The use of a portable near-infrared (NIR) spectrometer for detecting sea buckthorn juice SSC has not been explored. In this study, spectral data of 180 juice samples were collected using a portable NIR spectrometer. An SSC prediction model based on a mixture of experts convolutional neural network (MoE-CNN) was proposed. The MoE-CNN model was compared with traditional chemometric models in terms of prediction performance and feature extraction capability. The results showed that detecting the SSC of sea buckthorn juice using a portable NIR spectrometer combined with the MoE-CNN model is feasible. The optimal chemometric model, CARS-PLS, achieved RMSEP and RPD values of 1.42% and 2.67, respectively. The MoE-CNN model outperformed chemometric models and the CNN model, achieving an RMSEP of 1.26% and RPD of 3.02. Compared with CARS-PLSR, MoE-CNN adaptively weighted spectral features through MoE and feature fusion modules, effectively suppressing spectral noise and improving detailed feature extraction. These findings demonstrate that combining a portable NIR spectrometer with MoE-CNN is effective for rapid SSC detection in sea buckthorn juice. This study provides a new approach for the rapid detection of sea buckthorn juice SSC.