Automatic row alignment method and system for interplanting wheat harvester

Abstract

The application discloses a kind of automatic line-to-line methods and systems of interplanting wheat harvester, including obtaining seeding prior information;Get touch correction amount;Calculate touch correction confidence, statistical prior confidence and geometric prior confidence;According to touch correction information, seeding prior information and touch correction confidence, statistical prior confidence and geometric prior confidence, form the progressive fusion decision logic of structure prior determination parallel reference, statistical prior quantization handover line error, geometric prior constraint trajectory shape, touch correction compensation local deviation, finally form navigation path.Through the construction of multi-level prior model, dynamic confidence fusion and long-time online correction, the problem that the prior art is not sufficient to the information utilization of seeding link, poor adaptability to interplanting working condition is solved.

Description

Technical Field

[0001] This invention relates to the field of automatic navigation technology for agricultural machinery, specifically to an automatic row alignment method and system for wheat harvesters based on sowing prior and tactile correction. It is applicable to wheat harvesters in intercropping mode to achieve precise automatic row alignment and harvesting while avoiding damage to intercropped crops. Background Technology

[0002] Currently, wheat intercropping (such as wheat + cotton, wheat + corn, wheat + chili peppers, etc.) is an important planting model that can make full use of light and heat resources and improve land productivity. In intercropping, wheat harvesters need to navigate between rows of cotton, corn, or chili peppers, requiring extremely high precision in row alignment: the header must be aligned with the center line of the wheat rows, otherwise, missed harvesting or damage to the intercropped crops will occur. Traditional manual row alignment suffers from high labor intensity, limited operating speed, and fatigue from prolonged operation. Therefore, developing an automated row alignment assisted driving system is of great significance.

[0003] Existing automatic row alignment technologies mainly include: mechanical touch-based (relying on continuous tangibility of crops, prone to failure due to missing plants or broken rows, and severe overshoot at high speeds), mechanical form-following (only capable of ground-based form-following, unable to solve inter-row alignment), parallel trajectory navigation (assuming strictly parallel crop rows, only using the trajectory of the first row to fit the reference slope, not considering the cumulative deviation caused by the error of the handover row, and lacking robustness to missing plants or broken rows), visual navigation (highly affected by light and shadow, difficult to identify when wheat stalks are dense), and multi-sensor fusion (such as GNSS+IMU, tactile+GNSS, although fusing multiple real-time data, it is essentially reactive fusion). The common drawback of the above methods is that they do not systematically mine and utilize the multi-level prior information known in the sowing process (fixed row spacing, statistical distribution of handover row error, curvature gradual change characteristics), lack structured modeling and dynamic confidence evaluation of prior information, and the fusion weights are fixed or based on simple rules, lacking long-term online correction capabilities, resulting in poor adaptability and inability to effectively eliminate cumulative deviations under complex working conditions such as wheat intercropping and harvesting. Furthermore, existing methods do not consider the initial row alignment issue after the harvester turns around at the end of the field. When the row handover error is large, the tactile sensor may fail to align with the crop row, leading to system failure or misjudgment. Summary of the Invention

[0004] To address the technical problems existing in the prior art, this invention aims to provide an automatic row alignment method and system for intercropped wheat harvesters based on sowing priors and tactile correction. By constructing a multi-level prior model, dynamic confidence fusion, and long-term online correction, it solves the problems of insufficient utilization of sowing information and poor adaptability to intercropping conditions in the existing technology.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: This invention proposes an automatic row alignment method for intercropped wheat harvesters, comprising the following steps: Acquire prior information for sowing; the prior information for sowing includes complete trajectory data of the sowing process, real-time recorded data of the first row of manual driving, a reference trajectory formed by manual-assisted discrete point marking, as well as prior information on the structure of the sowing machine, prior information on the statistical aspects of sowing operations, and prior information on the geometric aspects of the sowing trajectory. Obtain haptic correction information; Calculate the haptic correction confidence, statistical prior confidence, and geometric prior confidence; Based on tactile correction information, seeding prior information, tactile correction confidence, statistical prior confidence, and geometric prior confidence, a progressive fusion decision logic is formed, which determines the parallel benchmark through structural prior, quantifies the handover line error through statistical prior, constrains the trajectory shape through geometric prior, and compensates for local deviations through tactile correction, ultimately forming a navigation path.

