The Damage Assessment (Drone Imagery) pretrained model available on ArcGIS Living Atlas of the World. This deep learning model is used to perform damage assessment on drone and aerial imagery.
The model is fine-tuned on the LADI v2 dataset, which contains 10,000 aerial images labeled by volunteers from the Civil Air Patrol. The Low Altitude Disaster Imagery (LADI) dataset was created to address the relative lack of annotated post-disaster aerial imagery in the computer vision community. Low altitude post-disaster aerial imagery from small planes and UAVs can provide high-resolution imagery to emergency management agencies to help them prioritize response efforts and perform damage assessments. To accelerate their workflow, computer vision can be used to automatically identify images that contain features of interest, including infrastructure such as buildings and roads, damage to such infrastructure, and hazards such as floods or debris.
For LADI v2, the authors used CAP volunteers who were trained in the FEMA damage assessment process and collected damage labels using the defined FEMA Preliminary Damage Assessment scale: unaffected, affected, minor, major, destroyed. These damage levels have specific criteria, helping reduce the subjectivity of identifying whether a structure is damaged. This model is a pretrained classifier trained on the LADI v2 dataset and serves as a basis for fine-tuning and potential deployments. The model performs multi-label classification and categorizes images as belonging to one or more of the following classes:
bridges_any
buildings_any
buildings_affected_or_greater
buildings_minor_or_greater
debris_any
flooding_any
flooding_structures
roads_any
roads_damage
trees_any
trees_damage
water_any
License requirements
To complete this workflow, the following are the license requirements:
- ArcGIS Desktop—ArcGIS Image Analyst extension for ArcGIS Pro
- ArcGIS Enterprise—ArcGIS Image Server
- ArcGIS Online—ArcGIS Pro or Professional Plus user type
Model overview
This model has the following characteristics:
- Input—High resolution individual drone images or an orthomosaic.
- Output—Feature class containing classified disaster.
- Compute—This workflow is compute intensive and a GPU with compute capability of 6.0 or higher is recommended.
- Applicable geographies—This model is expected to work well in all regions globally. However, results can vary for images that are statistically dissimilar from training data.
- Architecture—This model uses the google/bit-50 model architecture for classification.
- Accuracy metrics—This model has metrics for each class, as mentioned at the Arxiv site.
Access and download the model
Download the Damage Assessment (Drone Imagery) pretrained model from ArcGIS Living Atlas of the World. Alternatively, access the model directly from ArcGIS Pro, or consume it in ArcGIS Image for ArcGIS Online.
- Browse to ArcGIS Living Atlas of the World.
- Sign in with your ArcGIS Online credentials.
- Search for Damage Assessment (Drone Imagery) and open the item page from the search results.
- Click the Download button to download the model.
You can use the downloaded .dlpk file directly in ArcGIS Pro.
Release notes
The following are the release notes:
| Date | Description |
|---|---|
| October 2024 | First release of Damage Assessment (Drone Imagery) |