Marine mesoscale vortex classification recognition method based on deep learning

Abstract

The invention discloses a marine mesoscale vortex classification and recognition method based on deep learning. The method relates to the fields of physical oceans, computer graphic image processing and machine learning. An algorithm mainly comprises model construction, backward processing and model algorithm adjustment. The method comprises the steps of firstly, constructing a model, adopting multiple training strategies to train a mesoscale vortex classification model based on a convolutional neural network algorithm, establishing a vortex classification model, and realizing the efficient classification of mesoscale vortexes; and 2, carrying out backward processing, positioning the high-probability vortex pixels according to the probability density map outputted by the model, carrying out combination and rejection on the repeated vortex images, and recovering the wrongly classified data; and finally, adjusting a model algorithm, adding the error classification data into a training data set to retrain the model, establishing an identification model, and finally determining the position of the sea area where the vortex is located. Practice proves that the method improves the automatic classification and recognition efficiency of the vortex, and expands the application of deep learning in the ocean field.

Description

technical field [0001] The present invention relates to the fields of physical ocean, computer graphics and image processing, and machine learning, and in particular to a method for classification and recognition of marine mesoscale eddies based on deep learning. Classification and identification methods for mesoscale eddies. Background technique [0002] As a common natural phenomenon in the ocean, the mesoscale vortex can wrap and carry a large amount of water to migrate, causing vertical seawater mixing, which has an important impact on ocean kinetic energy, internal biogeochemical processes of seawater, and air-sea interaction. The time scale of the mesoscale vortex is several days to several years, and the space scale is tens of kilometers to hundreds of kilometers. The northern hemisphere rotates counterclockwise to form a cyclonic vortex, and the southern hemisphere rotates clockwise to form an anticyclonic vortex. The classification and recognition of mesoscale vort...

AI Technical Summary

Problems solved by technology

[0004] (1) The Okubo-Weiss (OW) parameter method, which has been used by oceanographers for a long time, can extract features from the ocean background field. This method has been proven to have many shortcomings

First, the threshold of W value needs to be formulated, but there is no uniform threshold for the whole world

Second, the estimation of the W parameter will also be affected by the noise of SSH

[0006] (3) WindingAngle (WA) winding angle method, which has been proven to have higher accuracy than the SSH method, but the computational complexity is too high

[0009] Due to the typical mesoscale vortex classification and recognition algorithms, most of them are calculated based on physical and geometric methods, the calculation amount is huge, and the algorithm effect is uneven

Images