Compared to ultrasonic sensors, LiDAR sensors have advantages, such as a higher measurement precision, a higher response speed, and a higher interference resistance. In this study, a new LiDAR sensor suitable for H_b detection was developed based on a low-cost, small-sized, low-power single-point laser rangefinder, TFmini Plus, manufactured by Benewake Co., Ltd. (Beijing, China), which measures distances based on the time-of-flight (ToF) principle. During the measurement process, the sensor periodically outwardly transmits a modulated near-infrared wave that is reflected upon contact with an object. The ToF is calculated by determining the difference in phase between the incident and reflected waves. On this basis, the distance from the tested target is determined. This single-point ranging sensor is composed primarily of a front shell, a transmitting lens, a receiver lens, a circuit board, screws, lenses, an ultrasonic welded structure, a back shell, a tail cable, a tail cable fastener, location columns, a mounting hole, a barbed structure, a protective shield, mounting grooves, and a partition structure (Figure 1a). Equipped with a unique optical and electrical design, this sensor can measure distances in a stable, accurate, and highly sensitive manner. The parameter indicators of the LiDAR sensor are shown in Table 2.

The single-point LiDAR sensor outputs signals at the transistor-transistor logic (TTL) level and thus has a limited signal transmission distance. H_b detection sensors installed at the two ends of a boom transmit signals over a distance exceeding 10 m, but the TTL transmission mode cannot ensure signal validity. To address this problem, a signal processing module was developed based on an STM32F103RCT6 single-chip microcomputer (SCM). This module can convert the signal output of the LiDAR sensor to a 4–20 mA signal output. Figure 1b shows the circuit board of this module. This signal processing module primarily consists of a main control module, a power module, and a current module. The main control module, composed of an ST STM32F103RCT6 SCM with a 72-MHz central processing unit (CPU) and 1-MB flash memory, is capable of simultaneously supporting serial port and controller area network (CAN) communication. The program download module has a serial wire debug (SWD) download interface that allows the user to develop and debug programs. The power module uses MPS MP1584DN and AMS AMS1117-3.3 switch mode voltage regulators and can supply a 12 V direct current (DC) and output 5 V and 3.3 V. The current output module, composed of a Texas Instruments voltage-to-current converter chip, is capable of converting a sensor signal to a 4–20 mA current signal output. During operation, the power module converts a 12 V DC input voltage to 5 V and 3.3 V to supply power to the main control and current output modules. The main control module acquires the distance information measured by the LiDAR sensor through communication at the TTL level via the serial port and subsequently converts it to a voltage signal and outputs it to the current output module via a pin of the SCM. The current output module converts the voltage signal to a current signal output. The detection range of the LiDAR sensor of 0.1–12 m corresponds to a current signal range of 4–20 mA. To improve the stability and accuracy of the output current signal in the signal conversion process, a PID control algorithm was incorporated into the signal conversion algorithm. In addition, to enhance the ranging performance of the LiDAR sensor in complex operating conditions, an adaptive algorithm for multiple application settings and targets was added to the signal conversion process, and multiple adjustable configurations and parameters were included. These measures can ensure that the LiDAR sensor performs well in ranging operations in complex settings and meets agricultural requirements in complex applications. Finally, the LiDAR sensor was encapsulated to facilitate the mounting and protection, as shown in Figure 1c,d. The boom sprayers often operate in complex field conditions. The encapsulated sensor was subjected to environmental temperature testing at −40 to 85 °C, vibration testing, and Ingress Protection 67 testing to determine if it could meet the operating requirements.

The LiDAR sensor needs to be checked prior to use. Figure 2 shows a photograph of the experimental setting for checking the LiDAR sensor. The LiDAR sensor was placed in an unobstructed, open space. A computer was connected via a universal serial bus (USB) serial port to a current signal acquisition card. The computer interface could display the current signal output by the LiDAR sensor in real time. A paper box was used as an object for detection in the experiment. A tape measure was placed directly below the central axis of the beam transmitted by the LiDAR sensor to record the distance over which the paper box moved. During the experiment, the paper box was placed directly in front of the sensor. The paper box was moved at distance intervals of 10 cm within the measuring range (0.1–2 m) of the LiDAR sensor. The current signal output by the LiDAR sensor was recorded when the paper box was moved. Three repeats were performed for each process. The average of the three measurements was used to fit a detected distance-output current curve.

