How ranjitha1065/a-system-and-method-for-ai-based-automatic-road-network-extraction-from-satellite-imagery- works
RoadAI is a road-segmentation product: users submit satellite imagery for U-Net predictions, inspect and export masks and overlays, or compare pre- and post-disaster road networks. Its backend serves inference, while a separate offline workflow prepares paired datasets, trains candidate checkpoints, and evaluates model quality. The frontend and inference wiring is visible in sampled source; training and deployment capabilities are represented only where the README documents them, so their internal call paths are not expanded.
Main entry points: app.py Flask API · train.py Candidate training · PredictionCard.jsx Image prediction · evaluate.py Model evaluation
Components 19
User interface
- App navigation
src/App.jsx - Home workflow
src/pages/Home.jsx - Image prediction
src/components/PredictionCard.jsx - Result comparison
src/components/ComparisonView.jsx - Disaster comparison
src/components/DisasterAnalysisCard.jsx
Inference service
- Flask API
backend/app.py - Mask cleanup
utils.py
Model lifecycle
- Dataset pairing
preprocess.py - Candidate training
train.py - Model evaluation
evaluate.py - Model configuration
config.py
Models and files
- U-Net inference
models/best_model.keras - Upload files
backend/uploads - Inference outputs
backend/outputs - Paired dataset
dataset.py - Production checkpoint
models/best_model.keras - Candidate checkpoint
models/best_model_improved.keras
Other components
- Analyst
- Road imagery dataset
Connections 21
- Analyst → App navigation: navigates
- App navigation → Home workflow: shows
- Home workflow → Image prediction: opens prediction
- Image prediction → Flask API: POST /predict
- Flask API → U-Net inference: predicts with
- Flask API → Mask cleanup: cleans mask
- Flask API → Upload files: stores uploads
- Flask API → Inference outputs: writes visualizations
- Flask API → Production checkpoint: loads by default
- Analyst → Disaster comparison: submits image pair
- Disaster comparison → Flask API: POST /disaster-analysis
- Image prediction → Result comparison: shows results
- Road imagery dataset → Dataset pairing: supplies imagery
- Dataset pairing → Paired dataset: pairs splits
- Paired dataset → Candidate training: provides training data
- Candidate training → Candidate checkpoint: saves candidate
- Candidate training → Model evaluation: validation selects threshold
- Model evaluation → Production checkpoint: evaluates checkpoint
- Model evaluation → Candidate checkpoint: compares checkpoint
- Model configuration → Candidate training: configures run
- Model configuration → Flask API: selects checkpoint