Smartphone Polarization Imaging for Water Pollution

ISEF Category: Embedded Systems

Subcategory: Optics  ·  Difficulty: Advanced  ·  Setup: School Lab  ·  Time: Full Year

The Hook

A thin oil film can hide in plain sight, yet it can change how light bounces off water. That makes polarization imaging a smart way to spot what your eyes miss. With a smartphone, a rotating polarizer, and a CNN, you can turn glare patterns into data.

What Is It?

Polarization is a property of light that describes the direction of its waves. A linear polarizer acts like a filter for that direction. When you rotate the filter, shiny surfaces and thin films change brightness in different ways. Your phone camera can capture those changes, then a CNN, which is a type of machine learning model that learns image patterns, can sort clean water from water with microplastics or oil sheen.

Think of it like wearing sunglasses that only let certain light directions through. As you turn the lenses, some reflections fade and others pop out. That extra contrast can help reveal tiny floating particles or a slick on the surface. Your project combines optics, embedded control, and computer vision in one system.

Why This Is a Good Topic

This is a strong science fair topic because you can test one clear idea, whether polarized imaging improves detection of water contaminants. You also get a real-world connection to water quality, spill monitoring, and environmental screening. A student can learn optics, data collection, image labeling, model training, and performance testing without needing a professional lab.

Research Questions

• How does polarizer angle affect the contrast of microplastics in water images?
• What is the effect of rotating the analyzer on CNN detection accuracy for oil sheen?
• Does adding polarization images improve classification compared with regular smartphone photos?
• To what extent does sample turbidity change the model's false positive rate?
• Which polarizer position gives the clearest separation between clean water and contaminated water?
• How does the type of microplastic shape affect detection performance under polarized light?

Basic Materials

• Smartphone with a rear camera and manual exposure control.
• Two linear polarizer films cut to fit a phone attachment.
• Small stepper motor with driver board.
• Microcontroller board such as Arduino or ESP32.
• Battery pack or USB power source.
• 3D-printed or cardboard phone mount.
• Clear sample containers with flat sides.
• Distilled water.
• Household oil for sheen testing.
• Common plastic fragments or glitter-sized plastic test particles.
• Tripod or stable stand for consistent imaging.
• Black poster board or matte backdrop.
• Metric ruler for scale in images.

Advanced Materials

• Research-grade linear polarizer film.
• Stepper motor with encoder feedback.
• Microcontroller development board with precise timing control.
• Interchangeable optical mount or 3D-printed rotation stage.
• Polarizing reference targets or calibration targets.
• Spectral or color reference card.
• Controlled light box with diffuse LED illumination.
• Microscope slides or shallow imaging cells.
• Standard microplastic samples with known size ranges.
• Image analysis workstation.
• Annotated image dataset for CNN training.
• Transfer learning model base such as a lightweight CNN architecture.

Software & Tools

• ImageJ: Measures brightness, contrast, and particle features in captured images.
• Python: Processes image data, trains models, and evaluates classification results.
• Google Colab: Runs CNN training in a browser without a local GPU.
• Label Studio: Helps you tag image regions or classes for supervised learning.
• Arduino IDE: Uploads control code to the microcontroller that rotates the polarizer.

Experiment Steps

1. Define the exact detection task, such as binary classification of clean versus contaminated water or multi-class detection of different contaminant types.
2. Design the imaging geometry so the phone, light source, sample container, and rotating polarizer stay fixed across trials.
3. Plan a calibration set that links polarizer angle and image intensity to a repeatable numeric signal.
4. Decide how you will label images and split them into training, validation, and test sets to avoid data leakage.
5. Build controls that separate true contamination signals from changes caused by lighting, container shape, or water clarity.
6. Choose the performance metrics that matter most, such as accuracy, precision, recall, and false positive rate.

Common Pitfalls

• Letting ambient room light change between image sets, which shifts reflection patterns and confuses the model.
• Using samples that are too similar to the background, which makes the CNN learn container edges instead of contamination signals.
• Collecting images at slightly different phone distances, which changes scale and breaks feature consistency.
• Training and testing on frames from the same sample, which causes data leakage and inflates accuracy.
• Ignoring turbidity or bubbles, which can mimic particle texture and create false positives.

What Makes This Competitive

A stronger version of this project goes beyond a simple yes-or-no classifier. You can test whether polarization actually improves detection under harder conditions, like murky water or mixed contaminants. You can also compare image features across polarizer angles, then report which angles matter most and why. Careful controls, clean data splits, and a thoughtful error analysis make the project much more convincing.

Project Variations

• Test whether the same polarization setup detects oil sheen better than floating microplastics in the same water background.
• Compare a hand-rotated polarizer with an MCU-controlled stepper to see whether automation improves repeatability.
• Train separate models on raw RGB images and polarization-enhanced images, then compare detection performance.

Learn More

• NASA Earth Observatory: Search for articles on water surface reflectance, remote sensing, and imaging of oil slicks.
• NIH PubMed: Search review articles on microplastics detection methods and optical sensing in water.
• NOAA National Ocean Service: Find background on oil spills, surface sheen, and water monitoring basics.
• ImageJ Documentation: Learn how to measure image intensity, contrast, and particle size from microscope or phone images.
• MIT OpenCourseWare, Introduction to Computer Science and Programming Using Python: Use the Python materials to build a simple image analysis pipeline.