Microplastic Classification With Smartphone Images

ISEF Category: Materials Science

Subcategory: Other  ·  Difficulty: Intermediate  ·  Setup: Home Setup  ·  Time: 1 to 2 Months

The Hook

A grain of sand can hide a plastic fragment smaller than a pencil tip. Your phone camera can help find it. With the right image setup, machine learning can sort tiny particles by shape and color. That makes this a real research project, not just a photo exercise.

What Is It?

Microplastics are tiny plastic pieces found in sand, water, and soil. In this project, you use smartphone-microscope images and a CNN, short for convolutional neural network, to tell different plastic types or particle classes apart. Think of the CNN like a very picky pattern matcher. It learns tiny visual clues, such as edges, texture, and color patches, then uses those clues to make predictions.

The big idea is simple. You collect or use open-source images, train a model on labeled examples, and test how well it classifies unknown particles. You can compare different image setups, different preprocessing methods, or different model architectures. Your real job is not just to get a model running. Your job is to find out what image quality and data choices help the model make better calls.

Why This Is a Good Topic

This topic works well for science fair research because you can measure performance clearly, compare design choices, and connect your work to a real pollution problem. Microplastics matter in beaches, waterways, and food chains, so your project has an obvious real-world link. You can learn data labeling, image preprocessing, model training, and error analysis without needing a university lab.

Research Questions

• How does image magnification affect CNN accuracy for classifying microplastic types?
• What is the effect of different background colors on model performance for beach-sand microplastic images?
• Does adding image preprocessing, such as sharpening or contrast correction, improve classification accuracy?
• To what extent does training on open-source microplastic datasets improve prediction on your own beach samples?
• Which CNN architecture gives the best balance of accuracy and simplicity for small microplastic image sets?
• How does class imbalance change false positives for rare microplastic types?

Basic Materials

• Smartphone with a camera and manual focus controls.
• Clip-on microscope lens or basic smartphone microscope attachment.
• White, black, and neutral sample cards.
• Small clear containers or petri dishes for sample mounting.
• Tweezers and a fine paintbrush for handling particles.
• Digital scale for labeling sample masses if needed.
• Free image analysis software.
• Laptop or desktop computer for model training.
• Access to open-source microplastic image datasets.
• Notebook or spreadsheet for tracking labels and results.

Advanced Materials

• Smartphone microscope with consistent lighting and a stable mount.
• Stereo microscope for ground-truth comparison.
• Standard reference particles or known polymer fragments.
• Stereomicroscope camera adapter.
• Controlled lighting box with fixed color temperature.
• Imaging stage with metric scale reference.
• Image annotation software.
• Machine-learning workstation or cloud notebook with GPU access.
• Curated open-source dataset of labeled microplastic images.
• Statistical analysis software for confusion matrices and significance tests.

Software & Tools

• Google Colab: Runs Python notebooks for training CNNs without needing a powerful local computer.
• Python: Lets you clean images, train models, and score classification results.
• TensorFlow or PyTorch: Provides tools for building and testing CNN models.
• ImageJ: Helps you crop, measure, and standardize microscope images before training.
• Excel or Google Sheets: Tracks samples, labels, and performance metrics in a simple table.

Experiment Steps

1. Define the particle classes you want to separate, and make sure each class has clear visual labels.
2. Plan one image setup that keeps lighting, scale, and focus as consistent as possible.
3. Build a clean training set and a separate test set so your evaluation stays fair.
4. Decide which preprocessing changes you will compare, such as cropping, resizing, or contrast correction.
5. Train a baseline CNN first, then compare it with one or two modified versions.
6. Analyze misclassified images to see whether errors come from blur, background noise, or weak labels.

Common Pitfalls

• Training on images from one lighting setup and testing on another, which makes the model learn the background instead of the particle.
• Mixing up plastic fragments, sand grains, and shell bits, which poisons the labels and lowers accuracy.
• Using too few examples per class, which makes the CNN memorize images instead of learning general features.
• Letting the same particle appear in both training and test folders, which inflates the score.
• Ignoring false positives on similar-looking sand debris, which hides where the model actually fails.

What Makes This Competitive

A stronger project goes beyond a simple accuracy score. You can compare image conditions, test generalization on your own beach samples, and explain why the model succeeds or fails on certain particle types. Good confusion matrix analysis, careful holdout testing, and a clean discussion of bias will make your work look much more like real research. A novel comparison, such as phone images versus microscope images, can push the project further.

Project Variations

• Classify microplastics by shape instead of polymer type, such as fibers, fragments, and films.
• Compare model performance on beach sand, river sediment, and roadside dust.
• Test whether grayscale images perform as well as color images for microplastic classification.

Learn More

• PubMed: Search for review articles on microplastics, image analysis, and environmental monitoring.
• NOAA Marine Debris Program: Find background on microplastic pollution and field sampling context on the NOAA website.
• NASA Earthdata: Explore remote sensing and environmental data methods that can inspire your sampling and validation plan.
• USGS Publications Warehouse: Search for studies on microplastics in sediments and water.
• MIT OpenCourseWare: Look for free computer vision and machine learning course materials that explain CNN basics.
• Nature and Scientific Reports: Search the journals for peer-reviewed microplastic detection and classification studies.