[0006] As a further technical solution, the acquisition of prior information on sowing also includes As a further technical solution, the tactile correction information is calculated based on the offset detected by flexible tactile sensors symmetrically arranged on both sides of the front header of the harvester. The offsets detected by the left and right sensors are defined as follows: and Then the lateral deviation is defined as follows:

[0007] As a further technical solution, the calculation process for the tactile correction confidence level is as follows: Let the contact detection signals of the left and right tentacles be respectively and s when in contact L =1、s R =1, otherwise s L =0、s R =0; simultaneously define the volatility of the whisker signal. Let N be the standard deviation of the signals from the most recent N sampling points; then the formula for calculating the tactile correction confidence level is: , when =1 and =1 and

[0008] , when( =1 or =1) and

[0009] , when =0 and =0 in The volatility threshold is calibrated based on sensor characteristics.

[0010] As a further technical solution, the statistical prior confidence level The calculation process is as follows: The initial confidence level of the statistical prior for sowing operations is 1.0, and it is dynamically adjusted based on the observed average slope deviation. Let the average slope deviation from the start of the run to the current position be...

[0011] in, ξ is the instantaneous lateral deviation (cm), ξ is the distance traveled (m); x is the distance traveled from the current starting point. The statistical prior confidence level is then:

[0012] Where λ is the attenuation coefficient. The standard deviation of the slope error for the handover line (cm / m) represents the typical inter-line deviation magnitude. When the average slope deviation... Less than hour, Approaching 1; when Much larger hour, It decreases and approaches 0.

[0013] As a further technical solution, geometric prior confidence... The calculation method is as follows: When the rate of change of curvature satisfies the equation hour, ;otherwise ; Where s is the current arc length position, κ is the curvature; κ(s) is the curvature of the trajectory at arc length s. The step size is the arc length. The maximum rate of change of curvature, As a further solution: the aforementioned navigation path as follows: (8) in For parallel navigation baselines, For the preset threshold, Lateral deviation for geometric extrapolation:

[0014] In the formula, s is the current arc length position. The step size is the arc length. The heading angle of the previous position. The curvature of the previous position. This represents the lateral deviation from the previous position. The tangent direction of the navigation baseline; The specific integration strategy is as follows: (1) That is, when the tactile information is reliable, the fusion weight is determined according to the confidence level; when At that time, fully trust haptic correction and adjust the path in real time; when In some cases, a compromise should be made, taking into account both prior knowledge and modifications; (2) When tactile information is unreliable, if at this time That is, if the prior information is reliable, we can revert to the prior baseline for sowing; (3) When and That is, when both touch and the a priori of sowing are unreliable, if Then, use the geometric prior of the sowing trajectory for smooth extrapolation; if This indicates that the actual trajectory does not satisfy the smoothness assumption, and the system should not rely on geometric extrapolation. At this point, all information is unavailable, and an alarm is triggered to prompt manual intervention.

[0015] As a further technical solution, at the end of each row operation, the coordinates of the current row's endpoint and heading are recorded. Based on the prior knowledge of the seeder's structure, the predicted position of the next row's starting point is automatically calculated: the harvester is laterally shifted by a fixed row distance from the endpoint, while maintaining the heading. After automatically navigating to the predicted starting point, a tactile search is performed. Proceed at low speed while making small, alternating turns to the left and right, monitoring the contact status of the left and right tentacles, and searching for a stable contact position on both sides; the search distance does not exceed a preset threshold and the time does not exceed 10 seconds; if a stable contact is successfully found, enter the normal fusion navigation mode; if the search fails, an alarm prompts manual intervention.

