Granulation is a physiological disorder of juice sacs in citrus fruit, which made juice sacs become hard and dry and damaged the internal quality of citrus fruit. Honey pomelo is a thick-skinned citrus fruit, and it is hard to identify the granulation levels by observing the outer peel and fruit shape. In this study, a rapid and non-destructive detection method based on near-infrared transmittance spectroscopy combined with machine vision technology was used to classify honey pomelo by the granulation levels. 600 honey pomelos in different growth stages were harvested and divided into five granulation levels according to the granulation changes of samples. Spectral data of samples were recorded in the range of 900~1700nm, which were combined with chemometrics to research the chemical changes of inner quality caused by granulation. Machine vision technology can be used to study the physical changes of external characteristics caused by granulation. Therefore, comparison of the traditional method, this method has better predictive performances in classification models. In particular, the predictive performances of accuracy, sensitivity, and specificity were respectively not less than 0.9700, 0.9231, and 0.9874 in the SPA-KNN (successive projections algorithm-K nearest neighbor) predicted model. The results showed that this method could be used for classification and evaluation of granulation, and had a great potential. The method provides a reference and theoretical basis for the online sorting and inner quality detecting of thick-skinned fruits.