Key Technologies in Intelligent Seeding Machinery for Cereals: Recent Advances and Future Perspectives

Abstract

The operational performance of cereal seeding machinery influences the yield and quality of cereals. In this article, we review the existing literature on intelligent technologies for cereal seeding machinery, encompassing active controllable seeding actuators, intelligent seeding rate control, and intelligent seed position control systems. In this manuscript, (1) the characteristics and innovative structures of existing motor-driven seed-metering devices and ground surface profiling mechanisms are expounded; (2) state-of-the-art detection principles and applications for soil property sensors are described based on different soil properties; (3) optimal seeding rate decision approaches based on soil properties are summarized; (4) the research state of seeding rate measuring and control technologies is expounded in detail; (5) trajectory control methods for seeding machinery and seeding depth control systems are described based on measurement and control principles; and (6) the present state, limitations, and future development directions of intelligent cereal seeding machinery are described. In the future, more advanced multi-algorithm and multi-sensor fusion technologies for soil property detection, optimal seeding rate decisions, seeding rates, and seed position control are likely to evolve. This review not only expounds the latest studies on intelligent actuating, sensing, and control technologies for intelligent cereal seeding machinery, but also discusses the shortcomings of existing intelligent seeding technologies and future developing trends in detail. This review, therefore, offers a reference for future research in the domain of intelligent seeding machinery for cereals.

1. Introduction

1.1. Background and Motivation

Cereals refer to the seeds of plants in the grass family, such as wheat, rice, corn, and millet. Cereals are a crucial part of staple foods all over the world, as they can provide 50–80% of the heat energy, 40–70% of the protein, and more than 60% of the vitamin B1 required by the world’s population [1]; therefore, there is great demand for cereals globally. According to the official statistics from the Food and Agriculture Organization (FAO) of the United Nations [2], the global total planting area for cereals was approximately 4.75 × 10⁷ hm² in 2022, and the yield of cereals was approximately 1.835 × 10³ kg/hm². The planting area and yield of cereals, as a staple food, need to be sustainably increased to meet the demands of the world’s population.

Seeding is one of the most crucial agronomic procedures in cereal cultivation [3], and an appropriate seeding rate can boost the production and growth quality of cereals [4]. If the cereal seeding rate is lower than the theoretical one, it is likely that land resources will be wasted. Additionally, if the cereal seeding rate exceeds the theoretical one, the yield will be diminished due to a shortage of water and nutrients. Therefore, precise control of the cereal seeding rate is of great significance.

Common cereal seeding methods include broadcast seeding [5], drill seeding [6], and precision seeding [7]. Broadcast seeding involves scattering seeds manually onto the soil surface and subsequently covering the scattered seeds with soil [8]; although this process is easy to perform, it is challenging to attain a uniform distribution of seeds.

For drill seeding, it is necessary to construct ditches at uniform intervals, into which the cereal seeds are subsequently deposited. Drill seeding, which is frequently employed for the direct seeding of wheat and rice, can maintain stable row spacings and optimal ventilation and light conditions for seeds. Furthermore, it also facilitates field management practices such as intertillage and plant protection [9,10]. Precision seeding refers to placing several seeds into holes spaced away from others according to specific pre-defined row and seed spacing parameters. Precision seeding is conducive to mechanized fertilization and field management due to the consistent growth of cereal seeds.

Various types of seeding machinery have been developed to implement the above seeding methods [11,12]. The seeds in the seeding machinery are discharged via seed-metering devices; however, in conventional seeding machinery, a ground wheel and chain transmission system are used to drive the seed-metering device, such that the seeding rate of the device cannot be regulated according to different soil fertility conditions [13]. In addition, the working depth of conventional seeding machinery is fixed after being lowered with the suspension mechanism of the tractor and cannot be controlled in line with the soil height [14], leading to different seeding depths and affecting the quality of the cereal. In summary, due to the lack of an intelligent monitoring and control system, conventional seeding machinery is unable to regulate the operational parameters in accordance with the soil properties, seeding position, and operational conditions [15], which ultimately affects the cereal yield and the efficiency of the subsequent agronomic process.

To optimally control the cereal seeding process, some intelligent sensing, decision-making, and control systems and algorithms have been used in conventional cereal seeding machinery [16,17]. Using intelligent technologies, conventional cereal seeding machinery can control the seeding rate according to the on-site soil fertility and control the seeding position and depth based on the seeding trajectory and down force of the seeding machinery, enabling the seeds to be placed at the ideal seed and row spacings and seeding depths [18,19].

The objective of this article was to review the latest intelligent technologies applied in the field of cereal seeding machinery. Some reviews have already expounded the differences between map- and sensor-based site-specific seeding and emphasized the importance of data fusion algorithms. For example, Šarauskis et al. [20] mainly focused on variable rate seeding (VRS) methods and their effectiveness under different field conditions. Munnaf et al. [21] reviewed multi-sensor and data fusion technologies for site-specific seeding methods. This study emphasized the various actuators, sensors, and controllers employed in intelligent seeding machinery. The unique perspective of this review is its description of how these intelligent technologies facilitate automatic and accurate cereal seeding, while the aforementioned reviews focus on how to flexibly change seeding strategies according to the specific conditions of different regions.

In addition, Du et al. [17] reviewed the research progress of VRS technologies and equipment for corn. The crops considered in our study include not only corn, but also other cereals such as wheat, rice, beans, and millet. As the seeding technologies for cereals have some similarities, we review the advanced intelligent technologies for different cereals.

Considering the aforementioned comparisons, this article provides a more comprehensive perspective and focuses on the intelligent seeding machinery technologies, which are conducive to improving the precision and efficiency of seeding processes. Although intelligent seeding machinery has been comprehensively and profoundly explored in previous research in numerous countries across the globe, and a succession of intelligent sensing, decision-making, and control algorithms for seeding machinery have been proposed, the research findings have not been summarized elaborately and systematically, which is disadvantageous for the development of subsequent intelligent seeding machinery. Therefore, we review the relevant references and research materials with the aim of offering a comprehensive perspective on the research status of key technologies in intelligent cereal seeding machinery.

1.2. Reference Indexing Methods

The literature summarized in this review was collected by searching the Web of Science and EI village databases. We aimed to acquire the latest research outcomes in the field of intelligent seeding machinery over the last decade (2014–2024), encompassing reviews and research articles. The keywords were set in accordance with the content of each section. For Section 3, the topics set in the search engine were {“seed-metering device” AND “cereal”}, {“seeding machinery” AND “wheat” OR “rice” OR “corn”}, and so on. For Section 4, the following search topics were set: {“soil moisture sensor”}, {“soil nutrient sensor”}, {“seeding rate” AND “decision method”}, {“seed” AND “optoelectronic” OR “piezoelectric” OR “machine vision”}, {“seeding rate” AND “control”}, and so on. For Section 5, the entries set in the search engine were {“seeding machinery” AND “navigation”}, {“seeding machinery” AND “trajectory plan”}, {“seeding machinery” AND “trajectory control”}, {“seeding depth” AND “control”}, and so on. Furthermore, some specific cereals—such as wheat, rice, corn, and beans—were also included in the searches. After organization, 255 articles and product web links were summarized in this review; among them, 246 papers were published from 2014 to 2024.

1.3. Manuscript Organization

The remainder of this manuscript is organized as follows: Section 1 expounds on the research background, motivation, and reference indexing methods of key technologies used in intelligent cereal seeding machinery. Section 2 elaborates the relationships between the key technologies of intelligent cereal seeding machinery. Section 3 introduces the seeding actuators developed for intelligent cereal seeding machinery. Section 4 reviews the key technologies for controlling the seeding rate, such as soil property sensing, the theoretical seed rate decision-making model, and seeding rate monitoring and control. Section 5 provides a review of seed relative position control technologies, whereby the lateral and longitudinal positions of a seed are controlled by a tractor, and the vertical position is controlled via a seeding depth control technique. Section 6 examines the influences of intelligent seeding machinery at different farming scales. Section 7 discusses the present state, limitations, and future development trends of intelligent cereal seeding technologies. Finally, Section 8 concludes the whole review.

