DDM Autoencoder

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.

Files

• 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 it
  • History.mat: a dictionary file compatible with MATLAB with the error losses for the training set and test set over epochs
  • DDMextract.mat: a dictionary file compatible with MATLAB with the decompressed DDM data saved
  • output.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

Usage

Default settings

1. Run python ddmautoencoder.py in command line

   1. Make sure you have the necessary libraries specified in requirements.txt, otherwise, run the following in your command line pip install -r requirements.txt
   2. Make sure you have the files mentioned in the ‘Files’ section in the same directory as ddmautoencoder.py

Custom settings

1. 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 the ddms.nc and saves them into a matrix dataset DDMtrain.mat

Output

Components