[Development of prediction models for determining N content in citrus leaves based on hyperspectral imaging technology]

Abstract

The present study presents prediction models for determining the N content in citrus leaves by using hyperspectral imaging technology combined with several chemometrics methods. The steps followed in this study are: hyperspectral image scanning, extracting average spectra curves, pretreatment of raw spectra data, extracting characteristic wavelengths with successive projection algorithm and developing prediction models for determining N content in citrus leaves. The authors obtained three optimal pretreatment methods through comparing eleven different pretreatment methods including Savitzky-Golay (SG) smoothing, standard normal variate (SNV), multiplicative scatter correction (MSC), first derivative (1-Der) and so on. These selected pretreatment methods are SG smoothing, detrending and SG smoothing-detrending. Based on these three pretreatment methods, the authros first extracted the characteristic wavelengths respectively with successive projection algorithm, and then used the spectral reflectance of the extracted characteristic wavelengths as input variables of partial least squares regression (PLS), multiple linear regression (MLR) and back propagation neural network (BPNN) modeling. Hence, the authors developed three prediction models with each pretreatment method, and obtained nine models in total. Among all the nine prediction models, the two models based on the methods of SG smoothing-detrending-SPA-BPNN (R(p): 0.8513, RMSEP: 0.1881) and detrending-SPABPNN (R(p): 0.8609, RMSEP: 0.1595) were found to have achieved the best prediction results. The final results show that using hyperspectra data to determine N content in citrus leaves is feasible. This would provide a theoretical basis for real-time and accurate monitoring of N content in citrus leaves as well as rational N fertilizer application during the plant's growth.