RFL-CDNet: Towards Accurate Change Detection via Richer Feature Learning

This software implements RFL-CDNet: Towards Accurate Change Detection via Richer Feature Learning in PyTorch. For more details, please refer to our paper.

Abstract

​ Change Detection is a crucial but extremely challenging task of remote sensing image analysis, and much progress has been made with the rapid development of deep learning. However, most existing deep learning-based change detection methods mainly focus on intricate feature extraction and multi-scale feature fusion, while ignoring the insufficient utilization of features in the intermediate stages, thus resulting in sub-optimal results. To this end, we propose a novel framework, named RFL-CDNet, to utilize richer feature learning for change detection. Specifically, we improve the capability and utilization of feature learning via introducing deep supervision information at the intermediate stages. Furthermore, we design the Coarse-To-Fine Guiding (C2FG) module and the Learnable Fusion (LF) module to further improve feature learning and learn more discriminative feature representations. The C2FG module aims to seamlessly integrate the side output from previous coarse-scale into the current fine-scale prediction in a coarse-to-fine manner, while LF module assumes that the contribution of each stage and each spatial location is independent, thus designing a learnable module to fuse multiple predictions. Experiments on several benchmark datasets show that our proposed RFL-CDNet achieves state-of-the-art performance.

Installation

• Install PyTorch 1.7.1+ and other dependencies:

  pip/conda install pytorch>=1.7.1, tqdm, tensorboardX, opencv-python, pillow, numpy, sklearn

Run demo

• Generate the train.txt, val.txt and test.txt

  python write_path.py
  

• A demo program can be found in demo. Before running the demo, download our pretrained models and best models from Baidu Netdisk (Extraction code: emby) . Then launch demo by:

  python eval.py
  

Train a new model

• Generate the train.txt, val.txt and test.txt:

  python write_path.py

• Submit the train.sh:

  sbatch train.sh

Evaluation

• Run the following script to compute the evaluation metrics including P, R, and F1 for the model.

  python eval.py

• Run the following script to visualize the predicted change maps of the model.

  python visualization.py

Results

Here gives some examples of change detection results, comparing with existing methods on CDD Dataset in Figure (a), and Figure(b) is the results on WHU Dataset.

(a)	(b)

Evaluation of RFL-CDNet on different datasets with SNUNet, STANet, and DASNet as baseline:

Methods	P(%)	R(%)	F1-score(%)
FC-EF	60.29	62.98	61.61
FC-Siam-diff	62.51	65.24	63.85
FC-Siam-diff	64.81	56.42	60.33
STANet	62.75	69.47	65.94
DASNet	60.58	77.00	67.81
SNUNet	68.88	72.09	70.45
BIT	70.84	70.11	70.48
Ours	70.78	74.64	72.66

​ Tabel 1. WHU Cultivated Land Dataset

Methods	P(%)	R(%)	F1-score(%)
FC-EF	80.75	67.29	73.40
FC-Siam-diff	54.20	81.34	65.05
FC-Siam-diff	48.84	88.96	63.06
STANet	77.40	90.30	83.35
DASNet	83,77	91.02	87.24
SNUNet	91.28	87.25	89.22
BIT	86.64	81.48	83.98
Ours	93.02	90.62	91.80

​ Tabel 2. WHU Dataset

Methods	P(%)	R(%)	F1-score(%)
FC-EF	84.68	65.13	73.63
FC-Siam-diff	88.81	62.20	73.16
FC-Siam-diff	87.57	66.69	75.72
STANet	83.17	92.76	87.70
DASNet	93.28	89.91	91.57
SNUNet	92.40	90.13	91.25
BIT	94.86	95.32	95.09
Ours	96.32	96.46	96.39

​ Tabel 3. CDD Dataset

Dataset Preparation

• Dataset structure

  - whu_cultivated_land
      ├─train
      ├─val
      ├─train.txt
      └─val.txt
  - CDD
      ├─train
      ├─val
      ├─test
      ├─train.txt
      ├─val.txt
      └─test.txt
  - whu_building
      ├─train
      ├─test
      ├─train.txt
      └─test.txt
  

  train.txt, val.txt and test.txt contains the image pairs of each dataset. Each line is organized in /path/to/img_A /path/to/img_B /path/to/label.

• Data structure

  - whu_cultivated_land
      ├─A  # images of t1 phase
      ├─B  # images of t2 phase
      └─Out  # binary change maps
  

• Our Processed Dataset Download

  • Whu_cultivated land : BaiduYun (passward: whu1)

  • CDD : BaiduYun (passward: whu1)

  • Whu_Building : BaiduYun (passward: whu1)

Acknowledgements

The authors would like to thank the developers of PyTorch, SNUNet, STANet, and DASNet. Please feel free to contact us if you encounter any issues.