Abstract: Fertilizers play an important role in soil fertility for crop growth. However, the excessive use of fertilizers often cause serious damage to the environment. Precision fertilization can be expected to significantly reduce the excessive use of fertilizers in farmland. An accurate and rapid detection of the fertilizers is also required for precision fertilization during operation. Previous research has been conducted on the stable, accurate, and reliable detection of high-flow granular fertilizers during field fertilization, in order to improve the quality of field fertilization and the utilization rate of fertilizers. However, fertilization agronomy is often required for the optimal speed in the field operation of the rapeseed direct seeder. Furthermore, there is a high application amount of granular fertilizers, particularly with the high frequency of the fertilizer discharge. At the same time, the sensor is dependent mainly on the type of fertilizer and environment. It is quite difficult to realize the high-precision real-time detection during mechanical direct seeding. Among them, capacitance has been widely used in detection devices during sowing, fertilization, and harvesting in precision agriculture. The high sensitivity can be expected for the simple design, stable and reliable operation. The device with the capacitance can share outstanding advantages, in terms of the mass flow rate of the granular fertilizer application. But it is still required for the high stability and dynamic measurement accuracy in the field environment. This study aims to improve the real-time accuracy of the detection during rapeseed direct seeding, due to the large amount of discharge of granular fertilizer, diversified types, and the sensitive components of the sensor. A non-contact capacitive device was also developed to detect the discharge and application of the granular fertilizers in the direct seeding machines. A self-calibration function was then used for the detection. The key structural parameters of the capacitive plates and isolation sleeves were designed to extract from the capacitive signals using simplified Kalman filtering. The key parameters of the filtering were determined after experiments. There was a relationship between the mass flow rate of the granular fertilizers and the capacitive signals. A three-point fitting calibration was proposed to detect the mass flow rate of granular fertilizers suitable for the different temperature environments and types of granular fertilizers. An 8-channel device and a terminal were then developed for the discharge and application of the granular fertilizers in rapeseed direct seeding machines. The bench tests show that the detection accuracy rate of the device was less than 97.35% under the environmental temperatures of 15, 25, and 35 °C. Specifically, the accuracy rates were less than 96.45% and 96.23% for the Xiangyan and Haosite compound fertilizers, respectively; The accuracy rate of the system was less than 95.62% when the speed of the seeding was 4-12 km/h and the average flow rate of the fertilizer discharge was 9.52-29.05 g/s. The field tests show that the accuracy rate of the system was less than 93.14% when the discharge amount of the granular fertilizer was 9.82-28.99 g/s. There was at least 3 percentage points higher than that of the existing granular fertilizer flow. There was accurate, integrated, and stable data transmission between the system and the cloud database. The vibration of the machine and dust also shared little impact on the detection. This system can provide an effective means for the evaluation of fertilization quality.