Detection of Bering Sea Slope Mesoscale Eddies Derived from Satellite Altimetry Data by the Attention Network
- Attresunet
attresunet_onlysla.py and attresunet_suv.py are used to train the Unet neural network with the addition of an attention mechanism.These two scripts use SLA data and SLA-geostrophic velocity fusion data respectively.
testatt_onlysla.py and testatt_suv.py are used to evaluate the model and output detection results.
- Danet
danet_onlysla.py and danet_suv.py are used to train the dual Attention Network.These two scripts use SLA data and SLA-geostrophic velocity fusion data respectively.
testdan_onlysla.py and testdan_suv.py are used to evaluate the model and output detection results.
- pre&post_process
compare.m: Comparison of the oceanic eddies detected by the four different algorithms in the BSs region on 09 September 2016.
cmm.m: This script is used to complement the output of multiple models with each other.
label_vg_bsc.m:This script is used to make the VG algorithm's eddy detection result into a label for model training.
modeltrack.m:This script is used to convert the classification results from the model output into eddy coverage areas.
radius_lifetime.m:This script is used to count the radius and lifetime of the eddy (A circle with a radius equal to the equal area of the eddy).
snapshot.m:This function is used to plot the eddy detection results for a particular day.
sshset_bsc.m:This script is used to interpolate SLA.
tracksizelife_bsc.m:This script is used to process and analysis the Lagrangian eddy tracking results.
uvgeos_vg_bsc.m:This script is to prepare the input data for the VG algorithm.