rlogh/lanternfly_swatter_training

Spotted Lanternfly Classification Dataset

Dataset Description

This dataset contains images for binary classification of spotted lanternflies (Lycorma delicatula), an invasive species causing significant damage to agriculture and ecosystems in the United States. The dataset is designed to train machine learning models to identify dead or squashed lanternflies from photographs, supporting community-driven environmental monitoring and pest management efforts.

Dataset Summary

The Spotted Lanternfly Classification Dataset consists of three distinct splits:

• Original: 166 high-quality photographs of verified spotted lanternflies
• Augmented: 1,660 images generated through data augmentation techniques
• Negatives: 1,826 images of non-lanternfly insects and objects

The dataset was created to support the development of AI-powered identification systems for citizen science applications, enabling accurate and automated detection of this invasive species from user-submitted photographs.

Supported Tasks and Leaderboards

• Image Classification: Binary classification (Lantern Fly vs Non-Lantern Fly)
• Object Detection: Identification of spotted lanternflies in images
• Environmental Monitoring: Community-driven pest tracking and management

Dataset Structure

Data Instances

Each data instance contains:

• image: High-resolution photograph (minimum 224x224 pixels, typically 512x512 or higher)
• filename: Unique identifier for the image file

Data Fields

• image: Image data in various formats (JPG, PNG)
• filename: String identifier for the image file

Data Splits

Split	Examples	Size (MB)	Description
original	166	28.6	Original verified spotted lanternfly photographs
augmented	1,660	742.4	Augmented versions of original images
negatives	1,826	469.4	Non-lanternfly images for negative examples

Dataset Creation

Curation Rationale

This dataset was created to address the critical need for automated identification of spotted lanternflies, an invasive species that threatens agriculture and natural ecosystems. The dataset supports:

1. Community Science: Enabling citizen scientists to contribute to invasive species monitoring
2. Research Applications: Providing high-quality data for ecological and entomological research
3. Pest Management: Supporting early detection and rapid response efforts
4. Educational Purposes: Training materials for species identification

Source Data

• Original Images: Collected through systematic field photography of dead/squashed spotted lanternflies
• Augmented Images: Generated using rotation, flipping, brightness/contrast adjustments, and random cropping
• Negative Examples: Curated from diverse sources including the Insecta dataset and other insect collections

Data Collection Process

1. Field Collection: Systematic photography of spotted lanternflies in various states of damage
2. Quality Control: Manual verification of species identification and image quality
3. Data Augmentation: Application of 4x augmentation factor to balance class distribution
4. Negative Sampling: Careful curation of non-lanternfly examples to ensure diversity

Personal and Sensitive Information

No personal or sensitive information is included in this dataset. All images contain only biological specimens and environmental contexts.

Considerations for Using the Data

Social Impact of Dataset

This dataset supports environmental conservation efforts by enabling:

• Early Detection: Rapid identification of new infestation areas
• Public Engagement: Making species identification accessible to non-experts
• Research Support: Providing high-quality data for scientific analysis
• Policy Development: Informing management strategies and resource allocation

Discussion of Biases

Potential biases in the dataset include:

• Geographic Bias: Images may be concentrated in specific regions
• Temporal Bias: Collection may favor certain seasons or times
• Condition Bias: Focus on dead/squashed specimens may not represent live insects
• Photographer Bias: Different collection methods and equipment

Other Known Limitations

• Limited representation of nymph stages
• Potential overrepresentation of certain damage states
• Variable lighting and background conditions
• Limited geographic diversity in some splits

Additional Information

Dataset Curators

• Primary Curators: Lanternfly Tracker Development Team
• Institution: Carnegie Mellon University (Hackathon 2024)
• Contact: Available through Hugging Face dataset page

Licensing Information

This dataset is released under the Creative Commons Attribution 4.0 International License (CC BY 4.0).

Citation Information

@dataset{spotted_lanternfly_classification_2024,
  title={Spotted Lanternfly Classification Dataset: Binary classification dataset for invasive species identification},
  author={Lanternfly Tracker Team},
  year={2024},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/rlogh/lanternfly_swatter_training},
  license={CC BY 4.0}
}

Contributions

We welcome contributions to improve this dataset. Please contact the curators for information about contributing additional images or improving existing annotations.

Acknowledgments

• Carnegie Mellon University for hosting the hackathon
• The citizen science community for data collection support
• Entomological experts for species verification
• The open-source community for data augmentation tools

---

This dataset supports the fight against invasive species through technology and community engagement.