Deep Learning Approach for Eddy Detection in Bay of Bengal using SPA-YOLO

Abstract

Eddy detection is crucial for understanding ocean dynamics and their impact on marine ecosystems. This paper introduces a new method based on the You Only Look Once (YOLO) deep learning algorithm for identifying ocean eddies in the Bay of Bengal. The model is trained on satellite-derived Sea Surface Height (SSH) and Sea Surface Temperature (SST) datasets to identify and categorize eddy structures, utilizing YOLO's real-time object detection capabilities. Our approach combines preprocessing stages, such as data normalization and augmentation, to improve the accuracy and resilience of the model. Additionally, we integrate a Spatial Attention (SPA) module into YOLO, creating SPA-YOLO, which enhances the model's ability to focus on relevant spatial features within the data. This integration allows for more precise identification of cyclonic and anticyclonic eddies by emphasizing critical regions in the input data. The trained SPA-YOLO model outperforms other approaches in terms of precision and recall. Experimental results highlight the model's efficiency in processing large-scale oceanographic data, providing timely and accurate eddy detection. This research contributes to the advancement of ocean monitoring systems, offering a scalable and dynamic solution for marine researchers and policymakers. The application of SPA-YOLO in this context underscores the potential of deep learning techniques in enhancing the understanding of complex oceanographic phenomena, thereby supporting efforts in climate research, marine biodiversity conservation, and sustainable ocean resource management.