Objective This study is aimed to address the issue that the spectral information collected in the non-destructive testing of pomelo fruit (Citrus maxima) is weak, and the detected internal quality of pomelo fruit is not accurate.

Method The optical simulation of the optical path of spectral nondestructive testing was carried out by using ZEMAX software, and tested and verified by the physical device. In order to further verify the nondestructive detection effect of the optical path on the internal quality of pomelo fruit, the spectral data of 132 pomelo fruits in 400−1 100 nm band were collected, and their soluble solid contents were determined physically and chemically for training modeling and testing verification.

Result The optimal parameters of the optical path were as following: The number of lamp beads was 7, the distance between the fruit and the light source was 2 cm, and the distance between the detector and the fruit was 0.5 cm. Through Savitzky-Golay (SG) smoothing and standard normal variate (SNV) preprocessing, competitive adaptive reweighted sampling (CARS) feature selection, a partial least squares regression (PLSR) prediction model was established. The R² and RMSE of the training set were 0.81 and 0.85, and those of the test set were 0.81 and 0.89, respectively.

Conclusion The optical path parameters of pomelo fruit obtained in this study are applicable for non-destructive detection of pomelo fruit internal quality, yielding satisfactory detection results. The results provide a reference for the internal non-destructive testing of other large fruits.