The moisture content of pellet feed directly affects the quality of pellet feed. At present, the drying method is widely used for the detection of moisture in pellet feed. The shortcomings of this method are long detection time and single detection means. In order to increase the pellet feed moisture detection method and realize the non-destructive detection of pellet feed moisture, the STM32F103ZET6 single-chip microcomputer was used as the control chip for the pellet feed moisture detector, which mainly included the capacitance detection module, the temperature detection module and the weight detection module. The capacitance detection module uses parallel plate capacitance sensor and range expansion detection circuit with a digital capacitor converter AD7745 chip as the core. The weight detection module uses a strain resistance sensor and an A/D conversion circuit whose core is HX711 chip. The temperature detection module uses DS18B20 temperature sensor. After the initialization of each module, the capacitance, weight and temperature of the tested sample were sequentially collected, and the capacitance and weight therein were converted into relative permittivity and bulk density. The obtained relative permittivity, bulk density, and temperature were substituted into a moisture content calculation subroutine based on the binary iterative method to obtain a moisture content detection value of the sample and the detection result was displayed on the OLED display. The self-made pellet feed moisture detector was used to analyze the influence of moisture content (9%~18%), temperature (10~30 ℃) and bulk density (558.3~662.5 kg/m3) on the relative permittivity of pellet feed. The prediction model between relative permittivity and moisture content, temperature and bulk density was established. 12 samples of pellet feed with moisture content ranging from 9% to 18% were randomly prepared. The actual relative permittivity at different temperatures was measured by filling the capacitance sensor in any way. Then, the bulk density, temperature, and moisture content were substituted into the established model to obtain the predicted relative permittivity. The actual relative permittivity was compared with the predicted relative permittivity to verify the prediction effect of the established model. The detection accuracy of the dielectric pellet feed moisture detector based on the parallel plate capacitive sensor was tested. The results showed that the relative permittivity of pellet feed increased with the increase of temperature, moisture content and bulk density. The determination coefficient of the established relative permittivity and moisture content, temperature and bulk density model was 0.996 8. There was a good linear correlation between the measured relative permittivity and the predicted relative permittivity, the coefficient of determination was 0.992 9, indicating that the established model could describe the relative permittivity and bulk density, temperature and moisture content relationship well; The coefficient of determination between the measured value of the moisture content of the pellet feed and the detected value of the designed detector was 0.990 3. Compared with the drying method, the absolute measurement error of the measured value and the detected value was within ±0.6%. The research provides a new method and technical support for fast and non-destructive on-line detection of pellet feed moisture content.