Design and Analysis of Sowing Depth Detection and Control Device for Multi-Row Wheat Seeders Adapted to Different Terrain Variations

Abstract

To address the issue of reduced sowing depth detection accuracy caused by varying soil topography during the operation of wheat row drills, an indoor bench test device suitable for wheat row drills was developed. The device integrates a laser sensor and an array sensor for terrain and sowing depth detection. The laser sensor provides the detected sowing depth values, while the array sensor captures different terrain features. The actual sowing depth values are obtained through the indoor experimental setup. The experiment includes three types of terrain: convex, concave, and flat. The terrain slope matrix is obtained using the array sensor, and terrain feature values are extracted. The laser sensor is then used to obtain the detected sowing depth, and the actual sowing depth is manually measured. PCA analysis is conducted to correlate terrain feature values with sowing depth deviations. Results indicate that under different terrain conditions, the slope mean and slope standard deviation are the main components affecting sowing depth deviations. Compared to using a single sensor, this system enables more accurate sowing depth measurement by analyzing terrain features. The device provides valuable data support for controlling sowing depth under varying terrain conditions in subsequent operations.

1. Introduction

Sowing depth is a critical technical parameter in crop planting. Factors such as soil type, soil moisture, and seeder structure significantly impact the planting depth [1,2]. For multi-row wheat seeders, the presence of large amounts of straw residue on the soil surface or varied surface undulations often results in poor ground evenness during operation, which directly affects the actual sowing depth. Thus, the stability of sowing depth is a vital indicator of a seeder’s performance and can directly reflect the quality of a no-till seeder [3,4].

Sowing depth consistency is defined as maintaining a uniform depth throughout the sowing process. Consistent and appropriate sowing depth and compaction ensure optimal soil-to-seed contact, facilitating water uptake and promoting rapid and uniform seed germination [5]. Consequently, achieving stable and consistent sowing depth remains a significant challenge. Sowing depth detection, a key component of precision agriculture technology, directly affects crop growth, yield, and quality. With the rapid advancement of measurement and control technologies, sowing depth detection is shifting towards higher precision and quality [6]. Traditionally, sowing depth was measured manually, a process that is both time-consuming and labor-intensive, with no capacity for real-time or continuous monitoring. With the increasing demand for mechanization and automation in agriculture, manual sowing depth measurement is becoming obsolete, highlighting the need for more efficient and accurate automated detection technologies [7].

Current sowing depth detection methods can be categorized into contact and non-contact methods [8]. Contact methods measure sowing depth using mechanical or profiling mechanisms. While straightforward and easy to implement, these methods are significantly influenced by factors such as soil surface irregularities, compaction, and particle size, leading to lower measurement precision and consistency [9,10]. Moreover, contact methods cannot achieve continuous measurement, limiting their application in modern agriculture.

Non-contact methods utilize advanced sensor technologies, including laser sensors, ultrasonic sensors, and FLEX sensors, to achieve efficient, real-time, and precise sowing depth detection [11,12]. For example, Mapoka et al. proposed a ground-penetrating radar-based non-destructive sensing technology for detecting corn seeds in closed furrows and measuring planting depth, exploring its potential as a supplement to traditional methods [13]. Nielsen et al. designed an innovative sowing depth measurement system integrating linear displacement sensors and ultrasonic sensors to measure the furrow opener’s distance relative to the frame and the frame’s distance relative to the ground. By fusing signals from both sensors, the system accurately calculates the furrow opener’s actual working depth. Similarly, Lee et al. developed a tillage depth detection system that uses optical sensors to measure the distance to the ground, tilt sensors to monitor the tractor’s pitch angle, and lift-arm sensors to determine the lift arm position. By integrating data from these three sensors and applying a mathematical model, the system accurately measures tillage depth [14]. These technologies avoid mechanical contact and measure the distance between the seeder and the ground without disturbing the soil surface, enabling sowing depth calculations. Among non-contact methods, laser and ultrasonic sensors are prominent due to their high precision and fast response, making them critical components of sowing depth detection systems [15]. Laser sensors, in particular, use pulse-based laser emission and reception to measure the relative distance between the soil surface and the seeder accurately, even under irregular and undulating soil conditions [16].

However, while laser sensors offer high precision, they are susceptible to environmental factors such as soil surface reflectivity, moisture changes, and lighting conditions. Furthermore, single-sensor systems often struggle with complex terrain or soil conditions, leading to error accumulation and measurement instability [17]. As a result, multi-sensor data fusion systems for sowing depth detection have become a research focus. By combining laser sensors with array sensors, the system leverages the precision of laser sensors and the terrain feature extraction capabilities of array sensors, enabling more accurate real-time sowing depth monitoring [18].