[0016] As a further technical solution, when the statistical prior of the sowing operation and the tactile correction information continue to conflict, it indicates that the statistical prior itself needs to be updated. The specific steps for updating are as follows: (1) Conflict detection The average slope deviation up to the present is used as the conflict evaluation index, as in equation (7). A persistent conflict is determined when the following conditions are simultaneously met: A. The average slope deviation exceeds the threshold, i.e. β is a preset coefficient; This represents the average slope deviation from the starting point of the trip to the current position. The standard deviation of the slope error at the handover point; B. Within the preset window length, the tactile correction amount The proportion of identical numbers exceeds a preset threshold; C. The aforementioned conflict states A and B occur repeatedly in the same position interval across multiple consecutive rows.

[0017] (2) Correction execution Once persistent conflict is detected, the accumulated [data / information] is utilized. For the data points, the slope difference is estimated using linear regression through the origin:

[0018] This represents the distance traveled from the starting point of the current row to the i-th sampling point. This refers to the tactile lateral deviation at the corresponding sampling point; Before performing the origin return, the controller automatically zeroes the tactile deviation at the starting point (after the driver manually aligns the starting point, the system records and deducts the starting point deviation), thus ensuring... ; Update reference slope:

[0019] K new The updated slope; K ref For reference slope; Simultaneously update the statistical error:

[0020] in The slope difference estimated from the most recent N rows. i The value ranges from 1 to N, and std is the sample standard deviation.

[0021] As a further technical solution, a speed-adaptive pure tracking algorithm is used for path tracking control as follows: Pre-aiming distance Dynamically adjust based on speed:

[0022] in This is the lower limit of speed (typical value 1 m / s). Minimum aiming distance; v For real-time speed; Derive the target front wheel steering angle from geometric relationships:

[0023] in For the heading deviation, in the formula: For heading angle;

[0024] This refers to the wheelbase.

[0025] Secondly, the present invention also provides an automatic row alignment system for intercropped wheat harvesters, comprising: The sowing prior information acquisition module is configured to acquire sowing prior information, including sowing machine structure prior, sowing operation statistical prior, and sowing trajectory geometric prior; The tactile correction amount acquisition module is configured to acquire tactile correction amounts; The confidence calculation module is configured to calculate the haptic correction confidence, statistical prior confidence, and geometric prior confidence. The navigation path generation module is configured to form a progressive fusion decision logic based on tactile correction information, seeding prior information, tactile correction confidence, statistical prior confidence, and geometric prior confidence. This logic involves determining the parallel benchmark using structural priors, quantifying the handover line error using statistical priors, constraining the trajectory shape using geometric priors, and compensating for local deviations using tactile corrections. Finally, a navigation path is generated.

[0026] The beneficial effects of this invention are as follows; This invention makes full use of sowing trajectory information: by exploring the prior knowledge of the sowing machine structure, the prior knowledge of sowing operation statistics, and the prior knowledge of sowing trajectory geometry, a complete information utilization chain from deterministic constraints to real-time observation is formed, which solves the problem that existing technologies do not make full use of the known information in the sowing process.

[0027] This invention improves the adaptability to handover line errors: when the statistical prior fails locally, tactile correction takes over and updates the statistical prior parameters through long-term correction, thus solving the problem of cumulative deviation in the pure GPS parallel trajectory method under handover line errors.

[0028] This invention improves the adaptability to missing plants and broken rows: when tactile correction temporarily fails, it reverts to statistical prior and uses geometric prior to smoothly extrapolate, maintaining the row matching capability, thus solving the problem of the failure of the pure mechanical touch method in the missing plant area.

[0029] This invention improves high-speed operation performance: the speed-adaptive pure tracking algorithm enables the system to maintain good line-tracking accuracy even at high speeds.

