Rip current is a powerful narrow flow of water that moves away from the shore, posing a danger to swimmers and surfers.
They form due to interactions between waves and uneven underwater features.
They're strong & fast, pulling even strong swimmers away.
Rip currents have long posed a serious threat to beachgoers and swimmers. Despite numerous preventive measures throughout the period, the fatality rate underscores the need for a robust rip current detection system. Recently deep learning models have shown promising results in rip current detection, outperforming traditional methods. However, these models still exhibit some accuracy limitations due to insufficient data distribution. To address this challenge, we incorporate a novel largest dataset comprising over 110,215 Korean coastline images. Through the comparative study of the state-of-the-art models, we aim to analyze the detection accuracy of each model and gain a deeper understanding of their intensity over rip current detection. In comparison to the other rip current datasets, the evaluation results on our proposed dataset demonstrate a remarkable elevation in accuracy, reaching 79.4 mAP. Further, we employ the EigenCAM (Eigen Class Activation Maps) to interpret the intense regions of the rip currents and to gain a deeper comprehension of rip current explainability. This comprehensive analysis marks a significant step toward improving rip current safety and understanding. (Choi et al., 2024)
Sample rip current detection
This image is an illustration of predictions of visible rip currents on the coastline. Eigen-CAM method interprets the predicted rip currents on Haeundae Beach. (a) The original image, (b) Bounding box prediction of YOLOv8 model, (c) Visual explanations of Eigen-CAM over the same model and image.
The latest model YOLOv8 demonstrates exceptional capabilities in rip current detection, even when multiple rip currents are present. The YOLOv8 outperforms the Faster RCNN and YOLOv5 models, especially the nano-sized model.
Plots showing the mean Average Precision values of different models with different sizes. In this PR curve plot, while the x-axis denotes recall, the y-axis refers to precision.
The below table provides the box losses and object losses over training and validation datasets. The YOLOv8 nano model has lesser loss values compared to other models. After the YOLOv8n model, the YOLOv5n model achieves less loss values comparatively.
Rip currents have long posed a serious threat to beachgoers and swimmers. Despite numerous preventive measures throughout the period, the fatality rate [1] underscores the need for a robust rip current detection system. Recently deep learning models have shown promising results in rip current detection, outperforming traditional methods. However, these models still exhibit some accuracy limitations due to insufficient data distribution. To address this challenge, we incorporate a novel largest dataset [2] comprising over 110,215 Korean coastline images. Through the comparative study of the state-of-the-art models, we aim to analyze the detection accuracy of each model and gain a deeper understanding of their intensity over rip current detection. In comparison to the other rip current datasets, the evaluation results on our proposed dataset demonstrate a remarkable elevation in accuracy, reaching 79.4 mAP. Further, we employ the EigenCAM (Eigen Class Activation Maps) to interpret the intense regions of the rip currents and to gain a deeper comprehension of rip current explainability. This comprehensive analysis marks a significant step toward improving rip current safety and understanding.
@inproceedings{choi2024explainable,title={Explainable Rip Current Detection and Visualization with XAI EigenCAM},author={Choi, Juno and Rajendran, Muralidharan and Suh, Yong Cheol},booktitle={2024 26th International Conference on Advanced Communications Technology (ICACT)},pages={1--6},year={2024},organization={IEEE},doi={https://doi.org/10.23919/ICACT60172.2024.10471913}}
