Research On Ocean Eddies Identification Method Based On Deep Learning

The ocean is of great significance to the survival and development of human society,so we need to fully understand the ocean,and then we can better promote the development of the ocean.Ocean eddy is an important marine phenomenon that is common in the ocean.It always moves at a high speed in a spiral shape in the ocean,and at the same time contains a huge amount of energy.In addition,the movement of the vortex is often accompanied by the delivery and mixing of seawater temperature,nutrients and energy.Therefore,eddies play an important role in the changes of marine ecosystems and global climate,and the identification of marine eddies has become one of the research hotspots in the field of physical oceans.The traditional identification of ocean eddies strongly relies on the measured data of ships and instruments such as the Acoustic Doppler Current Profiler(ADCP),SelfSinking and Floating Profile Detection Buoy(Argo),and Temperature and Salt Depth Meter(Conductance,Temperature,Depth,CTD).Discontinuous,only eddies on specific sea areas and routes can be studied.Advanced remote sensing satellites have outstanding advantages such as all-day,all-sea,and ready access,providing a wealth of data sources for ocean eddy research.Therefore,this paper uses multiple remote sensing fusion data sources and combines deep learning technology to improve the existing ocean eddy recognition algorithm,and proposes a faster recognition algorithm for ocean eddy based on deep learning.First,this paper analyzes the traditional recognition algorithms based on flow field geometric features and based on height anomalies,and conducts simulation experiments using data in the study area.Both recognition algorithms take fixed geometric flow field feature constraints as the key to eddy recognition,and the setting of threshold value needs to rely on expert experience,and the recognition results show that both are very easy to miss and wrong inspection,and the recognition is Inefficient and ineffective.Secondly,this paper delves into the recognition algorithm based on deep learning.Since the key to improving the accuracy of ocean vortex recognition lies in the accurate extraction of eddy features,for the problem that traditional algorithms cannot efficiently obtain features and rely on expert experience to set thresholds,this paper uses the characteristics of deep learning that can extract the high-level features of eddy through hierarchical abstraction to design a deep learning-based ocean.Eddy recognition model,using improved dense convolution to accurately extract the highlevel features of ocean eddies,and using cross-layer fusion techniques to improve the reuse rate of features,fully capture edge information,and improve detection accuracy.Compared with traditional ocean eddy recognition algorithms,the deep learning-based ocean eddy recognition method proposed in this paper can effectively separate and detect eddies in close proximity,with higher quality and efficiency of recognition.