[0030] This invention is adapted to complex intercropping conditions: a specially designed flexible tendril sensor can reliably sense standing wheat plants in the presence of intercropped crops, providing accurate local corrections and solving the problem that visual and traditional tactile methods are difficult to distinguish between wheat rows and intercropping rows in intercropping modes.

[0031] This invention features complementary prior information from multiple sources: by integrating prior information from three sources—sowing trajectory, first-row driving, and manual marking—the system can still function normally under extreme conditions such as missing sowing data or abnormal first-row conditions, significantly enhancing the system's robustness and adaptability. Attached Figure Description

[0032] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0033] Figure 1 A flowchart of a navigation framework based on seeding priors and tactile corrections; Figure 2 This is a top view of the cutting platform of the present invention; Figure 3 This is a side view of the cutting table of the present invention; Figure 4 It is a system workflow diagram; Detailed Implementation It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0034] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, unless otherwise expressly indicated by the invention, the singular form is also intended to include the plural form. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof. For ease of description, the words "up," "down," "left," and "right" appearing in this invention only indicate that they are consistent with the up, down, left, and right directions of the accompanying drawings themselves, and do not limit the structure. They are merely for the purpose of facilitating the description of this invention and simplifying the description, and do not indicate or imply that the device or component referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0035] As described in the background section, existing technologies have shortcomings. To address these technical problems, this invention aims to solve the issues of existing automatic row alignment technologies failing to fully utilize known trajectory information during the sowing process and exhibiting poor adaptability to intercropped wheat harvesting. It provides an automatic row alignment method and system for intercropped wheat harvesters based on sowing priors and tactile correction. The essential difference between this invention and existing technologies lies in that it does not use the reference slope of the first row fitting as a fixed parallel benchmark. Instead, it constructs a multi-level, multi-source prior information system (including sowing trajectory, first row record, and manual marking), and extracts three levels of priors from it: seeder structural prior (fixed row spacing), sowing operation statistical prior (normal distribution of handover row error), and sowing trajectory geometric prior (gradually changing curvature). Based on this, adaptive fusion of multi-source information is achieved through dynamic confidence assessment and long-term online correction. In contrast, existing parallel trajectory navigation methods treat subsequent lines as strictly parallel, ignoring handover line errors; although multi-sensor fusion methods integrate multiple real-time data, they are essentially reactive fusions, lacking structured modeling and confidence assessment of prior information, and do not have the ability to perform long-term online correction.

[0036] The harvester upon which the above method relies includes an RTK positioning module, flexible whisker sensors, a controller, and a steering actuator; among which RTK positioning module: Installed on the wheat harvester to acquire real-time position and heading information; Flexible tendril sensors: Symmetrically arranged on the front of the harvester's header, with a spacing equal to the wheat row spacing. The tendril length and diameter are designed according to the wheat plant height (tendril length 25cm, diameter 2mm, installation height slightly higher than the top of the wheat stubble), used to sense standing wheat plants and output lateral offset; Controller: Connected to the RTK positioning module and tactile sensors, configured to execute the above method steps; Steering actuator: Connected to the controller, including a steering ratio valve and an angle sensor, used to drive steering according to the target front wheel angle output by the controller.

[0037] This invention uses multi-level prior information from the sowing process as a benchmark, and leverages real-time tactile perception for local corrections, forming a progressively integrated decision-making logic: "structural priors determine parallel benchmarks, statistical priors quantify handover row errors, geometric priors constrain trajectory shapes, and tactile corrections compensate for local deviations." Specifically, the seeder's structural priors provide a deterministic benchmark for strict row parallelism; the statistical priors quantify the probability distribution of handover row errors across different passes; the geometric priors of the sowing trajectory utilize the gradually changing curvature characteristics to provide smooth constraints; and real-time tactile correction information is used to dynamically compensate for local lateral deviations. These four types of information are integrated hierarchically according to their deterministic, statistical, geometrical constraint, and real-time observation levels, collectively forming a complete basis for row-specific decisions.

