https://github.com/dwade-eng/hermes---ai
I'm currently developing Hermes AI, a modular deep learning framework for post-disaster flood, damage, and debris segmentation — leveraging datasets like FloodNet, RescueNet, and SpaceNet-8. My focus is on blending satellite imagery, drone footage, and multi-head segmentation models to support FEMA critical infrastructure recovery operations.
https://github.com/dwade-eng/hermes---ai
colab-notebook computer-vision deeplabv3 drone-plugin excel gis machine-learning python3 resnet-50 uav-images
Last synced: about 1 month ago
JSON representation
I'm currently developing Hermes AI, a modular deep learning framework for post-disaster flood, damage, and debris segmentation — leveraging datasets like FloodNet, RescueNet, and SpaceNet-8. My focus is on blending satellite imagery, drone footage, and multi-head segmentation models to support FEMA critical infrastructure recovery operations.
- Host: GitHub
- URL: https://github.com/dwade-eng/hermes---ai
- Owner: dWADE-ENG
- Created: 2025-09-21T23:14:42.000Z (9 months ago)
- Default Branch: main
- Last Pushed: 2025-09-21T23:39:59.000Z (9 months ago)
- Last Synced: 2025-09-22T01:13:36.340Z (9 months ago)
- Topics: colab-notebook, computer-vision, deeplabv3, drone-plugin, excel, gis, machine-learning, python3, resnet-50, uav-images
- Language: Jupyter Notebook
- Homepage: https://gems-ai.io/
- Size: 2.91 MB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Hermes---AI
I'm currently developing Hermes AI, a modular deep learning framework for post-disaster flood, damage, and debris segmentation — leveraging datasets like FloodNet, RescueNet, and SpaceNet-8. My focus is on blending satellite imagery, drone footage, and multi-head segmentation models to support FEMA critical infrastructure recovery operations.
# 🛰️ Hermes AI
*Geospatial Damage Intelligence for a New Era of Disaster Response*
**Hermes AI** is a modular deep learning framework built for **UAV-based damage mapping**, **flood segmentation**, and **infrastructure resilience assessment** in post-disaster scenarios. Leveraging high-resolution aerial imagery and satellite data, Hermes integrates multi-head semantic segmentation models to automate damage detection at scale for use by FEMA, public agencies, and disaster management contractors.
Hermes is the flagship project of **GEMS AI**, a Florida-based emergency tech company focused on applying computer vision, AI, and GIS to real-world disaster relief operations.
---
## 🌪️ Why Hermes?
In the critical hours after a hurricane, flood, or earthquake, the ability to rapidly map structural damage and flooded roads can mean the difference between chaos and coordinated response.
Traditional damage assessments are:
- **Slow**: Manual field inspections can take weeks.
- **Subjective**: Paper-based PDA systems vary across jurisdictions.
- **Inaccessible**: Delays in data centralization and sharing.
Hermes solves this by:
- Using UAVs and satellite imagery to **detect damage autonomously**.
- Integrating **multi-class segmentation models** to label water, debris, roofs, and structural failure.
- Fusing with GIS to allow **real-time, map-based visualization** for response teams and agencies.
---
## 🧠 Core Architecture
Hermes AI uses a multi-task, multi-head architecture where each model performs a specific geospatial task:
| Sub-Model | Task | Dataset | Backbone |
|------------------|-------------------------------------------|-------------------|------------------|
| `FloodNetHead` | Flood segmentation | FloodNet, SpaceNet8 | DeepLabV3+ (ResNet-101) |
| `DamageNetHead` | Building damage classification (none/minor/major/destroyed) | xBD, RescueNet | DeepLabV3+ |
| `DebrisNetHead` | Road debris segmentation | Custom UAV labels | UNet |
| `FireNetHead` | Burn scar detection (WIP) | FireNet, MODIS | EfficientNet |
All models are modular and can be run **individually or as an ensemble**, depending on the deployment scenario.
---
## 🧪 Key Features
✅ Multi-task segmentation (flood, damage, debris)
✅ Pre-/post-event change detection (using dual inputs)
✅ UAV + Satellite compatibility (.tif, .jpeg, .png)
✅ Google Colab A100-ready training notebooks
✅ Export to GeoTIFF and ArcGIS integration
✅ Designed for on-site field operation via local inference or cloud-streamed inputs
---