MARINESITU INC — Department of Energy SBIR Phase I: 18b

Description

Environmental monitoring of marine energy converters (MECs) often requires continuous optical monitoring to detect and classify interactions with marine animals. Currently available underwater camera systems however are designed for short term deployments, lack software for real-time data processing, and are prohibitively expensive for marine energy developers. There is a need for lower-cost underwater camera systems with user-friendly software for automated data processing to reduce the cost burden of environmental monitoring for the marine energy industry. The proposed research effort addresses the need for lower-cost and user-friendly environmental monitoring tools through the development and demonstration of a modular stereo-optical camera system. This camera system is based on that of the Adaptable Monitoring Package (AMP) developed at the University of Washington (UW), which has been specifically tailored for long-term monitoring at marine energy sites. Modularization of both the system hardware and software will enable a camera system that is well suited for a wide range of user-specified monitoring missions, while minimizing the system cost for both hardware and data management. To achieve the desired system capabilities, we are proposing to perform a hardware redesign to modularize the system components and fabricate a first commercial prototype. Software will be developed to enable either cabled or uncabled deployments with deployable machine learning algorithms for target detection, tracking, and classification. Testing of the prototype system with this software package will demonstrate the monitoring capabilities in the presence of an operating turbine at UW and the cost-savings of the system hardware and data management. Underwater optical monitoring represents the most effective method for species level classification of marine animals and their interactions with marine energy devices, however the associated high data bandwidth and manual processing burden have hindered the use of this tool. Currently, however, the field of automated optical image processing through machine learning is advancing at an unprecedented rate and we propose to bring these new techniques to the marine energy industry with this system. By implementing a deployable machine learning system for environmental monitoring data mortgages and manual review time will be drastically reduced, allowing for rapid identification of targets and events of interest. Through broad adoption of these techniques, we will be able to evaluate environmental risks and mitigate against any negative impacts. Beyond the marine energy industry this system and software will be applicable across the blue economy, including for ocean observations, underwater vehicles, and fisheries.