Underwater Object Recognition for Offshore Wind Farm Environmental Impact Assessment

Abstract

In recent years, Taiwan has actively built lots of offshore wind turbines in the western area of Taiwan due to its renewable energy policies. However, the construction of these turbines may potentially create a variety of issues for the marine ecosystems. Thus, it is necessary to evaluate each potential site for offshore wind turbines to decrease the impacts on the ecosystems. To achieve this, this paper proposes an underwater environmental monitoring architecture, using side-scan sonar imagery combining image noise filtering and YOLOv3 real-time object recognition technology to assist with the selection of the potential site of wind farms. The experimental results show this approach only needs 0.0021 seconds to process each sonar image with an average accuracy of 72.3% in the detection of fish schools.