Near-infrared Reflectance Spectroscopy for Total Nutrient Analysis in Organic Fertilizer Using Linear Support Vector Machine
A new method was explored to predict the total nutrient content in organic fertilizer products using near-infrared reflectance spectroscopy (NIRS) with linear support vector machine regression (SVR), compared with partial least-squares regression (PLSR). 120 commercial organic fertilizer samples were collected from 22 provinces in China. Spectra of orient and dried (passed 1 mm screen) samples were scanned with a SPECTRUM ONE NTS (PerkinElmer, New Jersey, USA) from 10 000 to 4 000 cm-1, respectively. NIRS-PLSR models for total nutrient in organic fertilizer products samples on orient and dried basis were developed with the following results: the determination coefficient of validation (R_v~2), the standard error of prediction (SEP) and RPD (SD/SEP) on orient and dried basis were 0.96, 7.95 g/kg, 2.47 and 0.93, 8.02 g/kg, 3.58, respectively. The validation results of NIRS-SVR model for total nutrient content on dried basis were R_v~2 0.93, SEP 7.38 g/kg and RPD 3.88.Results showed the feasibility and potential of NIRS to predict total nutrient content in organic fertilizer products, and NIRS-SVR method on dried basis is the best choice.
Organic fertilizerNutrientSupport vector machinePartial least squaresNear-infrared reflectance spectroscopy
国家科技期刊平台
NSTL
万方数据
维普
