Investigation Of The Measurement Of Na, Ca, And Mg In Greenhouse Organic Substrate Via Visible/Near-Infrared Spectroscopy

Hao Fei-Lin,Shen Ming-Wei

2011-01-01

Abstract:In order to investigate the feasibility of measuring Available Calcium (Ca), Magnesium (Mg) and Sodium (Na) in greenhouse organic substrate via visible/near-infrared spectroscopy (VNIRS), the reflective spectrum of three measuring status-natural loose, pressed and dry powder of a typical SE Chinese greenhouse substrate were measured and studied. Mixture Uniform Design is adopted in experiment, 10 pre-treatments and three regression methods: Partial Least Square (PLS), Radical Base Function of Artificial Neural Network (ANN-RBF) and Support Vector Regression (SVR) were used for data treatment, regression and prediction. Results show that for the three regression methods, the 1st Derivatives, Range Normalization and Standard Normal Variation (SNV) are not suitable pre-treatment methods, the effects of the remaining pre-treatments do not differ much and show no significant enhancements comparing with the no pre-treatment case; The regression and prediction ability of PLS and ANN-RBF is more easily affected by measuring status, of which the dry powder measurement status performs the best prediction, while SVR shows a small difference. The ANN-RBF has the highest test coefficient of determination (r(2)) in Na, Ca, Mg's prediction under substrate dry powder measurement status -0.9996, 0.9989 and 0.9954 respectively, and has the lowest Mean Square Error of Prediction (MSEP) except for the prediction of Mg under substrate natural loose measurement status, which has relative lower r2 than that of SVR; SVR shows an stable high prediction ability and is less effected by the substrate measurement status, for all three elements, the tested prediction r(2) is >= 0.99; While the PLS performs worst with r(2).>= 0.99 only for Ca and Na prediction under substrate dry powder measurement status. As for the three elements in substrate, Na, Ca and Mg could be precisely measured via VNIRS by adopting appropriate pre-treatment and regression methods, but Mg has a relative lower r(2).These instructions give you basic guidelines for preparing papers for conference proceedings.