2022 Volume 78 Issue 2 Pages I_877-I_882
It is an important issue for port management to monitor the drifting conditions of marine debris in port waters because many marine debris are easy to drift in the port waters. In this study, we attempted to construct an automatic marine debris detection model using deep learning from sea surface images for the purpose of constructing a simple monitoring method for marine debris and applied the model to sea surface images taken over a long period of time in port areas to monitor marine debris. The sea surface images were taken at an anchorage of Naruo-hama located at the inner part of Osaka Bay. We created teacher images of drifting debris from the sea surface images, verified the detection accuracy of multiple trained models with different combinations of teacher image types and training models, and constructed a model capable of detecting wood and other debris from the sea surface images. The model was applied to images of the sea surface taken between July and December 2020 to examine the inflow and outflow of marine debris and revealed that most of the marine debris was wood.
