Application of Wavelet Transform (WT) on canopy hyperspectral data for soybean Leaf Area Index (LAI) estimation in the Songnen Plain, China

Though hyperspectral data can provide more information compared with multi-spectral data, the major problem is the high dimensionality which needs effective approaches to extract useful information for practical purpose, and requires large numbers of training samples to meet statistical requirements. The use of Wavelet Transformation (WT) for analyzing hyperspectral data, particularly for feature extraction from hyperspectral data, has been extremely limited. WT can decompose a spectral signal into a series of shifted and scaled versions of the mother wavelet function, and that the local energy variation of a spectral signal in different bands at each scale can be detected automatically and provide some useful information for further analysis of hyperspectral data. Therefore, in this study, WT techniques was applied to automatically extract features from soybean hyperspectral canopy reflectance for LAI estimation; and compared the model prediction accuracy to those based on spectral indices (PCA). 144 samples were collected in 2003 and 2004, respectively in the Songnen Plain at two study regions. It is found that wavelet transforms is an effective method for hyperspectral reflectance feature extraction on soybean LAI estimation, and the best multivariable regressions obtain determination coefficient ( R2) above 0.90 with RMSE less than 0.30 m2/m2. As a comparison study, Vegetation Index (VI) method applied in this study, and wavelet transform technique performs much better than VI method for LAI estimation. Further studies are still needed to refine the methods for estimating soybean bio-physical/chemical parameters based on WT method.