Maturity Classification and Quality Determination of Cherry Using VNIR Hyperspectral Images and Comprehensive Chemometrics

Sweetness and acidity are the two most important indicators to evaluate cherry quality. In this research, hyperspectral imaging (380–1030 nm) technology was applied to visually detect the sweetness and acidity of cherry. To improve the imaging performance, two spectral pretreatment methods (wavelet transform, standard normal variable transformation and detrend), three feature selection methods (successive projection algorithm, genetic algorithm, and shuffled frog leaping algorithm), and four regression modeling methods (principal components regression, partial least squares regression, least square-support vector regression, convolutional neural network) were employed and compared. After the chemometrics related to hyperspectral imaging had been examined, the least square-support vector regression models based on the feature bands, which were selected by the shuffled frog leaping algorithm, showed the best performance, and the determination coefficients of prediction (R2P) for cherry sugar content and acidity reached 0.976 and 0.906, respectively. Based on the spectral principal components, three classifiers were built to classify the maturity level: support vector machine, backpropagation neural network, and radial basis function neural network. Excellent results with a classification accuracy of 100% were obtained. This research provides a complete and useful scheme to evaluate the quality and classify the maturity of cherry.