Quality attributes prediction and maturity discrimination of kiwifruits by hyperspectral imaging and chemometric algorithms

Abstract The nondestructive detection of the quality attributes and maturity of kiwifruits by using hyperspectral imaging technique and chemometric algorithms were investigated. Two models of the partial least square regression and principal components regression were built up based on full variables. The multiple linear regression (MLR) was designed to develop the simplified detection models based on feature variables chosen by the successive projection algorithm and competitive adaptive reweighted sampling (CARS). Results showed that the optimal CARS‐MLR model was designed to determine the quality attributes with RPD above 2.2. Particularly, about SSC ( R 2 P = 0.896, RMSEP = 0.628%, RPD = 3.121), about firmness ( R 2 P = 0.871, RMSEP = 1.065 kg/cm 2 , RPD = 2.809). The partial least square discriminant analysis and simplified k nearest neighbor models were designed to discriminate the kiwifruits maturity stages with a classification accuracy of 93.3% and 98.3%. This research demonstrates that the hyperspectral imaging technology coupled with chemometric algorithms is promising for the quality assessment and maturity discrimination of kiwifruits. Practical applications The quality and maturity of kiwifruits have a direct relationship with the indicators of the SSC, firmness and color. Traditional methods for predicting the quality and maturity of kiwifruits are destructive, time‐consuming, and unusually affected by subjective factors. Hyperspectral imaging technique has the advantages of nondestructive, rapid, non‐pollution, and so on. The results demonstrated that the hyperspectral imaging technique coupled with chemometric algorithms can predict the quality and maturity of kiwifruits, which provides a theoretical basis to develop a real‐time detection system to predict the quality and maturity of fruits.