Physiological disorder of citrus granulation is one of the qualitative issues and problems in consumer markets and the phenomenon of granulation in agricultural products is not visually detectable, so the detection of this disorder on a large scale requires the use of precise and rapid techniques. One of the applicable methods for non-destructive examination of internal tissue of agricultural products is visible and near-infrared spectroscopy technique, which was used in this research with Vis-NIR method and artificial intelligence for the detection of granulation in oranges. In this research, 200 samples of Valencia oranges were prepared, Then, VIS/NIR signals in the wavelength range of 200-1100 nanometers were recorded for each sample. Spectroscopy was performed on the samples With contact Then, the amount of reflectance was calculated in the interaction mode, and absorption spectra were normalized using the Min-Max method. Subsequently, the recorded spectra were smoothed by the moving average (MA) method, and the Stavisky-Golay algorithm was applied to each spectrum, and finally, 7 statistical features were extracted from each spectrum. In this research, five levels of granulation phenomenon, including levels A, B, C, D, and E were examined. The degree of granulation of each sample was defined based on the dryness and appearance of the dried area using a destructive method. Support vector regression (SVR) and support vector machine (SVM) were used to estimate the moisture content of oranges and to detect the levels of granulation in oranges, respectively. With the change in the levels of granulation, the changes in weight and moisture content of oranges were significant, so that with an increase in the level of granulation in oranges, the weight of samples and their moisture content decreased. Also, the examination of ViS-NIR spectra showed that with the development of granulation in oranges, the absorption in the 400-950 nanometer range significantly decreased, and the difference in absorption intensity for different levels of granulation was maximum in the 570-850 nanometer range. The results of granulation level detection showed that the highest accuracy was observed for SVM with polynomial kernels, with a sensitivity of 98%, 91%, 97%, 96%, and 95% for levels A, B, C, D, and E of granulation, respectively.