Logo Lanfrica

exosidus/AquaGuard_Ethiopia

Domaine:

climategeospatial

Type de record:

softwareproject
Créateur:
exo
Hôte:
AquaGuard Ethiopia - AI-powered flood detection platform for the Awash River Basin using SegFormer semantic segmentation # AquaGuard Ethiopia — AI-Powered Flood Detection Platform AquaGuard Ethiopia is a state-of-the-art Sentinel-1 SAR satellite monitoring platform specifically tuned for early detection, mapping, and analysis of seasonal flood events in the **Awash River Basin, Ethiopia**. The platform combines the **SegFormer** transformer-based semantic segmentation model with a **multi-cue heuristic analysis pipeline** to accurately map both clear and muddy/brown floodwaters under all-weather and lighting conditions. --- ## 📸 Platform Overview ### 1. Home Page / Earth Observation Platform A clean, premium landing page introducing the platform, detection capabilities, and the 4-step processing pipeline. ### 2. Analysis Dashboard An intuitive dashboard allowing researchers and disaster management teams to upload raw Sentinel-1 SAR imagery and configure detection parameters. ### 3. Detailed Results & Visualizations Displays pixel-level flood mapping, calculated area stats in square kilometers, detection confidence, and interactive location views. Includes Side-by-Side and overlay comparison modes. ### 4. Interactive Awash River Basin Map An active standalone informational page outlining the geography of the Awash River Basin, key stats (110k km² area, 1,200 km length, 10M+ affected population), and the underlying technology stack. --- ## 🛠️ Technology Stack & Architecture - **Front-End UI**: Responsive HTML5/Vanilla CSS with immersive visual styling (glassmorphism tabs, dark layout themes) and interactive GIS mapping leveraging **Leaflet.js**. - **Application Server**: Python-based **Flask** server serving static results, uploaded imagery, template rendering, and REST endpoints. - **Deep Learning Model**: Nvidia's **SegFormer-B0** pre-trained on ADE20K semantic segmentation dataset, fine-tuned for land/water features. - **Computer Vision Pipeline**: **OpenCV (cv2)**, **NumPy**, and **Pillow** processing pipelines featuring: - Multi-spectral and channel-ratio thresholdin …