The Delay Doppler Map (DDM) Autoencoder is a deep learning algorithm that aims to compress and decompress spatial image data while retain as much of the information as necessary.
- ddmautoencoder.py: the main file that is used to run the autoencoder. This will create 4 files in the same directory:
model_weights.h5
: a file with each of the optimized parameter weights saved in itHistory.mat
: a dictionary file compatible with MATLAB with the error losses for the training set and test set over epochsDDMextract.mat
: a dictionary file compatible with MATLAB with the decompressed DDM data savedoutput.png
: a side-by-side comparison between the input and extracted images
- model_weights.h5: the pre-trained model weights
- DDMtrain.mat: a matlab file containing 5000+ samples of anomaly filtered DDMs for training the autoencoder
- DDMtest.mat: a matlab file containing the test set for the autoencoder
- ExtractDDM.m: a matlab script to convert the
DDM.nc
file into a workable dataset - requirements.txt: a list of python libraries used for the autoencoder
- Run
python ddmautoencoder.py
in command line- Make sure you have the necessary libraries specified in
requirements.txt
, otherwise, run the following in your command linepip install -r requirements.txt
- Make sure you have the files mentioned in the 'Files' section in the same directory as
ddmautoencoder.py
- Make sure you have the necessary libraries specified in
- Create custom DDM dataset
a. The MERRByS website hosts a large collection of publically available L1b data. The data can be simply but slowly downloaded after registering a user account.
b. One of the satellite engineers and researcher Philip Jales has a repository for processing the raw data downloaded from the MERRByS website
c.
ExtractDDM.m
is a custom script that extracts exclusively the DDM data from theddms.nc
and saves them into a matrix datasetDDMtrain.mat