The detection performance of the LiDAR sensor was mainly affected by the moving speed and the uneven wheat stubble canopy. To investigate the effect of moving speed and uneven wheat stubble canopy on the detection performance of the LiDAR sensor, an experimental platform for H_b detection by the LiDAR sensor was constructed, as shown in Figure 3c. This experimental platform was composed primarily of a sensor fixation bar, a sliding track, a stepper motor, a synchronous belt, a sliding block, a stepper motor driver, and a controller. The total length of the experimental platform was 6.5 m. The effective length of the sliding track was 6.0 m. The sliding track was fixed in a bridge mode. The sliding block, with a length of 0.3 m, was used to fix the sensor fixation bar. Four CZ-7166 limit switches (Testo Instruments International Trading (Shanghai, China) Co., Ltd.), including two outer limit switches and two inner limit switches, were set at the two ends of the sliding track. The two outer limit switches, which served as safety limit switches, prevented the sliding block from impacting the two ends of the sliding track. The two inner limit switches, which served as position limit switches, limited the range of movement of the sliding block to 5 m. Of the mobile devices, the KH-01 stepper motor controller (Relong Automation Equipment (Zhejiang, China) Co., Ltd.) precisely controlled the HB80D stepper motor (Hongbaoda Industry (Shenzhen, China) Co., Ltd.) via the HB860D driver (Hongbaoda Industry Co., Ltd.) to drive the sliding block to steadily move along the track. The speed and range of the sliding block could be precisely controlled. During the experiment, to accurately record H_b values detected by the LiDAR sensor at different locations in the horizontal direction, an S85-MH-5-Y03-OOI laser rangefinder with a measuring range of 10 m and a current signal output of 4–20 mA (Datalogic Industrial Automation Company Ltd. (Shenzhen, China)) was fixed at one end of the sliding platform. The laser-reflecting plate and the LiDAR sensor were placed at the same horizontal location. The laser rangefinder recorded the horizontal location of the LiDAR sensor by measuring the location of the laser-reflecting plate. An SG-AD-Modbus-8I data acquisition card (DAC) (Sange Electronic Technology (Tianjin, China) Co., Ltd.) was used to simultaneously collect measurement data from the laser rangefinder and the LiDAR sensor. The DAC communicated with the upper computer via a RS485 serial port. A general acquisition system for analogue signals designed by Zhai et al. [26] that can collect experimental data and store the data in a database to facilitate subsequent experimental analysis was used in the upper computer.

To examine the effects of the movement speed of the LiDAR sensor (V_LiDAR) on the detection performance, a step detection experiment was designed. In this experiment, 70 paper boxes (0.30 m × 0.20 m × 0.10 m) were stacked to form a flight of eight steps, as shown in Figure 3a. The LiDAR sensor was fixed at one end of the sensor fixation bar and could detect vertically downward (Figure 3a). Prior to the experiment, the sliding block of the experimental platform for H_b detection was manually pushed to change the distance detected by the LiDAR sensor. A tape measure was used to measure the distance of the LiDAR sensor from the ground and each step, and the corresponding measurements were recorded. During the experiment, the center of the step-like ground for detection was placed on a straight line parallel to and at a horizontal distance of 1.0 m from the center line of the track of the sliding platform. The sliding block was used to drive the LiDAR sensor to move along a straight line at a uniform speed. The LiDAR sensor was located directly above the surface of the steps for detection at the site throughout its movement with the sliding block. In addition, the position limit switches were used to control the LiDAR sensor in such a way that the LiDAR sensor took measurements at distance intervals of 5.0 m. According to the experimental design in Zhai et al. [26], the measurements taken by the LiDAR sensor when moving at a speed of 0.05 m/s were treated as true values. The sliding block was controlled to move at speeds of 0.3, 0.6, 0.9, and 1.2 m/s. The measurements taken by the LiDAR sensor when moving at various speeds were collected. Three repeats were performed for each test. The average of the three measurements taken by the LiDAR sensor was used as the ultimate measured value.

