Variable-force compacting method and device for seeder based on fusion of vision and sensor

Abstract

The invention discloses a variable-force pressing method and device of a seeder based on vision and sensor fusion. The method comprises the following steps: acquiring an earth surface soil image through an industrial camera, extracting soil structure characteristics, and constructing a multi-source information fusion model to calculate a target ballast force in combination with ballast force and soil water content information acquired by a pressure sensor and a humidity sensor; and through feedforward calculation of the electric push rod and PID closed-loop control, self-adaptive adjustment of the ballast force is realized. And meanwhile, a model optimization and self-learning mechanism is introduced, so that the accuracy and adaptability of ballast force adjustment are improved. The matched device comprises a rack, a press wheel, an electric push rod, a camera, a pressure sensor, a humidity sensor and other structures, and the problems that an existing seeder press device is insufficient in press force adjusting precision and low in intelligent level are solved. The pressing force can be adjusted in a self-adaptive mode according to different soil conditions, the soil pressing uniformity and the sowing quality are improved, and the device is suitable for intelligent pressing operation of agricultural machines such as a sowing machine.

Description

Technical Field

[0001] This application belongs to the field of agricultural machinery and intelligent control technology, specifically relating to a variable force soil compaction method and device for a seeder based on the fusion control of multi-source information such as image recognition, pressure sensor and soil moisture sensor, which is used to realize adaptive compaction pressure adjustment during the sowing process and improve sowing quality and crop emergence consistency. Background Technology

[0002] With the continuous improvement of agricultural mechanization, sowing technology plays a crucial role in increasing crop yield and production efficiency. During the sowing process, soil compaction devices are an important component for ensuring stable sowing depth, improving seed coverage quality, and ensuring uniform seedling emergence. The core objective of soil compaction is to enhance seed germination by applying pressure to make the contact between the seed and the soil more compact. However, existing compaction devices in seeders still face certain technical challenges. Traditional mechanical fixed-force compaction methods, while simple in structure and low in cost, offer a fixed compaction effect and cannot adapt to varying working environments with different soil textures, moisture contents, and uneven surface surfaces. Furthermore, while existing spring-suspended compaction systems can mitigate the impact of surface undulations to some extent, their adjustment precision and adaptability are limited, making it difficult to meet the precise pressure requirements under different operating conditions. Hydraulic or pneumatic adjustable compaction devices, although capable of actively adjusting compaction pressure, are complex systems with high costs and place high demands on energy efficiency, control system stability, and maintainability.

[0003] With the continuous advancement of sensor and electronic control technologies, an increasing number of agricultural machinery and equipment are integrating various sensors, such as pressure sensors, downpressure sensors, and displacement sensors, to achieve real-time monitoring of soil compaction pressure or sowing depth. Based on feedback data, they can perform closed-loop control and adjustment of the actuators to improve operational quality. For example, Chinese patent document CN119717915A discloses a method and control device for controlling the sowing depth and compaction of individual seeding units. This method uses real-time data collected by pressure sensors and downpressure sensors to adjust the extension and retraction lengths of the compaction hydraulic cylinder and the downpressure hydraulic cylinder, thereby achieving variable force compaction and sowing depth control for individual seeding components. This improves the adaptability and operational quality of compaction operations to a certain extent. The aforementioned Chinese patent document CN119717915A may also be referred to as "Document 1" below.

[0004] Existing seeder compaction control methods based on pressure sensor feedback (including the technical solution disclosed in Reference 1) can adjust the compaction effect to a certain extent, but they have many shortcomings. The main problem is that traditional methods rely on local contact stress signals collected by pressure sensors, which cannot accurately reflect key soil characteristics such as soil surface condition, clod size distribution, roughness, and density. Therefore, this control method has poor adaptability to soil conditions and is prone to over- or under-compacting, especially in scenarios where compaction quality is greatly affected by clod size distribution and soil moisture content. Existing technologies have not yet provided a technical solution that can combine soil surface image recognition, pressure sensor, and moisture sensor data to accurately drive the actuator to achieve adaptive variable force compaction, which urgently needs further improvement and refinement. Summary of the Invention

[0005] The purpose of this application is to address the common problems in existing seeder compaction systems, such as fixed or manual adjustment of compaction pressure, poor adaptability to changes in soil conditions, insufficient precision and stability of compaction pressure adjustment, and low levels of automation and intelligence. This application provides a variable force compaction method and device for seeders based on vision and sensor fusion. By introducing visual feedforward sensing from an industrial camera, mechanical feedback from a pressure sensor, and moisture adjustment from a humidity sensor, an intelligent closed-loop control system of "visual feedforward + mechanical feedback + moisture adjustment" is constructed. This system enables real-time calculation and adaptive control of the target compaction pressure under different soil conditions, making compaction pressure adjustment more precise and stable, thereby improving the quality of sowing operations and seedling emergence consistency.

[0006] To achieve the above objectives, the first aspect of this application provides a variable force compaction method for seeders based on vision and sensor fusion. This method is applied to a variable force soil compaction system including an industrial camera, an electric push rod, a pressure sensor, a humidity sensor, a compaction wheel, and an embedded intelligent control platform. The method includes the following steps:

[0007] S1. System initialization: The image processing and operation control system is built through the embedded intelligent control platform. The image recognition model is loaded, and the zero point / range parameters of the industrial camera (including calibration parameters), pressure sensor and humidity sensor are initialized. Communication connection with the electric actuator control module is established to ensure that the system has stable data acquisition and command issuance capabilities.

