In the 6GF-4 forest fruit non-destructive testing and sorting equipment, an online citrus sugar content detection and sorting system was designed based on visible/near-infrared spectroscopy technology. The system mainly includes transmission devices, spectral acquisition devices, control systems, and sorting devices, which achieves spectral information collection during citrus movement, and synchronously calculates through the established sugar content model, achieving online sorting based on the obtained sugar content values. A dual lens optical path has been designed in the spectral acquisition device, to change the size of the light spot transmitted on citrus fruits. By comparing and analyzing the integration time and spot size of the experimental parameters, it is found that the optimal collection parameters for the system were the integration time of 100 ms, the spot size set to small, and the sample movement rate of 5 pieces/s. An SPXY-CARS-PLSR online detection model was established for citrus sugar content. The determination coefficients is 0.938 and 0.836 for the calibration and prediction sets of the model, respectively. The root mean square error is 0.273 °Brix and 0.418 °Brix for the calibration and prediction sets, respectively. 25 citrus samples not involved in modeling were used for online detection and sorting of external validation sets. It is found that the detecting sugar contents are within the error range of 1 °Brix with the accuracy of 92%. When the sample is divided into 4 levels, the system’s sorting accuracy is 92%; When the samples are divided into three levels, the system’s sorting accuracy can reach 100%.