2. Relationship Between the Key Technologies of Intelligent Cereal Seeding Machinery

The ultimate aim of intelligent cereal seeding machinery is to make full use of environmental and machinery information to place a certain quantity of cereal seeds at the desired seeding rate and relative position using intelligent sensing, decision-making, and control technologies [20]. As shown in Figure 1, the ultimate aim can be divided into two sub-goals; namely, an appropriate seeding rate and an accurate seed position [21,22]. To achieve the above sub-goals, three key technologies of intelligent cereal seeding machinery must be utilized: active controllable seeding actuators, intelligent seeding rate control, and intelligent seed position control.

As shown in Figure 1, the total aim—namely, achieving the optimal seeding rate and seed position—can be divided into two sub-goals. The first sub-goal involves the achievement of an appropriate seeding rate. Achieving the appropriate seeding rate requires the development of motor-driven seed-metering devices as a physical basis. Soil property sensing technology provides the decision-making factors, such as soil moisture content (SMC), soil organ matter (SOM), and soil electricity conductivity (SEC), for optimal seeding rate decision models. The theoretical seeding rate in a plot, as the principle aim of an intelligent seeding rate control system, can be determined using the optimal seeding rate decision technology. The seeding rate monitoring technology is then used to measure the actual seeding rates, serving as feedback to the seeding rate control system.

The other sub-goal is to achieve intelligent seed position control, which includes vertical, longitudinal, and lateral positions. The vertical position depends on the vertical seed position control technology. First, ground surface profiling mechanisms, which can be actively adjusted via a hydraulic, pneumatic, or electrical system, should be used for the development of actuators for intelligent seeding depth control. The soil height is measured using the topographic measurement technology. The topographic sensor then sends the actual soil height to the seeding depth control system, which is used to adjust the vertical position of the seed outlets. In addition, if the trajectory of the seeding machinery deviates from the theoretical route, then the seed spacing and row spacing will be abnormal, as depicted in Figure 2. Developments in seeding machinery include technology for controlling the trajectory and pose of the seeding machinery. Based on the above technologies, the sown seeds can be placed at their ideal positions.

Intelligent cereal seeding machinery can be integrated on the basis of the above active controllable seeding actuator intelligent seeding rate and position control systems. Based on the aforementioned intelligent technologies, intelligent seeding machinery can monitor soil and machinery information to determine the optimal seeding rate and drive the seed-metering device to control seeding rates in real time, according to the determined optimal seeding rate. Moreover, the intelligent seed position control technique could govern the longitudinal and lateral positions of seeds through regulating the trajectory of the seeding machinery, as well as controlling the vertical position in the soil via the seeding depth control method.

3. Active Controllable Seeding Actuators

3.1. Motor-Driven Seed-Metering Devices (MDSDs)

Seed-metering devices are the core components of seeding machinery, which are used to separate aggregated seed populations into a uniform and discrete seed flow [23]. An orderly discrete seed flow not only ensures a desirable yield and quality, but also significantly decreases the difficulty of monitoring and control through minimizing the mis-seeding and re-seeding phenomena [24,25]. Conventional seed-metering devices are controlled by ground wheels, the slipperiness of which may cause the rotational speed of the seed-metering device to be unstable. Hence, an increasing amount of seeding machinery has employed electric motors to drive the seed-metering devices. In the design process of motor-driven seed-metering devices, the effects of the motor should therefore be taken into consideration.

Based on the source of the contact force between cereal seeds and mechanical structures, MDSDs can be primarily categorized into two types. One is the mechanical seed-metering device, whose processes of filling, moving, and discharging seeds are dependent upon mechanical forces and gravity [26,27]. The other is the airflow-assisted seed-metering device, in which seeds are drawn into the holes by negative air pressure and subsequently discharged due to either gravitational force or positive air pressure [28,29,30]. Typical mechanical and air-assisted seed-metering devices are shown in Figure 3.

In the context of intelligent seeding, in order to reduce the difficulty of seed flow monitoring and control, the seed flow discharged by the MDSD should preferably be as accurate and uniform as possible [31]. A uniform seed flow can alleviate the overlapping phenomenon, thereby reducing the monitoring difficulty [32]. Additionally, most seeding control algorithms consider that the seed quantity discharged per seeding plate rotation is constant. Moreover, if the seeding flow is more uniform, the effect of seeding rate control will be better, and, hence, various innovative seed-metering devices have been proposed to enhance seed uniformity [33,34].

Regarding mechanical seed-metering devices, most studies have focused on structural innovation and parameter optimization [33,35,36]. Several innovative seed-metering devices with staggered teeth located on the periphery of the seed-metering plate have been developed [26,27,37]. The staggered arrangement of convex teeth allows for the filling, moving, and discharging of seeds in an alternate order. When a cavity completes the discharging process, the adjacent cavity begins to discharge the seeds. The structure of staggered convex teeth can, therefore, facilitate a reduction in the miss-seeding phenomenon and an improvement in seeding uniformity.

Unlike conventional seed-metering devices, a shaftless spiral transmission device has been proposed for conveying and discharging cereal seeds [38]. The spiral conveyor can alleviate the effects of the vibrations and collisions of cereal seeds during both the conveying and discharging processes. Additionally, an electromagnetic vibration-type MDSD for millet sowing has been developed [39]. The maximum coefficients of variation for seeding uniformity per row and for total seeding uniformity were 3.57% and 2.39%, respectively, and the seed damage rate was less than 0.5%.

Moreover, an ejection-type seed-metering device has been developed, in which high-speed rotating blades accelerate seeds and propel them into the soil [40]. The application of the ejection seed-metering device did not require coulters and seeding tubes, thereby reducing the impacts between the seeds and seeding machinery.

The discrete element method (DEM) is a numerical computational approach that can be used to simulate particle movement behaviors under given circumstances, which plays an important role in the structural optimization of MDSDs. The research processes employed for the structural optimization of MDSDs via the DEM are expounded below. First, the shape and size of cereal seeds should be set or imported [41]. The mechanical parameter values from particle to particle and from particle to MDSD then need to be calibrated [42]. Moreover, the geometry and kinematic parameters of the MDSD should be established and set in the EDEM software package (https://altair.com/edem/, (accessed on 17 November 2024)). In the next step, the seed motion processes can be simulated in the simulator section. After obtaining the simulation results, researchers can visually analyze the seed motion parameters, such as force, torque, and velocity. Based on these seed motion parameters, the structure of the MDSD can be optimized until cereal seeds can be discharged according to the desired parameters [43].

Gao et al. [44] studied the effects of the inlet velocity of particles, feeding rates, and the angle of the seed cleaning element on leakage and multiple seeding of a seed-metering device. They found that, under the conditions of a seeding plate rotational speed of 194.5 rpm and seeding rate of 14 seeds/s, the minimum amount of leakage and multiple seeding can be ensured. Dong et al. [45] conducted a DEM simulation to analyze the effects of operating parameters on seed dynamic characteristics, filling performance, and operational energy consumption for a corn MDSD. Moreover, Deng et al. [46] applied the average kinetic energy as an evaluation criterion and used the EDEM software package to simulate and analyze the disturbances caused by three types of seed-metering disks. The results showed that, when the rotational speeds were between 1.16 and 3.49 rad/s, the MDSD presented the minimum number of disturbances.

Airflow-assisted seed-metering devices are commonly utilized in precision seeding [30,34,47]. Aiming at enhancing seeding stability, some novel pneumatic seed filling and cleaning structures have been proposed [29,47]. Specifically, a trapezoidal opening was incorporated at the base of the hole insert, while an airflow outlet was integrated into the front shell to facilitate both the seed-filling and seed-cleaning processes [48]. Moreover, an airflow-assisted seed-metering device combined seed spoons and airflow to regulate seeds in a stable posture during the seed-moving process, thereby ensuring that seeds are steadily held in the seeding holes [46].

Existing research on pneumatic MDSDs has mainly focused on the interactions between airflow and cereal seeds. The computational fluid dynamics (CFD) method has, thus, been commonly combined with the DEM to optimize the structure and fluid field parameters of the MDSD [49]. Wang et al. [50] designed a non-contact self-suction wheat-shooting device and performed CFD-DEM coupling simulations to study the seed-filling performance. The simulation results showed that, when the airflow velocity was over 6.5 m/s and the rotational speed was greater than 800 rpm, the seed-filling performance could meet the considered requirements. Xing et al. [30] designed a double-cavity pneumatic MDSD and studied the movement details and state of rice seeds in the airflow using CFD-DEM technology. The best seeding performance of the developed MDSD was achieved when the inlet air velocity was 3 m/s, the inlet angle was 60 degrees, and the speed of the seed sucking plate was 20 r/min.