[0008] S2. Multi-source data synchronous acquisition: An industrial camera installed in front of the press wheel acquires surface images at a fixed frame rate and generates timestamps. At the same time, a pressure sensor acquires the actual pressing pressure (or equivalent contact pressure) between the press wheel and the ground. (In the formula, (Indicates actual town pressure), humidity sensor collects soil volumetric water content. (In the formula, (representing soil volumetric water content), the platform aligns images and sensor data by timestamp to form a fused data frame;

[0009] S3. Soil Feature Extraction and State Assessment: The embedded intelligent control platform performs target detection or segmentation on surface images, extracts soil state features such as soil block size distribution, surface roughness, and porosity, and forms a soil feature vector. (In the formula, (Represents the soil state feature vector) and outputs the soil state level. or soil condition score (In the formula, Indicates soil condition category / grade. (This represents the overall score of soil condition).

[0010] S4. Target town pressure calculation, embedded intelligent control platform based on soil characteristics. (or One of them), real-time pressure and soil volumetric water content The target town pressure is calculated through a fusion mapping model. (In the formula, (Indicates target town pressure), and applies physical constraints. (In the formula, and (These are the minimum and maximum allowable suppression pressure thresholds, respectively).

[0011] S5. Electric push rod feedforward solution: Based on the mechanical balance relationship between the press wheel, wheel frame, electric push rod, and machine frame, the target press pressure is... Converted to ground normal load requirements of the press wheel (In the formula, (representing the target normal load or target contact force), further determining the feedforward target elongation of the electric actuator. (In the formula, (Indicates the target elongation of the push rod feedforward), and sends control commands to the electric push rod control module;

[0012] S6. Closed-loop feedback control: The system continuously reads the actual pressure feedback from the pressure sensor. Pressure on the target town The error was obtained by comparison. (In the formula, (Indicating pressure error), the push rod correction amount is output using a PID control algorithm. (In the formula, (This represents the correction value for the push rod elongation), forming the final control target for the push rod. (In the formula, (Indicates the final target elongation of the push rod), driving the electric push rod to extend and retract to achieve closed-loop stable control of the pressure.

[0013] S7. Model optimization and self-learning: the system records data during long-term operations. , , , Including control output data, combined with indicators such as steady-state error, fluctuation amplitude, response time, or operational effectiveness evaluation of the pressure, the parameters of the fusion mapping model are optimized through online updates or self-learning mechanisms, thereby improving the accuracy, robustness, and environmental adaptability of pressure regulation under different soil conditions.

[0014] During the sowing operation, steps S2 to S6 are executed cyclically according to a preset control cycle to achieve real-time tracking and control of the target pressure. Step S6 feeds back the updated results of the fusion model parameters to step S4 for the calculation of the target pressure and push rod control in the next control cycle.

[0015] In one possible implementation, the image recognition model in step S3 is a deep learning object detection model or instance segmentation model, used to identify soil clod targets and output bounding boxes or masks to extract soil structural features and achieve classification and evaluation of surface soil conditions. The soil clod size distribution can be characterized by calculating the equivalent diameter of the soil clods and forming statistics, where, for the ... The area of ​​each soil block target is denoted as (In the formula, For the first (pixel area or mapped area of ​​each soil block), its equivalent diameter Defined as: and for all Statistics such as mean, standard deviation, or quantiles are generated and denoted as... (In the formula, (Equivalent diameter statistic of soil blocks), used to comprehensively characterize the size and dispersion of surface soil blocks.

[0016] In one possible implementation, the surface roughness index (In the formula, A roughness index is a measure of surface roughness used to quantify the complexity and undulation of surface texture. As an optional implementation, it is used to... It can be obtained from the root mean square of the gray-level gradient magnitude within the region of interest: In the formula, The number of pixels in the region of interest. For the first grayscale value of each pixel. This represents the grayscale gradient at that pixel. A larger value indicates a more complex and undulating surface texture. As another optional implementation method, It can also be obtained by combining features of the gray-level co-occurrence matrix (contrast, entropy, homogeneity, etc.).

[0017] In one possible implementation, porosity (In the formula, The porosity (the percentage of pore area) is used to characterize the porosity and looseness of the Earth's surface. The formula for calculating the porosity is: In the formula, The area of ​​the pore region. This represents the total area of ​​the region of interest. Lower porosity generally indicates a denser surface layer; higher porosity generally indicates a looser surface layer.

[0018] In one possible implementation, a soil condition score is defined to comprehensively evaluate soil condition. (In the formula, The comprehensive score for soil condition is calculated using the following formula: In the formula, These are the weighting coefficients. The normalized mapping function is used; preferably, the normalized function can be linearly normalized, for example... (In the formula, , These are the minimum and maximum statistical diameters obtained from calibration, respectively; the other terms are similar. Furthermore, it can be determined according to... The range of intervals divides the land surface into large soil block areas, medium soil block areas, small soil block areas, or loose, moderate, and dense soil block areas, with the threshold determined by field trials or historical data.

[0019] In one possible implementation, a pressure sensor is mounted on the press wheel frame or its force transmission component to collect the contact pressure signal between the press wheel and the ground in real time. This contact pressure signal, after filtering and A / D conversion, is input to an embedded intelligent control platform to obtain the current actual press pressure. This is used for target town pressure calculation and closed-loop feedback control.

[0020] In one possible implementation, a humidity sensor is mounted on a bracket near the compaction wheel to collect soil volumetric moisture content in real time. The humidity sensor can operate based on capacitive or time-domain reflectometry (TDR) principles. The collected humidity data, after filtering, is input into an embedded intelligent control platform to provide a characterization of soil moisture status for pressure regulation. Preferably, the humidity sensor... Normalization (In the formula, (Normalized soil volumetric water content), used to unify the dimensions of the fusion model.

