Automated UAS Damage Assessment Using Deep Learning
Primary Investigator (PI) Name
Adeel Khalid
Department
SPCEET – Engineering Technology
Abstract
Following a natural disaster, many homes are left damaged. A rapid assessment of structural damage to homes is extremely important; however, it can take some time to inspect each home and evaluate damage with only volunteers and disaster relief programs. The purpose of this paper is to integrate vision processing into a UAS (Unmanned Aerial System) to assist in the evaluation of damage to homes, to help automate the process of damage assessment so volunteers and damage relief programs can spend more time tending to people affected by natural disasters. Our methodology consists of first collecting data from a city in Georgia following Hurricane Hellen and annotating the data to train a model on damaged production homes. Each home has been given a label during the annotation process: Destroyed, Major, Minor, and Affected. This system helps relief programs properly evaluate where repairs need to be made to prioritize homes that are severely damaged before homes that may have sustained minor damage. We hope this study will help support relief programs in the future to accelerate relief efforts, which will give relief programs more time to tend to those in need who may have been hurt during the disaster.
Disciplines
Architectural Technology | Artificial Intelligence and Robotics | Other Aerospace Engineering | Risk Analysis
Automated UAS Damage Assessment Using Deep Learning
Following a natural disaster, many homes are left damaged. A rapid assessment of structural damage to homes is extremely important; however, it can take some time to inspect each home and evaluate damage with only volunteers and disaster relief programs. The purpose of this paper is to integrate vision processing into a UAS (Unmanned Aerial System) to assist in the evaluation of damage to homes, to help automate the process of damage assessment so volunteers and damage relief programs can spend more time tending to people affected by natural disasters. Our methodology consists of first collecting data from a city in Georgia following Hurricane Hellen and annotating the data to train a model on damaged production homes. Each home has been given a label during the annotation process: Destroyed, Major, Minor, and Affected. This system helps relief programs properly evaluate where repairs need to be made to prioritize homes that are severely damaged before homes that may have sustained minor damage. We hope this study will help support relief programs in the future to accelerate relief efforts, which will give relief programs more time to tend to those in need who may have been hurt during the disaster.