Code for Deeply synergistic optical and SAR time series for crop dynamic monitoring

This is the prototype code for MCNN-seq that publish in the RSE journal (Deeply synergistic optical and SAR time series for crop dynamic monitoring). Please cite the following paper if you find to be useful.

@article{ZHAO2020111952,
title = {Deeply synergistic optical and SAR time series for crop dynamic monitoring},
journal = {Remote Sensing of Environment},
volume = {247},
pages = {111952},
year = {2020},
}

How to use it?

Step 1 read the configs

check the configs.py for inputs, depth, units, outputs, etc. Also, be aware of the length of the S1 and S2 input time-series.

Step 2 run train.py

The train.py starts with dataset parsing which includes training and testing datasets. Then, the 1D-CNN was followed to filter the time-series for more robust feature generation. After that, the LSTM was also included for contextual information extraction in the temporal domain. Finally, the model can be saved as offline files.

Step 3 run predict.py

Suppose you have the well-trained model to transfer "SAR time-series" into "optical time-series", then you can do it by running this file.

TIPS: Use your own dataset?

Attend to the dataset.py file, where you can construct your own dataset for training and testing.