[0021] In one possible implementation, after obtaining soil feature parameters from image recognition and sensor data, the target hill pressure is calculated using a fusion mapping model. As an optional implementation, the target town pressure can be calculated using the following formula: In the formula, , , For adjustment coefficients, Scoring is given based on soil condition. For the actual pressure on the town, To normalize soil volumetric water content, This is the baseline pressure compensation term. It is used in the model. As a moisture regulation term, it indicates that the higher the soil moisture content, the lower the target compaction pressure should be to avoid excessive compaction of wet soil; when the soil is relatively dry... Increasing the target soil compaction pressure can appropriately improve the quality of contact between the soil covering and the seeds. Furthermore, constraints can be imposed on the target soil compaction pressure: In the formula, , These are the minimum and maximum stabilization pressure thresholds allowed by the system, respectively.

[0022] In one possible implementation, the target town pressure is obtained. Then, the system calculates the target elongation of the push rod feedforward based on the mechanical equilibrium relationship. As an optional implementation, the target pressing pressure is converted into the normal load requirement of the pressing wheel. : In the formula, This represents the equivalent contact area between the pressure wheel and the ground. Furthermore, the force transmission coefficient of the wheel frame-push rod mechanism is considered. (In the formula, (This is the geometric force transmission coefficient of the mechanism, used to characterize the proportional relationship between the output force of the push rod and the normal load of the press wheel). The required output force of the push rod can be expressed as: (In the formula, (This is for the push rod output force requirement). Combining this with the push rod displacement-output force characteristics or the system equivalent stiffness model, the target feedforward elongation of the push rod can be derived: In the formula, This represents the initial elongation of the push rod under no-load conditions. The equivalent stiffness constant of the pressing system, This is the equivalent constant load (or equivalent constant force term) generated on the ground by the self-weight of the press wheel and related structures.

[0023] In one possible implementation, the embedded intelligent control platform continuously reads the actual pressure feedback from the pressure sensor during operation. Pressure on the target town The error was obtained by comparison. The PID algorithm is used to calculate the push rod correction amount. To form the ultimate control target The system sends commands to the electric actuator control module via communication protocols (such as CAN, RS485 / Modbus, UART, etc.) to achieve closed-loop control of the pressure. Preferably, the PID controller includes output limiting and integral anti-saturation mechanisms to avoid overshoot and oscillation caused by integral accumulation when the actuator stroke reaches the upper and lower limits. Furthermore, in the event of sensor malfunction, image recognition failure, or data asynchrony, the system switches to a degraded control strategy (e.g., maintaining the previous effective target value or using only pressure closed-loop control) to ensure operational continuity and safety.

[0024] In one possible implementation, the system continuously records soil feature vectors during long-term operation. Soil volumetric water content Actual pressure on towns Pressure on target towns In addition to the corresponding control output data, and combined with the steady-state error of the pressure, the pressure fluctuation amplitude, the system response time and the evaluation index of the operation effect, the parameters of the fusion mapping model are optimized through online parameter updates or self-learning mechanisms, thereby improving the accuracy, robustness and environmental adaptability of the system in the control of pressure under different soil conditions.

[0025] This application also provides a variable force compaction device for a seeder based on vision and sensor fusion for implementing the above-mentioned method. The device includes a frame, a compaction wheel, a compaction wheel frame, an electric push rod, a scraper blade, a fixed bracket for connecting the electric push rod to the frame, an industrial camera, a pressure sensor, a humidity sensor, and an embedded intelligent control platform. One end of the electric push rod is hinged to the fixed bracket or frame, and the other end is hinged to the compaction wheel frame. By controlling the extension and retraction of the electric push rod, the compaction wheel frame is driven to swing / rotate, thereby changing the normal load of the compaction wheel on the ground to achieve active regulation of the compaction pressure. When the electric push rod extends, it pushes the compaction wheel frame downwards, causing the compaction wheel to contact the ground surface. The increased contact tightness increases the pressing pressure; when the electric push rod retracts, the pressing wheel frame moves upward, reducing the contact pressure between the pressing wheel and the ground, thus decreasing the pressing pressure; the industrial camera is installed in front of the pressing wheel and faces the ground to collect images of the soil surface in front of the pressing area; the pressure sensor and humidity sensor are installed on the pressing wheel frame or a nearby support to collect the pressing wheel's contact pressure signal and soil volumetric water content signal; the embedded intelligent control platform is electrically or communicatively connected to the camera, pressure sensor, humidity sensor, and electric push rod control module to perform image recognition, data fusion, target pressing pressure calculation, and closed-loop control of the electric push rod.

[0026] Compared with the prior art, this application has at least the following beneficial effects: First, by acquiring soil characteristics (such as soil block size distribution, surface roughness, and porosity) through image recognition algorithms, and combining the data input from pressure and humidity sensors, the system can calculate and dynamically adjust the target pressure in real time. This effectively avoids excessive soil compaction or insufficient compaction, improving sowing quality and seedling emergence consistency; secondly, it adopts a "push rod feedforward target elongation" method. +PID correction The closed-loop control structure compares the target pressure with the actual pressure and continuously corrects the output, so that the pressure stably tracks the target value, improving the accuracy and consistency of the operation. Thirdly, through model optimization and self-learning mechanism, the system can continuously optimize and integrate model parameters and control strategies in long-term operation, improve the response accuracy and robustness to different soil conditions and dynamic working conditions, thereby enhancing the adaptability and engineering practicality of the equipment in complex working environments. Attached Figure Description

[0027] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and the originals and elements are not necessarily drawn to scale.

