AquaBotix: AI-Powered UV Detection and Filtration of Harbor Microplastics

Microplastics are tiny plastic particles smaller than 5 mm that pollute oceans, lakes, and even drinking water. Current estimates suggest there are over 171 trillion microplastic particles floating in the world’s oceans, with hundreds of thousands of tonnes entering freshwater systems annually1. Because microplastics are so small, they are difficult to detect and remove using traditional filtration systems. Many current methods rely on expensive laboratory equipment, complex procedures, and manual analysis, making large-scale monitoring slow and inefficient2.

Detection difficulty: Microplastics are nearly invisible to the naked eye. In a 1-liter water sample, there can be thousands of microplastic particles, many smaller than 100 micrometers, that are impossible to see without advanced imaging tools3.

Limited filtration systems: Modern water treatment methods show high microplastic removal rates, but initial treatment stages capture as little as 1.8% to 54.5% of particles, leaving a significant portion to enter drinking water supplies and natural waterways4.

Time-consuming analysis: Researchers often need hours or days to examine samples under microscopes, significantly slowing down studies. For example, analyzing 100 samples manually could take a week, limiting the frequency and coverage of monitoring5.

High cost: Advanced detection equipment, such as micro-FTIR spectrometers, can cost $50,000–$100,000 per unit, making widespread deployment in rivers, lakes, or drinking water systems economically unfeasible6.

Environmental and health impact: Microplastics have been detected in over 90% of tap water samples worldwide and accumulate in aquatic species, potentially entering the human body through drinking water, fish, and seafood7.

To develop this solution, I followed a two-step process: Training the AI "brain" and building the physical "body" of the robot. My objective was to make a cost-effective solution to collect and filter microplastics from local waters. Initial research focused on microplastics and their reaction to UV fluorescence, and concluded that 85-90% of microplastics glow under UV light8.

The AI Brain

The AI was developed in two distinct phases to ensure it could recognize plastic in the real world. It was trained using Python and YOLOv5:

Phase 1: I captured 180 original images of various microplastics (fragments of bottles, cups) glowing under UV light to teach the computer what "fluorescent plastic" looks like.

Phase 2: Using Data Augmentation, I expanded my dataset to 500 images. By digitally adding "noise," changing brightness, and flipping images, I trained the AI to be much more accurate, even in messy or dark water. This is useful in conditions where the visibility is not the greatest.

The Robot Body

I built two versions of the hardware to move from a basic proof-of-concept to a powerful cleaner:

Prototype 1: A proof of concept using a 5V centrifugal pump powered by a standard phone power bank.

Prototype 2: I upgraded to a "Heavy Duty" design. I switched to a 12V diaphragm pump for massive suction and a high-capacity LiFePO4 battery for longer run times.

I conducted 10 trials for both the Phase 1 and Phase 2 AI models to evaluate variables such as accuracy rate and glow strength. Testing was measured across two plastic types with varying glow strengths (opaque and transparent utensils) and two environments: a controlled black background and simulated water tanks. Additionally, the robots underwent 10 trials in the water to determine their total capture efficiency and the consistency of the UV-induced fluorescence.

The objective of AquaBotix was to determine whether a low-cost, AI-integrated robotic system could reliably detect and extract microplastics from water surfaces. Data collected from 100 controlled trials confirms a substantial performance leap between the initial proof-of-concept and the final heavy-duty solution. This supports the hypothesis that accessible hardware paired with trained computer vision can address real-world microplastic pollution.

1. AI Detection Evolution (The Brain)

The transition from Phase 1 to Phase 2 proved that data augmentation is critical for real-world reliability and accuracy. I used mean Average Precision (mAP) at 50% IoU to measure the AI’s ability to recognize plastic fragments under UV light (accuracy).

Phase 1 (The Baseline): Trained on 180 images, the model reached a mAP of 75%. While it successfully identified high-visibility items (White Fork) at 100%, it struggled in simulated water environments, with clear cups dropping to 78% accuracy, revealing the model's vulnerability to reflections and surface glare, which are unavoidable conditions in real-world deployment.

