Self-Attentive Generative Adversarial Network for Cloud Detection in High Resolution Remote Sensing Images
Self-Attentive Generative Adversarial Network for Cloud Detection in High Resolution Remote Sensing Images
复制标题
用于高分辨率遥感图像云检测的自注意力生成对抗网络
DOI：
10.1109/lgrs.2019.2955071
复制
发表时间：
2020-10
影响因子：
4.8
通讯作者：
Matthieu Molinier
中科院分区：
文献类型：
--
作者：
Jun Li;Yisong Wang;Zhaocong Wu;Zhongwen Hu;Matthieu Molinier
云检测是遥感图像处理中的一个重要环节。大多数基于卷积神经网络（CNN）的云检测方法需要像素级标签，这是耗时且昂贵的注释。为了克服这一挑战，这封信提出了一种新的半监督云检测算法，通过训练一个自关注生成对抗网络（SAGAN）来提取云图像和无云图像之间的特征差异。我们的主要思想是将视觉注意力引入到生成"真实的"无云图像的过程中。SAGAN的训练基于三个指导原则：扩展云区域的注意力地图，将其替换为翻译的无云图像，减少注意力地图以符合云边界，以及优化自我关注网络以处理极端情况。SAGAN训练的输入是图像和图像级标签，这比基于CNN的现有方法更容易，更便宜，更省时。为了测试SAGAN的性能，在Sentinel-2A Level 1C图像数据上进行了实验。实验结果表明，该方法仅使用训练样本的图像级标签就取得了很好的效果。
Cloud detection is an important step in the processing of remote sensing images. Most methods based on convolutional neural networks (CNNs) for cloud detection require pixel-level labels, which are time-consuming and expensive to annotate. To overcome this challenge, this letter proposes a novel semisupervised algorithm for cloud detection by training a self-attentive generative adversarial network (SAGAN) to extract the feature difference between cloud images and cloud-free images. Our main idea is to introduce visual attention into the process of generating “real” cloud-free images. The training of SAGAN is based on three guiding principles: expansion of attention maps of cloud regions which will be replaced with translated cloud-free images, reduction of attention maps to coincide with cloud boundaries, and optimization of a self-attentive network to handle the extreme cases. The inputs for SAGAN training are the images and image-level labels, which are easier, cheaper, and more time-saving than the existing methods based on CNN. To test the performance of SAGAN, experiments are conducted on the Sentinel-2A Level 1C image data. The results show that the proposed method achieves very promising results with only the image-level labels of training samples.
登录
查看更多内容
影响因子：
13.5
作者：
Robert F. Cahalan;L. Oreopoulos;L. Oreopoulos;G. Wen;G. Wen;A. Marshak;A. Marshak;S. Tsay;T. DeFelice
通讯作者：
Robert F. Cahalan;L. Oreopoulos;L. Oreopoulos;G. Wen;G. Wen;A. Marshak;A. Marshak;S. Tsay;T. DeFelice
DOI：
10.1109/jstars.2017.2686488
发表时间：
2017-08-01
影响因子：
5.5
作者：
Xie, Fengying;Shi, Mengyun;Zhao, Danpei
通讯作者：
Zhao, Danpei
影响因子：
4.8
作者：
Y. Zhan;Jian Wang;Jianping Shi;Guangliang Cheng;Lele Yao;Weidong Sun
通讯作者：
Y. Zhan;Jian Wang;Jianping Shi;Guangliang Cheng;Lele Yao;Weidong Sun
影响因子：
5
作者：
Fisher, Adrian
通讯作者：
Fisher, Adrian
DOI：
10.1109/igarss.2017.8127438
发表时间：
2017-07
期刊：
2017 IEEE International Geoscience and Remote Sensing Symposium (IGARSS)
影响因子：
--
作者：
Gonzalo Mateo-García;L. Gómez-Chova;Gustau Camps-Valls
通讯作者：
Gonzalo Mateo-García;L. Gómez-Chova;Gustau Camps-Valls