A prediction model for soybean meal protein content was developed using low-field NMR and near-infrared spectral data fusion for rapid protein content detection during soybean meal production. Firstly, the low-field NMR and near-infrared spectral data were collected from test samples. Secondly, the two collected signals were preprocessed and the Successive Projections Algorithm (SPA) was used to extract the characteristic variables of the low-field NMR and near-infrared spectra. The partial least squares method, BP (Back Propagation) neural network and Sparrow Search Algorithm (SSA) were employed to optimize the BP neural network (SSA-BP). The selected characteristic variables were fused to establish a prediction model for soybean meal protein content. The SSA-BP model, constructed by fusing low-field NMR and near-infrared feature layer data, showed the best performance, with a calibration set determination coefficient of 0.983 0, RMSE of 0.127 3, validation set determination coefficient of 0.956 4, and RMSE of 0.203 9. In summary, this method enables achieve rapid, non-destructive and accurate quantitative detection of soybean meal protein content while verifying, feasibility and effectiveness of low-field NMR and near-infrared data fusion.