diana21170648/cloudnet-ds-theory

Research on Multi-temporal Cloud Removal Using D-S Evidence Theory and Cloud Segmentation Model

Cloudnet-DS-theory

Research on Multi-temporal Cloud Removal Using D-S Evidence Theory and Cloud Segmentation Model.

We proposed a D-S evidence theory based multi-temporal cloud removal method and Cloud Segmentation Model, which is introduced in "Research on Multi-temporal Cloud Removal Using D-S Evidence Theory and Cloud Segmentation Model". We applied cloud-net[1] in our method. If you want to train the network yourself, you can find information in here. The pretrained weights can be downloaded here. The data we used were provided by government, so we can't publish our data. We are sorry for that.

Requirements

Python 3.6
Tensorflow 1.9.0, 1.10.0, 1.12.0
Keras 2.2.4
Scikit-image 0.15.0
*If you are using latest GPU like RTX 3090, loading pretrained weights may cost long time.

How to use

python main_test.py

How to remove cloud in images

1. Collect a series of cloudy pictures in same location.
2. Apply 'Cloud-Net\single_test' to get cloud segmentation results.
3. Follow instructions in D-S.ipynb to get final result.

This is not the final version of our method, we will improve it afterwards.

[1] S. Mohajerani and P. Saeedi, "Cloud-Net: An End-To-End Cloud Detection Algorithm for Landsat 8 Imagery," IGARSS 2019 - 2019 IEEE International Geoscience and Remote Sensing Symposium, Yokohama, Japan, 2019, pp. 1029-1032. doi: 10.1109/IGARSS.2019.8898776. Arxive URL: https://arxiv.org/pdf/1901.10077.pdf, IEEE URL: URL: http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=8898776&isnumber=8897702