In modern agriculture, the application of pesticides remains a critical yet challenging task. Traditional manual spraying methods are not only inefficient but also pose significant health risks to operators due to direct exposure to harmful chemicals. Moreover, these methods often lead to low pesticide utilization rates, with substantial amounts drifting into the environment, causing pollution and resource waste. To address these issues, we propose a novel remote-controlled agricultural drone system based on the STM32 microcontroller. This system aims to enhance spraying efficiency, reduce human harm, and promote precision agriculture through automated and targeted operations. Our design integrates advanced sensors, control algorithms, and communication modules to achieve stable flight and precise spraying, making it a viable solution for modern farming needs.
The core of our agricultural drone system is the STM32F407 microcontroller, which serves as the flight controller. It processes data from various sensors, such as gyroscopes, accelerometers, barometers, magnetometers, and ultrasonic modules, to maintain stable flight姿态. Additionally, the system includes a remote control unit for manual operation, a pesticide spraying mechanism controlled via relays, and GPS modules for location tracking. The entire setup is designed to be robust, cost-effective, and scalable for different agricultural scenarios. In this article, we will delve into the system architecture, hardware and software design, experimental validation, and the broader implications of using such agricultural drones in precision farming.
The overall structure of our agricultural drone system can be divided into three main components: the remote control-receiver device, the flight control unit, and the pesticide spraying apparatus. The remote control unit uses a 2.4G automatic frequency-hopping digital system to send commands to the receiver, which then relays signals to the STM32 microcontroller. The flight control unit comprises the STM32F407 microcontroller along with sensors like the MPU6050 gyroscope, SPL06 barometer, AK8975 magnetometer, and ultrasonic sensors for altitude measurement. The pesticide spraying system includes a water pump, tank, relay switch, and GPS module for location-based activation. The workflow involves two primary processes: flight stabilization and pesticide spraying. For flight, the microcontroller receives control signals from the remote and sensor data, processes it using filtering algorithms, and outputs PWM signals to electronic speed controllers (ESCs) that regulate motor speeds. For spraying, the remote sends a PWM signal to trigger the relay, activating the pump when specific conditions are met, such as reaching a target location via GPS coordinates.
| Component | Model/Specifications | Key Parameters |
|---|---|---|
| Remote Control | FS-iA6 Remote Controller | 6 channels, 1024-level data resolution, 6V input voltage, GFSK modulation |
| Receiver | FS-iA6 Receiver | 6 channels, -105 dBm sensitivity, GFSK modulation |
| Microcontroller | STM32F407 | Cortex-M4 core, 82 I/O ports, 3 I2C interfaces, multiple PWM outputs, 1.8-3.6V operating voltage, up to 168 MHz frequency |
| Gyroscope | MPU6050 | Integrated 3-axis gyroscope and accelerometer, I2C interface up to 400 kHz, digital motion processor (DMP) |
| Barometer | SPL06 | I2C interface, resolution up to 10 cm |
| Magnetometer | AK8975 | 3-5V operating voltage, built-in A/D converter, sensitivity of 0.3 μT/LSB |
| GPS Module | SKG12D | Tracking sensitivity -162 dBm, supports NMEA and UBX protocols |
| EEPROM | AT24C02 | 2K-bit memory, I2C interface for data storage |
The hardware connections are critical for ensuring seamless communication between components. The STM32 microcontroller interfaces with sensors via I2C buses: the MPU6050 gyroscope provides motion data, the SPL06 barometer offers altitude information, and the AK8975 magnetometer gives heading direction. An EEPROM memory chip stores calibration data and configuration settings. Remote control signals are input through specific GPIO pins (PE9, PE11, PE13, PE14, PD12) on the STM32, which use input capture to decode PWM values. Based on processed data, the microcontroller outputs PWM signals via pins PB1, PB0, PA7, and PA6 to control four brushless motors through ESCs. The GPS module connects via a UART serial interface, enabling location tracking and data transmission to a ground station. Additionally, status indicators like RGB LEDs are connected to pins PE0, PE1, and PE2 to display operational modes, such as armed, flying, or error states. This intricate hardware network ensures that the agricultural drone can perform autonomously or under remote guidance.
