Detection and identification of plastics using SWIR hyperspectral imaging
Mehrubeoglu, M.; Van Sickle, A.; Turner, J. (2020). Detection and identification of plastics using SWIR hyperspectral imaging, in: Ientilucci, E.J. et al. Imaging Spectrometry XXIV: Applications, sensors, and processing, 24 August – 4 September 2020. Proceedings of SPIE, the International Society for Optical Engineering, 11504: pp. 1-11. https://dx.doi.org/10.1117/12.2570040
In: Ientilucci, E.J.; Mouroulis, P. (Ed.) (2020). Imaging Spectrometry XXIV: Applications, sensors, and processing, 24 August – 4 September 2020. Proceedings of SPIE, the International Society for Optical Engineering, 11504. SPIE: Bellingham. ISBN 9781510638143; e-ISBN 9781510638150. https://dx.doi.org/10.1117/12.2581605, more
In: Proceedings of SPIE, the International Society for Optical Engineering. SPIE: Bellingham, WA. ISSN 0277-786X; e-ISSN 1996-756X, more
|  |
| Available in | Authors |
|
Document type: Conference paper
|
| Author keywords |
hyperspectral imaging, SWIR hyperspectral imaging system, NIR imaging spectroscopy, macroplastics, microplastics, plastic debris detection, identification of plastics, semantic segmentation |
| Authors | | Top |
- Mehrubeoglu, M.
- Van Sickle, A.
- Turner, J.
|
|
|
| Abstract |
Most plastics are typically transparent in the visible spectral range, rendering them challenging to detect using silicon-based vision sensors. In this work a SWIR hyperspectral imaging system is used to collect the SWIR hyperspectral signatures as well as spatial information of a variety of plastics outdoors to test this technology for plastic debris detection and identification in future marine and environmental applications. In this study, hyperspectral imaging data have been collected from plastic samples including CPVC, PVC, LDPE, HDPE, PEEK PETG, PC, PP, PS, and Polyester in a natural environment. The data is acquired using a SWIR hyperspectral imaging system sensitive to 900 - 1700 nm wavelength range. Four spectral indices based on labeled spectral signatures have been identified and used as features to separate plastic materials and for classification of pixels. Semantic segmentation based on plastic materials is achieved in an independent scene with multiple plastic samples using shortest Euclidean distance to labeled feature cluster centers through multi-variate data analysis. The results show the capability of this technology and technique to detect and classify different plastics in natural environments under different light conditions. |
|