[0028] Figure 1 A flowchart of a variable force compaction method for a seeder based on vision and sensor fusion provided in this application embodiment;

[0029] Figure 2 A schematic diagram of the structure of a variable force compaction device for a seeder based on vision and sensor fusion provided in this application embodiment;

[0030] Figure 3 A flowchart of an algorithm based on a visual fusion sensor provided for an embodiment of this application. Detailed Implementation

[0031] The present application will be further described below with reference to the accompanying drawings and specific embodiments.

[0032] like Figure 1 As shown, the overall process of the variable force compaction method for a seeder based on vision and sensor fusion provided in this application includes steps S1 to S7. The system initializes the visual model and multi-sensor parameters to achieve synchronous acquisition and fusion processing of surface images, compaction pressure, and soil moisture content; calculates the target compaction pressure based on visual feature extraction and multi-source information fusion, and achieves real-time stable adjustment of compaction pressure by combining a feedforward-feedback composite control strategy; at the same time, it continuously optimizes model parameters through model optimization and self-learning mechanisms to improve environmental adaptability.

[0033] Step S1, System Initialization. Based on the embedded intelligent control platform, a system framework integrating image processing and operation control is constructed. By loading the image recognition model, the zero point and range parameters of the industrial camera (including calibration parameters), pressure sensor, and humidity sensor are initialized, and a communication interface with the electric actuator control module is established to ensure the stability and reliability of system data acquisition and control command issuance.

[0034] Step S2: Synchronous acquisition of multi-source data. The system acquires surface images and generates timestamps at a fixed sampling frequency using an industrial camera installed in front of the press wheel; simultaneously, pressure sensors acquire the actual pressing pressure (or equivalent contact pressure) between the press wheel and the ground in real time. Humidity sensors collect soil volumetric water content. The embedded intelligent control platform synchronizes and aligns image data and sensor data based on timestamps to construct multi-source fusion data frames.

[0035] In one possible implementation, a pressure sensor is mounted on the press wheel frame or its force transmission component to collect the contact pressure signal between the press wheel and the ground; the contact pressure signal is filtered and converted by an A / D converter before being input into an embedded intelligent control platform to obtain the current actual press pressure. , to be used as input parameters for target town pressure calculation and closed-loop feedback control.

[0036] In one possible implementation, the humidity sensor is mounted on a bracket near the compaction wheel for real-time acquisition of soil volumetric moisture content. The humidity sensor operates based on capacitive or time-domain reflectometry principles. The acquired humidity signal, after filtering, is input into an embedded intelligent control platform as a soil moisture state characterization parameter during the pressure regulation process. Preferably, the humidity sensor... Normalization (In the formula, (Normalized soil volumetric water content), used to unify the dimensions of the fusion model.

[0037] Step S3: Soil Feature Extraction and State Assessment. The embedded intelligent control platform performs target detection or semantic segmentation on the acquired surface images, extracting feature parameters characterizing soil structure and surface morphology, including soil block size distribution, surface roughness, and porosity, to construct a soil state feature vector. (In the formula, (Represents the soil state feature vector) and outputs the soil state level. or soil condition score (In the formula, Indicates soil condition category / grade. (This represents the overall score of soil condition).

[0038] In one possible implementation, the image recognition model used in step 3 is a deep learning-based target detection model or instance segmentation model, which is used to identify surface soil targets and output corresponding bounding boxes or mask information to extract soil structure features and classify and evaluate the surface soil state.

[0039] In one possible implementation, the size distribution of soil blocks can be characterized by calculating the equivalent diameter of the soil blocks and forming a statistic, where, for the th The area of ​​each soil block target is denoted as (In the formula, For the first (pixel area or mapped area of ​​each soil block), its equivalent diameter Defined as: and for all Statistics such as mean, standard deviation, or quantiles are generated and denoted as... (In the formula, (Equivalent diameter statistics of soil blocks) is used to comprehensively characterize the size and dispersion of surface blocks.

[0040] In one possible implementation, the surface roughness in step 3 is expressed as a surface roughness index. (In the formula, The roughness index quantifies the complexity and undulation of surface texture (as a characterization of surface roughness). As an optional implementation, the roughness index... It can be obtained from the root mean square of the gray-level gradient magnitude within the region of interest: In the formula, The number of pixels in the region of interest. For the first grayscale value of each pixel. This represents the grayscale gradient at that pixel. A higher value indicates a more complex surface texture and more pronounced spatial undulations. As another optional implementation method, It can also be obtained by combining features of the gray-level co-occurrence matrix (contrast, entropy, homogeneity, etc.).

[0041] In one possible implementation, the porosity in step 3 (In the formula, The porosity (the percentage of pore area) is used to characterize the porosity and looseness of the Earth's surface. The formula for calculating the porosity is: In the formula, The area of ​​the pore region. The total area of ​​the region of interest. Porosity reflects the density or looseness of the surface structure; a lower porosity indicates a higher degree of surface density, while a higher porosity indicates a relatively loose surface structure.

[0042] In one possible implementation, a soil condition score is defined to comprehensively evaluate soil condition. (In the formula, The comprehensive score for soil condition is calculated using the following formula: In the formula, These are the weighting coefficients. The normalized mapping function is used; preferably, the normalized function can be linearly normalized, for example... (In the formula, , These are the minimum and maximum statistical diameters obtained from calibration, respectively; the other items are calculated similarly. Furthermore, a comprehensive score based on soil condition can be used... The range of values ​​is used to classify the surface condition into large, medium, or small soil blocks, or corresponding to different state levels such as loose, moderate, and slightly dense. The threshold values ​​for each classification are determined through field experiments or calibration methods based on historical operational data.