Phase 2 (The Strengthening): After expanding to 500 images through augmentation (adding noise and modifying original images), the mAP increased to 91%

The Result: Phase 2 maintained near-perfect accuracy (around 97-100%) even for “Weak Glow” items like clear cups. This proves that a larger, more diverse dataset allows the AI to recognize plastic from background “noise” or water reflections.

2. Engineering Milestones (The Body)

The prototype also underwent a major change from a proof-of-concept into a functional utility tool.

Prototype 1 (Initial Test): Utilizing a 5V centrifugal pump and a standard USB power bank, this model proved that the electronics could survive on water. However, data revealed an Efficiency Dead-end: as microplastics increased to 15 fragments per trial, collection efficiency dropped to a failing 35% due to insufficient suction and rotor clogging.

Prototype 2 (Heavy Duty): I upgraded the system to a 12V diaphragm pump. This provided the constant suction needed to maintain a high capture rate regardless of debris volume.

The Result: Prototype 2 achieved a 94% capture rate in high-density trials, showcasing a 2.6x increase in efficiency over Prototype 1’s 35%. By utilizing a 12V LiFePO4 battery, I maintained a constant suction rate for 3x longer runtimes while also reducing weight compared to traditional lead-acid batteries.

How the Solution Works

The AquaBotix system operates through a synchronized three-step process:

Induce: The robot scans the water surface using high-intensity UV-A lights. This causes approximately 85–90% of plastics to fluoresce, making them stand out against the dark water.

Identify: The onboard camera feeds a live stream to the AI model with a 91% accuracy. When the AI detects a "glow" pattern that matches the characteristics of microplastics, it triggers a filtration system to collect and remove microplastics.

Extract: The 12V pump activates, pulling the water onboard and through a specialized filtration chamber. The water passes through a mesh, but any plastic fragments are caught on top. This allows microplastics to be captured while purified water flows through.

The success of AquaBotix leads to a critical conclusion: expensive, stationary lab equipment is no longer the only path to fighting microplastic pollution. By proving that a low-cost robot can detect and remove sub-5mm particles in real time, this project demonstrates a shift from passive observation to active restoration.

Scalability is attainable. The transition from a 5V to a 12V system proved that high-efficiency filtration doesn't require industrial-sized power sources. Capture efficiency jumped from a failing 35% to 94% in high-density scenarios, making it feasible for small municipalities to deploy local solutions without multi-million dollar infrastructure budgets.

AI & Visibility. Data augmentation was the single most impactful lesson. Expanding the training set from 180 to 500 images raised the mAP from 75% to 91%, training the system to see what the human eye cannot and turning invisible pollutants into actionable targets. Where current microplastic analysis often requires $50,000 spectrometers9, AquaBotix provides a real-time alternative at a fraction of the cost.

Global goals, local action. This project directly supports United Nations Sustainable Development Goals 6 (Clean Water and Sanitation) and 14 (Life Below Water). By removing microplastics before they enter the food chain, we protect both aquatic biodiversity and human health10.

Ultimately, AquaBotix shows that the plastic crisis is not an unbeatable problem. With the right combination of AI and accessible engineering, we can build a future where our waterways are cleaner, safer, and more sustainable.

Moving beyond the prototype phase, the next step involves strengthening the hardware to help it adapt to turbulent harbor conditions and saltwater corrosion. Future robots will use marine-grade HDPE, which is durable, recyclable, and corrosion-resistant for open-water deployment. I plan to expand the AI’s training library to include "weathered" plastics, ensuring high detection accuracy in murky water.

Future models will feature swarm capabilities, allowing multiple robots to communicate and map pollution areas collaboratively, increasing efficiency and effectiveness. Finally, I aim to integrate solar-charging decks to make the system fully energy-autonomous, creating a truly "set-and-forget" solution for long-term waterway restoration.

Thanks to everyone who helped me throughout this project:

Family: For always being there for me, supporting me, and making sure I had what I needed.

Teachers: For all the encouragement and feedback you gave me whenever I needed help. You've made a huge difference.

Coding Teacher: A special shout-out to my coding teacher, who taught me many valuable skills three years ago. Those lessons were key to making this project happen.

Simcoe County Regional Science Fair Team: For your ongoing support and for giving me a great platform to present my project.

