Geo-Darshan 🌍

A web-based geospatial visualization tool for analyzing cluster segmentation data and creating land cover classifications.

Features

Core Capabilities

• Cluster Animation: Navigate through different segmentation levels (k-values)
• Land Cover Classification: Label clusters with hierarchical land use categories
• Object Detection: LLM-powered detection of buildings, roads, vegetation, and other features from satellite imagery
• Visualization: Dynamic overlay rendering with adjustable opacity
• Export: Generate GeoTIFF and mapping files for analysis

Object Detection

• Workflow Modes: Manual (upload image) or Automatic (capture from map)
• Models: Gemini 2.5 Flash/Pro (requires API key)
• Output: GeoJSON features with confidence scores and geometry
• Region Selection: Draw rectangular bounds on map for analysis

Map Features

• Multiple base layers (Satellite, Street Map)
• Cluster-specific color mapping
• Labeled regions composite layer
• Map controls

Data Management

• Manifest-based data loading from folders
• Support for multi-temporal GeoTIFF datasets
• Color legend generation
• Local storage for label persistence

Tech Stack

• Frontend: Vanilla JavaScript + Svelte components
• Mapping: Leaflet with GeoRaster extensions
• Data Processing: TensorFlow.js for composite generation
• Geospatial: GeoTIFF library for raster handling
• Build: Vite with ES modules

Project Structure

geo-darshan/
├── app/
│   ├── components/          # Svelte UI components
│   ├── js/                  # Core JavaScript modules
│   ├── index.html          # Cluster animation viewer
│   ├── land-cover-viewer.html  # Land cover display
│   └── stores.js           # Svelte reactive stores
├── scripts/                # Data processing utilities
└── README.md

Getting Started

Prerequisites

• Node.js 18+
• Modern web browser with WebGL support

Installation

# Clone repository
git clone <repository-url>
cd geo-darshan

# Install dependencies
bun install

# Start development server
bun run dev

Data Format

The application expects a folder containing:

dataset/
├── manifest.json           # Dataset metadata
├── color_legend.json      # Cluster color mappings
├── k5_segmentation.tif    # Segmentation rasters
├── k10_segmentation.tif
└── ...

Manifest Structure

{
  "segmentation_keys": ["k5", "k10", "k15"],
  "files": ["k5_segmentation.tif", "k10_segmentation.tif", "k15_segmentation.tif"],
  "metadata": {
    "bounds": [minx, miny, maxx, maxy],
    "shape": [height, width]
  }
}

Usage

Basic Workflow

1. Load Data: Use "Select Data Folder" to load your segmentation dataset
2. Navigate: Use animation controls to explore k-values
3. Label Clusters: Click clusters to assign land use classifications
4. Visualize: Toggle labeled regions layer to see composite results
5. Export: Generate GeoTIFF and mapping files

Land Use Hierarchy

The tool supports 4-level hierarchical classification:

1. Broad Categories (e.g., Natural, Agricultural)
2. General Types (e.g., Forest, Cropland)
3. Specific Uses (e.g., Deciduous Forest, Rice Paddies)
4. Detailed Classification (e.g., Mature Oak Forest)

License

Apache License 2.0