Comparison of Two Data Smoothing Techniques for Vegetation Spectra Derived From EO-1 Hyperion |
| |
| Authors: | Anshu Miglani Shibendu S Ray D P Vashishta Jai Singh Parihar |
| |
| Institution: | (1) ABHG/EPSA, Space Applications Center, ISRO, Ahmedabad, India;(2) H. N. B., Garhwal University, Srinagar, Uttarakhand |
| |
| Abstract: | Hyperspectral data are generally noisier compared to broadband multispectral data because their narrow bandwidth can only
capture very little energy that may be overcome by the self-generated noise inside the sensors. It is desirable to smoothen
the reflectance spectra. This study was carried out to see the effect of smoothing algorithms - Fast-Fourier Transform (FFT)
and Savitzky–Golay (SG) methods on the statistical properties of the vegetation spectra at varying filter sizes. The data
used in the study is the reflectance spectra data obtained from Hyperion sensor over an agriculturally dominated area in Modipuram
(Uttar Pradesh). The reflectance spectra were extracted for wheat crop at different growth stages. Filter sizes were varied
between 3 and 15 with the increment of 2. Paired t-test was carried out between the original and the smoothed data for all the filter sizes in order to see the extent of distortion
with changing filter sizes. The study showed that in FFT, beyond filter size 11, the number of locations within the spectra
where the smooth spectra were statistically different from its original counterpart increased to 14 and reaches 21 at the
filter size 15. However, in SG method, number of statistically different locations were more than those found in the FFT,
but the number of locations did not changing drastically. The number of statistically disturbed locations in SG method varied
between 16 and 19. The optimum filter size for smoothing the vegetation spectra was found to be 11 in FFT and 9 in SG method. |
| |
| Keywords: | |
| 本文献已被 SpringerLink 等数据库收录！ |
|
