Bee Wing Vein Classifier

ISEF Category: Animal Sciences

Subcategory: Systematics and Evolution  ·  Difficulty: Intermediate  ·  Setup: School Lab  ·  Time: 1 to 2 Months

The Hook

Two bee species can look almost identical at first glance, yet their wing veins can still give them away. That makes wings a little like fingerprints. If you can train a model to read those patterns from phone photos, you can turn a hard ID problem into a clean test.

What Is It?

This project asks whether wing shape can help identify bee species that look almost the same. Morphometrics means measuring shape, not just guessing by eye. In this case, you would measure parts of the wing, such as vein angles, landmark positions, or outline shape, and use those numbers to train a machine learning model.

Think of it like sorting keys that all look similar except for tiny teeth patterns. A person can miss those details, but a classifier can compare many small measurements at once. Smartphone photos make the project practical, since you do not need a full research lab to collect images and test whether the pattern is strong enough to separate species.

Why This Is a Good Topic

This is a strong science fair topic because it has clear inputs, clear outputs, and real measurement. You can test whether shape features actually separate species, not just whether a model can memorize images. The topic connects to pollinator ID, biodiversity, and species monitoring, and it teaches image analysis, data cleaning, and model evaluation in a way you can handle as a student.

Research Questions

• How does image resolution affect species classification accuracy?
• What is the effect of lighting consistency on misclassification rate?
• Does landmark-based morphometrics outperform raw pixel features on a small bee dataset?
• To what extent does a model trained on one collection site generalize to a new site?
• Which feature set separates cryptic bee species best, vein angles, landmark ratios, or wing outline shape?

Basic Materials

• Smartphone with a macro lens clip-on or close-focus camera setting.
• Stable light source, such as a desk lamp with diffused white light.
• Labeled bee wing images from a museum collection, field guide, or open database.
• Laptop or desktop computer for sorting images and recording labels.
• Spreadsheet software and a simple image editor for checking focus and landmarks.
• Printed scale bar or ruler for image calibration.

Advanced Materials

• Dissecting microscope with a camera adapter for standardized wing imaging.
• Reference bee specimens with confirmed species IDs.
• Calibration micrometer slide or scale bar for pixel-to-millimeter conversion.
• Computer with Python, scikit-learn, and ImageJ for feature extraction and classification.
• R with the geomorph package for geometric morphometric analysis.

Software & Tools

• ImageJ: Marks landmarks, measures vein angles, and checks image quality.
• Python: Organizes the dataset and runs classification models.
• scikit-learn: Builds classifiers, cross-validation splits, and confusion matrices.
• Google Colab: Gives you a free notebook environment for Python if your laptop is slow.
• R with geomorph: Tests geometric shape differences with standard morphometrics methods.

Experiment Steps

1. Define the bee species pair or species complex you will compare, and confirm that every image has a trustworthy label.
2. Choose one measurement path first, such as landmark-based shape features, vein angles, or a simple image classifier.
3. Design a split that keeps images from the same specimen or collection batch out of both training and test sets.
4. Build a baseline model, then compare it with a second model so you can measure whether added shape detail helps.
5. Plan how you will report accuracy, false matches, and which species pairs stay hard to separate.

Common Pitfalls

• Photographing wings at mixed angles, which warps vein geometry and confuses shape measurements.
• Letting the same specimen appear in both training and test sets, which inflates accuracy.
• Using images from only one phone or one light setup, which makes the model learn the setup instead of the species.
• Keeping class sizes uneven, which can push the classifier toward the most common bee species.
• Measuring damaged, folded, or blurry wings, which adds noisy landmarks and weakens separation.

What Makes This Competitive

A strong version of this project goes beyond a single accuracy score. You can test whether the classifier still works across new phones, new lighting, or a second collection site. You can also compare shape features against raw-image models and report where each one fails. That kind of careful validation shows strong biology, image analysis, and model thinking.

Project Variations

• Try a different bee group, such as bumble bees, mason bees, or sweat bees, and test whether the same features still separate species.
• Compare landmark-based morphometrics with a convolutional neural network to see which approach works better on a small dataset.
• Add a site-based split, then test whether a classifier trained in one region can identify bees from another region.

Learn More

• PubMed: Search review articles on bee wing morphometrics, geometric morphometrics, and species identification.
• GBIF: Find specimen records, species distributions, and collection metadata for labeled bee samples.
• Biodiversity Heritage Library: Read historical bee taxonomy keys and species descriptions from scanned reference works.
• Fiji documentation: Learn landmark marking, calibration, and image measurement from the ImageJ community site.
• scikit-learn documentation: Review classification metrics, cross-validation, and confusion matrices for small datasets.