Abstract: Abstract: Grain yield sensor is one of the vital equipment for precision agriculture. There are various commercial products and experimental prototypes concerning grain yield sensor, in which impact-type grain mass flow sensor is the most popular due to its high reliability and the features of being easily installed. Nevertheless, there is still a great gap in grain yield technology between China and developed countries. In order to further improve the technical level and integration of grain yield system in China, the grain mass flow sensor consisting of sensor of double parallel beams and a signal acquisition and processing module designed on embedded system was proposed. One of double parallel beams was impacted by the grain flow, and the other was applied as a reference beam that was only excited by the vibration of the sensor frame. The signal acquisition and processing module mainly consisted of AD623 chips, two order Butterworth filter, A/D module and STM32F405 processor. The signals from sensor of double parallel beams were amplified by AD623 chips, and filtered by two order Butterworth filter before entering into A/D module of HY16F188 chip. Then the digitized signals from HY16F188 were sent to STM32F405 through the serial communication. By adaptive interference cancellation algorithm running on the STM32F405 processor, the signals from double beams were processed to eliminate the vibration noise of the impact force of the grain flow. The pure grain mass flow signal was obtained and output by STM32F405 through the serial port. The upper-computer software was designed on MFC(microsoft foundation classes) to display and save grain yield data, and generate grain yield map. To verify the performance of the grain yield sensor, no-load vibration experiment, calibration experiment, and yield experiment were carried out in the actual rice field in Jiangdu District, Yangzhou City, Jiangsu Province in November 2016. The no-load vibration experiment was conducted under the 3 kinds of conditions: in-situ small throttle condition, in-situ big throttle condition, and running condition. The results showed that the variation interval for no-load output of the sensor reduced by 97.4% compared to raw signal. The calibration method and procedures were proposed to obtain the correction coefficient in the calibration experiment, which realized the linear correspondence between grain mass flow signal and actual quality. Yield experiment was performed in farmland NO.1116, NO.1117 and NO.1118. Among them, NO.1116 and NO.1118 were subdivided into 3 small fields, NO1117 into 2 small fields, and 8 experiments were respectively conducted in the 8 small fields. Yield distribution maps were generated after the experiments, which showed the distribution of grain yield intuitively. However, there were some problems of discontinuity in yield map and excessively high yield value in individual areas, since interpolation and filtering were not performed for yield data. Yield experiment results showed that the maximum measurement error of the grain mass flow sensor based on embedded system was less 7.4%. At the same time, it was found that the higher the grain yield, the higher the accuracy of the grain flow sensor. Grain mass flow sensor designed on the embedded system was independent on upper-computer, which made it easier to migrate to other upper platforms. If the factors, such as humidity, the speed of the hoisting device and the speed of the agricultural vehicle, were taken into account, the performance of the system could be further improved. Future work will be focused on fusing these factors to improve the algorithms' stability, and accuracy of the grain mass flow sensor based on our proposed embedded system.