Damage-Informed Drone Reconnaissance

From Building Damage Recognition to Drone Mission Planning

After disasters, responders must rapidly identify damaged buildings and prioritize follow-up inspections. We connect AI-based damage recognition with drone mission planning to support targeted disaster reconnaissance.

01 — The Problem

Seeing the Damage Is Not Enough.

Drones can capture extensive imagery of disaster-affected areas, but interpreting building damage and deciding where to inspect next remain challenging. Damage assessment and drone mission planning are often disconnected.

Aerial view of tornado destruction across a neighborhood, showing damaged buildings and debris.
Tornado damage. Image: NWS/NOAA.

02 — Building Damage Recognition

Identifying Damage from Aerial Imagery.

We developed a deep-learning model to detect and classify building damage, trained on 24,496 annotated damage instances and achieving 71.9% mean average precision on test data.

Annotated damage instances
24,496
Mean average precision
71.9%
Explore all six damage levels
Considerable damage
Example of the considerable tornado-related building damage category.

03 — Damage Mapping

Turning Damage Recognition into Spatial Information.

In a Chattanooga-area tornado case study, recognition results were used to manually map 193 damaged buildings and their damage severity, providing spatial information for subsequent mission planning.

Original Google Earth map of the Chattanooga tornado study area, with its boundary outlined in red.
The same neighborhood with 193 damaged buildings marked by damage category; map constructed manually from recognition results.
Original Inference
Drag to compare · Use arrow keys when focused.
Original geographic map → Inferred building damage map.
Imagery: Google Earth · © 2022 Maxar Technologies.
Damaged buildings mapped in the case study
193

The damage map was manually constructed from recognition results.

04 — Damage-Informed Mission Planning

Prioritizing Where Drones Inspect Next.

We integrated building damage information into drone task assignment and route optimization, prioritizing more severely damaged buildings while accounting for drone availability, flight speed, and battery constraints.

Low resources

Published low-resource planning result: routes for 2 drones through damaged-building locations, at 2.1 meters per second with 10-minute batteries.
Available drones
2 UAVs
Flight speed
2.1 m/s
Battery life
10 minutes

Moderate resources

Published moderate-resource planning result: routes for 3 drones through damaged-building locations, at 4.6 meters per second with 20-minute batteries.
Available drones
3 UAVs
Flight speed
4.6 m/s
Battery life
20 minutes

High resources

Published high-resource planning result: routes for 4 drones through damaged-building locations, at 6.1 meters per second with 30-minute batteries.
Available drones
4 UAVs
Flight speed
6.1 m/s
Battery life
30 minutes

Damage-informed routes under different drone resource constraints.

Mission planning was evaluated computationally using a real disaster case study, without autonomous field deployment.

Research Contribution

Connecting Damage Assessment with Reconnaissance Planning.

This research demonstrates how AI-based damage recognition can inform drone mission planning, helping direct follow-up inspections toward more severely damaged buildings.

Publications

Journal Article · 2023

Da Hu, Shuai Li, Jing Du, Jiannan Cai. Automating Building Damage Reconnaissance to Optimize Drone Mission Planning for Disaster Response. Journal of Computing in Civil Engineering, 37(3), 2023.

Research Report · 2022

Shuai Li, Amirsalar Moslehy, Da Hu, Mengjun Wang, Nicholas Wierschem, Khalid Alshibli, Baoshan Huang. Drones and Other Technologies To Assist in Disaster Relief Efforts. Tennessee Department of Transportation, RES2021-05, 2022.