We can't find the internet
Attempting to reconnect
Something went wrong!
Hang in there while we get back on track
Plastic polymer classification with shortwave infrared reflectance: Opportunities and limitations for remote sensing of marine debris
Summary
Scientists are testing whether satellites and sensors can identify what type of plastic is floating in our oceans by reading light patterns bounced off the debris, without needing to physically collect samples. This matters because plastic pollution breaks down into microplastics that enter our food and water supply, and knowing exactly which plastic types are accumulating where could help target cleanup efforts and better understand health risks. The technology shows promise but still has real limitations, especially in noisy, real-world conditions, so more refinement is needed before it can be widely used.
Marine plastic debris or litter poses urgent environmental, socio-economic, and human health challenges, yet tools for polymer-specific detection with optical remote sensing remain limited. To advance optical classification capabilities, the newly curated and comprehensive MArine Debris spectral reference Library (MADLib) collection was used to evaluate three existing algorithms and develop two proof-of-concept approaches for polymer type classification with hyperspectral shortwave infrared (SWIR) reflectance measurements. A stepwise classification framework was designed to identify seven polymer types from hyperspectral reflectance data, known as the six-step polymer identification (SSPI) algorithm. SSPI achieved an overall classification accuracy of 89%, which declined to 73 ± 1% when simulated Gaussian noise was added to the measurements. Although performance declined under simulated noisy conditions, the framework establishes a foundation for future refinement through other avenues, such as spectral matching or machine learning methods. A multispectral polymer type classifier (MSPC) approach was also developed and evaluated using machine learning methods to leverage diagnostic features in SWIR reflectance data from only eight moderately-broad bands. A shallow neural network showcased strong performance for a subset of the seven most represented polymers in MADLib under the relatively restrictive testing procedures explored, achieving consistent and noise-robust class accuracies over 80%. The MSPC approach highlights the promise of compact multispectral systems for efficient and scalable polymer type identification. Overall, this work highlights both the potential and current limitations of reflectance-based polymer type identification and guides the next generation of remote sensing approaches for marine plastic debris.