Datasets:

LemkinAI
/

Multimodal_Atrocity_Identification_Dataset

# Satellite Imagery Dataset - MAID

## Dataset Overview

The satellite imagery component of the **Multimodal Atrocity Identification Dataset (MAID)** contains **19,950 georeferenced satellite images** covering global conflict zones and human rights violation sites. This dataset combines professional validation from international human rights organizations with comprehensive geospatial analysis.

## Dataset Statistics

- **Total Records**: 19,950 georeferenced locations
- **Total Size**: ~38GB
- **Image Types**: Multi-spectral satellite imagery (Panchromatic, RGB, Pansharpened, RGBN)
- **Coverage**: Global conflict zones with professional human rights validation
- **Professional Validation**: Amnesty International, Displacement_Documentation, ConflictZone_Monitor

## Data Structure

```
image_dataset/
├── metadata/
│   ├── unified_mass_atrocity_dataset.jsonl    # 19,950 records with coordinates
│   ├── unified_mass_atrocity_dataset.csv      # CSV format for analysis
│   └── integration_report.json                # Quality metrics and validation
├── high_resolution/                           # 18GB professional imagery
│   └── [1,524 high-resolution satellite images]
└── low_resolution/                            # 20GB additional coverage  
    └── [1,523 low-resolution satellite images]
```

## Data Sources

### Integrated Sources (19,950 total records)
1. **Event_Database**: 15,542 records - Armed conflict events with satellite validation
2. **Satellite_Analysis**: 1,366 records - UN satellite analysis of conflict zones
3. **Damage_Assessment**: 850 records - Building damage assessment imagery
4. **Environmental_Impact**: 671 records - Flood damage analysis with before/after imagery
5. **HR Visual Dataset**: 1,521 records - Professional human rights validation
   - Amnesty International: 189 verified sites
   - Displacement_Documentation: 981 refugee camp and displacement sites  
   - ConflictZone_Monitor: 351 artisanal mining conflict zones

## Image Quality Metrics

### Visual Evidence Quality
- **98.4%** of imagery has <10% cloud cover
- **Multi-spectral coverage**: Panchromatic, RGB, Pansharpened, RGBN bands
- **Professional validation**: Verified by international human rights organizations
- **Average confidence score**: 95%
- **Coordinate accuracy**: 100% coordinate completeness in HR visual dataset

### Geographic Coverage
- **Global scope**: Covers all major conflict zones and human rights violation sites
- **Cross-validation**: 1,521 matches within existing coordinate framework
- **Geospatial correlation**: All coordinates linked to corresponding satellite imagery

## Metadata Schema

Each record in the dataset contains:

```json
{
  "record_id": "unique_identifier",
  "latitude": "decimal_degrees", 
  "longitude": "decimal_degrees",
  "event_type": "conflict_type_classification",
  "severity_score": "0.0_to_1.0_scale",
  "confidence": "validation_confidence_level",
  "source_organization": "validating_organization", 
  "date_acquired": "image_acquisition_date",
  "image_bands": ["available_spectral_bands"],
  "cloud_cover": "percentage_cloud_coverage",
  "ground_sample_distance": "meters_per_pixel"
}
```

## Usage Examples

### Loading Metadata
```python
import pandas as pd
import json

# Load JSONL format
records = []
with open('metadata/unified_mass_atrocity_dataset.jsonl', 'r') as f:
    for line in f:
        records.append(json.loads(line.strip()))

# Or load CSV format
df = pd.read_csv('metadata/unified_mass_atrocity_dataset.csv')
print(f"Total records: {len(df):,}")
```

### Accessing Satellite Imagery
```python
from pathlib import Path
import matplotlib.pyplot as plt
from PIL import Image

# High-resolution imagery path
high_res_path = Path('high_resolution/')
low_res_path = Path('low_resolution/')

# Example: Load specific image
sample_dirs = list(high_res_path.iterdir())[:5]
for img_dir in sample_dirs:
    rgb_image = img_dir / f'{img_dir.name}_rgb.png'
    if rgb_image.exists():
        img = Image.open(rgb_image)
        plt.figure(figsize=(10, 10))
        plt.imshow(img)
        plt.title(f'Sample: {img_dir.name}')
        plt.axis('off')
        plt.show()
```

### Geospatial Analysis
```python
import geopandas as gpd
from shapely.geometry import Point

# Create GeoDataFrame from coordinates
geometry = [Point(record['longitude'], record['latitude']) for record in records]
gdf = gpd.GeoDataFrame(records, geometry=geometry, crs='EPSG:4326')

# Geographic distribution analysis
print("Records by region:")
print(gdf.groupby('source_organization').size())

# Conflict hotspot analysis
conflict_density = gdf.dissolve(by='event_type').geometry.bounds
print("Geographic bounds by conflict type:")
print(conflict_density)
```

