Article and Figures Provided By Project Team: NOAA’s National Centers for Coastal Ocean Science (NCCOS), NOAA’s Marine Debris Program (MDP), and Oregon State University (OSU)
Marine debris, human-made material that is discarded or abandoned into the marine environment, is a pervasive problem plaguing shorelines around the world. Marine debris poses serious threats to wildlife, degrades coastal and marine environments, and can negatively impact the Blue Economy (e.g., tourism, shipping, and fisheries). NOAA’s Marine Debris Program (MDP), as the U.S. Federal lead for assessment, prevention, and removal of debris, works with partners across the Nation to conduct debris shoreline surveys to identify debris accumulations, locations, and sources as part of the Marine Debris Monitoring and Assessment Project (MDMAP). Data from these surveys have been used to assess spatial and temporal trends in shoreline debris, inform behavior change campaigns focusing on specific items and assess the effectiveness of legislation targeting specific items. In this project, NOAA’s National Centers for Coastal Ocean Science (NCCOS), NOAA’s MDP, and Oregon State University (OSU) partnered to investigate three emerging technologies with the potential to transform how marine debris shoreline surveys are conducted: uncrewed aircraft systems (UxS), machine learning, and polarimetric imaging (PI) cameras.
Advantages of UxS for marine debris shoreline surveys include the ability to rapidly and affordably collect data for large stretches of shoreline, including in coastal areas that are difficult to access on foot. After the UxS imagery is collected, further automation can be achieved through the use of machine learning (ML) algorithms for automated detection and type classification of debris items. The project team partnered with Ross Winans of ORBTL LLC to extend a deep learning model for marine debris detection and classification developed in Ross’ Master’s thesis work at University of Hawaiʻi at Mānoa for use with UxS imagery.
The final technology investigated by this project is polarimetric imaging. Low-cost PI cameras, which have recently become commercially available, go beyond the data provided by typical cameras by providing information on the polarization state of received light. Because objects tend to polarize reflected light differently based on their characteristics, such as shape and texture, the additional information provided by PI cameras can assist in detecting and distinguishing various types of objects. A key benefit of polarimetric imagery is its ability to improve detection of human-made objects, and it is this capability which makes PI cameras of particular interest for marine debris shoreline surveys. In this project, marine debris detection and type classification was investigated using a FLIR Blackfly S USB3 5-megapixel PI camera.
The project team conducted experiments at three different locations to test the combined capabilities of these emerging technologies for marine debris shoreline surveys. Initial proof-of-concept tests and procedure development were performed at the OSU Hinsdale Wave Laboratory. Next, the PI camera and UxS were used to collect data at the Neptune State Scenic Area on the Oregon coast. Once the procedures for UxS marine debris surveys and polarimetric imagery acquisition were sufficiently tested and refined, the final experiments involved data acquisition on two barrier islands on the Texas coast, Padre Island and San José Island, locations which and local MDMAP partners had identified as having high concentrations of marine debris (Figure 1).
Results to date have shown the PI camera to be highly beneficial in debris detection and type classification. In the research of OSU Master’s student and project team member, Kyle Herrera, the inclusion of image bands derived from the PI data was found to improve visual assessment of debris. Debris items were readily identifiable against rock, cobble and sand backdrops, even when partially buried, weathered, and difficult to detect in traditional RGB (non-PI) imagery. Furthermore, using a supervised image classification algorithm, the classification accuracy was found to improve by an average of 15 percentage points by including image bands derived from the PI-data. Meanwhile, the ML model developed in this work has been deployed in a cloud computing environment for use by MDP partners and scientists in the field. Ongoing work on this project is anticipated to contribute to expanded use of UxS, PI cameras, and machine learning to facilitate marine debris detection and to help guide strategies related to debris mitigation and removal.
This innovative technology and corresponding operations were funded and supported by the OAR Uncrewed Systems Research Transition Office (UxSRTO).