论文部分内容阅读
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。
Aiming at the problem of large amount of satellite cloud image data and relatively small transmission channel and storage space, this paper implements a high-quality reconstructed satellite cloud compression sensing method based on Tetrolet transform and temporal-spatial correlation of adjacent sub-cloud images. This method introduces the Tetrolet transform, which is good at expressing the directional texture and the edge information of the image, into the sparse representation of the compressed sensing, thus well reflecting the feature of rich satellite imagery and complex texture structure. At the same time, taking into account the correlation between satellite cloud image sequences , The time-adjacent satellite cloud images are grouped into image groups, the intermediate time cloud images are used as reference images, and the differences between the sub-cloud images and adjacent sub-cloud images are calculated. The measurement data are obtained by reasonably allocating the sampling rates between the reference images and the sequence- Under the compressed sensing framework, the Landweber algorithm with smoothing projection is used to reconstruct the image group of adjacent times. The experimental results show that the Tetrolet transform is suitable for the sparse representation of satellite image, and the temporal and spatial correlation of image group can significantly improve the visual effect and objective evaluation of reconstructed image. When the sampling rate is less than 0.2, infrared 1, water vapor and visible light Compared with the traditional method, the peak signal-to-noise ratio (PSNR) of the three-channel reconstructed cloud graph increases by 7.48 dB, 13.51 dB and 6.15 dB on average. Thus, this method can reconstruct high-quality satellite image by taking a few random measurements, which not only provides a feasible solution for low bit-rate compression of cloud image data, but also has the advantages of Reference meaning.