🛰️ GeoAI DisasterMapper
Automated Building Damage Detection & Climate Vulnerability Analysis
🛰️ GeoAI DisasterMapper
Automated Building Damage Detection & Climate Vulnerability Analysis
⚙️ Configuration
', unsafe_allow_html=True) # Model path model_path = st.text_input( "Model Path", value="model/best_model_resnet50.pth", help="Path to trained U-Net model" ) st.markdown("---") st.markdown('📍 Location
', unsafe_allow_html=True) # Disaster presets disaster_presets = { "Custom" : (0.0, 0.0, None), "Morocco Earthquake 2023" : (31.07, -8.41, "2023-09-08"), "Turkey Earthquake 2023" : (37.17, 37.03, "2023-02-06"), "Nepal Earthquake 2015" : (27.73, 85.33, "2015-04-25"), "Palu Tsunami 2018" : (-0.90, 119.88, "2018-09-28"), "Hurricane Michael 2018" : (30.19, -85.68, "2018-10-10"), } selected_preset = st.selectbox("Disaster Preset", list(disaster_presets.keys())) preset_lat, preset_lon, preset_date = disaster_presets[selected_preset] col1, col2 = st.columns(2) with col1: lat = st.number_input("Latitude", value=preset_lat, format="%.4f") with col2: lon = st.number_input("Longitude", value=preset_lon, format="%.4f") disaster_date = st.text_input( "Disaster Date (YYYY-MM-DD)", value=preset_date if preset_date else "" ) st.markdown("---") st.markdown('🌡️ Climate Settings
', unsafe_allow_html=True) start_year = st.slider("Climate Data Start Year", 2000, 2020, 2010) end_year = st.slider("Climate Data End Year", 2015, 2024, 2023) st.markdown("---") st.markdown("""Upload Satellite Images
', unsafe_allow_html=True) col1, col2 = st.columns(2) with col1: st.markdown("**Before Disaster**") pre_file = st.file_uploader( "Upload pre-disaster image", type=['png', 'jpg', 'jpeg', 'tif', 'tiff'], key="pre" ) if pre_file: pre_image = Image.open(pre_file) st.image(pre_image, caption="Before Disaster", use_column_width=True) with col2: st.markdown("**After Disaster**") post_file = st.file_uploader( "Upload post-disaster image", type=['png', 'jpg', 'jpeg', 'tif', 'tiff'], key="post" ) if post_file: post_image = Image.open(post_file) st.image(post_image, caption="After Disaster", use_column_width=True) st.markdown("---") if pre_file and post_file: if st.button("🔍 RUN DAMAGE DETECTION"): with st.spinner("Loading model and running inference..."): try: # Load model if not os.path.exists(model_path): st.error(f"Model not found at: {model_path}") st.stop() device = 'cuda' if torch.cuda.is_available() else 'cpu' model = load_model(model_path, device) # Run inference pre_image = Image.open(pre_file) post_image = Image.open(post_file) pred = run_inference(model, pre_image, post_image, device) # Get stats stats = get_damage_stats(pred) st.success("Inference complete ✓") # ── Damage metrics ──────────────────────────────────── st.markdown('Damage Statistics
', unsafe_allow_html=True) metric_cols = st.columns(5) damage_colors_hex = { 'background' : '#64748b', 'no-damage' : '#2ecc71', 'minor' : '#f1c40f', 'major' : '#e67e22', 'destroyed' : '#e74c3c' } for i, label in enumerate(LABELS): with metric_cols[i]: pct = stats.get(label, {}).get('percent', 0) st.markdown(f"""Damage Map
', unsafe_allow_html=True) colored_pred = prediction_to_colored_image(pred) fig, axes = plt.subplots(1, 3, figsize=(18, 6)) fig.patch.set_facecolor('#0a0e1a') for ax in axes: ax.set_facecolor('#0a0e1a') ax.axis('off') axes[0].imshow(np.array(pre_image.convert('RGB'))) axes[0].set_title('Before Disaster', color='white', fontsize=12, pad=10) axes[1].imshow(np.array(post_image.convert('RGB'))) axes[1].set_title('After Disaster', color='white', fontsize=12, pad=10) axes[2].imshow(np.array(post_image.convert('RGB'))) axes[2].imshow(colored_pred, alpha=0.6) axes[2].set_title('Damage Prediction Overlay', color='white', fontsize=12, pad=10) # Legend patches = [ mpatches.Patch(color=COLORS[i], label=LABELS[i]) for i in range(len(LABELS)) ] fig.legend( handles=patches, loc='lower center', ncol=5, facecolor='#1a2035', labelcolor='white', fontsize=10, bbox_to_anchor=(0.5, -0.05) ) plt.tight_layout() buf = io.BytesIO() plt.savefig(buf, format='png', dpi=150, bbox_inches='tight', facecolor='#0a0e1a') buf.seek(0) plt.close() st.image(buf, use_column_width=True) # ── GeoJSON export ──────────────────────────────────── st.markdown('Export Results