Additionally, some seed-conveying techniques, such as the positive-pressure conveying [51] and high-speed belt conveying systems, have been developed to enhance the conveying stability of seed-metering devices. A pneumatic-assisted seed conveying device was designed through analyses of the airflow field under different structural parameters [49]. The positive-pressure airflow helps two adjacent seeds to maintain their original interval after being discharged by the MDSD. The main challenge of using positive-pressure conveying technology is how to avoid collisions with the seeding tube. As the seeds conveyed via positive-pressure airflow have a fast velocity, their falling trajectory is difficult to control and may strike the seed tube. The development of MDSDs that can maintain a precise and uniform seed flow discharge lays a solid foundation for intelligent seeding rate control technology.

Some studies on MDSDs for cereals are listed in

Table 1. Performance of MDSDs developed for cereals.

MDSD Type	Objective Cereal Seed	Qualified Rate	Missing Rate	Overlapping Rate	Reference
Pickup finger precision seed-metering device	Corn	86.90%	3.51%	9.62%	Wang et al. [52]
High-speed air-suction seed-metering device	Corn	94.60%	5.10%	4.20%	Liu et al. [53]
Telescopic clip finger-type seed-metering device	Corn	95.40%	1.90%	2.70%	Geng et al. [54]
Air-suction double-row staggered seed-metering device	Corn	95.70%	2.80%	1.60%	Wang et al. [34]
Shaftless spiral seed discharge and seed delivery device	Soybean	92.60%	2.40%	5.03%	Zhu et al. [38]
Flexible mechanical soybean precision seed-metering device	Soybean	94.00%	3.00%	3.00%	Hou et al. [55]
Four-bar translational seed-metering device	Soybean	87.61%	5.75%	6.63%	Dun et al. [42]
Air-suction wheat precision hill-seed-metering device	Wheat	91.66%	2.36%	5.98%	Fang et al. [56]
Pneumatic centrifugal combined seed-metering device	Wheat	93.46%	2.78%	3.73%	Hou et al. [57]
Airflow wheat seed-metering device	Wheat	92.70%	3.47%	3.83%	He et al. [58]
Hybrid rice air-suction single-seed-metering device	Rice	86.91%	3.41%	10.23%	Zang et al. [59]

. According to the different objective seeds of the MDSDs, it can be observed that the qualified rates of the MDSDs for corn are relatively high. It is possible that this is due to the corn seeds having two flat sides, which means they can easily be sucked through the holes of the seed-metering plate. In contrast, the MDSDs for wheat and rice have relatively low qualified rates due to their small sizes and irregular shapes.

3.2. Ground Surface Profiling Mechanism (GSPM)

An uneven ground surface significantly affects the vertical position of the seeding outlets, resulting in cereal seeds being sown at inconsistent seeding depths [60]. Therefore, a GSPM—as the actuator for intelligent seeding depth control systems—can be applied to control the vertical position of the seeding outlets.

GSPMs can be categorized into two types based on their structures; namely, single-hinge GSPMs and parallel four-link GSPMs. The single-hinge GSPMs have a lightweight structure. Moreover, the motion trajectories of seeding outlets of single-hinge GSPMs are arcs with rotational centers at a hinge joint [16,61]. Nielsen et al. applied a single-hinge GSPM to control the working depth of a seeding unit [60]. In this research, a fixed spring was used to load the down force, and the experimental results showed that the single-hinge GSPM with a seeding depth control system provided a mean depth deviation from the target depth of less than 0.2 mm.

The use of a parallel four-link GSPM can ensure that the two planes remain parallel and achieve a vertical motion trajectory. In practical seeding operations, the parallel four-link GSPM is the most prevalently utilized mechanism in seeding machinery [62]. The crucial step in developing a parallel four-link GSPM is to establish a parallelogram configuration between the rack and the seeding unit; that is, both the upper and lower connecting rods must have an equal length and be parallel to each other [63].

The conventional GSPM modifies the down force provided by the seeding unit on the soil by adjusting the length of the spring [64]—the lower the down force, the shallower the seeding depth. This spring-type GSPM cannot actively regulate the appropriate down force according to the on-site soil conditions, thereby influencing the emergence and root development of the cereal seeds [24].

Recent research in this area has been placing increased focus on active GSPMs which integrate active actuators—for example, the implantation of a hydraulic cylinder, electric motor [65], pneumatic cylinder [66], or magnetorheological damper [67]—thus enabling real-time control of the seeding depth based on soil surface conditions [68]. Several universal GSPMs with various actuating devices are displayed in Figure 4. Figure 4a shows a spring-type GSPM with a simple structure, high reliability, and wide range of applications; moreover, it is also easy to install and adjust. The spring provides an unstable down force, such that the preload force of the spring-type GSPM needs to be adjusted empirically according to the ground conditions before operation. Figure 4b displays a hydraulic-driven GSPM, which can also provide a large down force and adjustable range. In addition, it has a fast response time and is thus suitable for high-speed operations. Figure 4c shows a pneumatic-driven GSPM, which requires an external air supply unit and can thus withstand high pressures and vibration impacts, with good vibration damping effect. Figure 4d shows an electric motor-driven GSPM, which has high control precision and a fast response but requires an extra power source during operation.

According to previous studies on the design of active GSPMs, the development process of an active GSPM can be summarized as follows. First, kinematic and dynamic models are established to ascertain the structural parameters of the active GSPM [69]. Contact models between the profiling wheel and soil surface might then be developed in order to precisely estimate the soil subsidence [70]. The stress and strain distributions within the developed active GSPM are then analyzed via the finite element method, which can provide a basis for its structural optimization. Subsequently, field experiments of the optimized active GSPM are conducted to verify its profiling performance. Based on the developed active GSPM, intelligent topographic measuring and seeding depth control algorithms are then developed for seeding depth regulation during operations, according to the soil surface height. Although the profiling performance of the developed GSPMs is generally acceptable, there are still some practical challenges faced by GSPMs. The most salient challenge is how to reduce the vibration of the GSPM in the operation processes, which may necessitate further study [71].

4. Intelligent Seeding Rate Control

4.1. Sensing of Soil Properties

Some soil properties, such as the soil moisture content (SMC), soil organic matter (SOM), soil electrical conduction (SEC), pH value, and salinity, can affect the germination, growth, and health of cereal seeds [72]. The optimal seeding rate should, thus, be determined in accordance with the on-site soil properties in order to utilize the maximum potential of the cultivated field [21]. Moreover, soil properties serve as the basis for determining the seeding rate in a certain field. The relationships between soil property sensing methods, optimal seeding rate decision models, and the implemented seeding machinery are displayed in Figure 5. As the most crucial soil properties, the sensing methods for SMC, SOM, and SEC are expounded below.

(1)

Soil moisture content (SMC)

The SMC is defined as the ratio of the weight of water present in the soil to that in dry soil [73,74]. Insufficient moisture can impede seed germination, as excessively dry conditions may prevent seeds from absorbing adequate water. Conversely, overly saturated soils can lead to seed rot due to oxygen deficiency [75]. Therefore, it is essential to regulate the optimal seeding rate based on the SMC of the cultivated field; in this line, the detection efficiency and accuracy of the SMC sensor are of utmost importance.

SMC sensors using detection methods such as frequency domain reflectometry [76], time domain reflectometry [77], and time domain transmission [78] are already well established in the market, with some examples shown in Figure 6.

To improve the sensitivity and accuracy of SMC sensors, researchers have applied various intelligent materials, such as graphene quantum dots [79] and flexible polyethylene terephthalate substrate [80]. A dual-probe SMC sensor was designed based on a porous medium formed by sintering a mixture of kaolinite, charcoal bone powder, activated carbon powder, and water [81]. When one probe was heated, another probe was used to detect the variation in temperature. As the SMC influences the thermal conductivity and specific heat of the developed probes, the SMC value can be acquired through analyzing temperature alterations.

The output of an SMC sensor is typically represented as an electrical signal. The relationship between the outputs of the SMC sensor and the actual SMC values is characterized as an SMC prediction model. In most cases, these SMC prediction models are developed considering a consistent soil type and roughness, which may compromise their robustness in practice [82]. Therefore, comprehensive SMC prediction models based on multiple sources of information—including the data from SMC sensors, satellite imagery, and drone photography—have been developed to improve the accuracy and robustness of conventional SMC detection systems [83]. For example, dual-frequency data from the global navigation satellite system have been fused to build SMC estimation models based on the Helmert variance component estimation (HVCE) method [84]. The study indicated that the HVCE fusion method can significantly enhance the accuracy of the SMC prediction model, especially for models based on machine learning techniques.