[0038] This embodiment proposes an automatic row alignment method for intercropped wheat harvesters, including the following steps: Step 1: Obtain prior information for sowing The prior information for sowing in this invention mainly falls into two categories. One category consists of dimensional parameters such as the width of the seeder and the spacing between the seed metering devices, which can be obtained by reading preset values ​​or through manual input. The other category is sowing estimation information, which can originate from one or more of the following methods in combination: (1) Complete trajectory data of the sowing process: Read the trajectory file recorded by the agricultural machinery RTK automatic driving system during sowing to obtain the historical location information of all crop rows in the plot. This data includes the starting point and ending point coordinates and trajectory point sequence of each row. Based on this, combined with the width of the seeder and the spacing of the seed metering device, the sowing passes can be segmented, and then the statistical characteristics of the handover row error between different passes can be extracted.

[0039] (2) Real-time recording of the first row of manual driving: In the plot to be harvested, the wheat harvester is first manually driven by the driver to complete the first row of operation. The controller records RTK positioning data at a preset frequency, and the reference slope is obtained by fitting after the operation is completed. .

[0040] (3) Manually assisted discrete marking: In extreme cases such as missing sowing trajectory data (e.g., not recorded or lost during sowing) or ineffective operation of the first row (e.g., irregular plots), operators are allowed to use a handheld high-precision marking device or vehicle-mounted marking system to mark several key points along the crop row. The system generates a baseline trajectory based on these points.

[0041] The three sources mentioned above can be used individually or in combination, depending on the actual situation. When multiple sources are available simultaneously, the system automatically performs data fusion and verification according to priority (seeding trajectory > first row record > discrete points).

[0042] Based on the above prior information, we further explore the following three levels of prior information: (1) Prior to the structure of the seeder: The wheat seeder uses multiple seed metering devices fixedly mounted on a crossbeam. The lateral distance between each seed metering device is determined by the mechanical structure and remains constant. Therefore, in one pass, the trajectories between each crop row are strictly parallel, and the row spacing is equal to the seed metering device spacing. This prior can be expressed as: This holds true for all i, j (1) in This represents the distance between the i-th and j-th rows within the same pass. The row spacing is fixed for the seeder. Based on this prior, the navigation baseline for all crop rows within the same pass has the same reference slope. .

[0043] (2) Statistical Prior of Sowing Operations: Sowing operations on the same plot consist of multiple passes, with overlapping row errors between adjacent passes. For the m-th pass (m≥2), its true slope... and There are discrepancies: (2) in Let the slope error of the m-th pass follow a zero-mean normal distribution: (3) The standard deviation of the handover line error (cm / m) reflects the typical magnitude of the handover line error. The following methods can be used for offline estimation: (a) If RTK trajectories were recorded when the plot was sown, multiple sowing trajectories can be directly fitted with straight lines to calculate the standard deviation between the slope of each trajectory and the mean; (b) In the early stage of the method operation, the deviation data accumulated by the tactile correction in the first few rows can be used for estimation.

[0044] (3) Geometric Prior of Seeding Trajectory: When the seeder is operating in the field, the trajectory curvature changes slowly due to factors such as the inertia of the steering system, the requirements for work quality, and the driver's operating habits, and there will be no abrupt changes. This prior can be expressed as: , holds for all s (4) Where κ(s) is the curvature of the trajectory at arc length s (m - ¹), The arc length step size is (m). The maximum rate of change of curvature (m) - ²), determined by the seeder's turning capability and operational quality requirements.

[0045] Step 2: Obtain haptic correction information For harvesting intercropped wheat: Specialized flexible tendril sensors are installed, symmetrically arranged on both sides of the front header of the harvester. The tendril length is designed to produce a measurable bending deformation upon contact with standing wheat stalks, while avoiding accidental contact with intercropped crops. The bending angle is measured by an angle sensor and converted into a lateral offset after calibration.

