Draken

The Problem

The Tisza’s pollution crisis is driven by a systemic "plastic flood" originating from the steep Carpathian regions of Ukraine and Romania. Due to a lack of waste management infrastructure and the absence of flat land for landfills, communal waste is frequently deposited in floodplains or illegal riverside dumps. During seasonal snowmelts or heavy rains, the rising river sweeps this accumulated waste downstream, carrying thousands of tons of PET bottles and industrial debris into Hungary, where it settles in protected ecosystem floodplains.

The core challenge is a lack of transboundary accountability; once the waste is in the water, it is difficult to prove its point of origin. Our project addresses this by using on-site ML cameras to log the exact timing and volume of debris waves, then cross-referencing that data with Sentinel satellite imagery to pinpoint which upstream dump sites were washed away. This shifts the strategy from costly, reactive cleanup to a data-driven model of source identification and international accountability.

The Team

We are a team of Eastern Hungarian engineers and innovators united by the Tisza river. Seeing the impact of trash pollution on our local ecosystem firsthand, we’ve combined our expertise in ML and satellite data to protect our home waters. We don't just solve complex problems; we solve the ones that matter to our community.

We have been doing IoT projects since 2016, our latest achievement is a HU-RO-UA collaboration sensor network with more than hundred air quality (pm2.5/pm10) sensors installed and operated across three cities. 

The Solution

Summary

This project is a multi-scale monitoring system designed to tackle water pollution by combining localized computer vision with global satellite intelligence. By bridging the gap between immediate detection and long-range source tracking, the system provides a comprehensive solution for environmental protection and watershed management.

Costs

The hardware unit costs around 800 EUR. Operating costs depends on the communication and server platforms, estimated around 500 EUR / year for 100 units.

•    Interreg: EU funding programme supporting cross-border and transnational cooperation projects, especially in regional development, innovation, and environmental sustainability.
•    LIFE Programme (Calls for proposals 2026): EU funding instrument dedicated to environmental protection, climate action, and energy transition projects.
•    European Union Agency for the Space Programme (Copernicus): Provides funding and opportunities for projects using Copernicus satellite data and services, especially in downstream applications and innovation.
•    GoFundMe: Online crowdfunding platform where individuals or teams can raise funds directly from the public for specific projects or causes.
•    YouTube channel advertising support: Financial support generated through ad revenue and sponsorships on a YouTube channel, typically requiring consistent content and audience growth

Tech stack

Edge Layer (Data Acquisition & Detection)

• Hardware: An In-situ SBC (Single Board Computer) connected to a Camera. This is currently a RaspberryPi4 on our demo equipment, using Galileo satellites for location and anti-theft.

• Intelligence: Performs Local Inference with huggingface resnet-50_plastic_in_river pyTorch model, meaning the machine learning model runs directly on the device to identify debris without needing to stream raw video to the cloud.

• Connectivity: Uses NBIoT/Kiénis for telemetry. It's a Low Power Wide Area Network (LPWAN) designed for low-bandwidth, long-range environmental sensing.

Cloud Infrastructure (Backend & Storage)

• Application Framework: Django serves as the primary backend engine, orchestrating data between the edge devices and the database.

• Database: TimescaleDB is used for spatiotemporal data storage. Since it's built on PostgreSQL, it’s optimized for handling the time-series nature of sensor alerts and geographical coordinates.

• Satellite Integration: A dedicated pipeline connects to the Copernicus API to pull Sentinel data, which then undergoes satellite image Inference to cross-reference edge detections with orbital views. Using Sentinel-1 VV and VH polarization and Sentinel-2 as complementary data.

Visualization Layer (Frontend)

• Web Server: NodeJS acts as the delivery layer for the user interface.

• User Interface: A Hexbin UI is used for map visualization. Hexagonal binning is particularly effective for spatial data because it reduces visual noise and makes density patterns (like debris clusters) much easier to spot compared to standard heatmaps.

Scalability

Based on infrastructure data from Eurostat:

• There are over 1,000,000 bridges total in Europe (including small road overpasses).
• Of those, roughly 150,000 to 200,000 are estimated to span inland rivers and canals significant enough to be monitored for environmental reasons.

Competitors/ ongoing similar Europian projects

• SWIM (Surface Water Information Management) is an EU programme supporting UN Sustainable Development Goal 6—ensuring access to water and sanitation—while aligning with key European policies like the Water Framework Directive. It leverages Copernicus Earth observation data processed with AI, combined with in situ measurements from the consortium’s proprietary WAMO (Water Monitoring) IoT platform, to improve surface water monitoring, with international collaboration including U.S. and Colombian partners.
• Ljasuk Dimitry aims to reconnect people with nature through his films and inspire a love of life. Raised in Hungary, he highlights the country’s natural treasures while also exposing environmental problems and actively working on solutions, including organizing and supporting clean-up initiatives.
• Petkupa: In the early years, the main focus was on building boats from plastic bottles and collecting waste during races. Their goal is nothing less than eliminating the root cause of the problem and stopping the flow of waste, which requires international cooperation—something we are actively working on. They now have strong backing, including the Ministry of Innovation and Technology, and based on their objectives, they are likely to support the implementation of our idea.

Possible further improvements

There are commercial satellites that provide higher spatial resolution and much more frequent imaging than the Sentinel satellites within the Copernicus Programme. Both Sentinel missions and systems such as ICEYE and Capella include SAR (Synthetic Aperture Radar) capability, enabling all-weather, day-and-night observations. The main difference is that commercial constellations typically offer finer spatial detail and higher revisit rates, which can improve the detection of illegal waste dumping sites by increasing location accuracy and lowering the minimum detectable size of deposits.

Related content