Updating April 27, 2025

Implementation codes of FS-Net and the training samples of Landsat data are underway

Implementation of

1. Requirements

(a). Tensorflow/Keras > 2.6
(b). GDAL

We suggest using Conda to set the environment.<br><br>

2.Run

python shuidao.py

We will upload some training samples so anyone downloads the codes and run them directly.<br><br>

3.Dataset

The training dataset locates in the folder of the dataset. For each image, 1 indicates paddy, 0 indicates nonpaddy, and 3 means background which should be masked when you generate the training samples, e.g. size of 256*256.

For better visualization, ArcGIS is suggested to open the image, and Layer Properties -> Symbology -> Unique values should be employed. One demo is shown below.