Prediction of nitrogen and phosphorus contents in citrus leaves based on hyperspectral imaging

Abstract

The nutritional status of citrus leaves is very important to the determining of fertilization plans. The spectrum technique is a quick, un-injured method and is becoming widely used for plant nutrient estimation. The possibility and method of using spectrum technique to estimate the nutrient of citrus leaf was explored in this study. A total amount of 135 leaves from the mature spring shoots of navel orange trees (C. sinensis Osbeck, “Newhall”) were collected and randomly grouped into two sets of samples: 100 leaves for the calibration set and 35 leaves for the prediction set. The hyperspectral images were scanned upper and lower side of each leaf and then the total nitrogen (N) and phosphorus (P) contents of each leaf were measured. The raw spectra data were extracted to generate average spectra curves, preprocessed with five different methods, and was used to build N and P content prediction models. The performances of the five preprocessing methods, i.e., Savitzky-Golay smoothing (SGS), standard normal variate (SNV), multiplicative scatter correction (MSC), first-derivative (1-Der) and second-derivative (2-Der), were tested with linear partial least squares (PLS) models and nonlinear least squares-support vector machine (LS-SVM) models. The results showed that the SG-PLS and PLS were the best for the N predicting (Rp=0.9049, RMSEP=0.1041) and P (Rp=0.9235, RMSEP=0.0514) in citrus leaves, respectively; the hyperspectral image data from leaf’s upper side predicting better for the contents of N and P. The study suggested that the hyperspectral image data from the upper side of the citrus leaves are suitable for nondestructive estimation of nutrient content. Keywords: citrus leaves, nitrogen, phosphorus, prediction, hyperspectral imaging DOI: 10.3965/j.ijabe.20150802.1464