Underwater hyperspectral imaging for in situ underwater microplastic detection
Hui Huang,Zehao Sun,Shuchang Liu,Yanan Di,Jinzhong Xu,Caicai Liu,Ren Xu,Hong Song,Shuyue Zhan,Jiaping Wu
DOI: https://doi.org/10.1016/j.scitotenv.2021.145960
2021-07-01
Abstract:<p>Microplastics (MPs) on the seabed threatening marine ecology or human health have drawn much attention. Most research focuses on the <em>in situ</em> detection of MPs in air, while the underwater environment, including light absorption and scattering of the water body, makes <em>in situ</em> MP detection challenging. This study proposed a method for <em>in situ</em> detection of underwater MPs (0.5–5 mm) using underwater VIS hyperspectral imaging (400–720 nm). The underwater spectral image correction model of the water body was calibrated by comparing the images of swatches in air and underwater. Different classifiers, including support vector machine (SVM), neural network (NN), least squares–support vector machine (LS–SVM), and partial least squares–discriminant analysis (PLS–DA), were investigated to identify MPs in air and underwater. Combined with the underwater spectral image correction model, all classifiers achieved promising results, and SVM outperformed all the other classifiers, with average precision (<em>PR</em>) = 0.9839, recall (<em>RE</em>) = 0.9859, and <em>F</em>1-score (<em>F</em>1) = 0.9849, for the identification of six types of MPs, where <em>F</em>1 increased by 3.01% over the raw underwater condition. The effects of particle size, color, and shape were studied, among which a detection limit of 0.5 mm was observed and proved to be possible to extend. MP identification on the lakebed verifies the potential of underwater hyperspectral imaging for <em>in situ</em> underwater MP detection, which may translate to seabed detection.</p><h3 class="u-h4 u-margin-m-top u-margin-xs-bottom">Main findings</h3><p>Microplastics (MPs) on seabed can be detected using underwater hyperspectral imageries in turbid water, with the assistance of spectral image correction.</p>
environmental sciences