[0043] Step S4, Target Town Pressure Calculation. After obtaining soil characteristic parameters from image recognition and sensor data, the target town pressure is calculated using a fusion mapping model. As an optional implementation, the target town pressure can be calculated using the following formula: In the formula, , , For adjustment coefficients, Scoring is given based on soil condition. For the actual pressure on the town, To normalize soil volumetric water content, This is the baseline pressure compensation term. It is used in the model. As a moisture regulation term, it indicates that the higher the soil moisture content, the lower the target compaction pressure should be to avoid excessive compaction of wet soil; when the soil is relatively dry... Increasing the target soil compaction pressure can appropriately improve the quality of contact between the soil covering and the seeds. Furthermore, constraints can be imposed on the target soil compaction pressure: In the formula, , These are the minimum and maximum stabilization pressure thresholds allowed by the system, respectively.

[0044] Step S5, electric actuator feedforward solution. After obtaining the target pressure... Then, the system calculates the target elongation of the push rod feedforward based on the mechanical equilibrium relationship. As an optional implementation, the target pressing pressure is converted into the normal load requirement of the pressing wheel. : In the formula, This represents the equivalent contact area between the pressure wheel and the ground. Furthermore, the force transmission coefficient of the wheel frame-push rod mechanism is considered. (In the formula, (This is the geometric force transmission coefficient of the mechanism, used to characterize the proportional relationship between the output force of the push rod and the normal load of the press wheel). The required output force of the push rod can be expressed as: (In the formula, (This is for the push rod output force requirement). Combining this with the push rod displacement-output force characteristics or the system equivalent stiffness model, the target feedforward elongation of the push rod can be derived: In the formula, This represents the initial elongation of the push rod under no-load conditions. The equivalent stiffness constant of the pressing system, This is the equivalent constant load (or equivalent constant force term) generated on the ground by the self-weight of the press wheel and related structures.

[0045] Step S6, closed-loop feedback control. During the operation of the embedded intelligent control platform, the system continuously reads the actual pressure feedback from the pressure sensor. Pressure on the target town The error was obtained by comparison. The PID algorithm is used to calculate the push rod correction amount. To form the ultimate control target The system sends commands to the electric actuator control module via communication protocols (such as CAN, RS485 / Modbus, UART, etc.) to achieve closed-loop control of the pressure. Preferably, the PID controller includes output limiting and integral anti-saturation mechanisms to avoid overshoot and oscillation caused by integral accumulation when the actuator stroke reaches the upper and lower limits. Furthermore, when the system experiences sensor malfunction, image recognition failure, or data asynchrony, the system will switch to a degraded control mode (e.g., maintaining the previous effective target value or only performing pressure closed-loop control) to ensure the continuity and safety of the operation.

[0046] Step S7, Model Optimization and Self-Learning. Under long-term operating conditions, the system continuously records soil feature vectors. Soil volumetric water content Actual pressure on towns Pressure on target towns In addition to the corresponding control output data, and based on the steady-state error of the pressure, dynamic response characteristics and operation quality evaluation indicators, the fusion mapping model is optimized through online parameter updates or self-learning methods to improve the pressure regulation performance of the system under different soil environments.

[0047] In one possible implementation, during the sowing operation, steps S2 to S6 are executed cyclically according to a preset control cycle to achieve real-time tracking control of the target pressure; preferably, step S6 feeds back the updated results of the fusion model parameters to step S4 for the calculation of the target pressure and push rod control in the next control cycle.

[0048] like Figure 2 As shown, in this embodiment, the variable force compaction device for the seeder based on vision and sensor fusion is connected to the seeder via a frame 3. This device includes a frame 3, a compaction wheel 6, a compaction wheel frame 5, an electric push rod 2, a scraper blade 7, a fixed bracket 1 for connecting the electric push rod to the frame, an industrial camera, a pressure sensor, a humidity sensor, and an embedded intelligent control platform. The industrial camera, pressure sensor, humidity sensor, and embedded intelligent control platform are... Figure 2 The components are not individually labeled; one end of the electric push rod 2 is hinged to the fixed bracket 1 or the frame 3, and the other end is hinged to the press wheel frame 5. By controlling the extension and retraction of the electric push rod 2, the press wheel frame 5 is driven to swing / rotate, thereby changing the normal load of the press wheel 6 on the ground to achieve active regulation of the pressing pressure; when the electric push rod 2 extends, the push rod pushes the press wheel frame 5 downward, increasing the tightness of the contact between the press wheel 6 and the ground surface, thereby increasing the pressing pressure; when the electric push rod 2 retracts, the press wheel frame moves upward, and the contact pressure between the press wheel 6 and the ground weakens, thereby reducing the pressing pressure; the industrial camera is installed in front of the press wheel and faces the ground to collect images of the soil surface in front of the pressing area; the pressure sensor and humidity sensor are installed on the press wheel frame or a nearby bracket to collect the press wheel's contact pressure signal and the soil volumetric water content signal; the embedded intelligent control platform is electrically or communicatively connected to the camera, pressure sensor, humidity sensor, and electric push rod control module to perform image recognition, data fusion, target pressing pressure calculation, and electric push rod closed-loop control.

