AI Method For Ocean Eddy Recognition And Detection Based On Multi-feature Fusio

Ocean eddies are a common and complex phenomenon of seawater flow,which are key driving factors for the transportation of salt,heat,carbon,and other nutrients in the ocean.Therefore,understanding the characteristics and motion mechanisms of ocean eddies helps to enhance our understanding of oceanography,and the detection and identification of ocean eddies is the most important step.A series of methods for identifying eddies based on physical or geometric features have been developed in traditional oceanography,but they have drawbacks such as relying on human labor.In recent years,due to the popularity of artificial intelligence,many deep learning frameworks have been applied in the field of eddy detection.The AI based method not only overcomes the shortcomings of traditional oceanography in identifying eddies,but also achieves high accuracy in identifying eddies.Following the Encoder-Decoder network architecture,this study conducted research on the artificial intelligence(AI)method for ocean eddy identification and detection using multi feature fusion.The main research contents are as follows:(1)We have constructed a eddy label dataset that supports AI classification detection.Based on the fact that ocean eddies can cause anomalies of sea surface temperature and sea level height,this paper integrates the sea surface temperature data and sea level anomaly data and constructs a eddy label dataset with the same resolution as the sea surface feature dataset by using the principle of ray method in mathematics based on the altimeter ocean eddy dataset.(2)A U-shaped eddy detection architecture(PSA-EDUNet)based on pyramid split attention module was proposed.To effectively achieve deep integration of low-level and high-level hierarchical structural features,the model introduces pyramid split attention modules and a hierarchical skip connection structure on the basis of a U-shaped network containing an Encoder-Decoder structure.At the same time,it overcomes the problem of a large amount of loss of feature information in input data in nonlinear connection modes,thereby enhancing the model’s learning ability.(3)Thorough and comprehensive experiments were conducted to verify the eddy identification performance of the proposed method.Taking the Kuroshio Extension and the South Atlantic Ocean areas with strong eddy activity as the research area:(1)Five networks with high detection performance,U-Net、Swin-UNet、Deeplav V3+、Eddy Net and TBCNN,were selected for subjective and objective comparative experiments;(2)Conduct ablation experiments based on pyramid split attention module and hierarchical skip connection structure,respectively.At the same time,selecting evaluation indicators from multiple perspectives to evaluate the eddy identification performance of all experimental results shows that the proposed method has excellent performance and can meet the high-performance requirements of ocean eddy identification and detection.This study conducted a large number of comparative and ablation experiments while proposing the model,which not only verified the excellent performance of the model in the field of eddy recognition,but also provided a reference for subsequent scholars to conduct relevant research,making a certain contribution to the development of the scientific research field of eddy identification and detection.