NSCT与GS变换的资源三号卫星数据融合方法研究与应用
作者简介:秦善善(1989-),女,硕士生,主要从事遥感应用研究。E-mail:shanqinshan@163.com
收稿日期: 2013-11-18
要求修回日期: 2014-03-28
网络出版日期: 2014-11-01
基金资助
国家自然科学基金项目(41201441)
广西空间信息与测绘重点实验室基金资助课题(12077115-18)
A Research on Fusion Method for ZY-3 Satellite Data Based on NSCT and GS Transform
Received date: 2013-11-18
Request revised date: 2014-03-28
Online published: 2014-11-01
Copyright
资源三号卫星遥感数据是一种新型遥感影像,目前尚未有一种专门适用于资源三号卫星融合的方法。因此,提出了一种非下采样Contourlet变换(NSCT)与Gram-Schmidt(GS)变换相结合的融合方法,将非下采样Contourlet变换增强的空间信息用于补充具有高保真度Gram-Schmidt变换融合算法在影像清晰度方面的不足。采用线性回归的方法模拟低分辨率全色影像,将高分辨率全色影像、低分辨率全色影像及两者差值的细节影像分别进行非下采样Contourlet变换,对所得的高低频系数采取不同的融合策略进行自适应融合处理,得到新的全色影像。由低分辨率全色影像取代GS正变换第一分量,非下采样Contourlet变换得到的全色影像取代GS反变换第一分量,进行Gram-Schmidt正交变换,得到融合影像。与大多存在光谱扭曲的传统融合方法相比较,本文方法在光谱保真度、空间清晰度及地物分类精度方面都有明显的优势,说明该融合方法是一种适合资源三号卫星数据多光谱与全色影像融合的方法。
关键词: 资源三号卫星; 非下采样Contourlet变换; Gram-Schmidt变换; 支持向量机; 低空间分辨率全色影像; 融合
秦善善 , 王世新 , 周艺 , 王福涛 , 刘文亮 . NSCT与GS变换的资源三号卫星数据融合方法研究与应用[J]. 地球信息科学学报, 2014 , 16(6) : 949 -957 . DOI: 10.3724/SP.J.1047.2014.00949
ZY-3 satellite data, which is a new type of remote sensing imagery, has not yet owned a specific images fusion method. This paper proposed a new fusion algorithm based on the nonsubsampled Contourlet transform and Gram-Schmidt (GS) transform according to the characteristics of CBERS-03 data. Firstly, the linear regression method was used to generate the low-resolution panchromatic image with respect to the low-frequency pixels. Then, the nonsubsampled Contourlet transform was applied respectively on the high-resolution panchromatic image, the low-resolution panchromatic image, and the image generated from their differences, to generate high frequency and low frequency coefficients. With these coefficients, a new panchromatic image was produced by using adaptive fusion strategies. At last, the fused image was obtained through GS orthogonal transformation, by replacing the first component of GS positive transformation with the low-resolution panchromatic image, and replacing the first component of GS inverse transformation with the newly generated panchromatic image. Compared with the traditional image fusion methods, which always have defects of spectral distortion, the proposed method had obvious advantages in terms of spectral fidelity, spatial resolution and classification accuracy. In conclusion, the proposed method is appropriate for images fusion on the ZY-3 multispectral and high resolution data.
Fig. 1 Algorithm flow chart图1 算法流程图 |
Fig. 2 Low-resolution panchromatic image图2 低分辨全色影像 |
Fig. 3 New high-resolution panchromatic image图3 新高空间分辨率全色影像 |
Fig. 4 Comparison of a partial view of LowPan and NewPan Images图4 低分辨率全色影像与新高分辨率全色影像局部对比图 |
Fig. 5 Fused images with NSCT_GS图5 NSCT_GS融合影像 |
Fig. 6 Fused images with different methods图6 不同融合算法得到的融合影像 |
Fig. 7 Bar chart of classification accuracy图7 分类精度柱状图 |
Tab. 1 The fusion results evaluation parameter list表1 融合结果评价参数表 |
融合方法 | 评价方法 | |||||
---|---|---|---|---|---|---|
Bia | UIQI | corre | PSNR | std | entropie | |
QIHS | 26.157 | 0.926 | 0.882 | 155.740 | 92.902 | 7.239 |
GS | 30.288 | 0.935 | 0.921 | 170.177 | 71.126 | 6.937 |
LP | 19.874 | 0.950 | 0.963 | 169.605 | 79.281 | 6.954 |
WAIHS | 18.833 | 0.954 | 0.967 | 172.427 | 91.118 | 6.824 |
PCA | 46.681 | 0.919 | 0.921 | 145.557 | 84.811 | 7.532 |
NSCT-GS | 13.376 | 0.958 | 0.966 | 172.748 | 91.947 | 6.992 |
Tab. 2 The classification accuracy of various types of surface features表2 各地物分类精度表 |
融合方法 | 分类精度(%) | ||||
---|---|---|---|---|---|
建筑物 | 水体 | 耕地 | 经济作物 | 森林 | |
NSCT-GS | 82.55 | 100.00 | 94.82 | 96.01 | 99.98 |
PCA | 76.45 | 99.76 | 92.97 | 98.54 | 96.70 |
QIHS | 67.13 | 99.51 | 87.49 | 97.75 | 99.74 |
WAIHS | 71.69 | 99.10 | 86.37 | 93.19 | 99.70 |
The authors have declared that no competing interests exist.
[1] |
|
[2] |
|
[3] |
|
[4] |
|
[5] |
|
[6] |
|
[7] |
|
[8] |
|
[9] |
|
[10] |
|
[11] |
|
[12] |
|
[13] |
|
[14] |
|
[15] |
|
[16] |
|
[17] |
|
[18] |
|
[19] |
|
[20] |
|
/
〈 | 〉 |