency response during urban flooding. Supervised vision methods often require extensive labeled datasets, while recent foundation vision-language models (VLMs) offer flexible visual reasoning but can inconsistently yield large errors in metric depth estimation. This paper proposes FloodVision, a knowledge-guided framework for estimating flood depth from a single RGB image. FloodVision integrates a general-purpose VLM with FloodKG, a domain knowledge base encoding canonical object dimensions and component landmarks (e.g., wheel arch, curb top) to encourage reasoning at the component level rather than treating objects as wholes. This injects explicit geometric grounding without task-specific training. Evaluated on 654 crowdsourced MyCoast New York flood images with resident-reported depths as proxy labels, FloodVision reduces the mean absolute error from 15.62 cm to 8.75 cm and the median error from 14.35 cm to 7.75 cm, with lower error than the VLM-only baseline in 69.3% of cases. The paper also discusses current limitations and future integration into urban digital twin systems.

cs.CV