Abstract: Uneven seeding can be caused by speed fluctuations in direct rapeseed seeding machines, due to the soil moisture content and tillage depth. In this study, a speed-adaptive seeding system was developed for precision rapeseed direct seeding machines using Beidou-radar fusion speed measurement. The control system was implemented on a hardware system, sensor communication, seeding rate fitting, and the data interaction visualization interface. A programmable logic controller (PLC) was taken as the processor and integrated components, such as an electric-driven multi-chamber centralized metering device, Beidou speed measurement module, speed measurement radar, and human-machine interface. An adaptive gain algorithm was employed for velocity measurement after confidence evaluation. Dual-source velocity information was fused from both Beidou and radar, thus providing for real-time output of the machine's forward speed. The rotational speed of the seed metering wheel was dynamically adjusted in response to velocity fluctuations. Real-time matching between the seeding rate and speed was achieved for the stability and uniformity of seed distribution. A systematic analysis was made to explore the impact of the forgetting factor on speed measurement performance. The speed measurement errors of the combined system were compared under different values (0.2, 0.3, and 0.4). The results indicated that there was a balance between noise resistance and dynamic response at the forgetting factor of 0.4. Mean absolute error (MAE) and the root mean square error (RMSE) of the speed measurement were within 0.099 and 0.113 m/s under various conditions, indicating real-time and accurate output of the speed. The seed metering wheel speed was dynamically adjusted to precisely match the seeding rate and speed. Seeding quality was improved under speed fluctuations. A series of tests was conducted on a centralized metering device to verify the control accuracy and stability of the speed-adaptive seeding system. The rapeseed variety ‘Hua you za 62’ was taken as the experimental subject. Bench test results demonstrated that the average seeding quantity error of the speed-adaptive system was 3.57%, which was lower than the minimum error of constant-speed seeding (4.55%). The coefficient of variation was less than 2.05% for total seeding quantity stability, indicating the high seeding uniformity under different conditions. Field trial results indicated that the average relative error was 3.17% for the real-time rotational speed of the metering wheel under various conditions (target seeding rates of 3 750, 4 500, and 5 250 g/hm²; speeds of 6 and 8 km/h). The rotational and theoretical speed exhibited high consistency along the time axis, indicating minimal dynamic lag. The robust performance fully met the real-time requirements of speed-adaptive control systems. Furthermore, the tracking curves of rotational speed shared no significant fluctuations under a complex field. Additionally, the average seeding rate error was 4.49%, and the average coefficient of variation was 9.41% for seeding uniformity, both of which complied with the technical requirements for mechanized rapeseed sowing. The findings can provide technical support to enhance the quality of rapeseed sowing in precision agriculture.