[M-IS02] Ocean plastics, an earth science perspective

[MIS02-09] Laboratory Measurement and Classification Result of Plastic Marine Debris Using Spectral Information

*Ahmad Shaqeer Mohamed Thaheer¹, Yukihiro Takahashi¹ (1.Department of Cosmosciences, Faculty of Science, Hokkaido University)

Keywords:Plastic Marine Debris, Spectral Analysis, Classification, Support Vector Machine

The annual global production of plastic has reached a staggering 400 million tons, with a significant portion entering marine environments due to human activities. This influx of plastic waste has transformed marine debris into a critical global environmental issue requiring urgent action. While visual inspections and in-situ measurements using calibrated equipment are effective, they are time-intensive and labor-intensive. Consequently, remote sensing through air- or space-borne spectral imaging has emerged as a promising alternative for addressing this challenge. However, remote sensing of plastics in the ocean poses several challenges, including variations in camera and light angles, water conditions, plastic debris characteristics, and interference from non-plastic materials. To address these challenges, this study conducted laboratory spectral measurements on 137 samples, comprising commercial-off-the-shelf (COTS) plastics, non-plastics, and marine debris. The marine debris samples were collected from three distinct locations in Japan: (a) Nokobiura Bay, Goto Islands, Nagasaki; (b) Aso-beach Ishikari, Otaru, Hokkaido; and (c) Amami Oshima, Kagoshima. To simulate real-world conditions and ensure robust analysis, the measurements included samples with varying conditions to represent their lifetime exposure, while pure plastic samples were used as reference spectra to verify plastic types. The measurements were performed using an InGaAs-LCTF hyperspectral imaging camera, covering a wavelength range of 600-1600 nm with a fixed resolution of 2 nm. Band selection was performed using the XGBoost algorithm to reduce the number of wavelengths required. A classification model was developed using a Support Vector Machine (SVM) with the Synthetic Minority Over-sampling Technique (SMOTE) to address class imbalance and Leave-One-Out Cross-Validation (LOOCV) for performance validation. Several indicators were applied to evaluate the effectiveness and reliability of the proposed classification model. This research highlights the potential of hyperspectral imaging and machine learning techniques in advancing marine debris identification and classification. The study was supported by Dr. Shin’ichiro Kako of Kagoshima University, Dr. Daisuke Matsuoka of the Japan Agency for Marine-Earth Science and Technology (JAMSTEC), and Mr. Shinya Kimura.