Abstract: In order to improve the accuracy of on-line detection technology of rice moisture content,the parallel plate capacitor was taken as the research object,and the fin type double plate detection method was adopted to optimize the detection value. Taking the plate thickness,plate spacing and relative area as experimental factors,the capacitance ratio sensitivity test was carried out by quadratic regression orthogonal combination test method. The optimal plate structure parameters were obtained as follows:plate thickness was 2. 98 mm,plate spacing was 101. 60 mm,and relative area was 32 583. 69 mm~2. The prediction and correction model of moisture content based on nonlinear autoregressive neural network NARX was established by Matlab software. The parameters of model structure and optimization algorithm were determined by comparative analysis. The error analysis showed that the NARX prediction model based on the quantitative conjugate gradient algorithm was the best. The hidden layer of the model was 1 layer,the number of neurons was 5,and the lag order was 3. Compared with the 105℃ constant weight method,the calibration error range was within ± 0. 5%,the maximum deviation was 0. 65%,the minimum deviation was 0. 26%,and the average deviation was 0. 44%. Compared with the static capacitance water meter,the deviation fluctuation of on-line test of rice drying production was small,which met the requirements of rice drying production.