A field ground detection experiment was performed in a rotary-tilled field, as shown in Figure 3b. In this experiment, the distance between the signal transmission plane of the LiDAR sensor and the ground was measured. The experimental platform for H_b detection was placed in a rotary-tilled field. The distance between the LiDAR sensor and the ground was set to 1.4 m. Soil was manually heaped to produce a small slope, approximately 0.5 and 0.4 m in length and height, respectively, on the ground that was a horizontal distance of 1.0 m from the centre line of the track of the sliding platform to simulate an undulating surface. The sliding block was used to drive the LiDAR sensor to move over 5.0 m along a straight line at a uniform speed directly above the horizontal line of the small slope, as shown in Figure 3b. During the experiment, a tape measure was first used to measure the distance between the LiDAR sensor and the ground at intervals of 0.1 m. Then, V_LiDAR was adjusted to 0.05 m/s. The measurements taken by the LiDAR sensor when moving at this speed were treated as the true values. Finally, the sliding block was controlled to move at speeds of 0.3, 0.6, 0.9, and 1.2 m/s. The measurements taken by the LiDAR sensor at various speeds were collected. Three repeats were performed for each test. The average of the three measurements taken by the LiDAR sensor was used as the ultimate measured value.

The autumn crops are generally planted without tillage after harvesting wheat in China. It is necessary to spray pesticides and remove weeds before no-tillage planting. To examine the effects of WS canopies on the H_b detection performance of the LiDAR sensor, an H_b detection experiment in a post-harvest WS field was designed, as shown in Figure 3d. Before the experiment, an unharvested wheat plot 5.0 m long and 0.3 m wide was randomly selected. A harvester was then used to harvest the wheat in the plot. The header height of the harvester was set to 0.35, 0.25, and 0.15 m for areas a, b, and c of the plot, respectively, as shown in Figure 3d. Post-harvest areas a, b, and c were 1.5, 2.2, and 1.3 m in length, respectively. During the experiment, the experimental platform for H_b detection was first placed in each WS area in such a way that the LiDAR sensor was directly above the selected WS canopy. After the experimental platform was placed, the LiDAR sensor was moved to the starting position. A tape measure was used to measure the distances of the LiDAR sensor from the WS canopy and the ground at intervals of 0.1 m until the LiDAR sensor moved to its final position. Subsequently, V_LiDAR was adjusted to 0.05 m/s, and the measurements taken by the LiDAR sensor when moving at this speed were treated as the true values. Finally, the sliding block was controlled to move at speeds of 0.3, 0.6, 0.9, and 1.2 m/s. The measurements taken by the LiDAR sensor at these speeds were collected. Three repeats were performed for each test. The average of the three measurements taken by the LiDAR sensor was used as its ultimate measurement.

The ultrasonic sensors are the main type of sensor used to detect H_b. To better illustrate the feasibility of using the LiDAR sensor to detect H_b, an experiment was designed to compare the LiDAR sensor with an ultrasonic sensor. A knapsack boom sprayer was selected for the experiment. The boom of this sprayer was 21 m in length and was divided into three sections, namely, left, middle, and right sections. The left and right sections of the boom were each further divided into two sections. The extension and contraction of the boom was controlled by a hydraulic cylinder (Figure 4). One LiDAR sensor and one ultrasonic sensor were placed at the left, middle, and right sections of the boom to monitor the changes in H_b in each section in real-time.

