Code and resources from Team Kyogre's winning participation in Smart India Hackathon, 2022 - Software Edition.
Project 1 - SatVision: AI-Driven Detection of Non-Residential Built-Up Clusters from Satellite Images
Link to Project Documentation and Related Resources.
The identification of non-residential built-up areas and detecting clusters of such regions can greatly aid strategic industrial expansions, developmental planning, and understanding the earth’s topography in general.
- Landcover Visualization Frontend.
- Landcover Visualization Backend.
- Real-time Satellite Image Acquisition.
- Machine Learning Experiments for Landcover Type Detection.
- Application Pipelineing Utilities.
The proposed solution aims to achieve detection of non-residential built-up areas clusters through AI-driven analysis of Medium Resolution Satellite Imagery. In a systematic approach, the solution aims to,
- Prepare annotated datasets for using standard satellite imagery collected by NASA, ESA and ISRO.
- Utilize existing annotated datasets to support model building.
- Deep CNNs to detect, segment and identify clusters of non-residential built-up regions.
- Evaluate its performance on highly-populated regions like Mumbai, Kolkata, Bangalore, and Delhi that pose a complex topography to detect patterns from.
- Leverage geographic metadata and topological constraints to refine the solution formulation at multiple steps, including annotated dataset preparation and post-processing the Deep CNN results.
- As a user endpoint, we build a GUI tool that visualizes landcover segmentation overlays on geographic maps generated through a novel
patchify-process-reconstruct
pipeline.
Project 2 - EyeSea: Real-time Automatic Marine Species Threat Alerting at Shoreline via Underwater Surveillance
Link to Project Presentation Video.
Link to Project Documentation and Related Resources.
- Image Preprocessing Algorithms for Enhancement and Color Restoration.
- Video Frame Processing Utilities.
- Deep Learning for Underwater Species Classification.
- High-definition underwater cameras are deployed using buoys, moored in the sea at an optimal distance from the coast.
- To detect, localize, and identify lethal marine species, a pattern recognition algorithm analyzes real-time optical image feed from high-definition underwater cameras.
- On detecting a potentially lethal threat approaching the coast, the system sounds a public alarm as well as a personalized alert on swimmers' wearable devices to warn them of a possible threat.