[0046] Let the offsets detected by the left and right sensors be respectively and (Positive values ​​indicate deflection to the left, negative values ​​indicate deflection to the right), define lateral deviation: (5) Step 3: Information Confidence Assessment (1) Soft-modified confidence

[0047] Define the effective contact states of the left and right tentacles. Let the contact detection signals of the left and right tentacles be respectively... and (s=1 when in contact, s=0 otherwise). The volatility of the tentacle signal is also defined. Let be the standard deviation (normalized value) of the signal from the most recent N sampling points. Then, the formula for calculating the haptic correction confidence level is: , when =1 and =1 and

[0048] , when( =1 or =1) and

[0049] , when =0 and =0(6) in The volatility threshold is calibrated based on sensor characteristics.

[0050] (2) Statistical prior confidence

[0051] The initial confidence level of the statistical prior for sowing operations is 1.0, and it is dynamically adjusted based on the observed average slope deviation. Let the average slope deviation from the start of the run to the current position be...

[0052] in, ξ represents the instantaneous lateral deviation (cm), and ξ represents the distance traveled (m). The statistical prior confidence level is then: (7) Where λ is the attenuation coefficient. The standard deviation of the slope error for the handover line (cm / m) represents the typical inter-line deviation magnitude. When the average slope deviation... Less than hour, Approaching 1; when Much larger hour, It decreases and approaches 0.

[0053] (3) Geometric prior confidence

[0054] When the rate of change of curvature satisfies equation (4), ;otherwise This confidence level is used to determine whether geometric extrapolation can be enabled.

[0055] Step 4: Weighted Fusion and Path Generation The merged navigation line equations are as follows: (8) in For parallel navigation baselines, For the preset threshold, Lateral deviation for geometric extrapolation: (9) In the formula, s is the current arc length position. Step size, The heading angle of the previous position. The curvature of the previous position. This represents the lateral deviation from the previous position. This refers to the tangent direction of the navigation baseline.

[0056] The specific integration strategy is as follows: (1) That is, when the tactile information is reliable, the fusion weight is determined according to the confidence level.

[0057] when At that time, fully trust haptic correction and adjust the path in real time; when In some cases, a compromise should be made, taking into account both prior knowledge and modifications; (2) When tactile information is unreliable, if at this time That is, if the prior information is reliable, we can revert to the prior baseline for sowing.

[0058] (3) When and That is, when both touch and the a priori of sowing are unreliable, if Then, the geometric prior of the sowing trajectory is used for smooth extrapolation. If This indicates that the actual trajectory does not satisfy the smoothness assumption, and the system should not rely on geometric extrapolation. At this point, all information is unavailable, and an alarm is triggered to prompt manual intervention.

[0059] Step 5: Automatic initial alignment after turning around At the end of each row operation, the system records the coordinates of the current row's endpoint and heading. Based on the seeder's structural priors (fixed row spacing), it automatically calculates the predicted position of the next row's starting point: laterally shifting one fixed row spacing from the endpoint while maintaining the heading. After the harvester automatically navigates to the predicted starting point, it performs a tactile quick search: Proceed at a low speed (e.g., 0.5 m / s) while making small, alternating turns (e.g., ±5°) to the left and right, monitoring the contact status of the left and right tentacles and searching for a stable contact point on both sides. The search distance should not exceed a preset threshold (e.g., 3 m) and the search time should not exceed 10 seconds. If a stable contact is successfully found, enter the normal fusion navigation mode. If the search fails, an alarm will prompt manual intervention.

[0060] Step 6: Long-term correction of statistical priors When the statistical priors for sowing operations and the tactile correction information continue to conflict, it indicates that the statistical priors themselves need to be updated.

[0061] (1) Conflict detection The average slope deviation up to the present is used as the conflict evaluation index, as in equation (7). A persistent conflict is determined when the following conditions are simultaneously met: A. The average slope deviation exceeds the threshold, i.e. β is a preset coefficient; B. Within the preset window length, the tactile correction amount The proportion of identical numbers exceeds a preset threshold; C. The above-mentioned conflict states occur repeatedly in the same position interval of multiple consecutive rows.

