Rip Current Detection – An Orientation-aware Machine Learning Approach

A rip current is a natural phenomenon that causes numerous fatal accidents all over the world. I detect and localize rip currents with a deep neural network called Faster R-CNN. I assembled a custom database of rip currents and used transfer learning, resulting in an accuracy of 85.19% (IoU threshold: 0.5) and an AP of 0.37. In addition, I developed an orientation-aware region proposal layer. Based on evaluation using the IoU measure, the findings revealed that the orientation-aware region proposal layer was 11.7% more accurate, allowing the algorithm to adapt to many positions and different perspectives. An automated rip current detection system is under development. This approach contributes to the early identification of the hazard, to preventing accidents and to protecting human lives.

Category: COMPUTING Country: HUNGARY Year: 2020


Boglárka Ecsedi