Besides the above HVEC and machine learning fusion model, several intelligent SMC prediction models based on intelligent algorithms have also been proposed. A random forest regression model for SMC prediction was developed by integrating the information obtained from synthetic aperture radar, optical, and thermal infrared remote sensing devices [85,86]. Additionally, a bi-directional long short-term memory neural network module was applied to optimize an SMC prediction model [87]. In addition, the particle filtering algorithm has been employed to capture the nonlinear relationships between high-resolution thermal data, radar remote sensing data, and the actual SMC [88]. These studies claimed that the integration of machine learning methods with multi-source detection data can significantly enhance the robustness and accuracy of SMC detection.

(2)

Soil organic matter (SOM)

SOM is a kind of stable polymer organic compound, formed by the remains of plants and animals being decomposed by micro-organisms [89]. Regarding its effects on cereal seeds, SOM can provide nutrients, but it can also promote root development in cereal seedlings. To make full use of SOM, the optimal seeding rate should be determined based on the SOM content of the cultivated field [90].

Conventional SOM detection methods are dependent on laboratory chemical analyses. Under controlled heating conditions, soil organic carbon is oxidized using potassium dichromate in sulfuric acid solution. The remaining potassium dichromate is subsequently titrated with a standard ferrous sulfate solution. The quantity of organic carbon is then determined by applying the oxidation correction coefficient to the amount of consumed potassium dichromate. This chemical analysis method for detecting SOM has acceptable accuracy, but it is also time-consuming and costly.

To reduce the detection time and cost, spectroscopic detection methods, such as near-infrared spectroscopy sensing [91,92] and hyperspectral remote sensing [93,94], have been proposed due to their non-destructive, rapid, and precise characteristics [95]. The near-infrared spectroscopy method enables the precise identification of absorption bands and reflectance rates associated with elements, atomic groups, and relevant substances in soil through comprehensive spectral data analysis [96,97]. The diffuse reflectance spectral data demonstrate variations that correlate with fluctuations in SOM. Moreover, the effect of high-spectral remote sensing technology for SOM detection fundamentally depends on the interactions between the electromagnetic radiation and SOM [98].

A key process in intelligent remote SOM sensing is to model the regression relationship between the actual SOM content and the brightness values of multi-band remote sensing images [99]. However, spectral methods might be susceptible to the soil particle size, moisture content, oxides, and other factors [100,101,102]. Furthermore, Polonen et al. identified that atmospheric clouds can significantly affect images generated using hyperspectral remote sensing techniques [103].

Considering that the detection accuracy of remote SOM sensing methods can be affected by optical disturbances, researchers have put forward machine olfactory systems with gas sensor arrays as the core information source. These systems were employed to detect the SOM content in soil pyrolysis gas [104] and, based on the olfactory characteristics of the soil pyrolysis gas, SOM content prediction models were developed [105]. To improve the accuracy of nonlinear regression, some intelligent techniques—such as the sparrow search, random forest, and multilayer perceptron machine algorithms—have been fused into SOM prediction models. Experimental results revealed that a machine olfactory system combined with machine learning models can provide desirable predictive accuracy in the context of practical SOM detection [105]. The state-of-the-art olfactory SOM sensor array and vis-NIR optical spectrometer that can be used for SOM detection are shown in Figure 7.

Twelve research studies on SOM measurement are listed in

Table 2. Typical sensors and data processing algorithms used for SOM detection.

Types of SOM Sensors	SOM Prediction Algorithms	Coefficient of Determination (R2)	RMSE	Reference
Vis-NIR spectroscopy	Partial least squares regression	0.97	0.38	Ba et al. [108]
Visible–shortwave infrared spectroscopy	Random forest	0.89	0.42	Bao et al. [109]
Vis-NIR and MIR spectroscopy	Spectral-guided ensemble modeling	0.86	3.72	Xue et al. [110]
Multi-sensor array	Multilayer perceptron random forest	0.94	0.81	Liu et al. [106]
HyMap airborne hyperspectral imaging sensor	Semi-supervised deep neural network regression	0.71	3.52	Ou et al. [95]
Vis-NIR spectroscopy	A hybrid PLS-SVM algorithm	0.79	5.33	Hong et al. [97]
Spectral photometer and standard reflector	Analytic hierarchy process and particle swarm optimization	0.99	1.23	Liu et al. [111]
Visible–thermal imaging camera	MobileNet V3-LSTM	0.91	2.60	Li et al. [112]
Carbon dioxide sensor	Multiple linear regression	0.974	1.49	Jia et al. [113]
Hyperspectrometer	Improved dual-branch CNN model and BLMultiCNN	0.91	3.09	Li et al. [101]
Fiber-optic spectrometer	Partial least squares regression	0.89	0.03	Liu et al. [114]
Near-infrared spectroscopy and Raman spectroscopy	Outer-product analysis and partial least squares	0.903	2.59	Bai et al. [91]

, most of which indicated relatively small root-mean-square errors (RMSEs) and desirable linear relationships. The reported SOM sensors and prediction algorithms may provide valuable solutions for subsequent developments.

(3)

Soil electrical conductivity (SEC)

SEC directly reflects the mixed salt content in soil, with factors such as salinity, moisture, temperature, organic matter content, and texture also influencing the SEC value. The SEC value is correlated with the soil structure and SMC. Previous studies have typically applied field experiments to determine the relationships between SEC values and cereal yields, and they presented the most appropriate SEC for maximum yield in the experimental field. Regarding SEC detection, the electromagnetic induction [115], time domain reflectometry [116], and current–voltage terminal [117] methods are commonly applied.

To achieve real-time measurement of SEC, it is advantageous to install SEC sensors on tractors or seeding machinery. A typical model of an SEC sensor based on the principle of electromagnetic induction method is the EM38 (Geonics Limited, Toronto, Ontario, Canada), which has the advantage of non-destructive detection. However, in practice, the EM38 SEC sensor might be affected by surrounding electronics, soil temperature, and other factors; therefore, it needs to be calibrated before each measurement [118]. The development of vehicle-mounted SEC sensors has thus become a focal point in the field of intelligent seeding machinery [119,120,121]. The classical EM38 and some commercial SEC sensors are shown in Figure 8.

Murata et al. have developed a millimeter-scale conductivity sensor array system for real-time SEC measurement, and an Internet of Things (IOT) technique was used to transmit the SEC values to seeding machinery [123]. Han et al. have developed a vehicle-mounted SEC detection system utilizing the voltage–current four-terminal method [124]. However, as the current–voltage terminal method is an invasive technique, the measurement accuracy may be decreased due to the unstable contact between the soil and electrodes [121].

Regarding SEC prediction models, the most classic is Archie’s equation. Some researchers have expanded the application scope of Archie’s equation by increasing the number of factors, for example, including water content and soil texture [125,126]. Other studies have extended beyond the scope of Archie’s theory and applied machine learning in the field of SEC detection. An in-field multi-sensor portable device was developed to detect SEC and, based on the device, a SEC prediction model using a backpropagation neural network was established [127]. Moreover, another study proposed a multivariable model using the radial basis function artificial neural network for SEC prediction [128]. Relevant experimental results have shown that the accuracy of SEC prediction models regressed via machine learning algorithms is better than that of those regressed using linear or quadratic models [129].

In

Table 3. Some sensors and data processing algorithms for SEC measurement.

Sensor	Data Processing Algorithm	Coefficient of Determination (R2)	RMSE	Reference
Identical time domain reflectometry probe	Time domain reflectometry	0.985	0.019	Agah et al. [116]
Near-infrared spectral sensor	Current–voltage six-terminal method	0.983	15.625	Wang et al. [118]
Thermo-time domain reflectometry sensor	Archie’s model	0.960	0.315	Fu et al. [125]
Conductivity meter	Conductivity algorithm based on Archie’s equation	0.930	0.190	Choo et al. [126]
Self-developed soil conductivity measurement device	BP neural network	0.997	0.120	Qi et al. [127]
Conductivity meter	Radial basis function artificial neural network	0.801	0.350	Motie et al. [128]
Cond 7310 Soil Tester	Extreme learning machine	0.884	3.071	Cao et al. [129]

, we summarize several studies focusing on SEC measurement, with most showing acceptable levels of regression performance (R² greater than 0.8). The reported SEC sensors and data processing algorithms may provide valuable solutions for subsequent SEC sensor development.