Array sensors, capturing high-resolution surface image data, effectively extract soil surface undulations and feature changes, offering robust adaptability to complex terrain. This technology complements laser sensors by providing additional terrain data, which helps correct laser measurement errors [19]. Consequently, the fusion of laser and array sensor data significantly enhances the system’s measurement accuracy and stability, ensuring precise and continuous sowing depth monitoring under complex environmental conditions [20].

The objective of this study is to design and develop a sowing depth detection system based on the fusion of laser and array sensors. By integrating multi-sensor data, the system achieves precise sowing depth measurement under various soil conditions, terrain variations, and environmental factors. Indoor bench tests validate the system’s adaptability and stability under different terrain conditions, while an error correction model is established to support the development and application of precision multi-row wheat seeders.

2. Materials and Methods

2.1. Materials and Test Equipment

2.1.1. Characteristics of Ground Surface in Wheat Row Sowing

The accuracy of sowing depth is critical to ensuring the quality of sowing, and ground surface characteristics directly affect the precision of sowing depth detection systems. Ground surface characteristics encompass various factors, including the physical state of the soil, terrain undulations, surface hardness, and moisture levels [21]. Traditionally, sowing depth detection systems rely on sensing soil surface conditions, using sensors to measure the distance between the seeder and the ground in real-time to estimate sowing depth [22]. However, these measurements are often influenced by changes in ground surface characteristics, particularly under uneven and undulating soil conditions, which can lead to accumulated measurement errors. Therefore, studying the effects of different terrain conditions on sowing depth and simulating various soil and terrain conditions through experiments are key to improving detection precision.

Factors such as soil surface condition, moisture, and particle composition significantly impact the performance of seeders, particularly in sowing depth control. Soil density and hardness directly affect the penetration capability of the seeder, determining the stability of sowing depth [23]. In practical applications, the physical properties of soil need to be measured accurately to integrate with the sowing depth detection system for real-time depth adjustment [24]. However, aside from soil properties, terrain undulations and slope variations are also critical factors affecting sowing depth. Under varying terrain conditions, slopes and uneven surfaces can reduce the stability of the seeder. In cases of steep slopes or uneven ground, the seeder may tilt, impairing sensor measurement accuracy and increasing errors in sowing depth detection. Consequently, obtaining precise terrain data and performing real-time adjustments are essential to mitigating these effects.

To systematically investigate the impact of different terrains on sowing depth, this study simulates three typical soil terrain models: concave, convex, and flat terrains (Figure 1). Each terrain model is selected based on its mechanism of influence on sowing depth, aiming to reveal the specific effects of varying soil surface characteristics and terrain undulations on sowing depth detection through experimental data [25].

The terrain conditions in the experimental bed were designed to simulate typical scenarios encountered during wheat sowing operations. Raised terrains can represent soil ridges formed by mechanical compaction, such as those left behind by tractor wheels or other heavy machinery. Sunken terrains mimicked natural depressions, erosion channels, or drainage paths, while flat terrains represented the ideal sowing conditions with uniformly leveled soil surfaces. These terrain models were selected to evaluate the adaptability and precision of the sowing depth detection system.

2.1.2. Methods and Systems for Wheat Sowing Depth Detection

(1) Sowing Depth Detection Principle

The sowing depth of a wheat row drill is controlled by adjusting the height of the depth control wheel. However, during no-till operations, factors such as surface straw and weeds, undulating soil slopes, and variations in soil compaction can affect the height of the depth control wheel, leading to discrepancies between the actual and preset sowing depths. As a result, the traditional depth control method may suffer from insufficient precision.

To address this issue, this study integrates a laser sensor into the seeder’s structure. By monitoring the changes in the distance between the laser sensor and the soil surface, variations in sowing depth can be indirectly measured. This approach effectively eliminates interference from external factors such as straw and weeds, providing a more accurate method for detecting sowing depth.

(2) Laser Sensor Principle and Measurement Method

The laser sensor is mounted on the seeder frame and integrated with the furrow opener. As the furrow opener penetrates the soil, the laser sensor descends correspondingly, reflecting changes in sowing depth. The depth control wheel, which maintains close contact with the ground, ensures that variations in the distance between the laser sensor and the soil surface accurately represent the actual sowing depth. Real-time data captured by the laser sensor enable precise depth calculations [26].

The measurement principle is illustrated in Figure 2. Let the initial distance between the laser sensor and the soil surface be L, and the actual distance during sowing be L′. The sowing depth ddd can be calculated using the formula: $\mathrm{d}=L-L^{\prime }$

By continuously calculating the difference between L and L′, the system can determine the sowing depth with high precision. This non-contact detection method effectively addresses challenges posed by varying soil conditions, slope changes, and external interferences, ensuring accurate sowing depth measurement. Moreover, it maintains efficient operational capability even under complex terrain conditions [27].

2.1.3. 3D Structural Design of Multi-Row Wheat Row Seeder

The wheat row seeder is an agricultural machine designed for precision sowing, primarily used for wheat. Its overall structure consists of several main components, including the frame, furrow opener, seed box, and seed metering device.

