Transferring spectral libraries of canopy reflectance for crop classification using hyperspectral remote sensing data

Abstract

Spectral library search is emerging as an automated method for exploiting finer spectral details available in hyperspectral remote sensing data. We report on the potential of transferring independent crop spectral libraries for classifying various agricultural crops using airborne hyperspectral image. Spectral libraries constructed from multi-season field reflectance measurements for five agricultural crops (alfalfa, winter barley, winter rape, winter rye, and winter wheat) are used for the per-pixel and per-field classification of HyMAP airborne hyperspectral image by the spectral library search method. Results obtained from this method are compared with the results obtained from the per-field object-oriented, and per-pixel support vector machines (SVM) supervised image classification using image-based training data. Results from the spectral library search approach (best overall accuracy: 82%) exhibit strong correlation with the results obtained from both the object-oriented and SVM-supervised classification approach. However, per-field object-oriented classification shows moderate increase in the classification performance. In spite of significant reduction in the overall accuracy, the resultant overall accuracy of about 82% obtained from the spectral library search is fairly high, given the large spatial and temporal differences maintained between the image data and the field reflectance measurements. Results indicate the existence of a meaningful spectral matching between image and reflectance library spectra for some of the crops considered, showing their potential for transferring reflectance spectral libraries for image classification. Incomplete library coverage and phenological variations are found to be the key issues that influence the prospect of transferring spectral libraries for image classification.