Abstract: Abstract: Aiming at the problem that combine harvester cannot obtain reliable grain flow information in complex noise background, wavelet transform (WT) is proposed to denoise the output signal of grain flow sensor. The method is as follows: using the LabVIEW data acquisition card to acquire the output signal of the grain flow sensor and analyzing the amplitude-frequency characteristics of signal, in which the energy-dominant frequency band is set as the grain effective signal, and the chaotic and the frequency band with flat amplitude change are set as the noise signal. The spectrum of original signal and noise is used to determine how to choose the wavelet function and decomposition scale. By analyzing the amplitude-frequency characteristics of the original signal, the energy-dominant center frequency in the grain flow is 40 Hz, and the noise is distributed in 2 frequency bands, of which the signal energy attenuation is faster, the bandwidth distribution is narrower and the change trend is consistent with the original signal in 500-600 Hz band, so the bands are the vibration noise of the yield platform. Above 800 Hz the signal changes slowly in energy attenuation, the bandwidth distribution is wide and the changes are random, so the bands are the high-frequency noise of the yield platform. In the research, the yield measuring system is designed to obtain the yield information. The Mallat algorithm is applied to decompose and remove the noise components of grain flow signal, the DB4 wavelet is chosen as the wavelet function and the decomposition scale is 6. The effective data of grain flow are reconstructed by wavelet and calibrated by the analog to digital converter channel of microcomputer. The maximum relative error of the grain mass is 1.68% under different grain flow after calibration. The position data are obtained by the Beidou Navigation Satellite System (BDS) with the pseudo-range differential position technology. The yield measuring system is installed on the combine harvester to simulate the rice operation experiment. The quantitative rice is transported to the grain tank and output at different rates per unit time, and the speed of driving wheel of the grain-conveying auger is gradually increased in the grain flow experiment. After the WT for the output signal of the flow sensor, reconstructing the base frequency of the grain flow information can be removed by the vibration noise. The relative error of the grain flow measurement is 6.18% and the average relative error is 5.37%. The yield data recorded in the production system are imported into the PC (personal computer) to plot the yield map. The map confirms that the BDS differential location data are reliable and the grain flow information in the test area is consistent with the experiment. Through the WT of the information output by grain flow sensor, the spectrum of the signal after the wavelet denoising is compared with the original signal, and the feasibility and the accuracy of the flow signal denoising are validated as well as the grain flow information reconstructing method based on the Mallat algorithm. This research can provide a reference for the original information denoising and the reconstructing of effective information in the complex operating environment of agricultural machinery.