The 3D modeling capabilities of SolidWorks provide a clear visualization of the frame’s shape and the interconnections between various components [28,29]. In this study, SolidWorks software was employed to model and demonstrate all parts and the overall structure of the multi-row wheat row seeder. Given the complexity of the seeder’s structural design, the precise integration of each component is critical for enhancing both sowing efficiency and depth accuracy.

Key component diagrams and the full assembly illustration are shown in Figure 3. These visualizations not only highlight the functional design of the seeder but also serve as a foundation for further optimization in structural adjustments and performance improvements.

2.1.4. Structural Design and Analysis of Multi-Row Wheat Row Seeder

(1) Based on the 3D model diagram, the final physical design of the multi-row wheat row seeder is shown in the figure. In Figure 4a, the frame serves as the backbone of the entire seeder, supporting all key components. The frame design must ensure excellent rigidity and strength to adapt to operations on complex terrains. In Figure 4b, the primary function of the furrow opener is to create seed furrows in the soil. Its structural types include conical, disc, and shovel types, among which the disc-type furrow opener is suitable for operations under straw-covered conditions due to its superior cutting performance. The design requirements for the furrow opener include high depth adjustment accuracy, strong wear resistance, and adaptability to various soil conditions and terrain variations.

The primary function of the depth control wheel is to regulate the furrow depth and ensure stable sowing depth. The design allows for the selection of either support wheels or press wheels, which achieve depth control through close contact with the ground. In this design, the precise adjustment of the depth control wheel is achieved by integrating a stepper motor and a reducer [30]. The seed box is used to store seeds for sowing, with a capacity designed to meet the requirements of continuous operations [31]. The seed metering device evenly and quantitatively dispenses seeds into the furrows. This device adopts a gravity-fed seed metering system, which is simple in structure and suitable for conventional sowing needs.

In Figure 4c, the primary function of the compaction device is to compact and cover the furrows, ensuring adequate soil-to-seed contact and improving germination rates. Common compaction devices include press wheels or compactors, which are designed with an emphasis on shape and pressure distribution to ensure effective compaction.

The configuration parameters of this multi-row wheat planter are shown in

Table 1. Configuration parameters of multi-row wheat drill.

Project	Design Value	Project	Design Value
Structual form	Mechanical	Operating speed range	1.0~1.6 m/s
Machine size	1820 × 1210 × 1550 mm	Line spacing	21 cm
Working lines	6 rows	Seed feeder form	Outer grooved wheel type
Furrow opener form	Double disc type	Sowing part transmission mode	Shaft drive

below:

(2) Design and Analysis of Automatic Sowing Depth Adjustment Structure

In the design of the automatic sowing depth adjustment structure for the wheat row seeder, a combination of a stepper motor, a reducer, and the depth control wheel is primarily utilized [32]. The control structure integrating the depth control wheel and the stepper motor ensures the stability of the furrow opener’s operating depth. Specifically, the stepper motor and the reducer work together to adjust the position of the depth control wheel, achieving depth control with millimeter-level precision [33]. By using a laser sensor to monitor the distance between the depth control wheel and the ground surface in real-time, the system can automatically adjust the wheel’s position to ensure accurate sowing depth.

This combined structure uses the stepper motor to perform precise step movements through pulse signal control, while the reducer amplifies torque to maintain stable and reliable operation during adjustments. The reducer model is NMRV40-86BJ, and the stepper motor is an 86BYG250H two-phase hybrid stepper motor. The parameter tables for both components are shown below in

Table 2. 86BYG250H stepper motor parameters.

Parameter	Parameter Range	Parameter	Parameter Range
Fuselage length	150 mm	Adaptive voltage	48 V
Torque	12 N	Rated current	5.6 A
Out shaft shaft Diameter	14 mm	Keyway	5 mm

and

Table 3. DMA860H stepper drive parameters.

Parameter	Parameter Range	Parameter	Parameter Range
Output current (peak)	2 A–7.2 A	Control signal input current	7–16 mA
Input supply voltage	AC18–80 V, DC24–110 V	Step pulse frequency	0–200 KHz
Subdivision setting	400–51,200	Support pulse	Monopulse

. The stepper motor is driven by a DMA860H stepper driver, which converts pulse signals from the controller into power signals for the motor, ensuring that the motor’s speed is proportional to the pulse frequency, thus enabling precise speed regulation and positioning [34].

The system operates on a 220 V AC power supply and is equipped with a switching power supply for power distribution. A real image of the switching power supply is shown in Figure 5.

2.2. Test Method

2.2.1. Design of Mobile Depth Detection Platform

The experimental platform of this study is centered around the seeder, and incorporates various components such as a stepper motor, soil trough, sensors, and screw guide rails, to build a mobile sowing depth detection platform under different terrain conditions, as shown in Figure 6. The platform mainly consists of the following parts:

(1) Multi-row wheat strip seeder: The core structure of the platform includes the frame, furrow opener, and depth-limiting wheel. The furrow opener is used to create sowing furrows, while the depth-limiting wheel is used to control the sowing depth, ensuring precise and stable sowing depth during the process.

