2005 Volume 14 Issue 4 Pages 261-272
An Airborne Imaging Spectrometer for Applications (AISA) system was used to obtain hyperspectral images over an experimental citrus orchard located at Nebukawa Agricultural Research Station, Japan during the months of April, May and June in 2003. The objective was to determine to what extent the reflectances in the 72 visible and near-infrared (NIR) wavelengths (from 407 to 898 nm) obtained in different months might be related to differences in fruit yield of citrus trees. Multiple linear regression and neural network algorithms were employed to develop yield prediction models. Correlation Coefficient Analysis (CCA), Principal Component Analysis (PCA) and Partial Least Squares (PLS) regression were used to reduce the number of input variables. Results indicated that the PLS method performed the best in extracting predictor factors that have a high correlation with citrus yield, followed by the CCA method. However, the PCA method did not yield a satisfactory result. The performance of neural network models was further verified by statistical analysis of the ensemble ten thousand models developed with random initialization of training parameters. It is demonstrated that there is a more evident correlation between the spectral characteristics in May and the citrus yield, compared to those in April and June.
