This deep learning model is designed for damage assessment using drone and aerial imagery. It has been fine-tuned on the LADI v2 dataset, which includes 10,000 annotated aerial images collected by Civil Air Patrol (CAP) volunteers. The Low Altitude Disaster Imagery (LADI) dataset addresses the lack of labeled post-disaster aerial data for computer vision applications. Captured from UAVs and small aircraft, this low-altitude imagery provides high-resolution views crucial for emergency response and prioritization.

The model helps automate the identification of key post-disaster features in aerial images—such as buildings, roads, flooding, and debris—that are essential for emergency management teams. Trained on damage annotations aligned with FEMA's standardized Preliminary Damage Assessment (PDA) scale (unaffected, affected, minor, major, destroyed), it ensures consistency and reduces subjectivity in damage classification.

This is a multi-label classifier that can assign one or more of the following labels to each image:

• 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

The model is pretrained and can be further fine-tuned or directly deployed to support disaster response and damage analysis workflows.