(2) Soil trough: The soil trough adopts a welded structure, with a length of 3000 mm, a width of 2000 mm, and a height of 100 mm. It has a volume capacity of 0.6 cubic meters of soil and is used to simulate different terrain conditions. The soil trough can present three forms of terrain: flat terrain, raised terrain, and sunken terrain, helping to simulate complex terrain changes in the actual sowing process.

(3) Drive device: The platform uses the KGX2000 dual linear rail four-slider high-end ball screw slide module as the drive device for the seeder. Once the slide is fixed, it can output a horizontal pulling force of up to 200 kg. The slide uses a high-precision ball screw and high-precision linear rail sliders, offering smooth and wear-resistant operation performance, ensuring efficient long-term operation.

(4) Connection structure: To ensure the stable connection between the slide and the seeder, a connecting device is designed to link the fixed plate above the slide to the multi-row wheat strip seeder. Two universal floating joints are installed at both ends of the connecting device, improving the stability and flexibility of the connection. This design reduces the impact of the vertical up-and-down fluctuations of the seeder on the screw guide rail under large torques, thereby enhancing the stability and accuracy of the entire system.

Figure 6. Movable sowing depth detection platform.

Figure 6. Movable sowing depth detection platform.

The installation positions of key components are crucial to ensure the precise operation and efficient performance of the wheat seed drill. Figure 7 illustrates the installation locations of critical parts such as the laser sensor, array sensor, stepper motor, and reducer.

2.2.2. Multi-Sensor System Hardware Integration

(1) Sensors are the core components for data acquisition and processing. With the rapid development of electronic technology, especially infrared, ultrasound, and array radar sensors, these technologies have gradually been applied to the agricultural field [35]. However, due to the complex soil environment during the sowing process, such as surface residues and weed coverage, traditional ultrasonic and infrared sensors are prone to interference in receiving echo signals, causing measurement errors and affecting the accuracy of sowing depth detection [36]. Additionally, devices like inclination sensors and encoders are difficult to directly apply on different machines. To address these issues, this study adopts the P8864-SMD-B15 array area sensor based on Time of Flight (ToF) technology.

Compared to ultrasonic and infrared sensors, the array area sensor offers higher precision and resolution, and is less susceptible to environmental light changes. Furthermore, this sensor can generate three-dimensional point cloud data, giving it a significant advantage in environmental modeling and obstacle detection [37]. By collecting 3D data, the system can more accurately reflect the subtle variations in the soil surface, thereby improving the precision and stability of sowing depth detection [38]. The parameters of the array sensor are shown in

Table 4. P8864-SMD-B15 area array sensor.

Parameter	Parameter Range	Parameter	Parameter Range
Visual angle FOV	16°	Indoor measuring range	60
Accuracy rating	±10 mm	Service voltage	+5 v
Weight	9 g

.

(2) The working principle of the P8864-SMD-B15 array area sensor is based on ToF technology. The sensor emits a laser beam and measures the time it takes for the beam to reflect from the target object and return to the sensor, allowing it to calculate the distance to the object. Specifically, the sensor emits a laser beam in its sensing area, and after the laser signal is reflected, the time of flight is measured, generating a high-resolution 3D point cloud map with N rows and N columns. Figure 8 shows the basic working principle of this sensor.

There are two common modes of ToF ranging technology: direct pulse modulation and indirect continuous wave modulation. In this study, direct pulse modulation is used. In this method, a narrow laser pulse is emitted, and the time interval Δt of the reflected echo signal is measured. A time-to-digital converter is then used to precisely quantify the pulse time interval, and the actual distance between the sensor and the target is calculated by combining the speed of light cc as follows: l = ${\displaystyle \frac{\mathrm{c}\ast \Delta \mathrm{t}}{2}}$ [39].

(3) The STP-23L laser sensor is a high-precision measurement tool, mainly consisting of a laser ranging core, a main control board, and a mechanical housing. Figure 9 shows the 3D structure of the laser sensor. The laser ranging core uses DTOF (Direct Time of Flight) measurement technology and can perform distance measurements up to 120 times per second. During each measurement, the emitter emits infrared laser light, which is reflected by the target object and received by the receiving unit [40]. The distance is then calculated using the time-of-flight method, which operates on the same principle as the array sensor. Due to its high resolution and stability, the laser sensor is very suitable for use in sowing depth detection systems, which are sensitive to environmental changes. After acquiring the distance data, the STP-23L sends the data in real time via serial communication. The STP-23L uses an SMT 4PIN 1 mm connector to connect to external systems, enabling power supply and data reception. The data communication of the STP-23L uses a standard asynchronous serial port (UART) for unidirectional transmission, and once powered on, the module sends data via the serial port. The parameter range of the STP-23L laser sensor is shown in

Table 5. STP-23L laser sensor parameter table.