(4)

Optimal seeding rate decision

The optimal seeding rate refers to the most appropriate quantity of seeds under specific conditions to attain both the desired crop yield and quality. Based on the maximum potential of soil, the optimal seeding rate decision model can recommend the optimal seeding rate for a certain field. Common inputs for the optimal seeding rate decision model are the soil properties and desirable yield, and the output of the decision model is the optimal seeding rate [130].

Previous studies reporting on the optimal seeding rate decision model have mostly used experimental methods [131,132], following which regression models between the soil property parameters, seeding rate, and cereal yield can be established based on the experimental results [133]. In subsequent research, variance analyses have been performed to predict the optimal seeding rate [134,135]. According to the previous research findings, the yield initially increased and then decreased with an increase in the seeding rate under the same soil fertility, resembling a parabolic curve [90]. According to the peak of the desirable yield–seeding rate curve, an optimal seeding rate in accordance with the soil fertility was obtained.

Conventional decision models are regressed based on preset multivariate linear or nonlinear models [136,137]. Nevertheless, preset multivariate nonlinear models may not completely regress the entire nonlinear characteristics among the soil property parameters, seeding rate, and cereal yield [138]. Hence, intelligent nonlinear regression algorithms, such as backpropagation neural networks [138], random forest [90], and decision tree [139], have been extensively used for the development of optimal seeding rate decision models. A seeding rate decision model has been developed using a gradient boosting decision tree algorithm to choose the optimal seeding rate, simulating corn yield responses to a series of SOM content values and seeding rates [90].

The predicted optimal seeding rate can be directly output to the intelligent seeding machinery, but it can also be combined with geographic position information to form seeding prescription maps in a geographic information system [140]. In the seeding prescription map, adjacent plots with similar seeding rates are combined and marked with the same color. Then, a vector diagram of the seeding prescription map is transmitted to the master control system of the intelligent seeding machinery before the seeding operation.

When the seeding machinery travels in a certain field, the intelligent seeding rate control system will regulate the rotational speed of the seed-metering shaft according to the optimal seeding rate. The seeding prescription map is considered the crucial link between the soil properties and the operational parameters of the seeding machinery [141]. An example optimal seeding rate prescription map based on the historical yield distribution is displayed in Figure 9.

4.2. Seeding Rate Monitoring Technique

The seeding rate refers to the quantity of seeds discharged from the seed-metering device per unit area. The objective of the seeding rate control system is to achieve an actual seeding rate approaching the theoretical one [143]. A block diagram of a universal seeding control system is shown in Figure 10.

As one of the key components of the seeding control system, the seeding rate sensor is used to feed back the actual seeding rates to the intelligent seeding rate control system [144,145,146]. Additionally, based on the time interval between two adjacent seeds, the seeding rate sensor can also determine the occurrence of mis-seeding or re-seeding phenomena [147,148]. According to the measurement principles, seeding rate sensors can be classified into two categories; namely, indirect estimation sensors and direct measuring sensors.

Some commercial seeding rate sensors are displayed in Figure 11. Figure 11a displays a Hy rate plus^TM LED seed sensor designed for larger seeds, with a robust photo cell and optics package for seed counting and for monitoring whether the seed tube is blocked or not [149]. Figure 11b displays an MC uni-seeder photocell sensor with optimized optics, allowing it to distinguish seeds under high-speed conditions. In addition, it exhibits good self-adaptivity, as it includes an algorithm that can compensate for the presence of dust or ambient light [150]. Figure 11c shows a DS 2000i seeding rate sensor, designed to count dropping or blown seeds in precision seeding machinery [151]. Figure 11d displays the PLANTirium sensor, a high-resolution imaging element enabling the detection of small-sized cereal seeds [152], in which a pattern recognition method is used to distinguish debris from seeds inside the seed tube. Figure 11e shows a MAXI-SEEDER-TIN-type seeding rate sensor, which uses photocells as the sensing elements. It can automatically adapt the trigger threshold to the dust level to guarantee long-term reliability [153]. Figure 11f shows a WaveVision seed sensor, which uses radio waves to sense through dust [154].

The rotational angle estimation method (RASM) is an indirect estimation method. The principle of the RASM is to measure the rotational angle of the seed-metering shaft for estimation of the seeding rate. In most cases, an encoder or Hall sensor is used to measure the rotational angle of the seed-metering shaft [155]. The RASM is easy to implement but cannot be used to estimate the difference in seed quantity in each rotation of the seed-metering shaft, as the RASM assumes that the amount of discharged seeds per rotation of the seed-metering shaft is constant. In addition, some malfunctions—such as no seeds in the hopper or a seed tube blockage—cannot be monitored using the RASM.

To overcome the shortcomings of the RASM, direct measuring sensors have been developed, including piezoelectric [156], capacitive [157], and photoelectric [158,159] sensors. In previous studies, piezoelectric sensors were employed to detect the seed quantity by monitoring the impacts of seeds on the piezoelectric film [160,161]. The seeding rates of barley and triticale have been measured using a piezoelectric sensor, applying the slide window algorithm to perform seed quantity judgment [162]. A deformation of the piezoelectric film can be transformed into an electrical impulse. The quantity of the electrical impulses per unit period constitutes the seeding rate [160]. However, seeds might rebound after colliding with the piezoelectric material, thereby causing a seed to be counted more than once [163].

The detection principle for capacitive seeding rate sensors is that, when seeds pass through the capacitive sensor, the dielectric constant of the dielectric medium between the electrodes changes, resulting in variations in capacitance. Capacitive seeding rate sensors can monitor multiple seeds at each sampling, as the dielectric constant of the dielectric medium is sensitive to the seed quantity [164]. A previous study enhanced the detection accuracy of the capacitive sensor through adjusting the sampling frequency based on the seeding rate [144]. In a practical application, the capacitance sensor was found to be robust against dust but susceptible to temperature, mechanical vibration, and parasitic capacitance [165].

The photoelectric sensor is the most prevalent technology for seeding rate measurement. An opposite-type photoelectric sensing unit comprises a transmitter and a receiver [158,166]. When the light beam emitted by the transmitter is obstructed by seeds, a shadow will be cast on the surface of the receiver, resulting in a change in sensor output [167,168]. An intelligent seed quantity judgment algorithm has been developed to detect the seed quantity according to the time period of seeds passing through the light beam [169]. The rapid response and high accuracy characteristics of photoelectric sensors enable them to obtain acceptable performance levels in seeding rate measurement. However, the measuring accuracy of the photoelectric seeding rate sensor might be affected by dust; thus, some studies have proposed the use of self-cleaning devices for improved resistance to dust [170]. Mussadig et al. [171] applied three objective identification algorithms—namely, CellProfiler, P-TRAP, and SmartGrain—to count cereal seeds from digital images. Wheat, barley, maize, rye, oat, sorghum, triticale, and rice seeds were tested in this study, with the results suggesting that the CellProfiler is a reliable image analysis program for counting seeds from digital images.

In

Table 4. Summary of recently developed seeding rate sensors.

Objective Seed	Measuring Principle	Measuring Accuracy	Reference
Wheat	Photoelectric	95.00%	Zhu et al. [143]
Wheat	Capacitive	97.74%	Chen et al. [154]
Corn	Photoelectric	98.50%	Ji et al. [148]
Corn	Photoelectric	97.40%	Huang et al. [144]
Corn	Capacitive	94.60%	Zhou et al. [165]
Corn	Photoelectric	99.00%	Che et al. [170]
Soybean	Photoelectric	98.00%	Zhang et al. [146]
Corn	Photoelectric	98.50%	Okopnik et al. [147]
Corn	Photoelectric	96.00%	Jing et al. [172]
Corn	Photoelectric	92.50%	Zhao et al. [173]
Corn and bean	Piezoelectric	95.00%	Rossi et al. [161]
Wheat	Piezoelectric	90.38%	Zhao et al. [160]

, we summarize 12 types of seeding rate sensors, most of which have a measurement accuracy higher than 90%. These measuring principles and results can provide references for subsequent studies on seeding rate sensors.

