基于FastICA盲源分离法去除土壤干扰的小麦生物量高光谱估算

Applying fast independent component analysis algorithm of blind source separation method to remove soil effects on hyperspectral data for wheat biomass estimation

高光谱技术是一种快速无损监测植被生物量的有效方法,但土壤背景的干扰一直是生物量监测的主要限制因素之一。本研究试图利用盲源分离(blind source separation,BSS)法分离出净植被光谱,达到消除土壤背景影响,提高小麦生物量估算精度的目的。本研究对110组小麦冠层光谱数据进行快速独立分量分析(fast independent component analysis,Fast ICA)处理,提取净植被光谱,并对比了Fast ICA处理前后所建的偏最小二乘回归(partial least squares regression,PLSR)模型估算精度。结果表明:Fast ICA算法可有效分离土壤光谱和植被光谱;且基于净植被光谱建立的小麦生物量估算模型精度得到明显提升,建模集RPDc(ratio of performance to deviation of the calibration)和交叉验证集RPDcv(ratio of performance to deviation of the cross calibration)分别由原始光谱的1.83和1.64提高至2.77和2.09;可见,Fast ICA可以作为有效的光谱数据预处理方法,显著提高小麦生物量的估算精度,为利用遥感技术进行大尺度、精准监测生物量提供了方法支持和理论依据。

Hyperspectral technique has been an effective method to monitor the vegetation biomass as a rapid and nondestructive approach. However, the accuracy of biomass estimation is always limited by the influence of soil background. The purpose of this study aimed to alleviate the effects of soil on spectra and improve the accuracy of wheat biomass estimation based on the extracted vegetation spectra by blind source separation （BSS） method. In this study, with the application of fast independent component analysis （FastICA）, pure vegetation spectra were extracted from the 110 groups of original fieldobserved canopy spectra, and the wheat biomass estimation accuracy were compared before and after FastICA with the partial least squares regression （PLSR）. The results showed that the FastICA method could separate the soil spectra and vegetation spectra effectively, and the accuracy of wheat biomass estimation was significantly improved based on the extracted vegetation spectra, as compared with the original spectral, with the improvement of the ratio of performance to deviation of the calibration （RPDc） and the ratio of performance to deviation of the cross calibration （RPDcv） from 1.83 and 1.64 to 2.77 and 2.09, respectively. These results indicated that FastICA method could be applied as an effective spectral preprocessing method to significantly improve the accuracy of biomass estimation, thus providing guidance for accurate regional monitoring of wheat biomass by hyperspectral technology.