复杂背景下提出的改进卷积神经网络云检测方法
An Improved Convolutional Neural Network Cloud Detection Method Proposed in Complex Background
DOI: 10.12677/CSA.2024.143059, PDF,    国家自然科学基金支持
作者: 高 琳*, 盖晨曦, 田育星：沈阳理工大学信息科学与工程学院，辽宁 沈阳；芦偲俊：中冶沈勘工程技术有限公司，辽宁 沈阳
关键词: 遥感；深度学习；编码解码结构；云提取；Remote Sensing； Deep Learning； Encoding and Decoding Structure； Cloud Extraction
摘要: 对遥感影像云检测面临的下垫面复杂、厚云与雪的光谱特征相似导致同谱异物等复杂问题导致的传统的单类地物提取方法在云提取上效果不佳，提出了一种复杂背景下的改进卷积神经网络的遥感云检测方法MSANet。通过在浅层加入膨胀卷积以扩大首次感知野范围，同时关注整体结构特征传递，在解码部分加入包含多头“软”注意力的空间信息建模模块，增强了网络对全局信息的感知能力。使用ZY-3、38-cloud、GF1_WHU数据集上进行验证实验，实验结果表明该方法在复杂背景下云雪混淆的遥感影像上表现优秀，模型能够更好地应对复杂背景下的云检测任务，有效的提高了模型的精度。
Abstract: In view of the complex underlying surface of remote sensing image cloud detection and the similar spectral characteristics of thick clouds and snow, which lead to the same spectrum foreign bodies, the traditional single-class ground object extraction method is not effective in cloud extraction. An improved convolutional neural network remote sensing cloud detection method MSANet under complex background is proposed. By adding expansive convolution in the shallow layer to expand the range of the first perception field, and paying attention to the overall structural feature transmission, the spatial information modeling module containing multiple “soft” attention is added to the decoding part, which enhances the perception ability of the network to the global information. Verification experiments were carried out on ZY-3, 38-cloud and GF1_WHU data sets. The experi-mental results show that the proposed method performs well on remote sensing images with cloud snow confusion under complex background, and the model can better cope with cloud detection tasks under complex background, effectively improving the accuracy of the model.
文章引用：高琳, 盖晨曦, 芦偲俊, 田育星. 复杂背景下提出的改进卷积神经网络云检测方法[J]. 计算机科学与应用, 2024, 14(3): 66-77. https://doi.org/10.12677/CSA.2024.143059

参考文献

[1] Li, X., Zheng, H., Han, C., et al. (2021) SFRS-Net: A Cloud-Detection Method Based on Deep Convolutional Neural Networks for Gf-1 Remote-Sensing Images. Remote Sensing, 13, Article 2910. [Google Scholar] [CrossRef] 
[2] 刘广进, 王光辉, 毕卫华, 等. 基于DenseNet与注意力机制的遥感影像云检测算法[J]. 自然资源遥感, 2022, 34(2): 88-96.
[3] 张家强, 李潇雁, 李丽圆, 等. 基于深度残差全卷积 网络的Landsat8遥感影像云检测方法[J]. 激光与光电子学进展, 2020, 57(10): 364-371.
[4] Shi, C., Zhou, Y., Qiu, B., et al. (2021) CloudU-Net: A Deep Convolutional Neural Network Architecture for Daytime and Nighttime Cloud Images’ Segmentation. IEEE Geoscience and Remote Sensing Letters, 18, 1688-1692. [Google Scholar] [CrossRef] 
[5] Ghaffarian, S. and Valente, J. (2021) Effect of Attention Mecha-nism in Deep Learning-Based Remote Sensing Image Processing: A Systematic Literature Review. Remote Sensing, 13, Article 2965. [Google Scholar] [CrossRef] 
[6] Guo, Y., Cao, X., Liu, B., et al. (2020) Cloud Detection for Satellite Imagery Using Attention Based U-Net Convolutional Neural network. Symmetry, 12, Article 1056. [Google Scholar] [CrossRef] 
[7] Van, P., Olbrich, S. and Arns, M. (2018) Predicting Sex from Brain Rhythms with Deep Learning. Scientific Reports, 8, Article No. 3069. [Google Scholar] [CrossRef] [PubMed]
[8] Yang, J., Guo, J., Yue, H., et al. (2019) CDnet: CNN-Based Cloud Detection for Remote Sensing Imagery. IEEE Transactions on Geoscience and Remote Sensing, 57, 6195-6211. [Google Scholar] [CrossRef] 
[9] Singh, R., Biswas, M. and Pal, M. (2022) Cloud Detection Us-ing Sentinel 2 Imageries: A Comparison of XGBoost, RF, SVM, and CNN Algorithms. Geocarto International, 38, 1-32. [Google Scholar] [CrossRef] 
[10] Qiu, S., Zhu, Z. and He, B. (2019) Fmask 4.0: Im-proved Cloud and Cloud Shadow Detection in Landsats 4-8 and Sentinel-2 Imagery. Remote Sensing of Environment, 231, Article ID: 111205. [Google Scholar] [CrossRef]