In order to realize the visible/near infrared spectroscopy nondestructive testing of the internal quality of orange fruits, Trace pro software is used to carry out optical simulation analysis on the designed orange fruit online testing and conveying tray model. referring to the luminance/illuminance value in the simulation result, the tray model with higher value is processed in kind, and is tested and verified by combining with the actual spectrum testing platform.The simulation results show that the optimal shape parameters of the achievement transfer tray were: outer diameter 80 mm, inner transverse diameter 55 mm, inner longitudinal diameter 50 mm, and thickness 20 mm. Different materials were used to process the tray for the detection of soluble solids content (SSC) of actual orange fruits. After the spectral data were preprocessed, the partial least squares regression (PLSR) prediction model was used, among which acrylic tray had the best prediction result. In order to further optimize the detection model, genetic algorithm (GA) and stability competition adaptive re-weighted sampling (SCARS) algorithms are used to extract spectral characteristic bands respectively, and a prediction model of PLSR of orange SSC is established. The SCARS algorithm has the best feature extraction method, and the prediction decision coefficient R is 0.920 9. The predicted rms error (RMSEP) is 0.468 3.