Parameter	Parameter Range	Parameter	Parameter Range
Measuring ranging	0.07–7.5 m	Ranging frequency	120 Hz
Accuracy rating	±15 mm	Service voltage	4.5–5.5 v
Machine size	46 × 18 × 20 mm

.

(4) The terrain data acquisition system is a key component for achieving precise wheat sowing depth detection and stable control. The main function of this system is to collect real-time terrain features and surface undulation information during the sowing process. Additionally, the system can adjust the depth-limiting wheel height by inputting pulses and controlling the stepper motor direction, providing accurate terrain input data for subsequent data analysis and control systems. The terrain data acquisition system specifically consists of the array area sensor, laser ranging sensor, and data acquisition and communication module [41].

The data acquisition module connects the laser and array sensors through high-speed communication interfaces (such as COM ports) to collect data from the laser sensors and array radar sensors in real-time. The collected data are then uploaded to the main control system for real-time analysis. The integrated module is shown in Figure 10.

The control software interface, shown in Figure 11, includes functions such as communication parameter settings, data reception monitoring, and real-time analysis. The main control system processes the sensor data and, combined with the seeder’s operating status, calculates the terrain undulation features. The array area sensor obtains an 8 × 8 or higher resolution terrain data matrix through multi-point sampling. The laser sensor synchronously collects the real-time distance to the ground, calibrates the array sensor data, and eliminates system errors.

The advantages of this acquisition system include high-precision data collection, and the combination of the array area and laser ranging sensors significantly improves terrain detection accuracy. The system is capable of acquiring real-time terrain data during the sowing process, providing feedback for sowing depth control. The system also has good adaptability to complex ground conditions such as residue coverage and soil undulation.

2.2.3. Sowing Depth Detection Platform Parameter Calibration and Effect

During the actual sowing process, irregularities on the soil surface, such as particles, pores, and residual wheat straw, can interfere with the normal measurement of sensors, thus affecting the detection accuracy. Therefore, to ensure the reliability of the sowing depth detection platform, calibration experiments for the laser sensor and array sensor are necessary.

(1) Sensor Calibration

To verify the ranging accuracy of the array sensor, this study conducted calibration experiments for the P8864-SMD-B15 array area sensor in a laboratory environment [42]. The sensor was fixed onto the frame, and its angle and distance to the target object were gradually adjusted to simulate various typical terrain conditions for data collection. The specific experimental scenarios included the following three (Figure 12).

(a) Smooth surface without straw: The performance of the sensor on a standard soil plane was tested by collecting 3D terrain data to evaluate its basic ranging capability.

(b) Surface covered with straw: The soil surface was completely covered with wheat straw, and data were collected directly from the sensor to detect the straw terrain, assessing the impact of straw interference on the measurement results.

(c) Slope variation: Soil surfaces with different slopes (e.g., 5°, 10°, 15°) were set, and data were collected on terrain at different incline angles to verify the sensor’s response performance to slope changes.

The experimental results show that straw coverage has little effect on the measurements of both the laser sensor and the array sensor. Therefore, subsequent experiments were conducted using soil without straw coverage as the standard environment for the study.

When the array sensor was placed at a distance of 250 mm from the smooth soil surface, the resulting distance heatmap is shown in Figure 13a. As can be seen from the figure, the overall measurement error of the sensor at this distance is small, and the accuracy is high. When the measurement distance was increased to 450 mm, the distance heatmap obtained is shown in Figure 13b. At this point, the data are slightly higher than the normal value, indicating that the measurement accuracy of the sensor begins to decline. Furthermore, by analyzing the change in measurement error as the array sensor’s measuring distance increased from 200 mm to 500 mm, it was found that the measurement error gradually increased with distance.

In the experiment, the array sensor was placed approximately 350 mm above the soil, and testing was carried out under a soil incline angle of 10°. The resulting terrain heatmap is shown in Figure 13c. The results indicate that, even on sloped terrain, the sensor is still able to accurately capture the undulation features of the terrain.

Considering that the actual distance between the seeder frame and the ground is about 500 mm, the measurement accuracy of the array sensor at this distance is insufficient to obtain accurate numerical values (as shown in Figure 14). However, it is still effective in extracting the terrain undulation features. Therefore, this study designed the system to use the array sensor for terrain feature extraction, without using its direct ranging capability. This choice not only meets the practical application needs but also maximizes the advantages of the array sensor in feature extraction.

(2) Stepper Motor Calibration

In this study, an indoor reciprocating motion test was conducted to measure the stepper motor’s motion error. The test used a computer serial port to send motion position commands to a microcontroller, performing tests with different motion distances. The typical sowing depth for wheat is between 30 and 50 mm; therefore, the test was set with five distance ranges, each differing by 5 mm. For each range, the test was repeated five times for a round-trip motion, and the average value was taken. The initial position of the electric cylinder was marked with a marker, and the position change after the movement was measured using a vernier caliper. The electric cylinder motion error measurement test is shown in Figure 15. The experimental results are shown in

Table 6. Stepper motor motion error test results.