Research Articles & Organizations: NOAA, the University of Waterloo, and the United Nations.

Alan Groombridge (BWG Library): For helping me design and print pieces needed for my prototype

Town of Barrie: Diane Moreau, CAO, Michael Prowse, and Mayor Nuttall for recognizing my project, supporting my vision, and inspiring my creative journey.

Works Cited

Plastic Pollution Coalition. (2025, June 26). How microplastics are changing the oceans. https://www.plasticpollutioncoalition.org/blog/2025/6/26/how-microplastics-are-changing-the-oceans

Puteri, M. N., Gew, L. T., Ong, H. C., & Ming, L. C. (2025). Technologies to eliminate microplastic from water: Current approaches and future prospects. Environment International, 197, Article 109397. https://doi.org/10.1016/j.envint.2025.109397

National Oceanic and Atmospheric Administration Marine Debris Program. (n.d.). Microplastics. NOAA Marine Debris Program. Retrieved April 26, 2026, from https://marinedebris.noaa.gov/what-marine-debris/microplastics

Sol, D., Laca, A., Laca, A., & Díaz, M. (2020). Microplastics removal in wastewater treatment plants: A critical review. Environmental Science: Water Research & Technology, 6(10), 2751–2773. https://doi.org/10.1039/D0EW00397B

Sartain, A., Wessel, C., & Sparks, E. (n.d.). Sampling & processing guide book [PDF]. NOAA Institutional Repository. https://repository.library.noaa.gov/view/noaa/41267/noaa_41267_DS1.pdf

LabX. (2024, May 29). The best infrared and FTIR spectroscopy systems: A buyer's review of price and features. https://www.labx.com/categories/infrared-ft-ir

Shoman, N., Solomonova, E., Akimov, A., Rylkova, O., & Mansurova, I. (2024). Activation of stress reactions in the dinophyte microalga Prorocentrum cordatum as a consequence of the toxic effect of ZnO nanoparticles and zinc sulfate. Microchemical Journal, 207, Article 111697. https://doi.org/10.1016/j.microc.2024.111697

Ho, D., Prabhakar, P., Karthikeyan, K. G., & Feng, H. (2025). Shedding light on the polymer's identity: Microplastic detection and identification through Nile Red staining and multispectral imaging (FIMAP). Journal of Environmental Chemical Engineering, 13, Article 117944. https://doi.org/10.1016/j.jece.2025.117944

University of Waterloo, Global Water Futures Observatories. (n.d.). Microplastics analysis laboratory. Retrieved April 26, 2026, from https://uwaterloo.ca/global-water-futures-observatories/analytical-services/microplastics-analysis-laboratory

United Nations. (n.d.-a). Goal 14: Conserve and sustainably use the oceans, seas and marine resources for sustainable development. United Nations Department of Economic and Social Affairs, Sustainable Development. https://sdgs.un.org/goals/goal14

United Nations. (n.d.-b). Goal 6: Ensure availability and sustainable management of water and sanitation for all. United Nations Department of Economic and Social Affairs, Sustainable Development. https://sdgs.un.org/goals/goal6

Images

Hawkins, J. (2025, July 13). Tons of invisible plastics are hiding in our oceans [Photograph by Wonderful Nature/Shutterstock]. BGR. https://www.bgr.com/science/tons-of-invisible-plastics-are-hiding-in-our-oceans/

Singh, A. (2025, June 2). What happens to microplastics in the ocean? [Photograph by xalien/Shutterstock]. AZoCleantech. https://www.azocleantech.com/article.aspx?ArticleID=1984

Carrington, D. (2025, August 28). Microplastics in hair study [Photograph by pcess609/Getty Images/iStockphoto]. The Guardian. https://www.theguardian.com/environment/2025/aug/28/microplastics-in-hair-study

Intarat, S., Budtakeer, S., Kraisornpornson, B., & Pradit, S. (2024). A new approach to classifying polymer type of microplastics based on Faster-RCNN-FPN and spectroscopic imagery under ultraviolet light. Scientific Reports, 14, Article 3439. https://doi.org/10.1038/s41598-024-53251-5