Farmland obstacle detection is a prerequisite for automatic driving of farm machinery. For farmland obstacle detection based on millimeter wave radar, an algorithm for invalid target filtering was proposed. Firstly, the target information output from the millimeter wave radar was analyzed and the farmland target obstacle information was extracted, and then the empty target, pseudo target and non-threat data were filtered out by using the invalid target filtering algorithm. The empty targets with zero target distance in the radar data were directly filtered out; the pseudo targets generated by the radar working performance or unstable echo signal were filtered by the radar effective target life cycle method; the non-threat targets exceeding the horizontal distance threshold and vertical distance threshold were directly filtered out. The experimental results showed that the average filtering rate of the proposed algorithm reached more than 85% in the static state. When the speed was 3km/h, the average filtration rate was 85.24% in the non-working state and 84.23% in rotary tillage. When the speed was 5km/h, 84.22% were not operated and 84.18% were rotary tillage. When the speed was 7km/h, the non-operation was 84.19%, and the rotary tillage was 84.16%. The experimental results showed that although the filtering rate was decreased to a certain extent with the increase of speed or hanging rototiller tool in the driving state, the filtering rate of the proposed algorithm could reach more than 84% either in the stationary state or in the driving state. This method can be used to detect obstacles in farmland under complex environment. Under driving conditions, the experimental results showed that the vibration of the tractor had a large influence on the filtering effect. In order to reduce the impact of vibration, a device for millimeter wave radar vibration damping was designed, through the experimental verification, it was showed that the device vibration damping effect was obvious, the average filtering rate in the unoperated state reached more than 84%, which can help to reduce the impact of tractor vibration.