Object Distance/mm	Mean Distance/mm	Average Error/mm
15	15.02	0.02
20	19.95	0.05
25	25.03	0.03
30	29.95	0.05
35	34.96	0.04
40	39.93	0.07

.

From Table 6, it can be seen that the maximum average error of the stepper motor’s motion is 0.06 mm. According to the “Seeders Quality Evaluation Technical Standards”, the qualified range for sowing depth is ±5 mm or ±10 mm from the current set sowing depth. In this study, the average motion error of the stepper motor is only 0.6% to 1.2% of the qualified range for sowing depth, thus meeting the experimental requirements.

Upon analyzing the mechanical structure of the stepper motor, the study concluded that the error is mainly due to two factors: the displacement of the motor’s internal magnetic poles and unstable driving current. Improvements during the manufacturing process of the electric cylinder are needed to reduce the errors caused by gaps.

3. Results and Discussion

3.1. Performance Test on the Effect of Different Terrain Conditions on the Sowing Depth of a Striped Seeder

3.1.1. Definition and Control of Experimental Variables

In order to explore the impact of different terrain conditions on the stability of the sowing depth of the striped seeder, three typical terrain conditions were selected as research objects:

Flat Soil Terrain (Figure 16a), simulating ideal terrain conditions where the operation depth of the opener is relatively stable. Raised Soil Terrain and Sunken Soil Terrain (Figure 16b,c), simulating soil uplift and depression to analyze the possible fluctuations of sowing depth.

By manually controlling and arranging the experimental site, it was ensured that the terrain conditions met the design standards. In addition, a system combining the depth-limiting wheel and laser sensor was used to monitor the sowing depth in real-time and collect the corresponding depth data.

3.1.2. Strip Seeder Performance Response Analysis

The performance response analysis of the striped seeder is the core part of evaluating the operation performance of the seeder under different terrain conditions. The objective is to study the impact of terrain undulation on the detection of sowing depth, quantify the performance of the seeder under various terrain conditions, and provide data support for design optimization.

Simulation experiments on flat, raised, and depressed terrains were conducted to analyze the influence of different terrain characteristics on sowing depth. Raised terrain may lead to shallower sowing depths, while depressed terrain may result in deeper sowing depths. These fluctuations are directly related to the contact state between the soil surface and the depth-limiting wheel, as well as the working depth of the opener.

The experiment uses the mean, standard deviation, and deviation of sowing depth as the main performance indicators. Combined with the sensor feedback data under different terrain conditions, the stability and uniformity of the seeder’s operation are evaluated. By extracting and analyzing the slope matrix characteristic values (such as the mean and standard deviation), a quantitative relationship between terrain factors and sowing depth deviation is established.

During the experiment, both the laser sensor and the array sensor were used to simultaneously collect sowing depth and terrain data, and the data were processed by filtering, smoothing, and normalization to ensure the accuracy of the analysis. Principal Component Analysis (PCA) was used to extract key features of the data, removing redundant information and providing a simplified model for predicting sowing depth deviations.

Based on the experimental analysis, the key parameters of the seeder (such as the position of the depth-limiting wheel and sowing speed) were adjusted to improve its ability to adapt to complex terrain conditions, thus ensuring the quality and efficiency of the sowing operation.

3.2. Experimental Equipment and Process

3.2.1. Experimental Conditions and Process

The experiment was conducted on the sowing depth detection platform built indoors. After the stepper motor was powered on, the sowing depth control device entered the operational state. Different terrains were set along the strip seeder’s path, and the characteristic data of each terrain were obtained by the array sensor through the detection system. According to the national standards for wheat sowing depth, the wheat sowing depth is specified to be between 3 and 5 cm. Therefore, in this experiment, the sowing depth was preset to 40 mm. The opener was lowered to approximately 40 mm by controlling the stepper motor’s lifting mechanism, and then the guide rail was operated, allowing the depth-limiting wheel to pass over various terrains.

During the experiment, the sowing depth detection value was measured by the laser sensor before the seeder moved. After the seeder ran, the actual sowing depth at the same position was measured manually with calipers. Figure 17 shows the actual value of 37.13 mm obtained by a manual measurement. To ensure accuracy, the actual sowing depth was obtained through multiple measurements at the same point, and the average of these values was calculated. The difference between the averaged actual sowing depth and the detected value from the laser sensor gave the sowing depth deviation value.