As can be seen from Table 4, the seeding rate sensors for corn seeds have a relatively high measuring accuracy due to the size of corn seeds and the precision seeding mode. In contrast, as wheat seeds are sown via drill seeding, if the photoelectric measuring principle is used to measure the wheat seeds, more than one seed may block the light at the same time, which may lead to incorrect measurements. Thus, the measurement accuracy of the seeding rate for wheat is relatively lower than that for corn and soybean when using the photoelectric principle. However, as capacitive sensors are not affected by multiple seeds, they have a relatively high measurement accuracy for the wheat seeding rate.

4.3. Seeding Rate Control Technique

Conventional seeding machinery regulates seeding flow rates using a mechanical transmission system that maintains a fixed ratio between the ground wheel and the seed-metering device. In this manner, the seed-metering device might lead to re-seeding or mis-seeding phenomena due to the slippage of the ground wheel. To tackle the aforementioned problems, some researchers have developed electric seeding rate control systems (ESCSs), which apply sensors to monitor the traveling velocity of seeding machinery and use an electric motor instead of a mechanical transmission system to drive the seed-metering devices [174,175]. Some commercial ESCSs are shown in Figure 12.

At the initial stage of development of ESCSs, most ESCSs were open-loop systems [180]. An open-loop ESCS, which can maintain a linear relationship between the rotational speed of the seed-metering device and the traveling velocity of seeding machinery, has been designed [181]. Due to the lack of feedback, open-loop ESCSs might be affected by the motor speed, manufacturing errors of seed-metering devices, and external disturbances, thereby leading to the possibility that the seeding rate becomes inaccurate and unstable [182].

To overcome the drawbacks of open-loop ESCSs, encoders or Hall sensors can be used to feed the rotational speed of the seed-metering device back to the ESCS [183]. He et al. developed an ESCS which applies encoders to monitor the rotational speed of the seed-metering devices, and they deployed a PID control algorithm to regulate the rotational speed of the motor [184]. Their experimental results demonstrated that the control accuracy of the seeding rate greatly improved, as the developed ESCS significantly decreased the instabilities related to motor speeds [185]. The universal architecture of a typical ESCS is presented in Figure 13.

Some intelligent optimization algorithms, such as the gray wolf optimizer [186] and fuzzy logical inference [187], have been applied to optimize the parameters of seeding rate control algorithms. An ESCS with a dual closed-loop (rotational speed loop and current loop) PID control algorithm has been developed [188,189]. The proportional, integration, and differential parameters of the dual closed-loop PID control algorithm can be self-tuned based on a fuzzy logical inference model. Using the parameter-variable PID control algorithm, the average qualified index of the ESCS was 90.89% and the average coefficient of variation in the motor speed was less than 10.0%.

An ESCS that controls the rotational speed of the seeding motor or seed-metering device is a type of semi-closed loop control system, as it does not feed the actual seeding rates back. If there are disturbances affecting the seed quantity discharged per seed-metering device rotation, the semi-closed loop control system cannot monitor and eliminate these disturbances. Hence, some researchers have integrated seeding rate sensors into ESCSs, making them full closed-loop control systems [190,191,192]. Liu et al. developed a seeding rate sensor based on the seed flow reconstruction technique, and then integrated this seeding rate sensor into an embedded ESCS [193]. To enhance the response speed and robustness of the control algorithm, an expert model was built to tune the parameters of the seeding rate control algorithm in real time. Concerning the ESCS with the expert optimization algorithm, its seeding accuracy and coefficient of variation were 94.12% and 6.77%, respectively.

Some ESCSs developed in recent years are summarized in

Table 5. Summary of recently developed ESCSs.

Seed Type	Loop Types	Control Algorithm	Control Accuracy	Reference
Corn	Closed loop	PID	96.50%	Zhao et al. [182]
Corn	Closed loop	PID	97.09%	He et al. [183]
Corn	Closed loop	Fuzzy PID	90.89%	Chen et al. [188]
Corn	Closed loop	Developed compensation algorithm for seeding lag	99.38%	He et al. [184]
Corn	Closed loop	Fuzzy PID	95.27%	Wang et al. [189]
Corn	Semi-closed loop	PID	97.64%	He et al. [185]
Wheat	Closed loop	PID	90.00%	Liu et al. [193]
Wheat	Semi-closed loop	Seed quantity real-time correcting algorithm	97.39%	Ding et al. [194]

. The associated hardware configurations, control algorithms, and experimental results can serve as valuable references for subsequent studies.

5. Intelligent Seed Position Control

5.1. Trajectory Control of Seeding Machinery

The accuracy of the motion trajectory for seeding machinery is crucial for achieving precise seed spacing and row spacing [195], and cereal seeds can only be accurately placed with the specified seed spacing and row spacing when the seeding machinery accurately follows the desired trajectory. Therefore, in the field of intelligent seeding machinery, an increasing number of studies have concentrated on the trajectory control of seeding machinery [196,197].

The schematic of a general seeding machinery trajectory control system is displayed in Figure 14. At present, the trajectory of seeding machinery depends upon controlling the tractor. The methods employed for trajectory control in seeding machinery can be mainly categorized into three primary types: kinematic [198], dynamic [199], and model-free [200] methods.

Universal kinematic methods, including the pure pursuit tracking (PPT), linear quadratic regulator (LQR), and model predictive control (MPC) methods, are particularly straightforward to implement and, thus, have been widely adopted [201]. The PPT method selects a series of points along the reference path as look-ahead points, enabling agricultural machinery to follow the designated path by sequentially approaching these look-ahead points [202,203]. Both the LQR and MPC methods perform extensive online iterative optimization to predict the future behaviors of agricultural machinery [204,205]. The LQR method is optimally applicable to linear systems without rigid constraints [206]. While LQR can also be utilized for nonlinear systems through linearization, this is typically confined to the vicinity of the specific operating points of the system.

As for the MPC method, it is crucial to develop the state space model of the seeding machinery. Subsequently, the objective function and constraint conditions must be formulated in accordance with the seeding environment. Subsequently, an MPC controller should be established to predict the future trajectory of the seeding machinery [207,208]. The MPC algorithm demonstrates robust anti-interference capabilities and high control precision; however, its extensive iterative computations make it time-consuming [205].

Compared to kinematic model control methods, motion trajectory tracking algorithms based on dynamic models have been developed to enhance control accuracy and robustness. Among these dynamic approaches, the sliding mode control algorithm is regarded as the most classical technique [209,210]. To mitigate the influence of unknown disturbances, some nonlinear disturbance observers have been constructed in accordance with seeding machinery dynamics models, and the observed disturbances were compensated by the sliding mode control algorithm [209]. While the sliding mode control method shows strong robustness and a rapid response, chattering may occur when system states traverse on the sliding mode surface, adversely impacting the tracking performance of agricultural machinery [211,212].

Due to the difficulty in measuring certain parameters, such as the friction coefficient between the tire and the ground and vehicle load distribution, developing an accurate kinematic or dynamic model of seeding machinery is still challenging. Therefore, an increasing number of studies have focused on model-free trajectory control methods to deal with the time-varying and nonlinear characteristics inherent to seeding machinery [213,214,215]. The PID controller is the most prevalently employed among the model-free trajectory control approaches, as it is relatively facile to implement in practice. Nevertheless, conventional PID controllers have fixed coefficients, which means that they are not adaptable to the field environment featuring time-varying parameters and external disturbances [216,217]. To address the aforementioned limitations, nonlinear trajectory control algorithms that integrate the PID controller and coefficient inference models have been proposed [218]. Based on the tracking error and error change rate, fuzzy logical inference models can identify the optimal proportional, integral, and differential coefficients of the PID controller in real time.

As a non-driven system, the movement of seeding machinery depends on the traction provided by the tractor [219], and the positions of the sown seeds rely on the trajectories of the seeding outlets. Most trajectory control algorithms are used to control the trajectory of the tractor center [220]. However, the actual motion trajectories of seeding outlets differ from that of the tractor center. Therefore, kinematic or dynamic models including the relative relationship between seeding machinery and the tractor (i.e., tractor–trailer models) have been proposed [221]. These models use the expected seeding machinery trajectory to determine the trajectory of the tractor, which is more practical for the control of seeding positions [222].

Table 6. Strengths and drawbacks of universal trajectory control methods.