To illustrate the signal flow, consider the PWM output for motor control. The duty cycle of the PWM signal determines motor speed, which can be modeled using the following relationship between thrust and PWM duty cycle: $$T = k \cdot D^2$$ where \(T\) is the thrust generated by a motor, \(D\) is the PWM duty cycle (ranging from 0 to 1), and \(k\) is a motor-specific constant derived from experimentation. For our agricultural drone, we calibrated each motor to ensure balanced thrust, critical for stable flight. Similarly, sensor data acquisition involves sampling analog signals and converting them to digital values. For instance, the barometer outputs pressure readings that are converted to altitude using the hypsometric formula: $$h = \frac{T_0}{L} \left[ \left( \frac{P_0}{P} \right)^{\frac{R L}{g M}} – 1 \right]$$ where \(h\) is altitude, \(T_0\) is sea level temperature, \(L\) is temperature lapse rate, \(P_0\) is sea level pressure, \(P\) is measured pressure, \(R\) is gas constant, \(g\) is gravity, and \(M\) is molar mass. Such formulas are implemented in software to achieve accurate positioning for the agricultural drone.
The software architecture is designed around a main loop and interrupt service routines (ISRs) for timely data processing. The main function initializes system clocks, interrupts, and peripherals like sensors and communication modules. After initialization, it enters a while loop, waiting for interrupts to trigger data processing. Two key interrupts are used: timer interrupts for periodic sensor reading and control updates, and external interrupts for handling remote control signals. The timer interrupt is set at a high frequency (e.g., 1 kHz) to ensure real-time responsiveness. Within this interrupt, sensor data is collected, filtered, and fused to estimate the drone’s state. The state estimation employs a Kalman filter, which combines predictions from gyroscope data with measurements from accelerometers and magnetometers. The Kalman filter equations are as follows:
Prediction step:
$$ \hat{x}_k^- = A \hat{x}_{k-1} + B u_k $$
$$ P_k^- = A P_{k-1} A^T + Q $$
Update step:
$$ K_k = P_k^- H^T (H P_k^- H^T + R)^{-1} $$
$$ \hat{x}_k = \hat{x}_k^- + K_k (z_k – H \hat{x}_k^-) $$
$$ P_k = (I – K_k H) P_k^- $$
where \(\hat{x}_k\) is the state estimate (e.g., orientation angles), \(P_k\) is the error covariance, \(A\) and \(B\) are state transition matrices, \(u_k\) is control input, \(Q\) and \(R\) are process and measurement noise covariances, \(H\) is measurement matrix, and \(z_k\) is sensor measurement. This filter effectively reduces noise from sensors like the MPU6050, providing smooth attitude estimates for the agricultural drone.
Following state estimation, control algorithms generate PWM outputs for motor control. We use a cascaded PID controller structure, with an outer loop for angle control and an inner loop for angular rate control. The PID control law for the outer loop is:
$$ \theta_{des} = K_{p,\theta} e_\theta + K_{i,\theta} \int e_\theta dt + K_{d,\theta} \frac{de_\theta}{dt} $$
where \(\theta_{des}\) is the desired angle, \(e_\theta\) is the angle error, and \(K_{p,\theta}\), \(K_{i,\theta}\), \(K_{d,\theta}\) are proportional, integral, and derivative gains. The inner loop then computes the desired angular rate:
$$ \omega_{des} = K_{p,\omega} e_\omega + K_{i,\omega} \int e_\omega dt + K_{d,\omega} \frac{de_\omega}{dt} $$
where \(\omega_{des}\) is the desired angular rate, and \(e_\omega\) is the rate error. These desired values are converted to PWM signals using a mixing algorithm that distributes thrust across four motors based on the drone’s kinematics. For a quadcopter, the relationship between motor speeds and body torques can be expressed as:
$$ \begin{bmatrix} \tau_x \\ \tau_y \\ \tau_z \\ T \end{bmatrix} = \begin{bmatrix} 0 & -l & 0 & l \\ l & 0 & -l & 0 \\ -c & c & -c & c \\ 1 & 1 & 1 & 1 \end{bmatrix} \begin{bmatrix} \omega_1^2 \\ \omega_2^2 \\ \omega_3^2 \\ \omega_4^2 \end{bmatrix} $$
where \(\tau_x, \tau_y, \tau_z\) are roll, pitch, and yaw torques, \(T\) is total thrust, \(l\) is arm length, \(c\) is a drag coefficient, and \(\omega_i\) are motor speeds. By inverting this matrix, we compute required motor speeds from desired torques and thrust, ensuring stable flight for the agricultural drone.
