基于降维正则化多项式的光谱反射率重建方法 下载: 1091次
Spectral Reflectance Reconstruction Based on Dimension Reduction Regularization Polynomials
1 西安建筑科技大学管理学院, 陕西 西安 710055
2 西安建筑科技大学信息与控制工程学院, 陕西 西安 710055
图 & 表
图 1. 多光谱相机
Fig. 1. Multispectral camera
下载图片 查看原文
图 2. 滤光片的透射率曲线
Fig. 2. Transmissivity curve of filter
下载图片 查看原文
图 3. 训练样本个数与重建光谱反射率的RMSE平均值之间的关系
Fig. 3. Relationship between training sample number and mean RMSE of reconstructed spectral reflectance
下载图片 查看原文
图 4. 利用L曲线方法得到的正则化参数。(a)训练结果1;(b)训练结果2
Fig. 4. Regularization parameters obtained by L-curve method. (a) Training result 1; (b) training result 2
下载图片 查看原文
图 5. 3种重建方法对20个测试样本的均方根误差
Fig. 5. RMSE of 20 testing samples with three reconstruction methods
下载图片 查看原文
图 6. 3种重建方法的光谱反射率曲线。(a) xISSD=0.022;(b) xISSD=0.121;(c) xISSD=0.351
Fig. 6. Spectral reflectance curves with three reconstruction methods. (a) xISSD=0.022; (b) xISSD=0.121; (c) xISSD=0.351
下载图片 查看原文
图 7. 壁画参考色块及多光谱图像。 (a)壁画参考色块标记;(b) 11个通道的多光谱图像
Fig. 7. Mural referential color patches and multispectral images. (a) Markings of mural referential color patches; (b) multispectral images with 11 channels
下载图片 查看原文
图 8. 不同重建方法得到的壁画参考色块的CIELAB色度分布空间
Fig. 8. CIELAB chromaticity distribution space of mural referential color patches obtained by different reconstruction methods
下载图片 查看原文
图 9. 壁画中6个参考色块的重建光谱曲线与实际光谱曲线。(a)色块1;(b)色块2;(c)色块3;(d)色块4;(e)色块5;(f)色块6
Fig. 9. Reconstructed and measured spectral reflectance curves of six referential color patches of mural. (a) Color patch 1; (b) color patch 2; (c) color patch 3; (d) color patch 4; (e) color patch 5; (f) color patch 6
下载图片 查看原文
表 1不同主成分个数时的累积贡献率、色差和均方根误差
Table1. CVC, color difference, and RMSE with different principal component numbers
k | CVC /% | Mean ΔE | Mean RMSE |
---|
5 | 98.87 | 0.717 | 0.082 | 6 | 99.95 | 0.430 | 0.058 | 7 | 99.89 | 0.424 | 0.042 | 8 | 99.99 | 0.263 | 0.033 |
|
查看原文
表 2不同多项式项数时的色差和均方根误差
Table2. Color difference and RMSE with different polynomial term numbers
n | Mean ΔE | Mean RMSE |
---|
6 | 0.482 | 0.032 | 9 | 0.383 | 0.023 | 12 | 0.361 | 0.020 | 14 | 0.345 | 0.020 | 18 | 0.341 | 0.019 |
|
查看原文
表 3PCA、PRE和DRRP方法的光谱反射率重建精度
Table3. Spectral reflectance reconstruction accuracies of PCA, PRE and DRRP methods
Method | RMSE | GFC /% | ISSD | ΔE |
---|
Mean | Minimum | Maximum | Mean | Minimum | Maximum | Mean | Minimum | Maximum | Mean | Minimum | Maximum |
---|
PCA | 0.058 | 0.012 | 0.106 | 98.20 | 93.53 | 99.63 | 0.146 | 0.064 | 0.637 | 0.430 | 0.172 | 0.931 | PRE | 0.020 | 0.011 | 0.042 | 98.17 | 92.53 | 99.74 | 0.160 | 0.021 | 0.556 | 0.361 | 0.064 | 0.766 | DRRP | 0.018 | 0.007 | 0.032 | 99.51 | 98.81 | 99.96 | 0.106 | 0.011 | 0.359 | 0.283 | 0.060 | 0.412 |
|
查看原文
表 4壁画中6个参考色块的光谱反射率重建精度
Table4. Spectral reflectance reconstruction accuracies of six referential color patches of mural
Number of referentialcolor patches | DRRP method | PRE method | PCA method |
---|
RMSE | GFC /% | ISSD | RMSE | GFC /% | ISSD | RMSE | GFC /% | ISSD |
---|
1 | 0.0123 | 99.63 | 0.0931 | 0.0145 | 98.43 | 0.1429 | 0.0562 | 97.78 | 0.1325 | 2 | 0.0112 | 99.21 | 0.1568 | 0.0178 | 98.15 | 0.2047 | 0.0593 | 97.01 | 0.1930 | 3 | 0.0317 | 99.18 | 0.1613 | 0.0347 | 98.07 | 0.2163 | 0.0712 | 96.81 | 0.2013 | 4 | 0.0194 | 99.42 | 0.1236 | 0.0239 | 98.21 | 0.1745 | 0.0674 | 97.23 | 0.1701 | 5 | 0.0376 | 99.09 | 0.1702 | 0.0462 | 98.82 | 0.2213 | 0.0820 | 97.86 | 0.2113 | 6 | 0.0115 | 99.78 | 0.1072 | 0.0177 | 98.65 | 0.1586 | 0.0489 | 98.01 | 0.1489 | Mean | 0.0206 | 99.39 | 0.1353 | 0.0258 | 98.39 | 0.1864 | 0.0675 | 97.45 | 0.1761 |
|
查看原文
王可, 王慧琴, 龙艳群, 王伟超, 赵丽娟, 杨蕾. 基于降维正则化多项式的光谱反射率重建方法[J]. 激光与光电子学进展, 2018, 55(5): 053004. Ke Wang, Huiqin Wang, Yanqun Long, Weichao Wang, Lijuan Zhao, Lei Yang. Spectral Reflectance Reconstruction Based on Dimension Reduction Regularization Polynomials[J]. Laser & Optoelectronics Progress, 2018, 55(5): 053004.