Waterbody Detection Via Deep Learning

This project explores the application of deep learning to waterbody detection.

Dataset

https://drive.google.com/file/d/1faVYayxdNFGx2m0IxswDncoKmIxusdf7/view

Running The Script

To run the script, simply execute main.py with the index of the GPU you want to train with specified as an optional parameter. If no GPU is specified, the script will default to GPU 0.

python3 main.py [GPU]

Configuration

The script expects an external file called config.json in which the use should specify the desired configuration. Below is an example of such a file and a table outlining the effects of each setting.

Example Configuration

{
  "timestamp": 1,
  "patch_size": 512,
  "experiment_tag": "unet_multispectral",
  "create_logs": true,
  "train": true,
  "test": true,
  "experiments": 1,
  "use_mixed_precision": true,
  "hyperparameters": {
    "model": "unet", 
    "bands": ["RGB", "NIR", "SWIR"],
    "backbone": null,
    "learning_rate": 0.00005,
    "fusion_head": "naive",
    "loss": "jaccard_bce",
    "batch_size": 4,
    "epochs": 50,
    "apply_transfer": false,
    "random_subsample": false, 
    "water_threshold": 0
  }
}

Available Settings

Setting	Effects
timestamp	The timestamp to use (1, 2 or 3)
patch_size	The desired size of the generated patches
experiment_tag	The human-readable tag with which to lable the experiment
create_logs	Indicates whether or not we want to create logs for the experiment
train	Whether or not we want to run the training loop
test	Whether or not we want to test the trained model on the test set
experiments	Indicate the number of identical experiments we want to run
use_mixed_precision	Indicate the number of identical experiments we want to run

Available Hyperparameters

Hyperparameter	Effects
model	The model we want to use
bands	The bands used as inpiut to the model
backbone	The model of the pre-trained backbone we want to use
learning_rate	The learning rate used by the optimizer
fusion_head	The type of fusion head to use to combine spectral bands
loss	The loss to use during training
batch_size	The size of batches used in training
epochs	The number of epochs to train for
apply_transfer	Whether or not to apply the PCT water transfer method
random_subsample	Whether or not to randomly sample patches for training
water_threshold	The threshold at which to stop transplanting water bodies if apply_transfer=true

Citation

Please cite our work if it is helpful for your research.

@article{rs15051253,
title={Water Body Extraction from Sentinel-2 Imagery with Deep Convolutional Networks and Pixelwise Category Transplantation},
author={Billson, Joshua and Islam, MD Samiul and Sun, Xinyao and Cheng, Irene},
journal={Remote Sensing},
volume={15},
year={2023},
number={5},
article-number={1253},
url={https://www.mdpi.com/2072-4292/15/5/1253},
issn={2072-4292},
doi={10.3390/rs15051253}
}