针对卫星云图数据量大，但传输通道和存储空间相对狭小的问题，本文基于Tetrolet变换，利用相邻时次云图的时空相关性，实现了一种高重构质量的卫星云图压缩感知方法。该方法将善于表达图像方向纹理及边缘信息的Tetrolet变换引入压缩感知的稀疏表示环节，从而很好地体现了卫星云图细节丰富、纹理结构复杂的特性；同时，考虑到卫星云图序列间的相关性，将时间相邻的卫星云图组成图像组，以中间时刻云图作为参考图像，计算其与相邻时次云图的差异，通过在参考图片及序列差异图片间合理分配采样率，获取测量数据，在压缩感知框架下，采用带平滑处理的投影Landweber算法，重构出相邻时次的图像组。实验结果表明，Tetrolet变换适用于卫星云图的稀疏表示，而且图像组时空相关性的利用可显著提高重构云图的视觉效果及客观评价指标；在采样率低于0.2时，红外1、水汽和可见光3个通道重构云图的峰值信噪比（PSNR）较传统方法平均提高了7.48 dB，13.51 dB和6.15 dB。由此可见，本文方法可以通过获取少数随机测量值，重构出高质量的卫星云图，不仅为云图数据的低比特率压缩提供了一种可行的解决方案，而且对于其他序列图像的压缩采样具有借鉴意义。

This study aims to balance the contradiction caused by large satellite cloud image data and the limitations of transmission channels and storage space. We proposed a compressed sensing approach with highly reconstructive quality that utilizes temporal-spatial correlation of adjacent sequence cloud images based on Tetrolet transform. The approach introduces Tetrolet transform into the sparse representation step of compressed sensing for its remarkable representation of important image information, such as directional texture and edge, which can depict the detail and texture structure of satellite cloud images well. Meanwhile, considering the correlation between sequences of the satellite images, three adjacent cloud images are bound to form a Group Of Pictures (GOP), in which the intermediate time cloud image is the reference picture, and the difference with its adjacent cloud images is calculated to form the difference images. Finally, GOP is reconstructed by the projected Landweber algorithm with smoothing operation under the compressed sensing framework. Experimental results show that Tetrolet transform is appropriate for the sparse representation of satellite cloud images. The use of the temporal-spatial correlation of GOP can improve the visual quality and quantitative measurements of reconstructed cloud images. When the sampling rate for cloud images of infrared one, water vapor, and visible light channel is lower than 0.2, the Peak Signal-to-Noise Ratio (PSNR) of the reconstructed images is improved by an average of 7.48 dB, 13.51 dB and 6.15 dB, respectively, compared with that obtained by traditional methods. High-quality cloud image can be reconstructed by obtaining random measurements using the proposed method. This method provides a feasible solution for the low-bit-rate compression of cloud images and is significant for the compressed sampling of other sequence images.