Abstract: In near-infrared spectroscopy analysis， existing calibration transfer methods are mostly based on standard samples and non-parametric induction models， which generally suffer from short model lifespan， limited model applicability. To address this problem， a variational inference calibration adaptation(VICA)method is proposed， which aligns the feature distributions of the source domain(master instrument)and a target domain(slave instrument)by a parametric method. VICA performs principal component analysis on the source domain data and establishes a variational regression model for the source domain features. During prediction， VICA first projects the target domain data into the source domain feature subspace， and then establishes a distribution difference function between the source and target domain features， and obtains the probability density model of the target domain by minimizing this function， achieving model transfer. Experimental comparison results show that VICA performs better in calibration transfer than most existing methods.

Key words: chemometrics; near-infrared spectroscopy; domain adaptation; calibration transfer; variational inference