[0049] In one possible implementation, the embedded intelligent control platform preferably employs an embedded platform with image processing and model inference capabilities, such as a small industrial computer running Linux or an embedded processor with a GPU / dedicated acceleration module, and connects to the electric linear actuator controller via a bus communication interface (such as RS485, CAN, Ethernet, etc.). The electric linear actuator controller receives target displacement commands, velocity, acceleration, and start / stop signals from the embedded intelligent control platform and drives the electric linear actuator 2 to complete the corresponding extension and retraction actions.

[0050] After the device is installed, the first step is system initialization (S1). An integrated system for image processing and operation control is built using an embedded intelligent control platform. The image recognition model is loaded and configured, and the zero-point and range parameters of the industrial camera (including calibration parameters), pressure sensor, and humidity sensor are initialized. At the same time, a communication interface with the electric actuator control module is established to ensure that the system has stable data acquisition and control command issuance capabilities.

[0051] During operation, the agricultural machinery moves along the designated route, while the variable-force compaction device of the seeder, based on vision and sensor fusion, moves synchronously with the seeder. The system acquires surface images and generates timestamps at a fixed sampling frequency using an industrial camera mounted in front of the compaction wheel; simultaneously, pressure sensors acquire the actual compaction pressure (or equivalent contact pressure) between the compaction wheel and the ground in real time. Humidity sensors collect soil volumetric water content. The embedded intelligent control platform synchronizes and aligns image data and sensor data based on timestamps to construct multi-source fusion data frames.

[0052] The embedded intelligent control platform performs target detection or semantic segmentation on the acquired surface images, extracts feature parameters characterizing soil structure and surface morphology, including soil block size distribution, surface roughness, and porosity, and constructs a soil state feature vector. It also outputs the soil condition level. or soil condition score .

[0053] The size distribution of soil clods can be characterized by calculating the equivalent diameter of the clods and forming a statistical measure, where, for the th The area of ​​each soil block target is denoted as (In the formula, For the first (pixel area or mapped area of ​​each soil block), its equivalent diameter Defined as: and for all Statistics such as mean, standard deviation, or quantiles are generated and denoted as... (In the formula, (Equivalent diameter statistics of soil blocks) is used to comprehensively characterize the size and dispersion of surface blocks.

[0054] Surface roughness is measured by the surface roughness index. Quantify the complexity and undulation of surface texture. As an optional implementation method, the roughness index... It can be obtained from the root mean square of the gray-level gradient magnitude within the region of interest: In the formula, The number of pixels in the region of interest. For the first grayscale value of each pixel. This represents the grayscale gradient at that pixel. A higher value indicates a more complex surface texture and more pronounced spatial undulations. As another optional implementation method, It can also be calculated by combining features of the gray-level co-occurrence matrix (contrast, entropy, homogeneity, etc.).

[0055] porosity The porosity is used to characterize the porosity and looseness of the earth's surface. The formula for calculating porosity is: In the formula, The area of ​​the pore region. The total area of ​​the region of interest. Porosity reflects the density or looseness of the surface structure; a lower porosity indicates a higher degree of surface density, while a higher porosity indicates a relatively loose surface structure.

[0056] After obtaining the characteristic parameters of soil structure and surface morphology (including soil block size distribution, surface roughness, and porosity) and soil volumetric water content... After the parameters, define the soil condition score. The calculation formula is as follows: In the formula, These are the weighting coefficients. The normalized mapping function is used; preferably, the normalized function can be linearly normalized, for example... (In the formula, , These are the minimum and maximum statistical diameters obtained from calibration, respectively; the other items are calculated similarly. Furthermore, a comprehensive score based on soil condition can be used... The range of values ​​is used to classify the surface condition into large, medium, or small soil blocks, or corresponding to different state levels such as loose, moderate, and slightly dense. The threshold values ​​for each classification are determined through field experiments or calibration methods based on historical operational data.

[0057] After obtaining soil feature parameters from image recognition and sensor data, the target pressure is calculated using a fusion mapping model. As an optional implementation, the target town pressure can be calculated using the following formula: In the formula, , , For adjustment coefficients, Scoring is given based on soil condition. For the actual pressure on the town, To normalize soil volumetric water content, This is the baseline pressure compensation term. It is used in the model. As a moisture regulation term, it indicates that the higher the soil moisture content, the lower the target compaction pressure should be to avoid excessive compaction of wet soil; when the soil is relatively dry... Increasing the target soil compaction pressure can appropriately improve the quality of contact between the soil covering and the seeds. Furthermore, constraints can be imposed on the target soil compaction pressure: In the formula, , These are the minimum and maximum stabilization pressure thresholds allowed by the system, respectively.

[0058] Pressure on the target town Then, the system calculates the target elongation of the push rod feedforward based on the mechanical equilibrium relationship. As an optional implementation, the target pressing pressure is converted into the normal load requirement of the pressing wheel. : In the formula, This represents the equivalent contact area between the pressure wheel and the ground. Furthermore, the force transmission coefficient of the wheel frame-push rod mechanism is considered. (In the formula, (This is the geometric force transmission coefficient of the mechanism, used to characterize the proportional relationship between the output force of the push rod and the normal load of the press wheel). The required output force of the push rod can be expressed as: (In the formula, (This is for the push rod output force requirement). Combining this with the push rod displacement-output force characteristics or the system equivalent stiffness model, the target feedforward elongation of the push rod can be derived: In the formula, This represents the initial elongation of the push rod under no-load conditions. The equivalent stiffness constant of the pressing system, This is the equivalent constant load (or equivalent constant force term) generated on the ground by the self-weight of the press wheel and related structures.