Trajectory Control Algorithm	Model Type	Strengths	Drawbacks	Representative Reference
PPT	Kinematic model	The PPT algorithm has a high dynamic response and fewer input variables.	The PPT algorithm is susceptible to unstable and time-varying systematic parameters	Wu et al. [223]
LQR	Kinematic model	The controlled systems of the LQR should have a time-invariable state-space model with a defined goal	The LQR algorithm is dependent on the accuracy of the model	Hou et al. [224]
MPC	Kinematic model	The MPC is suitable for complex scenarios with higher precision demands and variable seeding conditions	The performance of the MPC is limited by the complicate parameter-tuning processes and enormous real-time computational burden	Zhou et al. [208]
Sliding mode control	Dynamic model	The sliding mode control is suitable for scenarios where the controlled system has uncertainties such as disturbances and perturbations	The output chattering cannot be eradicated completely	Liu et al. [225]
PID	Model-free	The PID algorithm does not need a precise model, so it is simple to implement	The PID algorithm with fixed parameters might be affected by external disturbances of nonlinear and time-varying systems	Wang et al. [213]
Fuzzy PID	Model-free	Fuzzy PID improves upon the robustness and adaptability of the conventional PID algorithm and can control nonlinear systems	When the inputs deviate from the fuzzy set, the output may have discontinuities or oscillations	An et al. [226]

lists the universal trajectory control methods for intelligent seeding machinery, along with their strengths and drawbacks. Representative references are also included.

5.2. Intelligent Seeding Depth Control

A proper seeding depth can improve the water and soil conservation capacity of cereal seeds. In order to ensure that cereal seeds emerge evenly and make full use of light, temperature, and soil resources, the seeding depth needs to be controlled precisely. Based on the developed active GSPMs, seeding depth control systems and methods have been developed to control the vertical positions of seeding outlets [227].

The real-time measurement of topography is a necessary pre-condition for achieving accurate seeding depth. The topographic measurement method mainly encompasses geometry measurement methods [228] and force measurement methods [229]. Geometry measurement methods directly measure the distance from the sensor to the soil surface [230]; for example, ultrasonic sensors [231,232], infrared laser sensors [233], and angle sensors [234,235] are commonly used in geometry measurement methods. A multi-sensor fusion system has been proposed to fuse the information obtained by the angle, ultrasonic, and rotary sensors installed on the seeding machinery, then use the geometric relationship to estimate the vertical distance from the frame to the seed outlets [236].

The force measuring method senses the force from the soil, which is then used to estimate the seeding depth according to the calibrated down force and seeding depth conversion model. For this purpose, force sensors, such as polyvinylidene fluoride (PVDF) piezoelectric film [237], flex bending sensors [238], and pin shaft sensors [239], are installed on the machinery frame, which measure the reactive force from the soil surface. Afterwards, the relationship between the output of the force sensor and the actual seeding depth is established via calibration. As a universal seeding depth control method, the block diagram of the hydraulic down force control system is illustrated in Figure 15.

As for actuators, hydraulic cylinders can provide sufficient force to drive the GSPM, and proportional directional valves are used to control the lifting or lowering of seeding outlets by changing the direction of the oil flow. A hydraulic seeding depth control system was developed to control the coulter depth and generate a stable seed bed [60]. In detail, the three-position hydraulic control system was determined to be the most cost-effective solution, featuring a rapid response when compared with proportional control and proportional–integral–differential control. Subsequently, in order to enhance the stability of the hydraulic control system, the controller was designed on a cascade basis, where the inner loop governs the position of the lift cylinder, while the outer loop regulates the seeding depth.

Besides the hydraulic control system, pneumatic cylinders can also be used for seeding depth control. A pneumatic cylinder is fixed with a connecting rod, and the developed pneumatic control system drives the connecting rod to lift or lower the outlets by controlling the inner pressure of the pneumatic cylinder [239]. A PVDF piezoelectric film sensor has been applied to measure the deformation quantity of the tire, allowing the pneumatic spring to control the pressure on the ground based on the PVDF-derived deformation [237,241].

To enhance the response speed and robustness of seeding depth control systems, intelligent algorithms have been adopted. A hydraulic control system with a fuzzy adaptive PID control algorithm has been utilized to control the down force generated by the compact wheel, which controls for the soil thickness between the seeds and the soil surface.

In

Table 7. Recently developed seeding depth control techniques.

Topographic Monitoring Sensor	Actuator	Control Algorithm	Control Accuracy	References
Angle sensor	Hydraulic cylinder	Compensation PID	98.60%	Zhou et al. [242]
Flex sensor and Hall sensor	Pneumatic cylinder	Mamdani fuzzy algorithm	82.00%	Li et al. [243]
Angle sensor and force sensor	Pneumatic cylinder	Pneumatic cylinder-to-down force model	98.91%	Gao et al. [239]
Angle sensor	Electric cylinder	Adaptive Fuzzy PID	93.00%	Xue et al. [65]
Down force sensor	Hydraulic cylinder	Bang-bang	91.33%	Bai et al. [91]
Polyvinylidene fluoride	Pneumatic cylinder	PID	90.00%	Huang et al. [237]
Ultrasonic sensor, angle sensor, and rotary sensor	Hydraulic cylinder	Double closed-loop PID control	95.60%	Suomi et al. [236]

, we summarize several types of seeding depth control systems and methods, where all of their control accuracies are higher than 90%. These research methods and results can serve as references for further research on seeding depth control systems.

6. Influences of Intelligent Seeding Machinery at Different Farming Scales

In this section, the applied models, costs and benefits, practical problems, and solutions associated with intelligent seeding machinery at different farming scales—namely, large-scale commercial planting and small-scale farmer planting—are compared.

6.1. Application Models

Large-scale commercial planting farms frequently use large intelligent seeding machinery with GPS navigation systems for cereal seeding. This large seeding machinery can be controlled through the use of a touch-screen in the cab of the tractor. Furthermore, the machinery can automatically measure the actual seeding rate and seeding depth using sensors, allowing for their control according to the measured values.

In contrast, small-scale planting farmers usually use small intelligent seeding machinery with less than 12 planting units. This small seeding machinery usually has less sophisticated technology and is available at a relatively low price. The universal seeding machinery for small field seeding typically includes motor-driven seed-metering devices, rotational speed control systems for seed-metering devices, and passive ground surface profiling mechanisms. Without expensive soil properties sensors, the small seeding machinery discharges seeds at a fixed seeding rate set by the operator.

6.2. Costs and Benefits

The use of intelligent seeding technology is conducive to ensuring that cereal seeds are placed uniformly, thus improving the cereal yield and quality. Bullock et al. [244] applied optimal seeding rate decision technology to obtain the theoretical seeding rate and used variable seeding machinery to discharge cereal seeds at the theoretical rate. Their results showed that, according to different soil properties, the seeding rate of corn seeds ranged from 44,000 to 104,000 particles per hectare. Through applying the soil property sensing technology and seeding rate decision technology, the maximum yield increased to 18.3 Mg per hectare. Moreover, using the seeding rate control technology, the farmers could increase their income by USD 12 per hectare, compared to that when applying conventional seeding methods [20].

In addition, through the use of seeding rate control technology, the cost could be reduced due to decreased double-sown fields; in particular, as the seed-metering devices can be controlled individually, double-planted areas can be avoided. Furthermore, the seeding rate control technology can save 4.3% of seeds, as well as improving the cereal yield by 17% through reducing yield losses. The cost savings associated with the adoption of intelligent seeding technology ranged from USD 4 to 26 per hectare, depending on the distribution of field types in a farming operation [245].

As for the growth quality of cereals, Dong et al. [246] applied intelligent seeding rate control technology to regulate the seeding rates. Their experimental results showed that, when the sowing rate was controlled at 3400 seeds per tray precisely, the growth quality of the rice was the highest, with the average plumpness value increasing by 0.18 and the seedling strength index increasing by 0.42 per individual plant.

According to the above analyses, intelligent seeding machinery technologies can distribute seeds with desirable seed spacing, row spacing, and seeding depth, and the consequent high cereal yield and quality allow farmers to turn a better profit. If a large-scale farming company plans to select intelligent seeding machinery, it is recommended that advanced active controllable actuators, as well as intelligent seeding rate and seed position control systems, are installed, considering the large scale of the farming area and long operation times. In this way, the use of more expensive intelligent seeding machinery can allow for higher profits. However, if a small-scale farmer wants to buy intelligent seeding machinery, in order to balance between cost and performance, it is recommended that at least an intelligent seeding rate control system is incorporated in the seeding machinery. In this way, the profit can be maximized at a small cost.

