Determination of Soluble Solid Content in Multi-Origin 'fuji' Apples by Using FT-NIR Spectroscopy and an Origin Discriminant Strategy
Xiaona Li,Jichao Huang,Yingjun Xiong,Jun Zhou,Xiangyu Tan,Baohua Zhang
DOI: https://doi.org/10.1016/j.compag.2018.10.003
IF: 8.3
2018-01-01
Computers and Electronics in Agriculture
Abstract:Apple is widely planted all over the world. Origin variability influences the internal quality of apples because soil characteristics, light effects, nutrition, weather conditions, as well as growing management vary from orchard to orchard. However, if taking spectral variations caused by the origin variability of apple samples from different orchards into account, the fruit quality parameters could not be measured or predicted with high accuracy by using the current models without updates. To improve the practicability and accuracy of the prediction models, a multi-origin regression model for the determination of soluble solids content in apples from three origins by using FT-NIR spectroscopy and a model search strategy was developed in this paper. In this model, based on the wavelengths selected by competitive adaptive reweighted sampling algorithm (CARS), partial least squares discriminant analysis (PLS-DA) was trained and applied to identify the geographical origins of the apple samples. The results indicate that the samples spectra were correctly matched to the corresponding classes and a 98.1% correct classification was achieved. Partial least squares regression (PLS) was used to establish three single-origin calibration models for the determination of soluble solids content (SSC) in apples from three different origins, and meanwhile, CARS algorithm was also applied to select the most effective wavelengths for calibration models. Then, the multi-origin CARS-PLS model for determination of SSC in apples from three origins was developed combined with origin discriminant and the proposed model search strategy. It was concluded that the multi-origin CARS-PIS model achieved more satisfying results than the single-origin CARS-PLS models for the determination of SSC, with R-p and RMSEP values for the apple samples from three geographical origins being 0.921, 0.759, 0.924 and 0.661, 0.673, 0.547 (A) over cap degrees Brix, respectively. The above results indicate that it is promising to build a multi-origin CARS-PLS model to predict SSC for apples based on an origin discriminant approach to reduce the effect of geographical origin.