AN IMPROVED METHOD FOR CONTINUOUS DETECTION OF SCUM IN RIVERS USING U-NET

Abstract

 Continuous monitoring using fixed-point cameras is effective for early detection and understanding of the behavior of scum in urban tidal rivers. In previous studies, scum detection techniques using U-Net have been developed. However, it is difficult to apply the method to multiple locations because it requires a lot of effort and time to create the label images necessary for training. In this study, we developed a new learning method using dummy images, and evaluated its usefulness by comparing it with conventional methods using the following evaluation indicators: precision, recall, F-measure, and mIoU. As a result, we succeeded in detecting scum with higher accuracy than conventional methods, while significantly reducing the effort required to create labels, which is a bottleneck in conventional training methods. This method makes it possible to understand a wide range of spatio-temporal behavior of scum. Furthermore, by applying this method to suspended solids other than scum, it is expected to be used as a general-purpose technique for continuous monitoring of river debris.