3.2.2. Test Results and Analysis

The experimental results show that after applying sliding filtering smoothing to the terrain height data obtained by the array sensor, 3D surface maps of three typical terrains were successfully generated, visually highlighting the height variations across different terrain types. The color gradient from blue to red represents the variation in terrain height. The 3D surface of the raised terrain (Figure 18a) shows a prominent raised center with a downward trend at the edges, indicating significant terrain height variation. This demonstrates the array sensor’s ability to effectively capture abrupt height changes, which are crucial for studying the impact of raised terrain on sowing depth deviations.

The 3D surface of the depressed terrain (Figure 18b) exhibits a concave center with gradual elevation around the edges, forming a dish-like groove. The 3D surface of the flat terrain (Figure 18c) is nearly smooth, indicating minimal terrain height variation. Even in flat terrain conditions, the array sensor can still observe terrain undulations due to the low surface reflectivity and sensor accuracy limitations. This highlights the importance of further optimizing sensor calibration to reduce errors in flat terrain scenarios, ensuring consistent accuracy across all terrain types.

3.3. Data Analysis and Model Establishment

In this experiment, the data analysis involved converting the distance matrix obtained from the array sensor into a slope matrix. After denoising, smoothing, and normalization, the feature values (slope mean, slope standard deviation, slope range) were extracted. Each set of feature values corresponds to a deviation value of the sowing depth detection. Finally, Principal Component Analysis (PCA) was used to explain the variance of the feature values and to establish the model.

Firstly, the data obtained from the array sensor underwent preprocessing. Denoising was performed using a combination of median filtering and moving average filtering, aiming to eliminate noise interference and ensure the data were smooth and continuous. Then, normalization was performed using the Z-Score standardization method, which transformed the data into a form with a mean of 0 and a standard deviation of 1. The specific calculation formula is: $\mathrm{z}={\displaystyle \frac{x-\mu }{\sigma }}$. Where $x$ is the original data, $\mu$ is the mean, and $\sigma$ is the standard deviation.

Based on the denoised and normalized data, a 6 × 6 slope matrix was generated by calculating the local slope of the spatial points. The slope calculation method utilizes the gradient in the X and Y coordinate directions, derived using finite difference formulas. The calculation is based on a height matrix, as illustrated by the 3 × 3 grid in Figure 19, where each cell (i,j) represents a specific terrain height value. For example, the center cell (1,1) is surrounded by neighboring cells in all directions, and the height differences between these neighboring cells are used to compute gradients in the X and Y directions. These gradients are then combined to determine the slope at each internal point of the grid, providing a detailed representation of terrain steepness. To ensure data validity, the slope matrix calculation only considers non-boundary points, resulting in a 6 × 6 region for the final slope matrix. The slope calculation formula is as follows: $s=\arctan \left(\sqrt{{\left({\displaystyle \frac{\partial z}{\partial x}}\right)}^2+{\left({\displaystyle \frac{\partial z}{\partial y}}\right)}^2}\right)$

Based on the slope matrix, three key feature values were extracted: slope mean (${\mu }_{\mathrm{s}}$), slope standard deviation (${\sigma }_{\mathrm{s}}$), and slope range ($R_{\mathrm{S}}$). These features are used to describe the overall characteristics of the terrain slope: the slope mean reflects the overall slope level, the slope standard deviation describes the degree of slope fluctuation, and the slope range represents the difference between the maximum and minimum slopes, reflecting the degree of slope dispersion. These feature values were extracted as input variables and correlated with the corresponding sowing depth deviation values.

In the data analysis, Principal Component Analysis (PCA) was used to reduce the dimensionality of the feature values. PCA is a commonly used dimensionality reduction technique that transforms multiple correlated original variables into a set of new, uncorrelated variables (principal components) through linear transformations. The principal components are ordered by variance, with the first few principal components retaining the main information of the original data, while reducing the dimensionality of the data, removing redundant features, and improving computational efficiency [43]. In this study, because there may be some linear correlation between the slope mean, slope standard deviation, and slope range, performing multiple linear regression directly could lead to model instability. Therefore, PCA was used to reduce the dimensionality of the input variables, addressing the multi-collinearity problem between the features, and thereby constructing a more stable regression model.

In this experiment, the sowing depth detection values under different terrain conditions were compared, and the feature values of the slope matrix and their corresponding sowing depth deviation values are shown in the

Table 7. Slope matrix eigenvalues.

Slope Mean	Slope Standard Deviation	Slope Range	Sowing Depth Detection Value/cm	Sowing Depth Actual Value/cm	Sowing Depth Deviation Value/cm
30.8820	13.0808	48.0405	4.200	3.7329	0.4671
35.9596	11.5994	47.5991	4.100	3.4864	0.6136
30.9083	15.5122	52.1883	3.700	3.3794	0.3206
27.0983	14.6951	48.2988	4.000	3.4658	0.5342
23.2468	15.7124	51.4582	3.500	3.0183	0.4817
26.0006	13.8024	48.3614	4.100	3.3979	0.7021
29.4901	12.3199	45.4599	4.000	3.5491	0.4509

.

