Study On Estimation Of Fall Dormancy In Alfalfa (Madicago Sativa) By Near Infrared Reflectance Spectroscopy

This study explores the period when the highest accuracy is achieved in the estimate of fall dormancy (FD) in the alfalfa in Beijing. Its samples were standard American alfalfa that has the property of dormancy. The alfalfa samples were mowed four separate times, with each sample mowed once. As LSD significance test indicated, the sample mowed on October 10,2009 demonstrated significantly different levels of growth 25 days after the mowing among groups with different FD classes. Maverick, with a FD class of 1, showed a significant decline in growth rate, and grew 2.4 cm in the 25 days after the mowing. The Maverick plants concerned in this round of mowing were at the same level (h<5cm). Therefore the estimate of alfalfa fall dormancy in Beijing produces the most accurate result on October 10.Alfalfa samples mowed on October 10 were baked, ground and processed in other manners. An NIR analyzer was applied to measure the near infrared spectra of the processed samples. The FD of the alfalfa samples was estimated based on the spectra through approaches such as principal component regression (PCR), partial least squares method (PLS), the BP neural work, LVQ neural work, and the support vector machine (SVM). The conclusions are as follows:The conclusion 1:Estimate models were developed for alfalfa dormancy based on different spectral coverages through principal component analysis (PCA) and the BP neural network. The highest accuracy rate,87.27 percent, was achieved by the model developed based on the spectral coverages of 7000～8000cm-1 and 4000～7000cm-1. This model can be used to estimate fall dormancy in alfalfa and its estimates are significantly influenced by the selection of spectral coverages.The conclusion 2:Estimate models were developed for alfalfa dormancy based on different spectral coverages through principal component analysis (PCA) and the LVQ neural network. A good accuracy rate of 90.90 percent was achieved by the model based on full spectrum. Therefore the LVQ neural network can be applied to developing a model for estimating alfalfa dormancy.The conclusion 3:Estimate models were developed for fall dormancy in alfalfa based on different spectral coverages through principal component analysis (PCA) and support vector machine (SVM). The model based on full spectrum, when c=0.3392 and g=32, achieved an estimate accuracy rate of 98.182 percent, significantly higher than the accuracy rates of the models developed through other approaches. Therefore the model developed through the PCA and the SVM can be applied to initially estimating dormancy in alfalfa.This study is the first in China to apply near infrared spectroscopy to estimating fall dormancy in alfalfa. It provides a faster, more accurate and more convenient new approach for the estimate, and for the application of such an estimate to research and commercial activities.