[0062] (2) Correction execution Once persistent conflict is detected, the accumulated [data / information] is utilized. The slope difference is estimated using linear regression through the origin for the data points (where x is the distance traveled from the current row starting point): (12) Before performing the origin return, the controller automatically zeroes the tactile deviation at the starting point (after the driver manually aligns the starting point, the system records and deducts the starting point deviation), thus ensuring... .

[0063] Update reference slope: (13) Simultaneously update the statistical error: (14) in For the most recent N The slope difference of the row estimate, where std is the sample standard deviation.

[0064] Step 7: Path tracking control A speed-adaptive pure tracking algorithm is employed. Pre-aiming distance. Dynamically adjust based on speed: (15) in This is the lower limit of speed (typical value 1 m / s). This is the minimum aiming distance.

[0065] Derive the target front wheel steering angle from geometric relationships (wheat harvesters typically use front wheel steering): (16) in For heading deviation, This refers to the wheelbase.

[0066] Furthermore, based on the above method, this embodiment also provides an automatic row alignment system for intercropped wheat harvesters, including: The sowing prior information acquisition module is configured to acquire sowing prior information, including sowing machine structure prior, sowing operation statistical prior, and sowing trajectory geometric prior; The tactile correction amount acquisition module is configured to acquire tactile correction amounts; The confidence calculation module is configured to calculate the haptic correction confidence, statistical prior confidence, and geometric prior confidence. The navigation path generation module is configured to form a progressive fusion decision logic based on tactile correction information, seeding prior information, tactile correction confidence, statistical prior confidence, and geometric prior confidence. This logic involves determining the parallel benchmark using structural priors, quantifying the handover line error using statistical priors, constraining the trajectory shape using geometric priors, and compensating for local deviations using tactile corrections. Finally, a navigation path is generated.

[0067] The system also includes an automatic start-alignment module after turning around, configured to record the coordinates of the current row's endpoint and heading at the end of each row's operation. Based on the seeder's structural priors (fixed row spacing), the system automatically calculates the predicted position of the next row's starting point: shifting laterally by a fixed row spacing from the endpoint while maintaining the heading. After the harvester automatically navigates to the predicted starting point, it performs a rapid tactile search: advancing at a low speed (e.g., 0.5 m / s) while simultaneously making small, alternating left and right turns (e.g., ±5°), monitoring the contact status of the left and right tendrils to find a stable contact position on both sides. The search distance does not exceed a preset threshold (e.g., 3 m) and the time does not exceed 10 seconds. If a stable contact is successfully found, it enters the normal fusion navigation mode. If the search fails, an alarm prompts manual intervention.

[0068] Furthermore, the system also includes a long-term correction module for statistical priors, which is configured to perform conflict detection and correction.

[0069] Furthermore, the system also includes a path tracking control module.

[0070] This invention makes full use of sowing trajectory information: by exploring the prior knowledge of the sowing machine structure, the prior knowledge of sowing operation statistics, and the prior knowledge of sowing trajectory geometry, a complete information utilization chain from deterministic constraints to real-time observation is formed, which solves the problem that existing technologies do not make full use of the known information in the sowing process.

[0071] This invention improves the adaptability to handover line errors: when the statistical prior fails locally, tactile correction takes over and updates the statistical prior parameters through long-term correction, thus solving the problem of cumulative deviation in the pure GPS parallel trajectory method under handover line errors.

[0072] This invention improves the adaptability to missing plants and broken rows: when tactile correction temporarily fails, it reverts to statistical prior and uses geometric prior to smoothly extrapolate, maintaining the row matching capability, thus solving the problem of the failure of the pure mechanical touch method in the missing plant area.

[0073] This invention improves high-speed operation performance: the speed-adaptive pure tracking algorithm enables the system to maintain good line-tracking accuracy even at high speeds.