The experiment was performed in a rotary-tilled field (400 m × 50 m) and a WS field (560 m × 240 m) with a WS height (H_WS) of 20 cm at the National Research Station for Precision Agriculture, Beijing, China, as shown in Figure 5. During the experiment, the tank of the sprayer was filled half full with water. The height of the nozzle of the sprayer head was set to 110 cm above ground. The signal transmission plane of the LiDAR sensor and the nozzle were set on the same horizontal plane. The sprayer was mounted on a tractor. The Chinese standard GB/T 20183.3-2006 “Equipment for crop protection—Spraying equipment—Part 3: Test methods for volume/hectare adjustment systems of agricultural hydraulic pressure sprayers” is considered, which requires that the speed during the test be 1.5 m/s to 2.5 m/s, and studies have usually sprayed at a moving speed of 1.9 m/s in recent years [27]. The sprayer was operated at speeds of 2.0 m/s. The experiment was repeated three times. The changes in H_b during each test were recorded.

The K-means clustering algorithm was used to process the LiDAR sensor detection data. The algorithm first selects k objects at random from data objects as the initial clustering centres, calculates the distance between each object and these central objects according to the mean value of each clustering object, and re-divides the corresponding objects according to the minimum distance. Then, the algorithm re-calculates the average of each cluster. This process is repeated until each cluster no longer changes [28,29]. Usually, the criteria for the K-means algorithm use the squared error criterion function, which is defined as:

where E is the sum of the squared error in the data set of all the LiDAR sensor detection values, Ci is the cluster of LiDAR sensor detection values, x is the LiDAR sensor detection value, and xi is the average value of cluster Ci. k is the number of clusters.

Our previous study contains more details on this process [30]. The flow chart of the H_b calculation based on the K-means clustering algorithm is shown in Figure 6. During the working process, the H_b detection system reads the three LiDAR sensor signals in real-time, and it first judges the validity of the sensor detection signal. If the detection signal is abnormal, it will judge whether the sensor has entered the blind zone. Once the LiDAR sensor has entered the blind zone, the system will send a data exception command. Otherwise, the system will convert the obtained data into H_b values every 1 s and store them in an array form and then call the K-means clustering algorithm to preprocess the acquired H_b detection data, ultimately obtaining accurate H_b detection values.

The H_b values measured by the LiDAR sensor for the left, middle, and right sections of the boom were fitted to the output current data. Figure 7 shows the fitted curves for the left, middle, and right sections of the boom. The R² values of the equations are 0.9994, 0.9996, and 0.9998, respectively. Evidently, the three LiDAR sensors show good linearity and meet the operating requirements.

Figure 8 shows the distances detected by the LiDAR sensor when moving at different speeds. As demonstrated in Figure 8, the distances detected by the LiDAR sensor when moving at different speeds can reflect the changes in the step height. However, the distances detected by the LiDAR sensor at the junctions of the ground and the upper surface of the flight of steps formed by paper boxes differ from the true values. In addition, this difference becomes increasingly prominent as V_LiDAR increases. This is because the angle of spread of the beam transmitted by the LiDAR sensor is 3.6°. As the LiDAR sensor moves to directly above each junction of the ground and the upper surface of the flight of steps, the beam transmitted by the LiDAR sensor partially reaches the ground and partially reaches the upper surface of the flight of steps, thus affecting the distance detected by the LiDAR sensor. As the detection of the LiDAR sensor moves from the ground to the upper surface of the flight of steps, there is a detection delay in the LiDAR sensor. Moreover, as V_LiDAR increases from 0.3 to 0.9 m/s, the detection delay increases. As V_LiDAR increases from 0.9 to 1.2 m/s, the detection delay decreases. This is because as V_LiDAR increases, there is a decrease in the amount of data sampled over the same distance of movement, and the low amount of sampled data increases the detection delay. The junction between the ground and the upper surface of the flight of steps is relatively small. When moving at a speed of 1.2 m/s, the LiDAR sensor is unable to detect the junction; therefore, the detection delay is reduced. As the LiDAR sensor moves from the upper surface of the flight of steps to the ground, the distances detected by the LiDAR sensor are between those from the ground and the upper surface of the flight of steps. This is because the large difference between the heights of the ground and the upper surface of the flight of steps at this junction increases the probability for the beam transmitted by the LiDAR sensor to reach the sides of the flight of steps. A low V_LiDAR leads to a high probability that the beam transmitted by the LiDAR sensor reaches the sides of the flight of steps.

