Welcome to PLP2024!

We are excited to kick off this year’s Plastic Litter Project (PLP2024), an ambitious initiative dedicated to advancing marine litter detection from space. Our focus this year is on enhancing our detection algorithms and assessing the effectiveness of different satellite systems.


This year, we aim to refine our FML/FMD detection algorithm, which uses a combination of spatial feature detection (via U-net++) and pixel-based spectral classification. Our primary research goal is to determine whether Planet SuperDove is superior to Sentinel-2 for spatial feature detection using CNN models.


To achieve this, we will produce training and validation data from in-situ validated images of artificial targets that simulate marine litter formations. These targets will vary in size and composition to reflect real-world conditions. The data collected will help train our CNN models and validate their accuracy.


Our project timeline includes several key milestones, starting with an initial meeting with participants, followed by the construction and deployment of targets, and concluding with a summer acquisition campaign.


Thank you for joining us on this important journey towards a cleaner, healthier marine environment!

How Our FML Detection Algorithm Works:

Our detection system combines two powerful modules:

Spatial Feature Detection (U-net++)

This module helps identify key features in the data.

Pixel-based Spectral Classification (MF or SAM)

This module classifies pixels to spot FMD or potential FML formations, especially windrows.


Our Main Research Goal for 2024:

We aim to use the U-net/U-net++ CNN architecture as the core feature detection module in our marine litter detection algorithm.

Our Big Question:

Can the Planet SuperDove satellite outperform Sentinel-2 in detecting spatial features using CNN models? To find out, we’ll train our CNN model on field images and validate its accuracy using data from PLP2024. This will also help us understand the minimum detection thresholds for both satellites.

PLP2024 Objectives:


Create Training and Validation Data: We’ll produce data from in-situ validated images of artificial targets simulating FML windrows.


Target Details: These targets will include pure white HDPE mesh and mixed HDPE-natural debris, with dimensions ranging from 1.2 to 4.8 meters wide and 50 to 100 meters long. This aligns with Cozar et al. (2021), which found that 45% of windrows were ≤5 meters wide. Smaller windrows are more common in the Eastern Mediterranean and the Aegean.


Target Construction: Mixed targets will use reeds on HDPE sheets or wooden planks on HDPE sheets, while pure non-HDPE targets will use wooden planks.

Timeline Milestones


Kickoff meeting with students to introduce the project, participants, and workflow.


Gather natural debris materials and construct the targets.


Start deploying targets, with deployments continuing based on the acquisition schedule and weather.


Remove targets and wrap up the summer acquisition campaign (which may extend into September if needed).