The results of the PCA analysis demonstrated that the slope mean and slope standard deviation are the primary contributors to terrain-related sowing depth deviations. The first principal component, slope mean, accounted for 75.29% of the total variance, while the second principal component, slope standard deviation, contributed 23.11%. Together, these two components explained 98.51% of the total variance, indicating that they effectively captured nearly all the essential information about the terrain’s characteristics. Conversely, the third principal component, slope range, contributed only 1.60% and was determined to have negligible influence, allowing for its exclusion to reduce the model’s complexity without significant loss of representativeness.

The analysis also revealed a clear relationship between terrain characteristics and sowing depth deviations. On raised or depressed terrains, the actual sowing depth was consistently lower than the detected depth. This deviation was strongly correlated with the slope mean and, to a lesser extent, the slope standard deviation, while the slope range showed no significant impact. Specifically, the slope mean emerged as the most influential factor, with its value directly proportional to the magnitude of sowing depth deviation. The slope standard deviation, while less impactful, still contributed to variations in depth.

In summary, the findings underscore that the larger the terrain’s slope mean and slope standard deviation, the greater the deviation in sowing depth. These results highlight the importance of focusing on the slope mean as the dominant factor in predicting and correcting sowing depth deviations, providing a clear direction for future optimization of terrain-adaptive sowing systems.

3.4. Reflection on Experimental Results

While the experiments provided valuable insights into the relationship between terrain features and sowing depth accuracy, several limitations and considerations need to be acknowledged.

(a) Experimental Conditions and Representativeness

The experiments were conducted in a controlled indoor environment using a simulated soil trough, which, while effective for isolating specific variables, may not fully represent the complexities of real field conditions. Factors such as varying soil moisture, compaction heterogeneity, and natural debris were simplified or excluded, potentially limiting the applicability of the results to real-world scenarios.

(b) Sensor Performance and Stability

The laser and array sensors demonstrated consistent performance during the trials, but their stability under fluctuating environmental conditions, such as extreme temperatures or high levels of ambient light, was not comprehensively evaluated. These factors could influence the reliability of sowing depth detection in field applications, particularly on uneven or sloped terrain.

4. Conclusions

(1) Advantages of the Multi-Sensor Fusion Test Platform

The multi-sensor fusion test platform developed in this study achieves high-precision, real-time collection of terrain features and sowing depth values during the sowing process through the collaborative work of laser sensors and array sensors. The application of filtering and normalization algorithms effectively reduces the interference of complex terrains and variable operating conditions on sensor data, ensuring the platform’s excellent environmental adaptability. Additionally, the platform can capture multi-dimensional terrain feature data, such as the mean, standard deviation, and range of the slope matrix, providing crucial data support for sowing depth control and error correction. This also lays the technological foundation for the future development of automated sowing depth control systems.

(2) Impact of Different Terrain Conditions on Sowing Depth

This study shows that terrain slope is an important factor influencing sowing depth. As the slope increases, the sowing depth deviation significantly increases, especially under steep slope conditions. Furthermore, the mean, standard deviation, and range of the slope matrix can reflect, to some extent, the impact of terrain characteristics on sowing depth deviation. However, their explanatory power varies due to the complexity of data distribution, and under complex terrain conditions, other features (such as soil compaction and sowing speed) may need to be considered for further corrections. The study also found that the flatter the terrain, the smaller the sowing depth deviation, indicating that terrain flatness plays a key role in sowing depth stability.

(3) Main Findings from Experimental Analysis

The experimental analysis demonstrated that the key terrain characteristics influencing sowing depth accuracy are captured primarily by two principal components: slope mean and slope standard deviation. This insight provides a clear understanding of how terrain slopes directly affect sowing depth deviation. By focusing on these two features, we can better predict and correct depth variations, thereby enhancing the precision of sowing operations.

The integration of experimental results with physical models has significantly optimized the depth-limiting wheel control algorithm. This optimization is a crucial step towards developing intelligent sowing systems that can autonomously adjust to varying terrain conditions. Specifically, the insights gained from the slope mean and standard deviation enable the creation of adaptive control strategies for the depth-limiting wheel, ensuring that the seeding depth remains consistent, even in the presence of uneven terrain.

The results also pave the way for the implementation of real-time automated depth correction systems. Using terrain data and the extracted features, future systems can dynamically adjust the position of the depth-limiting wheel to compensate for terrain fluctuations, such as raised or sunken areas, which commonly occur in agricultural fields. This adaptive capability not only improves the accuracy and stability of sowing depth but also addresses operational challenges related to soil variability, residue coverage, and local slope changes.

In practical terms, these findings lay the groundwork for intelligent seeders capable of maintaining optimal sowing depths across diverse and unpredictable terrain conditions. This adaptive control system could revolutionize sowing operations by ensuring uniform seed placement, improving crop establishment, and ultimately enhancing yield consistency.