当前的云检测方法未能充分利用遥感图像的光谱特征和空间特征。光谱信息的不充分利用会导致错分具有和云相似光谱特征的目标，而空间信息的不充分利用会导致碎云和薄云难以识别。基于此，提出一种融合遥感图像光谱和空间信息的新型云检测深度网络(SSFF-Net)。SSFF-Net首先利用1×1的卷积核提取遥感图像的光谱特征，其次将Transformer引入到遥感图像空间上的编解码来学习远距离的特征，充分利用遥感图像的光谱和空间信息。SSFF-Net克服了光谱特征提取依赖于经验性的线性组合，并能减少空间位置信息损失。将模型在Landsat 8 Biome以及AIR-CD数据集上进行评估，结果表明SSFF-Net具有较好的云检测效果，精度分别达到97%和96%。
Current cloud detection methods fail to fully utilize the spectral-spatial features of remote sensing images. Insufficient use of spectral information results in misclassification of cloud with similar spectral feature, and insufficient use of spatial information makes it difficult to identify broken clouds or thin clouds. Motivated by these issues, we propose a spectral-spatial feature fusion network (SSFF-Net) for cloud detection which leverages the spectral-spatial information of remote sensing images. Firstly, SSFF-Net extracts the spectral features of remote sensing images with the 1×1 convolution kernel, then Transformer-based spatial encoder and decoder is applied to learn long-distance feature, which fully exploits the spectral and spatial information of remote sensing images. In this way, SSFF-Net overcomes the problem that spectral feature extraction depends on the empirical linear combination, and further reduces the loss of spatial position information. We evaluate our proposed model on the Landsat8 Biome and AIR-CD datasets. The results show that SSFF-Net has a good performance for cloud detection, with an accuracy of 97% and 96%, respectively.
[1] Zhang Y C, Rossow W B, Lacis A A, et al. Calculation of radiative fluxes from the surface to top of atmosphere based on ISCCP and other global data sets:refinements of the radiative transfer model and the input data[J]. Journal of Geophysical Research:Atmospheres, 2004, 109(D19):D19105. DOI:10.1029/2003JD004457.
[2] Sun L, Wei J, Wang J, et al. A universal dynamic threshold cloud detection algorithm (UDTCDA) supported by a prior surface reflectance database[J]. Journal of Geophysical Research:Atmospheres, 2016, 121(12):7172-7196. DOI:10.1002/2015JD024722.
[3] 刘庆凤, 刘吉平, 宋开山. 基于MODIS/NDVI时序数据的土地覆盖分类[J]. 中国科学院研究生院学报, 2010, 27(2):163-169. DOI:10.7523/j.issn.2095-6134.2010.2.003.
[4] 段雅鸣, 张锦水, 朱爽. 基于深度卷积神经网络的云检测方法[J]. 测绘通报, 2021(4):33-39. DOI:10.13474/j.cnki.11-2246.2021.0107.
[5] Zhu Z, Woodcock C E. Object-based cloud and cloud shadow detection in Landsat imagery[J]. Remote Sensing of Environment, 2012, 118:83-94. DOI:10.1016/j.rse.2011.10.028.
[6] Zhai H, Zhang H Y, Zhang L P, et al. Cloud/shadow detection based on spectral indices for multi/hyperspectral optical remote sensing imagery[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2018, 144:235-253. DOI:10.1016/j.isprsjprs.2018.07.006.
[7] Frantz D, Haß E, Uhl A, et al. Improvement of the Fmask algorithm for Sentinel-2 images:separating clouds from bright surfaces based on parallax effects[J]. Remote Sensing of Environment, 2018, 215:471-481. DOI:10.1016/j.rse.2018.04.046.
[8] Oishi Y, Ishida H, Nakamura R. A new Landsat 8 cloud discrimination algorithm using thresholding tests[J]. International Journal of Remote Sensing, 2018, 39(23):9113-9133. DOI:10.1080/01431161.2018.1506183.
[9] 张永宏, 蔡朋艳, 陶润喆, 等. 基于改进U-Net网络的遥感图像云检测[J]. 测绘通报, 2020(3):17-20,34. DOI:10.13474/j.cnki.11-2246.2020.0070.
[10] Jeppesen J H, Jacobsen R H, Inceoglu F, et al. A cloud detection algorithm for satellite imagery based on deep learning[J]. Remote Sensing of Environment, 2019, 229:247-259. DOI:10.1016/j.rse.2019.03.039.
[11] Chai D F, Newsam S, Zhang H K, et al. Cloud and cloud shadow detection in Landsat imagery based on deep convolutional neural networks[J]. Remote sensing of environment, 2019, 225:307-316. DOI:10.1016/j.rse.2019.03.007.
[12] Li Z W, Shen H F, Cheng Q, et al. Deep learning based cloud detection for medium and high resolution remote sensing images of different sensors[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2019, 150:197-212. DOI:10.1016/j.isprsjprs.2019.02.017.
[13] 赵敏钧, 赵亚伟, 赵雅捷, 等. 一种新的基于深度学习的重叠关系联合抽取模型[J]. 中国科学院大学学报. DOI:10.7523/j.ucas.2020.0026.
[14] Sha Y, Zhang Y, Ji X, et al. Transformer-Unet:raw image processing with unet[EB/OL]. arXiv:2109.08417v1. (2021-09-17)[2021-11-11]. https://arxiv.org/abs/2109.08417v1.
[15] Vaswani A, Shazeer N, Parmar N, et al. Attention is all you need[EB/OL]. arXiv:1706.03762v5. (2017-12-06)[2021-11-11]. https://arxiv.org/abs/1706.03762v5.
[16] Dosovitskiy A, Beyer L, Kolesnikov A, et al. An image is worth 16x16 words:transformers for image recognition at scale[EB/OL]. arXiv:2010.11929v2. (2021-06-03)[2021-11-11]. https://arxiv.org/abs/2010.11929v2.
[17] Carion N, Massa F, Synnaeve G, et al. End-to-end object detection with transformers[C]//Computer Vision-ECCV 2020, 2020:213-229. DOI:10.1007/978-3-030-58452-8_13.
[18] Liu Z, Lin Y T, Cao Y, et al. Swin transformer:hierarchical vision transformer using shifted windows[EB/OL]. arXiv:2103.14030v2. (2021-08-17)[2021-11-11]. https://arxiv.org/abs/2103.14030v2.
[19] Foga S, Scaramuzza P L, Guo S, et al. Cloud detection algorithm comparison and validation for operational Landsat data products[J]. Remote Sensing of Environment, 2017, 194:379-390. DOI:10.1016/j.rse.2017.03.026.
[20] He Q B, Sun X, Yan Z Y, et al. DABNet:deformable contextual and boundary-weighted network for cloud detection in remote sensing images[J]. IEEE Transactions on Geoscience and Remote Sensing, 5474, PP(99):1-16. DOI:10.1109/TGRS.2020.3045474.