Figure 9 shows the distances from the field ground detected by the LiDAR sensor at different detection speeds. As demonstrated in Figure 9, the distances detected by the LiDAR sensor can reflect the changes in the slope of the field ground well. Thus, the LiDAR sensor can be used to detect changes in H_b in rotary-tilled fields. Compared to the flight of steps used in the step detection experiment, the height of the surface of the slope in the field varies gently. There is no notable detection delay as V_LiDAR increases. However, the distances detected by the LiDAR sensor at the slope differ from the true values. In addition, this difference increases as V_LiDAR increases. This is because as V_LiDAR increases, there is a decrease in the amount of sampled data. The low amount of sampled data reduces the detection accuracy of the LiDAR sensor and, thereby, increases its detection error.

Figure 10 shows the distances detected by the LiDAR sensor at different H_WS and V_LiDAR values. As demonstrated in Figure 10, when H_WS remains unchanged, there is a small change in the distance detected by the LiDAR sensor as V_LiDAR increases from 0.05 to 1.2 m/s. This suggests that when H_WS remains unchanged, V_LiDAR has less of an impact on the distance detected by the LiDAR sensor. As H_WS increases, the distance detected by the LiDAR sensor increases. The detected distances can reflect the changes in H_WS. However, the distance detected by the LiDAR sensor is between the ground and the surface of the WS canopy. In addition, there is no notable pattern between the changes in the distance detected by the LiDAR sensor and H_WS. This finding suggests that H_WS affects the distance detected by the LiDAR sensor and that the distance detected by the LiDAR sensor is between the ground and the surface of the WS canopy.

To further analyse the distribution of the distance detected by the LiDAR sensor within the WS canopy, the distances detected by the LiDAR sensor at different H_WS values were subjected to a cluster analysis using the K-means clustering analysis algorithm, and the results are shown in Figure 11. The clustering results show that the detection signal transmitted by the LiDAR sensor partially reaches the inside of the WS canopy and partially reaches the WS roots and that the signal reflection sites are distributed in a random manner. Since the post-harvest wheat straws are thin-walled circular tubes, the WS canopy exhibits a sparse, porous structure that absorbs the detection beam and, thereby, causes the reflection sites of the signal transmitted by the laser within the WS canopy to change. Moreover, due to tillering, there are many branches at the WS roots near the ground, resulting in a relatively high density at the WS roots. Most of the signals transmitted by the LiDAR sensor reach the WS roots. Consequently, the distances from the WS detected by the LiDAR sensor are close to those from the ground.

To analyse the effects of H_WS on the detection performance of the LiDAR sensor for the WS roots and canopy, the mean errors of the distances detected by the LiDAR sensor from the WS roots and canopy in areas a, b, and c were statistically analysed, and the results are shown in Figure 12. As demonstrated in Figure 12, for each area, the mean error of the distances detected by the LiDAR sensor from the WS roots is lower than that of the distances detected by the LiDAR sensor from the WS canopy. As H_WS decreases, the mean detection errors of the LiDAR sensor for both the WS roots and canopy in each area gradually decrease. The maximum and minimum detection errors of the LiDAR sensor for the WS roots are 3.08% and 1.40%, respectively. There are no significant changes in the detection error of the LiDAR sensor between areas a and b. However, the detection error of the LiDAR sensor for area c is notably lower than those for areas a and b. This is because in areas a and b, the WS canopy is relatively high, and the litter layer at the WS roots had been relatively insignificantly damaged during the harvesting process. In contrast, in area c, the litter layer had been relatively significantly damaged during the harvesting process, so the beam transmitted by the LiDAR sensor mostly reaches the WS roots. As a consequence, the detection error of the LiDAR sensor for the WS roots is demonstrably lower in area c than areas a and b. For the WS canopies, the maximum and minimum detection errors of the LiDAR sensor are 18.11% and 7.65%, respectively, and the detection error of the LiDAR sensor changes considerably in areas a, b, and c. This is because as H_WS decreases, there is an increase in the density of the WS canopy. In addition, the presence of a relatively large amount of litter below a relatively low WS canopy improves the ability of the WS canopy to reflect the laser beam and, thereby, increases the detection accuracy of the LiDAR sensor for the WS canopy.

