Continuous Urban Change Detection from Satellite Image Time Series

This repository contains the official code for the following paper:

S. Hafner, H. Fang, H. Azizpour and Y. Ban, "Continuous Urban Change Detection from Satellite Image Time Series with Temporal Feature Refinement and Multi-Task Integration," (accepted to TGRS).

Setup

This section shows you how to setup the dataset and the virtual environment.

Dataset

The SpaceNet 7 dataset is available from this link. More information can be found in the SpaceNet 7 dataset paper.

We use a metadata file (metadata_conturbancd.json) that can be downloaded from here or from this repository. The metadata file should be placed in the root directory of the SpaceNet 7 dataset (see below).

We also generated raster labels using the code in this paper. The labels can be obtained from this link and should be placed in the respective study site folder.

The dataset directory should look like this:

$ SpaceNet 7 dataset directory
spacenet7  # -d should point to this directory
├── metadata_conturbancd.json  # Download this file and place it in the dataset directory
└── train
    ├── L15-0331E-1257N_1327_3160_13
    ...
    └── L15-1848E-0793N_7394_5018_13
        ├── images_masked
        ├── labels
        └── labels_raster  # Generated using the vector data in labels

Environment

1. Clone this repository

   git clone https://github.com/SebastianHafner/ContUrbanCD.git
   cd ContUrbanCD

2. Setup a virtual environment

   We use conda to setup the environment:

   conda create -n conturbancd python=3.9.7
   conda activate conturbancd

3. Install the dependencies

   Install pytorch according to the official guide.

   Install other dependencies using the requirements.txt file. Note that not all libraries specified in the file are required to run this code.

   pip install -r requirements.txt

Running our code

This section provides all the instructions to run our code. If you do not want to train your own models, we also provide our model weights (trained using this code base).

All scripts (train.py, evaluation.py, inference.py) require three arguments:

• The config file is specified using -c. This repo includes dataset-specific configs for the proposed method.
• The output directory is specified using -o. We use this directory to store model weights and evaluation and inference outputs.
• The dataset directory is specified using -d. This directory points to the root folder of the dataset.

Training

To train our network on SpaceNet 7, run the following script:

python train.py -c conturbancd_sn7 -o output -d spacenet7

Evaluation

To calculate the evaluation metrics, run this script:

python eval.py -c conturbancd_sn7 -o output -d spacenet7

The script outputs a .json file containing accuracy values.

Inference

To produce multi-temporal building segmentation outputs from a satellite image time series, run the following script:

python inference.py -c conturbancd_sn7 -o output -d spacenet7

Note: You have the option to set the edge setting in the MRF module using -e (degenerate, adjacent, cyclic, and dense). Dense is used as default option.

The resulting file (.np) is of dimension T x H x W with the first dimension denoting the temporal dimension.

seg_sits = np.load(output_file)  # T x H x X

# Building segmentations for second image in the time series
seg_img_t2 = seg_sits[1]

# Changes between first and last images of the time series
ch_first_last = np.not_equal(seg_sits[0], seg_sits[-1])

Credits

If you find this work useful, please consider citing:

@article{hafner2024continuous,
title={Continuous Urban Change Detection from Satellite Image Time Series with Temporal Feature Refinement and Multi-Task Integration},
author={Hafner, Sebastian and Fang, Heng and Azizpour, Hossein and Ban, Yifang},
journal={arXiv preprint arXiv:2406.17458},
year={2024}
}