| 摘要: |
| 针对现有的高光谱图像压缩感知重构算法对图像的空谱特性利用不够充分,导致重构图像质量不够高的问题，提出了一种高光谱图像变投影率分块压缩感知结合优化谱间预测重构方案。编码端以频段聚类方式将高光谱图像的所有频段分成参考频段和普通频段，对不同频段单独采用不同精度分块压缩感知以获取高光谱数据。在解码端，参考频段直接采用稀疏度自适应匹配追踪(SAMP)算法重构，对于普通频段，则设计了一种优化谱间预测结合SAMP算法的新模型进行重构：首先通过重构的参考频段双向预测普通频段，并对其进行压缩投影，然后计算预测前后普通频段投影值的残差，最后利用SAMP算法重构该残差，以此修正预测值。实验表明，相比同类算法，该算法充分考虑了高光谱图像的空谱特性，有效改善了重构图像质量，且编码复杂度低，易于硬件实现。 |
| 关键词: 高光谱图像 分块压缩感知 频段聚类 优化谱间预测 稀疏度自适应匹配追踪 |
| DOI： |
|
| 基金项目: |
|
| Hyperspectral image compressed sensing reconstruction based on spatial-spectral characteristics |
| CHEN Shanxue,HU Can,QU Longyao |
| () |
| Abstract: |
| The existing hyperspectral image compressed sensing reconstruction algorithm can not fully utilize the spatial-spectral characteristic of image so that the quality of the reconstructed image is not high enough.For this problem,a new compression scheme for hyperspectral images is proposed which is based on variable projection rate sub block compressive sensing and reconstruction of optimized inter spectral prediction.At the encoder,all bands of the hyperspectral image is divided into some reference bands and common bands by band clustering,different bands are used to separate the compressed sensing with different precision in order to obtain hyperspectral data.At the decoder,the reference band is reconstructed by using sparsity adaptive matching pursuit(SAMP) algorithm,and for reconstruction of the common band,a new model of optimized inter spectral prediction combined with SAMP algorithm is designed:firstly,the common band is predicted by means of the reconstructed reference band,and it is compressed and projected,then the residual error of the projection value of prediction before and after is calculated for the common band,finally,the SAMP algorithm is used to reconstruct the residual error,which is used to correct the prediction value.Experimental results show that compared with similar algorithms,the proposed algorithm fully considers the spatial-spectral characteristics of hyperspectral images,effectively improves the quality of reconstructed image,and the complexity of encoding is low,and the hardware implementation is easy. |
| Key words: hyperspectral image block compressed sensing band clustering optimized inter spectral prediction sparsity adaptive matching pursuit |
《电讯技术》编辑部 