[M-IS09] Ocean Plastics, an earth science perspective

[MIS09-P13] Marine plastic debris detection using band selection method from the hyperspectral sensors

*Ye Min Htay¹, Ahmad Shaqeer Mohamed Thaheer¹, San Lin Phyo¹, Yukihiro Takahashi¹ (1.Hokkaido University)

Keywords:marine debris, hyperspectral, band selection, plastics

Plastic marine debris became a threat to marine habitats because of waste mismanagement and practices. However, detection of the plastics through remote sensing is limited since most of the satellite missions are planned for land and ocean monitoring. In this study, the detection of marine debris method is established starting with the creating spectral library of common plastic debris such as low-density polyethylene (LDPE), high-density polyethylene (HDPE), polyethylene terephthalate (PET), rope, foam and metal using the InGaAs hyperspectral camera which has 400-1600nm wavelength range. However, the hyperspectral sensor is complex in data processing, high computation cost, and is expensive. Therefore, for the band selection methods, the most common methods of feature selection such as filtering methods and wrappers methods are used. Normalized difference snow index (NDSI) is used for the filtering method and support vector machine (SVM) machine learning method is used for the wrapper method. The selected bands from both methods are compared and the resulting bands are nearly identical. The spectral library showed that although the classification of debris type is possible up to 1200nm, the plastic-type classification required up to 1600nm. The classification proceeded based on the subset selected bands and supervised classification of random forest and K-mean unsupervised classification was applied and it showed promising results. The selected bands are compared with Sentinel-2 satellite bands and the most adjacent band of Sentienl-2 to selected subset bands are used for the classification and the large plastic target area can be detected.