To investigate the effects of V_LiDAR on the detection performance of the LiDAR sensor for the WS roots and canopy in each area, the detection errors of the LiDAR sensor moving at different speeds for the WS roots and canopy in each area were analysed. Table 3 summarizes the results. As demonstrated in Table 3, V_LiDAR has a greater impact on the detection performance of the LiDAR sensor for the WS canopy than that for the WS roots in each area. For the WS roots in areas a, b, and c, the maximum detection errors of the LiDAR sensors are 4.66%, 3.67%, and 2.44%, respectively, and the minimum detection errors are 2.43%, 2.45%, and 0.68%, respectively. For the WS canopies in areas a, b, and c, the maximum detection errors of the LiDAR sensors are 20.02%, 12.84%, and 8.50%, respectively, and the minimum detection errors are 16.80%, 11.48%, and 6.65%, respectively. When H_WS remains unchanged, V_LiDAR has a slight impact on the detection errors of the LiDAR sensor for the WS roots and canopy. In addition, when V_LiDAR remains unchanged, as the WS canopy height decreases, the detection errors of the LiDAR sensor for the WS roots and canopy decrease.

In summary, V_LiDAR has a small impact on the detection performance of the LiDAR sensor for the WS roots and canopy. However, V_LiDAR has a greater impact on the detection performance of the LiDAR sensor for the WS canopy than that for the WS roots. H_WS has a greater impact on the detection performance of the LiDAR sensor for the WS roots and canopy. Similarly, H_WS has a greater impact on the detection performance of the LiDAR sensor for the WS canopy than that for the WS roots. H_WS and V_LiDAR affect the detection error of the LiDAR sensor for the WS roots but to a relatively small extent. The detection error of the LiDAR sensor for the WS roots is less than 5.00% in all the cases (the maximum detection error is 4.66%). and the detection error of the LiDAR sensor for the WS roots is greater than 8.00% (the minimum detection error is 8.50%).

Figure 13 shows the H_b values detected by the LiDAR and ultrasonic sensors in the rotary-tilled and WS fields, respectively. As demonstrated in Figure 13, the H_b values detected by the two sensors show similar trends. Currently, ultrasonic sensors are the main type of sensor used to detect H_b, and detection data from ultrasonic sensors can reflect changes in H_b well, which further demonstrates that LiDAR sensors can satisfactorily reflect changes in H_b and can be employed to detect changes in H_b. In addition, the H_b values detected by the two sensors in the rotary-tilled field are basically the same, whereas there is a difference between the H_b values detected by the two sensors in the WS field. Moreover, in the WS field, the H_b values detected by the LiDAR sensor were lower than those detected by the ultrasonic sensor. For a boom sprayer, H_b changes less in the middle section of the boom than at the two ends of the boom. The changes in H_b monitored by the two sensors for the middle section of the boom can reflect the distances detected within the canopy. Figure 13b,e show the H_b values detected by the two sensors in the middle section of the boom. As demonstrated in Figure 13b,e, the distance detected by the LiDAR sensor is lower than that of the ultrasonic sensor within the WS canopy, and the distances detected by the LiDAR sensor for the WS canopy fluctuate to a relatively large extent, which further demonstrates that the distances detected by LiDAR sensor for the WS canopy include the WS canopy and roots.

Not applicable.