[0059] During the operation of the embedded intelligent control platform, the system continuously reads the actual pressure feedback from the pressure sensor. Pressure on the target town The error was obtained by comparison. The PID algorithm is used to calculate the push rod correction amount. To form the ultimate control target The system sends commands to the electric actuator control module via communication protocols (such as CAN, RS485 / Modbus, UART, etc.) to achieve closed-loop control of the pressure. Preferably, the PID controller includes output limiting and integral anti-saturation mechanisms to avoid overshoot and oscillation caused by integral accumulation when the actuator stroke reaches the upper and lower limits. Furthermore, when the system experiences sensor malfunction, image recognition failure, or data asynchrony, the system will switch to a degraded control mode (e.g., maintaining the previous effective target value or only performing pressure closed-loop control) to ensure the continuity and safety of the operation.

[0060] Under long-term operating conditions, the system continuously records soil feature vectors. Soil volumetric water content Actual pressure on towns Pressure on target towns In addition to the corresponding control output data, and based on the steady-state error of the pressure, dynamic response characteristics and operation quality evaluation indicators, the fusion mapping model is optimized through online parameter updates or self-learning methods to improve the pressure regulation performance of the system under different soil environments.

[0061] During the sowing operation, steps S2 to S6 are executed cyclically according to a preset control cycle to achieve real-time tracking and control of the target pressure. Preferably, step S6 feeds back the updated results of the fusion model parameters to step S4 for the calculation of the target pressure and push rod control in the next control cycle.

[0062] like Figure 3 As shown, the algorithm flow based on visual fusion sensors in this application includes steps S301 to S305. This process realizes closed-loop control of the rolling pressure based on multi-source information fusion, which can dynamically adjust the rolling pressure of the rolling wheel on the ground according to the surface soil condition and real-time pressure feedback, effectively improving the adaptability, stability and quality of sowing operations.

[0063] Step S301 involves multi-source data acquisition. An industrial camera is mounted in front of the press wheel to acquire surface images at a preset sampling frequency and generate corresponding timestamp information; a pressure sensor is mounted on the press wheel frame or its force transmission component to collect the actual pressing pressure or equivalent contact pressure between the press wheel and the ground in real time, denoted as... The humidity sensor is installed on a bracket near the compactor wheel to collect real-time data on soil volumetric moisture content, denoted as . The data collected by the aforementioned sensors are processed through signal conditioning, filtering, and analog-to-digital conversion before being input into the embedded intelligent control platform, providing the raw data foundation for subsequent fusion processing and control decisions.

[0064] Step S302 involves timestamp alignment and fusion processing of multi-source data. The embedded intelligent control platform performs time synchronization and fusion processing on multi-source data from industrial cameras, pressure sensors, and humidity sensors. Specifically, the platform aligns image data and sensor data based on a unified timestamp, eliminating sampling time sequence differences and constructing a data structure that includes surface image information and actual pressure. and soil volumetric water content The data is a multi-source fused data frame. This fused data frame serves as input to the control algorithm, used for subsequent feature analysis and control calculations. Through timestamp alignment and fusion processing, the consistency of multi-source information in both spatial and temporal dimensions can be effectively improved, thereby enhancing the accuracy and stability of control decisions.

[0065] Step S303 involves using a PID algorithm to correct the elongation of the electric actuator. The system executes control algorithm calculations based on fused data frames to correct the elongation of the electric actuator. Preferably, the embedded intelligent control platform will collect the actual pressing pressure in real time. Pressure on the target town By comparison, the pressure error of the ballast can be obtained. Based on this, a proportional-integral-derivative (PID) control algorithm is used to calculate the corrected control quantity of the push rod. Its expression is: ,in, , and These are the proportional, integral, and derivative control coefficients, respectively. Through the PID control algorithm, dynamic compensation for pressure errors can be achieved, improving system response speed and suppressing pressure fluctuations.

[0066] Step S304 involves the control and execution of the electric linear actuator. The system uses the calculated actuator control quantity for the control and execution of the electric linear actuator. The embedded intelligent control platform controls the target elongation amount based on the feedforward control. Correction amount from PID algorithm output The ultimate control objective for generating the electric linear actuator is: Subsequently, the embedded intelligent control platform sends control commands to the electric actuator control module via a communication interface (such as CAN bus, RS485 / Modbus, or UART communication protocol) to drive the electric actuator to extend and retract, thereby adjusting the actual pressing pressure of the press wheel on the ground to approach the target pressing pressure. .

[0067] Step S305 displays the results on the embedded intelligent control platform. The embedded intelligent control platform displays the system's operating status and control results. Specifically, this includes real-time display of surface images, soil condition assessment results, and target pressure. Actual pressure on towns The system displays information such as the current extension of the push rod and the system's operating status. Through a visual interface, operators can intuitively obtain information about the system's operation, facilitating work monitoring and parameter adjustment.

