Prediction of apple moisture content based on hyperspectral imaging combined with neural network modeling

The moisture content (MC) of apples directly affects their flavor and market value. In this study, the MC of apples was successfully predicted using hyperspectral imaging combined with neural network modeling. Experiments were conducted to compare seven spectral preprocessing methods, two feature extraction methods and to establish classification models respectively. The results show that the back-propagation neural network (BPNN) model built based on multiplicative scatter correction (MSC) preprocessing and competitive adaptive reweighted sampling (CARS) algorithm extraction of characteristic wavelengths is the most effective. The determination coefficients of the correction (RC ) and prediction (RP ) sets are 0.9875 and 0.9850, respectively. The root mean square error of correction set (RMSEC) and prediction set (RMSEP) are 0.4106, 0.4256, respectively. The relative percent difference (RPD) is 5.8026. Consequently, hyperspectral images with small sample combined with appropriate regression models can be used for prediction of MC of apples. This study can serve as a valuable reference for the nondestructive detection of fresh fruit MC.