In agricultural production, capacitive soil moisture sensor (CSMS) has received considerable attention, for it is relatively inexpensive and easy to operate, and can meet the application requirements of wireless sensor network. But CSMS is vulnerable to the soil temperature and salinity in the measurement process. Therefore, a CSMS was designed based on the principle of resistance capacitance series step response. It had two frequency square wave excitation signals, adopted peak detection technology, and took digital signal as output. An inverse function model was proposed to analyze the soil equivalent relative permittivity by fusing high and low frequency response signals. The experimental results in air and distilled water showed that the high and low frequency response stability time of the sensor was within 338~464ms, and it was suggested that the high and low frequency switching interval of the sensor was greater than 500ms. The calibration results in non-conductive liquid medium showed that the exponential model can well fit the relationship between the output signal and the relative permittivity of liquid at high and low operating frequencies, and the coefficient of determination R2 was greater than 0.98. In the range of 0~1000μS/cm, the maximum error of relative permittivity measurement caused by conductivity was 1.775 and the corresponding maximum quote error was 2.16%. The results of soil measurement showed that the output signals of high frequency and low frequency sensors were affected by soil conductivity respectively, and the change of output voltage signal can be greater than 100mV, and the corresponding volumetric water content error was greater than 10%. After the proposed data fusion processing of the high -frequency and low-frequency output signals, combined with the Topp model, the influence error of conductivity on the sensor was controlled within 3.2%.