Artificial Intelligence R&D (POC)
Computer Vision, multi-sensory threshold engines, and frame parsing mechanics.
AI-SABUD: Satellite Inundation & Built-environment Damage Assessment System for Odisha Resilience
Architected and engineered a cloud-native, deep learning-driven disaster response system designed to secure government grants and optimize relief workflows following severe cyclonic events along the Odisha coastline. Deployed an all-weather satellite processing pipeline utilizing Google Earth Engine to ingest multi-temporal Sentinel-1 Synthetic Aperture Radar (SAR) imagery. By completely bypassing cloud cover limitations, the system evaluates pre- and post-disaster dual-polarization (VV/VH) backscatter anomalies, applying a customized neural network to map heavy coastal inundation, track storm surges, and isolate structural destruction within minutes of satellite pass completion.
Metrics at a Glance:
Key Architectural Achievements:
- All-Weather SAR Telemetry Ingestion: Built an automated data streaming connector using the Google Earth Engine Python API to ingest C-band Sentinel-1 radar imagery, guaranteeing disaster monitoring capability during extreme storms when traditional optical satellites are blinded by dense cyclonic clouds.
- Dynamic Dimension Normalization Pipeline: Engineered a robust bilinear and nearest-neighbor interpolation layer inside the PyTorch Dataset using
torch.nn.functional.interpolate, correcting spatial sizing mismatches from satellite crops to ensure steady, crash-free batch processing. - Imbalanced Damage Variant Optimization: Mitigated severe class imbalances (where unaffected terrain represents >90% of imagery and destroyed infrastructure <2%) by implementing a penalty-weighted Cross-Entropy Loss matrix, forcing the neural network to prioritize rare structural anomalies.
- Geospatial Vectorization Mapping Engine: Programmed a standalone vector export script utilizing SciPy image labeling and GeoPandas to cluster contiguous pixel predictions and output standard WGS84 Geodetic Shapefiles (
.shp) ready for direct integration into government GIS pipelines (OSDI/SATARK). - Multi-District Dashboard & Regulatory Feed Integration: Developed an interactive Streamlit frontend utilizing Folium mapping to dynamically center views over Balasore, Bhadrak, Kendrapara, and Puri, while building an asynchronous XML parser to stream live Common Alerting Protocol (CAP) cyclone notices from the India Meteorological Department (IMD) via the SACHET/NDMA gateway.