基于数学形态滤波的植被光谱去噪方法研究

光谱维噪声使地物光谱扭曲或变形,中心波长偏移,影响地物信息提取和地表参量反演的精度。对光谱维噪声进行滤波处理,有利于改善遥感数据定量应用的效果。由于数学形态滤波的原理简单且较易实现,被应用到植被光谱以及有机化合物光谱的研究中。运用数学形态滤波对地面实测小麦光谱去噪,一方面对滤波后的光谱进行噪声和波形相似度的直观分析,另一方面通过植被指数反演小麦理化参量进行定量应用评价。结果表明,与传统Savitzky-Golay滤波相比,在可见-近红外波段范围内,数学形态滤波去噪后的光谱能够保持可见—近红外波段原始光谱的固有特征,叶面积指数和叶绿素的反演精度比去噪前有小幅提升,主要原因是实测光谱在该谱段范围的噪声影响很小;在短波红外波段范围内,数学形态滤波能有效去除短波红外大尺度噪声,提高叶片含水量的反演精度。而传统Savitzky-Golay滤波只能削弱短波红外大尺度噪声。广义形态滤波去噪后植被指数和叶片含水量之间的R2最高可达0.5130(去噪前0.3753),叶片含水量的反演值与实测值之间的R2最高可达0.4221(去噪前0.3097),RMSE为0.0243(去噪前0.0318),优于传统Savitzky-Golay滤波。

Research on Vegetation Spectrum Denoising Method based on Mathematical Morphology Filtering

Noise in spectral dimension makes the spectrum distorted or deformed,shifting the central wavelength,affecting the precision of extracting land cover information and inverting surface parameters.Therefore,spectral denoising is of great significance for improving the effects of quantitative applications of hyperspectral remote sensing.Because the principle of mathematical morphological filtering is simple and easy to implement,it has been used in the study of vegetation spectra and fluorescence spectra of organic compounds.Mathematical morphology filtering was used to remove spectral noise of wheat.Intuitive analysis was made according to the spectrum similarity and spectral denoising effects;Moreover,the quantitative evaluation in practical application was made by adopting different vegetation indices to invert the biophysical and biochemical parameters.The results show that,compared with the traditional Savitzky|Golay filter,the mathematical morphology filtering can keep the inherent characteristics in visible and near|infrared region,and improve the accuracy of inverting Leaf Area Index and Chlorophyll slightly,which is caused mainly by the low spectral noise in the range of the spectrum.The mathematical morphology filtering can remove the noise in shortwave infrared (SWIR) region effectively,improving the accuracy of inverting foliar water content of wheat.But the traditional Savitzky|Golay filter can only weaken the large scale noise in SWIR region.After generalized morphology filtering,the determination coefficient of regression (R²) between the vegetation index and foliar water content can reach 0.5130,while R² is 0.3753 before filtering;R² between inverted value and the measurement of foliar water content can reach 0.4221 with a root mean square error (RMSE) of 0.0243,while the corresponding values are 0.3097 and 0.0318 for R² and RMSE before filtering.The results of the mathematical morphology filter are better than those of the traditional Savitzky|Golay filter.