| title: Multi-Hazard Warning System | |
| emoji: 🔥 | |
| colorFrom: red | |
| colorTo: blue | |
| sdk: gradio | |
| sdk_version: "5.12.0" | |
| app_file: app.py | |
| pinned: true | |
| license: mit | |
| tags: | |
| - wildfire | |
| - aqi | |
| - multi-task-learning | |
| - deep-learning | |
| - pytorch | |
| - resnet50 | |
| - bilstm | |
| - gradcam | |
| - air-quality | |
| - fire-detection | |
| short_description: "Wildfire risk + AQI forecasting via MTL" | |
| # 🔥 Multi-Hazard Warning System | |
| **Multi-Task Learning for Joint Wildfire Risk Prediction & AQI Forecasting** | |
| A production-grade deep learning system that simultaneously predicts wildfire risk and forecasts Air Quality Index (AQI) for any user-selected region using Multi-Task Learning (MTL). | |
| ## 🏗️ Architecture | |
| - **CNN Branch**: ResNet-50 (4-channel: RGB + NIR) → 2048-dim spatial features | |
| - **LSTM Branch**: 2-layer Bidirectional LSTM → 256-dim temporal features | |
| - **Fusion**: Concatenation → Shared FC layers (2304 → 512 → 256) | |
| - **Task 1**: Wildfire Risk Heatmap (128×128 spatial probability map) | |
| - **Task 2**: 72-hour AQI Forecast (hourly PM2.5 predictions) | |
| ## 🎯 Features | |
| - Interactive Folium map with fire risk heatmap overlay | |
| - Grad-CAM explainability visualization | |
| - 72-hour AQI forecast chart with WHO thresholds | |
| - Location search via geocoding | |
| - Real-time weather data integration | |
| ## 🛠️ Tech Stack | |
| PyTorch · ResNet-50 · BiLSTM · Gradio · FastAPI · Folium · Plotly · OpenCV | |