## Integration with Text Dataset

This imagery dataset is designed to work seamlessly with the text component of MAID:

```python
# Cross-reference with text dataset
import sys
sys.path.append('../text_dataset')

# Load both datasets
text_records = json.load(open('../text_dataset/comprehensive_mass_atrocity_database_full.json'))
image_records = [json.loads(line) for line in open('metadata/unified_mass_atrocity_dataset.jsonl')]

# Find geographic matches
def find_nearby_incidents(text_lat, text_lon, image_records, threshold_km=10):
    from geopy.distance import geodesic
    matches = []
    for img_record in image_records:
        distance = geodesic((text_lat, text_lon), 
                          (img_record['latitude'], img_record['longitude'])).kilometers
        if distance <= threshold_km:
            matches.append((img_record, distance))
    return sorted(matches, key=lambda x: x[1])

# Example usage
text_incident = text_records['records'][0]
if 'location' in text_incident:
    nearby_imagery = find_nearby_incidents(
        text_incident['location']['latitude'],
        text_incident['location']['longitude'], 
        image_records
    )
    print(f"Found {len(nearby_imagery)} nearby satellite images")
```

## Use Cases

### Machine Learning Applications
- **Damage Assessment**: Training models to detect building destruction and infrastructure damage
- **Conflict Prediction**: Combining satellite imagery with temporal analysis for early warning systems
- **Multi-modal Analysis**: Cross-referencing satellite evidence with textual incident reports
- **Change Detection**: Before/after analysis of conflict zones and humanitarian crises

### Research Applications
- **Human Rights Documentation**: Visual evidence for international legal proceedings
- **Displacement Monitoring**: Tracking refugee movements and camp establishment
- **Environmental Impact**: Assessing ecological damage from conflicts and violations
- **Academic Studies**: Computational analysis of conflict patterns and geographic factors

## Technical Specifications

### Image Formats
- **File Format**: PNG (lossless compression)
- **Bit Depth**: 8-bit and 16-bit depending on source
- **Coordinate System**: WGS84 (EPSG:4326)
- **Naming Convention**: `{source}_{region}_{date}_{band}.png`

### Spectral Bands Available
- **Panchromatic**: High spatial resolution grayscale
- **RGB**: True color composite
- **Pansharpened**: Enhanced resolution color imagery
- **RGBN**: RGB + Near-infrared for vegetation analysis

## Quality Assurance

### Validation Process
1. **Coordinate Verification**: GPS coordinates validated against multiple sources
2. **Professional Review**: Human rights experts verified incident locations
3. **Cross-Reference**: Multiple satellite passes confirm location accuracy
4. **Metadata Completeness**: All records include comprehensive attribution data

### Quality Metrics
- **Spatial Accuracy**: <10m average positional error
- **Temporal Relevance**: Images acquired within 6 months of reported incidents
- **Spectral Quality**: Radiometrically calibrated imagery
- **Coverage Completeness**: 98.4% cloud-free imagery

## Ethical Considerations

### Responsible Use Guidelines
- **Academic Research**: Approved for scholarly analysis and publication
- **Human Rights Advocacy**: Supporting documentation of violations for legal proceedings
- **Policy Development**: Evidence-based humanitarian and conflict resolution policies
- **Technology Development**: Building AI systems for conflict prevention and response

### Prohibited Uses
- **Individual Identification**: No tracking or identification of specific persons
- **Military Targeting**: Not for operational military intelligence or targeting
- **Commercial Surveillance**: No commercial surveillance or monitoring applications
- **Privacy Violation**: Respect for civilian privacy and data protection standards

## Data Provenance

All imagery is sourced from:
- **Open-source satellites**: Publicly available satellite imagery
- **Professional organizations**: Validated by international human rights bodies
- **Academic partnerships**: University research collaborations
- **Government declassified**: Released government satellite analysis

## Citation

If you use this dataset in your research, please cite:

```bibtex
@dataset{maid_imagery_2024,
  title={MAID Satellite Imagery Dataset: Global Conflict Zone Analysis},
  author={Lemkin AI},
  year={2024},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/LemkinAI/Multimoda_Atrocity_Identification_Dataset},
  note={19,950 georeferenced satellite images with professional human rights validation}
}
```

## License

This dataset is released under Creative Commons Attribution 4.0 International (CC BY 4.0) license with the following requirements:
- **Attribution**: Cite dataset creators and contributing organizations
- **Academic Use**: Freely available for research and educational purposes
- **Responsible Use**: Adhere to ethical guidelines for human rights research
- **Non-Commercial**: Professional validation sources require non-commercial use

---

**Dataset Version**: 1.0  
**Last Updated**: November 2024  
**Total Size**: 38GB  
**Validation Status**: Professionally Verified