Claims

1. A variable force compaction method for seeders based on vision and sensor fusion, characterized in that, The method, applied to a pressing system comprising a pressing wheel, an electric actuator, an industrial camera, a pressure sensor, a humidity sensor, and an embedded intelligent control platform, includes: S1. System Initialization: The embedded intelligent control platform loads the image recognition model, establishes data communication between the industrial camera, pressure sensor, humidity sensor and electric actuator control module, and completes the initialization of camera intrinsic / extrinsic parameter calibration parameters, sensor zero point and range parameters and control cycle. S2. Multi-source data synchronous acquisition: An industrial camera acquires images of the ground surface in front of the press wheel at a fixed frame rate and generates a timestamp for each frame; a pressure sensor acquires the real-time contact pressure between the press wheel and the ground. Humidity sensors collect soil volumetric water content. Image data and sensor data are aligned with timestamps to form a fused data frame; S3. Soil Feature Extraction and State Assessment: Target detection or segmentation is performed on the surface image to obtain soil block regions and pore regions, and soil state feature vectors are calculated within a preset region of interest. ,in At least include: a) Equivalent diameter statistics of soil blocks : by the area of ​​the soil block Calculate the equivalent diameter and to Forming a mean, quantile, or discrete distribution; b) Surface roughness index Roughness characterization measures calculated from image grayscale gradient magnitude or texture statistics; c) Porosity : Obtained by the ratio of the area of ​​the pore region to the total area of ​​the region of interest; And based on Output soil condition level or soil condition score ; S4. Target Town Pressure Calculation: The embedded intelligent control platform will... , or Real-time contact pressure and soil volumetric water content Input the fusion mapping model to calculate the target town pressure. and to Apply physical constraints ; S5. Electric push rod feedforward solution: Based on the mechanical balance relationship formed by the press wheel—wheel frame—electric push rod and machine frame, the target press pressure... Converting this to the normal load requirement of the press wheel, and combining the device's geometric parameters with the push rod's stroke-load characteristics, the target elongation of the electric push rod is calculated. ; S6. Closed-loop feedback control: based on error... As input, a PID controller is used to output the push rod stroke correction amount. And set the push rod control command to The electric actuator moves to make the pressure wheel track the target pressure against the ground. ; S7. Model Optimization and Self-Learning: Recording during continuous operation. , , , The parameters of the fusion mapping model are updated online or adaptively adjusted based on the control output, the tracking error of the tamping pressure, the steady-state fluctuation, or the evaluation index of the operation effect. The updated parameters are used for the target tamping pressure calculation in the subsequent step S4 to improve the control accuracy and robustness under different soil conditions. Steps S2 to S6 are executed cyclically according to a preset control cycle.

2. The variable force compaction method for a seeder based on vision and sensor fusion according to claim 1, characterized in that, The image recognition model is a deep learning object detection model or instance segmentation model, used to identify soil clod targets and output soil clod contours or masks to calculate the equivalent diameter statistics of the soil clods. Surface roughness index and porosity .

3. The variable force compaction method for a seeder based on vision and sensor fusion according to claim 1 or 2, characterized in that, The soil condition score It is obtained from the following state evaluation function: in These are the weighting coefficients. The weighting coefficients are obtained from field trials or historical operational data, and are normalized mapping functions.

4. The variable force compaction method for a seeder based on vision and sensor fusion according to any one of claims 1 to 3, characterized in that, The fused data frame aligns image data with sensor data through timestamp synchronization, interpolation, or sliding window matching, and sets a maximum alignment error threshold. When the threshold is exceeded, the previous valid fused data frame is used or the system enters a degraded control mode.

5. The variable force compaction method for a seeder based on vision and sensor fusion according to any one of claims 1 to 3, characterized in that, The fusion mapping model is any one or a combination of a piecewise linear model, a fuzzy rule model, a regression model, or a state estimation model based on a Kalman structure, used to... Mapped as target town pressure .

6. The variable force compaction method for a seeder based on vision and sensor fusion according to any one of claims 1 to 3, characterized in that, Target elongation of the electric actuator in step S5 The target town pressure is determined by the following process: Converted to normal load requirement of the press wheel And based on the lever arm ratio or transmission coefficient between the push rod and the wheel frame, Convert this to the required output force of the push rod, and then solve it using the push rod displacement-output force characteristic curve. .

7. The variable force compaction method for a seeder based on vision and sensor fusion according to any one of claims 1 to 3, characterized in that, The PID controller includes output limiting and integral anti-saturation mechanisms, and freezes or recalculates the integral term when the push rod stroke reaches the upper or lower limit to ensure the stability of the pressure control.

8. The method according to any one of claims 1 to 3, characterized in that, The online update in step S7 uses sliding window least squares, recursive least squares or incremental gradient update to update the parameters of the fusion mapping model, and sets update trigger conditions. The trigger conditions include at least: changes in soil condition level, changes in humidity range or the steady-state error of pressure exceeding a threshold.

9. A variable force compaction device for a seeder based on vision and sensor fusion, characterized in that, Includes a frame, press rollers, press roller frames, electric actuators, scraper blades, industrial cameras, pressure sensors, humidity sensors, and an embedded intelligent control platform; among which: a) The press wheel is connected to the frame via a press wheel frame. The press wheel frame can rotate or swing relative to the frame to change the normal load of the press wheel on the ground. b) One end of the electric push rod is hinged to the frame or fixed bracket, and the other end is hinged to the press wheel frame. The extension and retraction of the electric push rod is used to drive the press wheel frame to rotate or swing, thereby realizing the adjustable pressure of the press wheel on the ground. c) The industrial camera is installed in front of the press wheel and facing the ground, for real-time acquisition of images of the ground surface in front of the press wheel; d) The pressure sensor is mounted on the press wheel frame or its force transmission component to collect the real-time contact pressure between the press wheel and the ground. ; e) The humidity sensor is mounted on a bracket near the compaction wheel and is used to collect soil volumetric water content. ; f) The embedded intelligent control platform is communicatively connected to the industrial camera, pressure sensor, humidity sensor, and electric actuator control module, respectively, and is used to extract soil state features from surface images. Real-time contact pressure With soil volumetric water content Calculate the pressure of the target town Based on the feedforward target elongation and the PID closed-loop feedback control, the electric push rod control command is output, enabling the press wheel to adaptively track the target press pressure relative to the ground press pressure. Simultaneously, it stores job data for self-learning updates of fusion mapping model parameters.

Images