Abstract: Variable rate spraying is one of the most high-efficiency and low-cost technologies in the sustainable development of precision agriculture. Recently, the online detection of the canopy has been a key challenge to realize variable spraying in an orchard. Many state-of-the-art technologies are emerging for the fast and accurate acquisition of canopy data, including stereo vision, ultrasonic, and laser sensors. Previous studies have shown that a high-resolution binocular camera can be utilized to measure the canopy structure of fruit trees. However, no attempt has been made to apply this camera for the variable spray in an orchard. In this study, a precise control system of variable-rate spraying was designed based on the 3WF-400Z orchard wind sprayer. This system was composed of binocular cameras, solenoid valves, a touch screen, an encoder, and a control program. The specific procedure was: first to detect the depth of grape canopy using binocular cameras, then to calculate the canopy volume with the forward speed of the sprayer, while to regulate the duty cycle of several solenoid valves in real time under the Pulse-Width Modulation (PWM), finally to realize variable spray using grape canopy volume. A calculation method was proposed for the canopy volume of the grape canopy, where the detection program was compiled using the characteristics of a binocular camera. An accuracy test of volume detection was carried out to verify the calculated data in a vineyard. Each sample was measured manually to compare with the detection data of the binocular camera. It was found that there was a strong correlation between the camera detecting and manual data, where the determination coefficient of linear fitting was 0.933 after 150 mm compensation, indicating the binocular camera suitable for the canopy volume detection in actual practice. An experimental calibration was utilized to determine the flow control model of nozzles in the sprayer. The simulated depth was also sent to the control system, where the same forward speed was set under the static condition, thereby verifying the consistency of variable rates. The results showed that the linear determination coefficient between the actual and theoretical flow was 0.990, when the volume of the canopy was larger than 0.036 m3, indicating an excellent real-time performance of the program, the high response ability of the hardware, and the good consistency of the variable spray. A field experiment showed that the variable spray system reduced significantly the diameter of droplets, while increased the density of the droplet, where the coverage rate remained basically unchanged. Specifically, the Volume Median Diameter (NMD) and Number Median Diameter (VMD) of droplets decreased by 87.71 and 182.79 mm, respectively, whereas, the density of droplets increased by 79.31/cm2. Combined with the observed spray at the experimental sites, it was found that the spray volume in the conventional mode exceeded the actual demand, resulting in the droplet condensation again after reaching the canopy surface, whereas, the variable mode cannot generate excessive spray to improve the droplet size and spatial distribution. The determination coefficients between the predicted and actual flow of left and right sprinklers were 0.897 and 0.877, respectively, indicating a strong correlation. The overall trend of actual flow and canopy volume was all the same, indicating the control system suitable for variable spraying according to the volume of the target canopy. The variable spray mode saved about 55.27% of pesticide, compared with the traditional constant spray. The finding can provide a sound reference for the application of orchard variable spraying, further to achieve a high efficiency pesticide application for an expected production level in modern agriculture.