⚡ Deep Learning Powered

Building Damage Assessment from
Satellite Imagery

Leveraging YOLOv8 and EfficientNet-B3 to automatically detect buildings and classify post-disaster damage severity.

Analyze Your Image Read Full Report

About the Project

This research develops and evaluates a two-stage deep learning pipeline that combines object detection for building localisation with classification models for damage severity predictions. The increasing frequency of natural disasters due to climate change necessitates automated systems for rapid damage assessment to support disaster response efforts.

🧠

Deep Learning Pipeline

A sophisticated two-stage architecture combining state-of-the-art object detection with image classification.

🗺️

YOLOv8 Detection

Fast and accurate building localization using YOLOv8s, outperforming Faster R-CNN and FCOS on the xBD dataset.

⚙️

EfficientNet-B3

Transfer learning with progressive fine-tuning achieves 87.8% test accuracy on damage classification.

🛰️

xBD Dataset

Trained and validated on the xBD benchmark, one of the largest satellite imagery datasets for disaster assessment.

Damage Classification Categories

No Damage
0.887
F1 Score
Minor
0.214
F1 Score
Major
0.143
F1 Score
Destroyed
0.723
F1 Score

Pipeline Performance

87.8%
Classification Accuracy
1.28s
Per Image Processing
0.503
End-to-End F1 Score

Full Report

Detailed methodology, experiments, ablation studies, and evaluation metrics.

Open Report PDF

How It Works

Our two-stage pipeline processes satellite imagery through building detection and damage classification for comprehensive damage assessment.

⬆️
01

Upload Image

Upload a post-disaster satellite image in common formats (JPEG, PNG, TIFF).

🧭
02

Building Detection

YOLOv8 identifies and localizes all buildings with bounding boxes.

🧠
03

Damage Classification

EfficientNet-B3 classifies each detected building into a damage severity level.

📊
04

Results

Receive an annotated image with color-coded damage levels and a detailed report.

Receive an annotated image with color-coded damage levels and a detailed report.

Try the Model

Upload a satellite image to see building damage assessment in action. The model will detect buildings and classify damage severity.

⬆️
Upload Satellite Image
Drag and drop your image here, or click to browse
Supports: JPEG, PNG, TIFF (max 50MB)

Or try a sample image:

Sample damage image 1 Sample damage image 2 Sample damage image 3
For best results, use high-resolution post-disaster satellite imagery (0.5m GSD or better).

Analyzing image... This may take a few moments

Analysis Results

Original Image

Original

Detection & Classification Results

Results
No Damage
Minor Damage
Major Damage
Destroyed
0
Buildings Detected
0
No Damage
0
Minor Damage
0
Major Damage
0
Destroyed

Damage Distribution

Individual Buildings