', unsafe_allow_html=True) geojson_path = export_geojson(pred) if geojson_path: with open(geojson_path, 'r') as f: geojson_data = f.read() st.download_button( label = "⬇️ Download GeoJSON", data = geojson_data, file_name = "damage_map.geojson", mime = "application/json" ) st.markdown("""Climate Vulnerability Analysis
', unsafe_allow_html=True) st.markdown("""Climate Summary
', unsafe_allow_html=True) mcols = st.columns(4) with mcols[0]: st.metric( "Avg Temperature", f"{summary['avg_temp']}°C" ) with mcols[1]: st.metric( "Max Temperature", f"{summary['max_temp']}°C" ) with mcols[2]: st.metric( "Avg Rainfall", f"{summary['avg_rainfall']} mm/day" ) with mcols[3]: warming = summary.get('warming_rate', 0) or 0 st.metric( "Warming Rate", f"{warming:+.3f}°C/year", delta=f"{'↑ Warming' if warming > 0 else '↓ Cooling'}" ) # Pre vs post disaster if disaster_date and 'temp_change' in summary: st.markdown('Pre vs Post Disaster Climate
', unsafe_allow_html=True) pcols = st.columns(3) with pcols[0]: st.metric( "Pre-disaster Avg Temp", f"{summary['pre_disaster_avg_temp']}°C" ) with pcols[1]: st.metric( "Post-disaster Avg Temp", f"{summary['post_disaster_avg_temp']}°C" ) with pcols[2]: change = summary['temp_change'] st.metric( "Temperature Change", f"{change:+.2f}°C", delta=f"{'Warmer' if change > 0 else 'Cooler'} after disaster" ) # ── Climate charts ──────────────────────────────────── st.markdown('Climate Trends
', unsafe_allow_html=True) location_name = selected_preset if selected_preset != "Custom" \ else f"{lat}°N, {lon}°E" chart_buf = plot_climate_trends( df, location_name, disaster_date if disaster_date else None ) st.image(chart_buf, use_column_width=True) # Download CSV csv = df.to_csv(index=False) st.download_button( label = "⬇️ Download Climate Data (CSV)", data = csv, file_name = "climate_data.csv", mime = "text/csv" ) # Vulnerability assessment st.markdown('Vulnerability Assessment
', unsafe_allow_html=True) warming_rate = summary.get('warming_rate') or 0 if warming_rate > 0.02: risk_level = "🔴 HIGH" risk_color = "#e74c3c" risk_msg = f"Region is warming at {warming_rate:.3f}°C/year — significantly above global average." elif warming_rate > 0.01: risk_level = "🟠 MEDIUM" risk_color = "#e67e22" risk_msg = f"Region shows moderate warming trend of {warming_rate:.3f}°C/year." else: risk_level = "🟡 LOW-MEDIUM" risk_color = "#f1c40f" risk_msg = f"Region shows relatively stable temperature trend." st.markdown(f"""Model Architecture
', unsafe_allow_html=True) col1, col2 = st.columns(2) with col1: st.markdown("""Training Curves
', unsafe_allow_html=True) curve_path = "outputs/training_curves.png" cm_path = "outputs/confusion_matrix.png" if os.path.exists(curve_path): st.image(curve_path, caption="Training Loss & Accuracy Curves", use_column_width=True) else: st.info("Place training_curves.png in outputs/ folder to display here.") if os.path.exists(cm_path): st.image(cm_path, caption="Confusion Matrix", use_column_width=True) else: st.info("Place confusion_matrix.png in outputs/ folder to display here.") st.markdown('Project Pipeline
', unsafe_allow_html=True) st.markdown(""" ``` xBD Dataset (USA/Guatemala disasters) ↓ Preprocessing → 256x256 chips with augmentation ↓ U-Net ResNet50 Training (Google Colab T4 GPU) ↓ Inference on unseen test disasters ↓ Apply to target disaster location ↓ GeoJSON damage map export ↓ NASA POWER climate vulnerability overlay ``` """) # ── Footer ──────────────────────────────────────────────────────────────────── st.markdown("---") st.markdown("""