AI-powered FRA Atlas and Decision Support System

Project Vision

This project aims to revolutionize the implementation and monitoring of the Forest Rights Act (FRA), 2006, by developing an AI-powered FRA Atlas and a WebGIS-based Decision Support System (DSS). The system will digitize legacy records, visualize FRA claims and granted titles, integrate satellite-based asset mapping, and provide actionable insights for targeted development through Central Sector Schemes (CSS).

Problem Statement

The Forest Rights Act (FRA), 2006, faces significant challenges in its implementation:

• Scattered and Non-Digitized Records: Legacy records (IFR, CR, CFR) are fragmented, non-digitized, and difficult to verify.
• Lack of Centralized Repository: There is no real-time visual repository (FRA Atlas) of FRA claims and granted titles.
• Missing Data Integration: Integration of satellite-based asset mapping with FRA data, and legacy data with the FRA Atlas, is absent.
• Absence of Decision Support: Decision-makers lack a DSS to layer CSS benefits (e.g., PM-KISAN, Jal Jeevan Mission, MGNREGA, DAJGUA) for FRA patta holders.

Objectives

1. Digitize and Standardize Legacy Data: Digitize and integrate legacy FRA claims, verifications, and pattas with the FRA Atlas, including FRA patta holders’ shapefiles.
2. Create an FRA Atlas: Develop an FRA Atlas showing potential and granted FRA areas using AI and satellite data.
3. Integrate WebGIS Portal: Implement a WebGIS portal for visualizing and managing spatial and socio-economic data.
4. AI-based Asset Mapping: Utilize Remote Sensing and AI/ML to map capital and social assets (ponds, farms, forest resources) in FRA-holding villages.
5. Build a Decision Support System (DSS): Create a DSS to recommend and layer CSS schemes based on mapped data, facilitating targeted development.

High-Level Architecture

graph TD
    subgraph Data Sources
        A[Scanned FRA Documents]
        B[Satellite Imagery]
        C[Geospatial Shapefiles]
        D[External Data Layers]
    end

    subgraph Backend Processing
        E[Data Ingestion Pipeline]
        F[OCR/NER Service]
        G[Computer Vision Service - U-NET]
        H[PostgreSQL + PostGIS Database]
        I[API & Decision Support System]
    end

    subgraph Frontend
        J[WebGIS Portal]
    end

    A --> E
    B --> E
    C --> E
    D --> E
    E --> F
    E --> G
    F --> H
    G --> H
    H --> I
    I --> J

Loading

Technology Stack

• Programming Language: Python 3.9+
• Web Framework: FastAPI (for high-performance APIs) or Django
• Database: PostgreSQL with PostGIS extension (Primary/Geospatial), MongoDB or Elasticsearch (Optional, for document store)
• AI/ML Libraries:
  • OCR/NER: PyTesseract, spaCy, Hugging Face Transformers
  • Geospatial & CV: GeoPandas, Rasterio, GDAL, Scikit-learn, PyTorch/TensorFlow
• Task Queues: Celery with RabbitMQ or Redis
• Containerization & Deployment: Docker, Docker Compose, Kubernetes, CI/CD pipeline (GitHub Actions)
• Cloud Services (Recommended): AWS, GCP, or Azure

Modules

Data Ingestion & Processing

This module handles the intake and preparation of various data sources:

• Geospatial Data: Scripts like import_state_boundaries.py, import_district_boundaries.py, and import_village_boundaries.py are used to import foundational geospatial data (e.g., shapefiles for state, district, and village boundaries) into the PostGIS database.
• External Data Layers: Integration of additional data such as forest data, groundwater data, and infrastructure data from sources like PM Gati Shakti portals.
• Document Ingestion: A pipeline to watch storage buckets for new scanned documents, triggering the digitization process via asynchronous tasks.

AI/ML Models

This section details the core artificial intelligence and machine learning components:

• OCR/NER Pipeline:

  • OCR Service: Converts scanned document images into raw text.
  • NER Model: Parses raw text to extract key entities like village_name, patta_holder_name, coordinates, claim_status, and area_in_hectares.
  • This pipeline is containerized and integrated as an asynchronous Celery task.

• Computer Vision for Asset Mapping:

  • Model Architecture: A U-Net model (defined in cv_models/model.py) is employed for semantic segmentation of satellite imagery.
  • Asset Identification: Trained to identify and segment agricultural land, water bodies (ponds, streams), homesteads, and forest cover.
  • Post-processing: A pipeline (implemented in cv_models/inference.py) converts raster output (pixel masks) into vector polygons (GeoJSON/WKT) and calculates their area.
  • The CV pipeline is containerized using cv_models/Dockerfile and executed as an asynchronous Celery task (defined in cv_models/celery_tasks.py).

API & Decision Support System

This module provides the interface for the frontend and the intelligence for policy recommendations:

• RESTful APIs: Developed using FastAPI/Django to serve data to the frontend, including digitized FRA claims, patta holder details, geospatial queries (e.g., assets within a bounding box), and summaries for villages.
• Decision Support System (DSS) Engine:
  • Rule Engine: Codifies eligibility rules for Central Sector Schemes (CSS) such as PM-KISAN, Jal Jeevan Mission, MGNREGA, and DAJGUA.
  • DSS Logic: Cross-references FRA holder data (land type, assets, water index) against CSS rules to identify eligible schemes.
  • AI Enhancements: Explores clustering models (e.g., K-Means) to group villages based on asset profiles for targeted policy-making.
  • DSS API: Provides an endpoint (/api/dss/recommendations) to output prioritized lists of recommended schemes with justifications.

Getting Started

Detailed instructions for setting up the development environment and running the application will be provided in a separate INSTALL.md or CONTRIBUTING.md file. However, a general overview includes:

1. Clone the Repository: git clone [repository-url]
2. Set up Docker: Ensure Docker and Docker Compose are installed.
3. Environment Configuration: Configure environment variables for database connections and API keys.
4. Build and Run Services: Use Docker Compose to build and run the backend services, including PostgreSQL/PostGIS, Celery, and the API server.
5. Data Ingestion: Run the data ingestion scripts to populate the database with initial geospatial and FRA data.

Project Roadmap

The project is structured into five phases:

• Phase 0: Foundation & Architecture: Setting up the development environment, defining architecture, database schema, and initial API specifications.
• Phase 1: Data Ingestion & Digitization Core: Building data ingestion pipelines, developing OCR/NER services, and creating initial read-only APIs.
• Phase 2: Geospatial Core & AI-Powered Asset Mapping: Importing foundational geospatial data, integrating external data layers, developing a tile server, and building Computer Vision models for asset mapping.
• Phase 3: The Decision Support System (DSS): Optimizing database queries, exploring AI enhancements for DSS, and building the rule-based DSS engine and API.
• Phase 4: Integration, Testing, and Deployment: Conducting end-to-end testing, finalizing deployment architecture, setting up monitoring, and optimizing ML models.

Target Users

• Ministry of Tribal Affairs
• District-level Tribal Welfare Departments & Line Departments of DAJGUA
• Forest and Revenue Departments
• Planning & Development Authorities
• NGOs working with tribal communities

Future Scope

• Incorporate real-time satellite feeds for monitoring Community Forest Resource (CFR) forests.
• Integrate IoT sensors for soil health, water quality, etc., in FRA lands.
• Enable mobile-based feedback and updates from patta holders.