NSF AI-SDM Damage Assessment Workshop

Awareness Training for Using Artificial Intelligence for Building and Road Damage Assessment from Drone Imagery

 Instructors
Tom Manzini, Stephen Johnson, Robin Murphy (Texas A&M)

Intended Participants
Emergency management/public safety professionals, no prior experience or knowledge of AI is required. 

Date/Time
Friday, September 19th, 2025, 1:00pm to 3:00pm

Description

The AI for Building and Road Damage Assessment training covers a combination of knowledge and practical skills that will enable the student to correctly interpret the damage assessment visualizations and tools provided by the Texas A&M CLARKE Damage Assessment Software package. The system was first used by FL-UAS1 during Hurricanes Debby and Helene. The CLARKE system uses computer vision and machine learning to classify visual damage and can process an orthomosaic image of 2,000 houses in roughly 7 minutes on a standard video-gamer laptop. The road damage assessment option also comes with a Google Maps-like route planner that shows the best routes to bypass road damage and the fastest routes through damage. Scenario-based practical training will focus on how to interpret the data products and understand where it has limitations and blind spots.  The overall goal is to enable emergency managers to familiarize themselves with the CLARKE system, how it fits their expected use cases, and to be able to use the output, which would be produced by Texas A&M and FL-UAS1 in the field, who would install and run the system. 

Mastery will be demonstrated via class participation and a passing grade on a written test. 

Course Learning Objectives

Upon completion of this course, students will be able to:

• Compare the benefits and limitations of drone aerial imagery compared to satellite, crewed aviation, and ground personnel for damage assessment

• Show the steps in using the CLARKE system and the role of the UAS teams, the AI team, and the emergency managers consuming the output, including

  • Describe the inputs and outputs of the CLARKE damage system  

  • Name the type of AI being used

• Correctly interpret the output of the automated building damage assessment, including how the damage labels correspond to FEMA standards. 

• Correctly interpret the output of the automated road damage assessment

• Request maps with routes that bypass road damage

• Specify or task how drone teams should fly (e.g., altitude, pattern, resolution) to obtain the imagery that is best suited for CLARKE

• Request and access the CLARKE damage assessment system 

  • Who to work with through FL-UAS1

  • How to integrate into existing GIS workflows, including ARCGIS, SARCOP, and TAK