Zero-Cost AI Governance Pipeline for PM Surya Ghar ☀️

EcoInnovators Ideathon 2026 Submission

📖 Overview

We have built a governance-ready, auditable digital pipeline to remotely verify rooftop solar installations. Unlike traditional solutions that rely on expensive commercial APIs (like Google Maps) or slow physical inspections, our system utilizes ArcGIS Wayback to fetch high-resolution satellite imagery at zero marginal cost, making it scalable for nationwide auditing.

🇮🇳 Relevance to PM Surya Ghar Yojana

The PM Surya Ghar scheme targets 1 crore households, requiring large-scale verification to prevent duplicate claims, false reporting, and post-installation tampering. Our pipeline enables remote, repeatable, and timestamped verification using historical imagery—reducing the need for costly physical inspections by >80% while significantly improving transparency.

🚀 Key Features

• Zero-Cost Data Ingestion: Uses freely accessible ArcGIS Wayback imagery, resulting in zero marginal data acquisition cost per audit (excluding compute).
• High Accuracy: 95.5% mAP50 detection rate using a fine-tuned YOLOv8-Segmentation model.
• Precise Quantification: Uses pixel-level segmentation to calculate the exact $m^2$ area of solar panels for subsidy estimation.
• Audit-Ready: Generates "Audit Overlay" images (Blue Polygons) and a qc_status JSON for human-in-the-loop verification.

📊 Potential Impact

• Scale: Enables verification of millions of rooftops without site visits.
• Cost: Drastically reduces the operational expense of subsidy distribution.
• Time-Travel: Allows retrospective audits (e.g., "Was this panel here last year?") using historical Wayback data.
• Integration: Outputs standard JSON suitable for district, state, and national dashboards.

📂 Repository Structure

• pipeline_code/: The 3-step Python pipeline (Fetch -> Classify -> Report).
• models/: The trained best.pt weights.
• output_examples/: Real results showing detection on Cochin Airport and residential roofs.
• docs/: Detailed Model Card.

🛠️ How to Run

1. Install Dependencies: pip install -r requirements/requirements.txt
2. Phase 1: Fetch Imagery python pipeline_code/1_fetch_wayback.py
3. Phase 2: Analyze python pipeline_code/2_classify_quantify.py
4. Phase 3: Generate Report python pipeline_code/3_generate_report.py

⚠️ Known Limitations

• Imagery Recency: Verification depends on ArcGIS Wayback update cycles (typically every few months).
• Occlusion: Dense rooftop clutter (water tanks, heavy tree cover) may cause occasional false negatives.
• Geographic Resolution: Area estimation accuracy ($m^2$) varies slightly with latitude due to satellite projection.
• Scope: Optimized for residential rooftops; may require fine-tuning for ground-mounted utility farms.

🏛️ Compliance & Data Usage

This system relies exclusively on publicly accessible satellite imagery provided via ArcGIS Wayback for non-commercial research, auditing, and policy evaluation purposes. No proprietary APIs are scraped or bypassed. Final deployment should comply with Esri’s terms of use and applicable government geospatial data policies.

⚖️ License & Citations

• Model: Ultralytics YOLOv8 (AGPL-3.0)
• Data Source: ArcGIS Wayback Imagery (Esri, Maxar, Earthstar Geographics)