Detection of plastic litter through remote sensing
The remote sensing of plastic litter in natural waters presents special challenges due to the spectral signal interference from the water and the multitude of plastic materials. On the basis of various experiments, performed under both controlled lab and mesocosm outside conditions, using different sensor systems (RGB, multi-spectral, hyperspectral) and sensing set-ups/platforms (UAV, fixed poles & near-surface set-ups), we want to define to which extent plastics near the water surface can be detected and quantified based on remotely sensed data. The suitability of different image classification methods, as spectral feature analyses combined with innovative machine learning techniques, will be assessed.
We will perform laboratory and outside hyperspectral experiments. laboratory-based mesocosm experiments will be performed to assess the appropriateness/feasibility of the detection technique and to unravel reflectance spectra of plastics. The spectral signature of typical types of plastic litter (such as plastic bottles of different colours and transparency levels, coffee cups, straws, bags, …) will be measured under different conditions: 1) dry lab conditions, 2) floating on the water, 3) submerged in (harbour) water at various depths. In this experiment also degraded/weathered/bleached beach litter will be collected and measured. These experiments give insight in the spectral characteristics of the typical marine plastic litter and how these optical properties change with the addition of water. We will develop a machine learning tool for automated detection of plastics acquired from fixed-pole set-ups and bridge fixed cameras, and we will perform a show case to illustrate the suitability for the use of drones for plastic observations.