The objective of this research was to create supervised classification models of pear ripeness with the use of hyperspectral imaging system in the visible and short near infrared (425–1000 nm) regions. Spectra and images of 450 pear samples were studied, which were selected from three ripeness stages (unripe, ripe, and overripe). Three classification algorithms—partial least square-discriminant analysis (PLS-DA), soft independent modeling of class analogy (SIMCA), and linear discriminant analysis (LDA) were compared. Successive projection algorithm (SPA) was used to select optimal wavelengths, and gray level co-occurrence matrix (GLCM) was applied to extract four textural feature variables. The overall results revealed that the best model was PLS-DA, of which the correct classification rate (CCR) was 87.86% with the input consisted of the spectra and texture feature of images. Hence, combining spectral with texture analyses were effective for improving pear ripeness prediction. In addition, the same dataset for ripeness classification got a good performance of firmness prediction by PLS in the regression analysis. In addition, using the same input of ripeness classification to make a study on firmness prediction by partial least square analysis revealed a potential for further research, with correlate coefficient of prediction set rpre of 0.84 and root mean square error of prediction of 0.78 N.

Keywords