SSC and pH for sweet assessment and maturity classification of harvested cherry fruit based on NIR hyperspectral imaging technology

The relationships between soluble solids content (SSC) and pH cherry fruit of different maturity stages has been investigated using near-infrared (NIR) hyperspectral imaging technology. Using 550 fruit, 11 hyperspectral images in the 874-1734 nm region were captured and compared with SSC and pH measured by standard methods. Two types of models based on full bands, namely principal components regression model and partial least squares regression model, showed similar predictive ability. To reduce the modeling complexity based on full bands, a genetic algorithm (GA) and a successive projections algorithm were employed to select feature bands; both algorithms were tested by multiple linear regression (MLR). By comparing the results of different modeling methods, GA-MLR was selected as the final modeling method with a ratio of standard deviation of prediction set to standard deviation of prediction error of 2.7 for SSC and 2.4 for pH. SSC and pH distribution maps were generated by inputting the feature bands of each pixel into GA-MLR models. Classification of fruit maturity stages was studied, and a linear discrimination analysis method produced a correct classification ratio of 96.4%. We conclude that it is feasible to detect the quality of cherry fruit by NIR hyperspectral imaging technology.