For pesticide spraying, the software includes an external interrupt routine that captures PWM signals from the remote control. When a specific channel’s PWM value exceeds a threshold (e.g., 1500 μs), the microcontroller sets a GPIO pin high, activating a relay that powers the water pump. This allows operators to manually trigger spraying during flight. Additionally, the GPS module provides real-time location data, which can be used to automate spraying at predefined coordinates. The GPS data is parsed from NMEA sentences, such as GGA for position and time. We implement a geofencing algorithm that compares current coordinates with target areas: if the agricultural drone enters a designated zone, spraying is automatically enabled. This enhances precision and reduces pesticide waste.
To validate our agricultural drone system, we conducted extensive experiments focusing on flight stability and spraying performance. The drone was assembled with all components mounted on a carbon fiber frame, weighing approximately 2 kg including a 500 mL pesticide tank. We first tested the Kalman filter’s effectiveness by comparing raw sensor data with filtered outputs. As shown in Figure 15 of the original text (simulated here), the filtered data exhibited significantly reduced noise, with standard deviation decreasing by over 70% for acceleration readings. This improvement is crucial for accurate attitude estimation, especially in windy conditions common in agricultural fields.
| Sensor | Raw Data Noise (σ) | Filtered Data Noise (σ) | Noise Reduction (%) |
|---|---|---|---|
| Accelerometer (X-axis) | 0.15 m/s² | 0.04 m/s² | 73.3 |
| Gyroscope (Y-axis) | 0.08 rad/s | 0.02 rad/s | 75.0 |
| Magnetometer (Z-axis) | 0.25 μT | 0.05 μT | 80.0 |
Next, we tuned the cascaded PID controller parameters using a systematic approach. The drone was tethered to allow rotation only in one axis, and we adjusted gains sequentially: first inner loop P and D gains, then outer loop P, I, and D gains. The objective was to achieve fast response with minimal overshoot. We used the Ziegler-Nichols method as a starting point and refined gains through iterative testing. The final parameters for roll control were: inner loop \(K_{p,\omega} = 0.8\), \(K_{d,\omega} = 0.05\); outer loop \(K_{p,\theta} = 5.0\), \(K_{i,\theta} = 0.1\), \(K_{d,\theta} = 0.5\). These values ensured that when the agricultural drone was tilted by 30 degrees and released, it returned to level within 0.5 seconds with less than 5% overshoot. Similar parameters were applied to pitch and yaw axes, with minor adjustments based on flight tests.
The spraying performance was evaluated through two tests: total spray volume per minute and coverage area at a fixed height. We used water as a substitute for pesticide and measured output over multiple trials. The results are summarized in Tables 3 and 4 below. The pump consistently delivered over 120 mL/min, which translates to a spraying rate suitable for typical agricultural needs. For instance, if a crop requires 12 g/m² of pesticide, our agricultural drone can cover approximately 0.6 m² per second at 0.9 m height, allowing it to treat large fields efficiently. We also tested the GPS-based automation by programming the drone to spray only within a 10 m x 10 m square area. The system successfully activated the pump upon entry and deactivated it upon exit, demonstrating precise location control.