6.3. Practical Problems

Large-scale commercial planting farms are becoming more and more likely to apply advanced intelligent seeding machinery in their practical seeding operations. However, as there is still no standard data transmission protocol for agricultural machinery, the existing intelligent seeding machinery lacks compatibility with other agricultural machinery and farm management systems, and also has low scalability potential due to the use of newly developed components.

As for small-scale planting, the price of state-of-the-art intelligent seeding machines is typically high, making them a considerable expense for most small farmers. Furthermore, even if small-scale planting farmers can afford the price of intelligent seeding machinery, the maintenance of intelligent seeding machinery is another challenge. Small-scale planting farmers generally do not have sufficient technical ability to maintain the sophisticated electronic sensors and control systems. If they use the professional services provided by manufacturers, the cost of the seeding machinery may increase sharply. This is an intractable problem for small-scale planting farmers deciding whether to use intelligent seeding machinery or not.

6.4. Solutions

For large-scale commercial planting farms, it is wise to select intelligent seeding machinery that meets industry standards and ensures they have seeding machinery with open data interfaces, allowing for communication with other agricultural machinery and management systems. Moreover, a centralized agricultural management platform should be established. This platform needs to integrate the hardware and software data provided by all types of agricultural machinery in order to achieve centralized data management and analysis.

Regarding small-scale planting farmers, it is recommended to actively apply for agricultural modernization project funding provided by the state or local government in order to reduce procurement costs. Additionally, several small-scale planting farms can buy and use high-end intelligent seeding machinery together, thus sharing the associated costs. Moreover, small-scale planting farmers may purchase the intelligent seeding machinery in installments through financial leasing programs provided by financial institutions in order to ease the pressure of one-time payment [20,244].

7. Present State, Limitations, and Future Development

7.1. Present State and Limitations

Intelligent seeding machinery for cereals is one of the mainstream technologies in modern agricultural equipment. Researchers have carried out extensive studies in this field. Although some achievements have been made, there are still some limitations. The present state and limitations of intelligent seeding machinery are discussed as follows.

(1)

Existing studies on MDSDs have mainly concentrated on improving the qualified rate; however, few studies have focused on their adaptability to different cereal seeds or paid attention to the seed breaking rate of developed MDSDs. These limitations may mean that a given MDSD can only be used for a certain cereal, increasing the idle rate of the seeder. Additionally, seed breaking phenomena may frequently occur in practical cereal seeding operations.

(2)

Due to complex field operating conditions, the vibration of the seeding unit can affect the stability of the seeding depth and reduce the control accuracy of the GSPM. The majority of studies on GSPMs relate to structural designs, but few have focused on dynamic analyses to reduce the influence of vibrations caused by uneven soil surfaces. This research gap may result in an uncontrolled seed falling trajectory during practical cereal seeding.

(3)

A majority of high-precision soil property sensors are of the probe type, which must be inserted into the soil and transmit information using the IoT paradigm. In actual cereal seeding operations, the intelligent seeding machinery needs to implement variable seeding in real time; however, only a few soil property sensors can be installed on the seeding machinery. Hence, the seeding machinery must use previous soil property parameters, potentially reducing the accuracy of seeding rate decisions.

(4)

Existing seeding rate sensors only can monitor the seed quantity out of the seed-metering device, but the distribution of seeds in the soil cannot be monitored. Hence, existing seed spacing monitoring technology judges the spacing between adjacent seeds through estimation according to the time interval, and the actual seed spacings under soil cannot be monitored.

(5)

The objective of existing seeding rate control technology is to discharge cereal seeds at uniform and accurate time intervals via developing innovative control algorithms. However, some types of cereal seeds, such as corn and rice, benefit from being sown in a certain direction and position. Few studies have focused on the direction and position control of cereal seeds.

(6)

Studies on seeding depth control systems have mainly focused on how to control the down force or seeding depth precisely. However, in reality, the theoretical down force of seeding machinery should be varied according to the soil moisture rate of the cultivated field. At present, existing seeding depth control systems do not regulate the theoretical down force or seeding depth according to the soil conditions, leading to deviation in the actual seeding depth from the theoretical value.

7.2. Future Developments

In order to overcome the aforementioned limitations of the existing intelligent seeding machinery, corresponding technical solutions are proposed below.

(1)

Future research on MDSDs should aim to improve their adaptability to different cereal seeds. For example, the innovative structural holes of a pneumatic MDSD can suck in several types of seeds with similar shapes, such as wheat, rice, barley, and so on, improving the utility of the intelligent seeding machinery. Furthermore, by means of DEM simulation and practical experiments, the structural optimization of MDSDs should be conducted to reduce the seed breaking phenomenon during the filling, conveying, and discharging processes.

(2)

Dynamic analyses should be performed when the GSPM is working on uneven soil. According to the analytical results, new structures or methods which can absorb vibrations affecting the GSPM could potentially be developed.

(3)

Vehicle-mounted soil property sensors (which may be based on high-spectrum and near-infrared spectrum technologies) should be developed, and soil property estimation models should be researched. Compared to probe-type sensors, non-contact optical soil property sensors are more suitable for use in high-speed seeding machinery during practical seeding operations.

(4)

Intelligent seed distribution sensors that can detect seeds under covered soil need to be developed. Ground-penetrating radar might serve as a reference for such a seed distribution sensor, allowing for the monitoring of actual seed spacing, row spacing, and seeding depth. In this way, intelligent seeding machinery can provide more information for subsequent agronomic processes.

(5)

Innovative structures for seed-metering devices that can adjust the positions of seeds should be developed. The positions of cereal seeds should be adjusted during the seed-filling process, following which belt conveying or airflow technologies can contribute to placing seeds into the soil in the desired pose.

(6)

A LiDAR sensor could be installed in front of the tractor to measure the soil height conditions. Moreover, point cloud processing algorithms for soil height measurement should be developed in the future. In this way, the seeding depth can be controlled according to the soil height, instead of the down force.

8. Conclusions

This review provided a detailed overview of key technologies used in intelligent cereal seeding machinery from the perspective of active controllable seeding actuators, intelligent seeding rate control, and intelligent seed position control systems. Concerning the active controllable seeding actuators, MDSDs and GSPMs were introduced. Some MDSDs with innovative structures were summarized, and the application of DEM and CFD-EDM optimization methods to determine the structural parameters of MDSDs was expounded. The working principle and development processes of GSPMs were introduced. Existing soil property sensing technologies, designed to guide the optimal seeding rate decision, were then reviewed. Based on feedback from the seeding rate sensors, intelligent seeding rate control technology was introduced, explaining how to discharge cereal seeds according to a uniform time interval. Intelligent seed position control encompasses two processes. One involves the trajectory control of the seeding machinery, while the other involves seeding depth control. The precise trajectory control of seeding machinery can facilitate the distribution of cereal seeds with appropriate row spacings and seed spacings, and intelligent seeding depth control can be used to adjust the working depth of the seeding unit according to the height of the soil surface.

At the end of this article, the present state, limitations, and future developments of intelligent seeding technology were discussed and analyzed. Researchers have attained remarkable results in the development of active controllable seeding actuators, intelligent seeding rate control, and seed position control systems. The present state and limitations of intelligent technologies are summarized below.

(a)

The existing studies on MDSDs have mainly focused on the improvement of seeding performance, but MDSDs still lack adaptability for different cereal seeds;

(b)

The development of GSPMs has focused on structural designs, while few studies have analyzed the seeding depth errors caused by mechanical vibrations and proposed relevant solutions;

(c)

Moreover, most soil property sensors are of the probe type and, thus, cannot be installed on the seeding machinery;

(d)

Existing seeding rate sensors can only monitor the seeds out of the seed-metering device, meaning that the distribution of seeds in the soil cannot be monitored;

(e)

Seeding rate control technology to control the distribution of seeds using innovative control algorithms has been considered, but few studies have considered seed pose control approaches;

(f)

Existing seeding depth control systems mainly focus on precisely controlling the down force or seeding depth; however, they cannot regulate the theoretical down force or seeding depth according to the soil conditions.

In the future, it is advisable to conduct research with the goal of developing more adaptable MDSDs, machinery-mounted soil property sensors, seed distribution sensors, and seed posture control methods. We hope that this review effectively presents the research status of key technologies for intelligent cereal seeding machinery from several aspects, which can serve as a reference for future research in this field.