A Standard-Free Calibration Transfer Strategy for a Discrimination Model of Apple Origins Based on Near-Infrared Spectroscopy

The nondestructive discrimination model based on near-infrared is usually established by detected spectra and chemometric methods. However, the inherent differences between instruments prevent from being used universally, calibration transfer often to solve these problems. Standard-sample requires additional standard samples build a mathematical mapping instruments. Thus, standard-free research hotspot in this field. Based spectroscopy (NIRS), new combined strategy of wavelength selection was proposed two portable spectrometers. Three learning (TL) algorithms—transferred component analysis (TCA), balanced distribution adaptation (BDA), manifold embedded alignment (MEDA)—were applied achieve transfer. Moreover, paper presents relative error (REA) method select wavelength. To optimal model, parameters accuracy, precision, recall were examined evaluate discriminatory capacities each model. findings show that MEDA-REA capable higher prediction accuracy (accuracy = 94.54%) than other transferring models (TCA, BDA, MEDA, TCA-REA, BDA-REA), it demonstrated has good transmission performance. REA shows potential filter wavebands for simplify transferable