Estimation of fractional vegetation cover using mean-based spectral unmixing method

Abstract

Mixed pixels have a significant impact on the accurate estimation of Fractional Vegetation Cover (FVC) using digital photos acquired by Unmanned Aerial Vehicle (UAV). A single threshold is inadequate for the separation of vegetation and background when images contain numerous mixed pixels. We propose a spectral unmixing method to measure FVC with UAV-acquired digital images. In this method, the spectral mean values of vegetation and background are obtained as a priori spectral knowledge from the photos taken at a very low flight altitude around 5 meters above ground level (AGL). Two thresholds with high confidence level derived from the a priori knowledge are determined to select pure vegetation and background pixels from the photos taken at high flight altitudes ranging from dozens to hundreds of meters AGL. For the mixed pixels, endmember spectra are undertook by mean values of those two pure components. Images with different aggregation levels were generated from a 10 meters AGL image. A comparison with four commonly used methods indicated that our method could robustly characterize the FVC in a good agreement with the ground truth, and the accuracy of FVC estimates over corn crops was around 0.01 in terms of root mean square error (RMSE) value. All aggregated images produced stable FVC estimates and the corresponding standard deviation (STD) was around 0.01 with relative average deviation (RAD) being less than 0.15.