In order to improve the detection accuracy of the wheat straw moisture content prediction model based on capacitance method, expand the detection range of moisture content and improve the adaptability of the model, taking wheat straw as the research object and using LCR digital bridge, the capacitance data of straw with 10.43%~25.89% moisture content were measured in the frequency range of 0.05~100kHz, volume density range of 90.03~179.42kg/m3 and temperature range of 25~40℃. The original data were preprocessed by using the successive projections algorithm (SPA) and principal component analysis (PCA) to extract characteristic frequencies, BP neural network was used to establish quantitative analysis models of straw moisture content, volume density, temperature and capacitance at full frequency and two characteristic frequencies respectively, and sparrow search algorithm (SSA) was introduced to optimize the BP neural network model. The experimental results showed that the prediction effect of the model based on full frequency was slightly better than that of the model based on SPA algorithm. Considering the model complexity and prediction performance, the BP neural network model (SPA-SSA-BP) optimized based on SPA algorithm and SSA algorithm was selected as the prediction model of wheat straw moisture content. The R2P, RMSEP and RPDP of prediction sets were 0.9832, 0.00550 and 7.715, respectively. The model was used to predict 13 straw samples with water content ranging from 10.62% to 25.59%, and the relative error of water content prediction results was within -5.27% to 5.52%, 96.8% of which was within ±5%. This showed that the model had high accuracy and good robustness, and the method can provide an idea and theoretical reference for other crop straw water content prediction. 