基于空间域压缩采样和谱域Karhunen-Loève变换的光谱成像与重构

唐意东; 黄树彩; 黄达

doi:doi:10.3788/AOS201838.0530004

光学学报, 2018, 38 (5): 0530004, 网络出版: 2018-08-30

基于空间域压缩采样和谱域Karhunen-Loève变换的光谱成像与重构下载： 654次

Spectral Imaging and Reconstruction Based on Spatial Compressive Sampling and Spectral Karhunen-Loève Transform

唐意东黄树彩 ^*黄达

作者单位

空军工程大学防空反导学院, 陕西西安 710051

图 & 表

图 1. Flow chart of 2D-CRPG algorithm

Fig. 1.

下载图片查看原文

图 2. 基于空域压缩采样和谱域KL变换的光谱成像

Fig. 2. Spectral imaging based on spatial compressive sampling and spectral KL transform

下载图片查看原文

图 3. 光谱图像各波段信噪比

Fig. 3. SNR of each band of spectral image

下载图片查看原文

图 4. 不同 Rspa 时2D-CRPG算法重构光谱图像各波段的PSNR。(a) K =10; (b) K =150

Fig. 4. PSNR of each band of reconstructed spectral image for 2D-CRPG algorithm with different Rspa . (a) K =10; (b) K =150

下载图片查看原文

图 5. K= 150条件下2D-CRPG算法在不同 Rspa 时的二维图像重构效果

Fig. 5. Two-dimensional image reconstruction effect of 2D-CRPG algorithm with different Rspa under condition of K= 150

下载图片查看原文

图 6. 不同 K 值时2D-CRPG算法重构光谱图像各波段的PSNR。(a) Rspa =40%; (b) Rspa =90%

Fig. 6. PSNR of each band of reconstructed spectral image for 2D-CRPG algorithm with different K values. (a) Rspa =40%; (b) Rspa =90%

下载图片查看原文

图 7. 不同算法重构光谱图像各波段的PSNR。(a) Rspa =40%; (b) Rspa =50%; (c) Rspa =60%; (d) Rspa =70%; (e) Rspa =80%; (f) Rspa =90%

Fig. 7. PSNR of each band of reconstructed spectral image for different algorithms. (a) Rspa =40%; (b) Rspa =50%; (c) Rspa =60%; (d) Rspa =70%; (e) Rspa =80%; (f) Rspa =90%

下载图片查看原文

图 8. Rspa =40%时不同算法重构光谱图像的假色图(波段250、222、10)。(a) 原图; (b) BCS, aAPSNR =28.86 dB; (c) CSC, aAPSNR =27.08 dB; (d) 2D-OMP, aAPSNR =26.41 dB; (e) Kro-OMP, aAPSNR =19.05 dB; (f) 2D-CRPG, aAPSNR =33.14 dB

Fig. 8. False color images of reconstructed spectral images for different algorithms with Rspa =40% (bands 250, 222, 10). (a) Original image; (b) BCS, aAPSNR =28.86 dB; (c) CSC, aAPSNR =27.08 dB; (d) 2D-OMP, aAPSNR =26.41 dB; (e) Kro-OMP, aAPSNR =19.05 dB; (f) 2D-CRPG, aAPSNR =33.14 dB

下载图片查看原文

图 9. Rspa =90%时不同方法重构光谱图像的假色图(波段250、222、10)。(a) 原图; (b) BCS, aAPSNR =34.54 dB; (c) CSC, aAPSNR =32.95 dB; (d) 2D-OMP, aAPSNR =35.41 dB; (e) Kro-OMP, aAPSNR =23.10 dB; (f) 2D-CRPG, aAPSNR =42.81 dB

Fig. 9. False color images of reconstructed spectral images for different algorithms with Rspa =90% (bands 250, 222, 10). (a) Original image; (b) BCS, aAPSNR =34.54 dB; (c) CSC, aAPSNR =32.95 dB; (d) 2D-OMP, aAPSNR =35.41 dB; (e) Kro-OMP, aAPSNR =23.10 dB; (f) 2D-CRPG, aAPSNR =42.81 dB

下载图片查看原文

图 10. 不同 Rspa 时不同算法的APSNR

Fig. 10. APSNR of different algorithms with different Rspa

下载图片查看原文

唐意东, 黄树彩, 黄达. 基于空间域压缩采样和谱域Karhunen-Loève变换的光谱成像与重构[J]. 光学学报, 2018, 38(5): 0530004. Yidong Tang, Shucai Huang, Da Huang. Spectral Imaging and Reconstruction Based on Spatial Compressive Sampling and Spectral Karhunen-Loève Transform[J]. Acta Optica Sinica, 2018, 38(5): 0530004.