| Trial | Duration (min) | Spray Volume (mL) | Rate (mL/min) |
|---|---|---|---|
| 1 | 1 | 133 | 133 |
| 2 | 1 | 128 | 128 |
| 3 | 1 | 120 | 120 |
| Average | 1 | 127 | 127 |
| Trial | Spray Area (m²) | Uniformity Index (%) |
|---|---|---|
| 1 | 0.63585 | 88.5 |
| 2 | 0.58059 | 85.2 |
| 3 | 0.66442 | 90.1 |
| Average | 0.62695 | 87.9 |
The uniformity index was calculated based on droplet distribution measured using water-sensitive papers. It indicates that the agricultural drone provides even coverage, which is essential for effective pest control. Compared to manual spraying, which often has uneven application due to human error, our system offers superior consistency. Additionally, the downward airflow from the drone’s propellers helps penetrate crop canopies, reaching undersides of leaves and roots—a key advantage highlighted in prior research.
Beyond basic functionality, we explored advanced features for the agricultural drone. For instance, we implemented a waypoint navigation system using GPS and onboard algorithms. The drone can follow predefined paths uploaded via a ground station, enabling autonomous missions over large fields. The path planning algorithm uses Bézier curves to generate smooth trajectories: $$ B(t) = \sum_{i=0}^n \binom{n}{i} (1-t)^{n-i} t^i P_i, \quad t \in [0,1] $$ where \(P_i\) are control points and \(n\) is the curve order. This minimizes abrupt turns, conserving battery life and ensuring stable spraying. We also integrated a telemetry system using 433 MHz radio modules to transmit real-time data like battery voltage, altitude, and spraying status to a laptop running Mission Planner software. This allows operators to monitor the agricultural drone remotely and make adjustments as needed.
Safety is a paramount concern for agricultural drones. Our design includes multiple fail-safes: if communication with the remote is lost, the drone enters a return-to-home (RTH) mode using GPS coordinates; if battery voltage drops below a threshold, it automatically lands; and if sensor data is inconsistent, it hovers and alerts the operator. These features mitigate risks in field operations. Moreover, the use of STM32’s hardware watchdog timer ensures that the system reboots in case of software hangs, enhancing reliability. We also conducted vibration analysis to ensure that motor oscillations do not interfere with sensor readings. By mounting sensors on damping materials, we reduced vibration-induced errors by over 50%, as quantified by power spectral density measurements.
The economic and environmental benefits of our agricultural drone are significant. Traditional manual spraying typically covers 0.1-0.2 hectares per hour, whereas our drone can cover 0.5-1.0 hectares per hour, depending on battery capacity and tank size. This represents a 5-10 times efficiency improvement. Additionally, pesticide usage can be reduced by up to 30% through targeted application, lowering costs and environmental impact. For example, if a farmer sprays 100 hectares annually, switching to our agricultural drone could save hundreds of liters of pesticide and reduce operator exposure to toxins. The initial investment in the drone system is offset by long-term savings and increased crop yields due to better pest management.
Looking ahead, there are several avenues for enhancing our agricultural drone. Future work could involve integrating multispectral cameras for crop health monitoring, using machine learning algorithms to detect pest infestations and adjust spraying in real time. We also plan to optimize battery life through solar charging modules and lightweight materials. Collaboration with agricultural experts will help tailor the system to specific crops like rice, wheat, or orchards. Furthermore, regulatory frameworks for drone operations in agriculture are evolving, and our design complies with emerging standards for safety and privacy.
In conclusion, we have developed a robust remote-controlled agricultural drone based on the STM32 microcontroller. Its hardware-software co-design ensures stable flight and precise pesticide spraying, addressing key limitations of manual methods. Through experiments, we validated its performance in terms of noise reduction, control stability, and spraying efficiency. The agricultural drone offers a scalable solution for precision farming, with potential to revolutionize agricultural practices by improving productivity, reducing health risks, and promoting sustainability. As technology advances, such drones will become integral to smart agriculture, enabling data-driven decisions and automated operations for a growing global population.