How cznmr/sih-162 works
ENTRO-26162 is a geospatial thermal-event intelligence prototype for Telangana. Its development workflow combines historical FIRMS detections with persistence, spatial behavior, OSM infrastructure, land-cover, and forest context; builds features and conservative seed labels; then trains a Random Forest. At runtime, the Streamlit app fetches current FIRMS detections, classifies candidates with confidence, and presents them on an interactive map. The README documents these capabilities; some implementation wiring is therefore represented from that documentation rather than sampled source.
Main entry points: build_feature_table.py Master feature table · historical_loader.py Historical detections · build_ml_dataset.py ML dataset · seed_labeling_v2.py Seed labels
Components 21
Data sources
- Historical detections
src/historical_loader.py
Spatial enrichment
- Persistence features
src/persistence_analysis.py - Spatial behavior
src/spatial_fire_behavior.py - OSM context
src/feature_engineering.py - Land-cover context
src/merge_dynamic_world.py - Forest context
src/forest_location_evidence.py
Feature and model development
- Master feature table
src/build_feature_table.py - ML dataset
src/build_ml_dataset.py - Seed labels
src/seed_labeling_v2.py - Random Forest training
src/train_classifier_v2.py - Classifier bundle
Live detection
- Live FIRMS fetch
app.py - Candidate classification
app.py - Interactive dashboard
app.py
Analysis and review
- District assignment
src/unknown_by_district.py - Evidence analysis
src/district_evidence_analysis.py
Other components
- Analyst
- NASA FIRMS
- OpenStreetMap
- Dynamic World
- Forest boundaries
Connections 24
- NASA FIRMS → Historical detections: supplies history
- NASA FIRMS → Live FIRMS fetch: provides detections
- Historical detections → Persistence features: feeds detections
- Historical detections → Spatial behavior: feeds detections
- OpenStreetMap → OSM context: provides infrastructure
- Dynamic World → Land-cover context: provides land cover
- Forest boundaries → Forest context: provides boundaries
- Persistence features → Master feature table: adds features
- Spatial behavior → Master feature table: adds features
- OSM context → Master feature table: adds context
- Land-cover context → Master feature table: adds context
- Forest context → Master feature table: adds context
- Master feature table → ML dataset: contributes features
- ML dataset → Seed labels: provides evidence
- ML dataset → Random Forest training: provides training data
- Seed labels → Random Forest training: provides labels
- Random Forest training → Classifier bundle: produces model
- Live FIRMS fetch → Candidate classification: passes detections
- Candidate classification → Classifier bundle: loads model
- Candidate classification → Interactive dashboard: displays predictions
- Analyst → Interactive dashboard: views results
- Master feature table → District assignment: provides locations
- Seed labels → District assignment: provides labels
- District assignment → Evidence analysis: summarizes districts