[0074] This invention is adapted to complex intercropping conditions: the flexible tendril sensor can reliably sense standing wheat plants in the presence of intercropped crops, providing accurate local corrections and solving the problem that visual and traditional tactile methods are difficult to distinguish between wheat rows and intercropping rows in intercropping modes.

[0075] This invention features complementary prior information from multiple sources: by integrating prior information from three sources—sowing trajectory, first-row driving, and manual marking—the system can still function normally under extreme conditions such as missing sowing data or abnormal first-row conditions, significantly enhancing the system's robustness and adaptability.

[0076] The following is a specific example to illustrate this: This embodiment uses a self-propelled wheat combine harvester of a certain brand as a platform for illustration, but the application of the present invention is not limited to this model.

[0077] 1.1 Positioning Module Employing the UM982 high-precision RTK positioning module and utilizing the "Qianxun Zhicun" centimeter-level high-precision positioning service, a horizontal positioning accuracy of 2cm is achieved. The RTK dual antennas are symmetrically mounted on the roof of the cab, with a baseline length of 1.5m, providing both positioning and orientation information. The positioning data output frequency is 5Hz and is transmitted to the controller via a CAN bus.

[0078] 1.2 Flexible Tentacle Sensor The sensor consists of two sets of tentacles arranged symmetrically on the left and right sides, each set containing: Tentacle body: made of spring steel wire, 25cm in length, 2mm in diameter, with rust-proof treatment on the surface; Angle sensor: MLX90316 Hall angle sensor, measuring range 0~90°, accuracy ±0.5°; Mounting bracket: The height is adjustable so that the tendril tip is positioned below the wheat plant, ensuring contact with upright wheat straw while avoiding contact with intercropped crops.

[0079] The distance between the left and right tendrils is equal to the harvesting width of one pass, and it is installed 25cm in front of the divider at the front of the harvester's header. The sensor sampling frequency is 10Hz, and the output is transmitted to the controller via the CAN bus.

[0080] Angle-Offset Calibration: Under static conditions, the tentacles are deflected by known angles (e.g., 10°, 20°, 30°...), and the corresponding lateral offset is measured. Calibration coefficients are obtained using linear fitting. .

[0081] 1.3 Controller The controller uses an STM32F407VGT6 microcontroller as the main processor, with a main frequency of 168MHz, and is equipped with: Two CAN interfaces are used to receive RTK positioning data and tactile sensor signals. Analog signal acquisition interface, used to acquire front wheel angle sensor signal (0~5V); Pulse acquisition interface, used to acquire signals from the vehicle speed sensor; The PWM output interface is used to drive the steering proportional valve.

[0082] The controller has 2MB of built-in Flash and 256KB of RAM for program storage and runtime data caching.

[0083] 1.4 Steering actuator The front wheel steering system is hydraulically driven and equipped with: Steering proportional valve: ATOS DKZOR type, response frequency 10Hz, control current 0~1A; Front wheel angle sensor: WDD35 potentiometer-type angle sensor, measuring range ±45°, accuracy ±0.2°; Steering cylinder: Double-acting hydraulic cylinder with a stroke of 120mm.

[0084] 2. Acquisition and Initialization of Prior Information 2.1 Fitting the first row reference slope In newly cultivated fields awaiting harvest, the wheat harvester is first manually operated by a driver to complete the first row-to-row operation. During the operation, the controller records RTK positioning data points at a frequency of 5Hz. The sampling interval is approximately 0.14m (calculated based on an operating speed of 2.5m / s).

[0085] After the first line of operation is completed, the controller processes the recorded trajectory data: (1) Segmented median filtering: The trajectory points are grouped into groups of 100 points per M (corresponding to a journey of about 14m), and the median point of each group is calculated. Suppose that a total of N median points are obtained.

[0086] (2) Least squares fitting: Perform linear fitting on N median points and calculate the reference slope. :

[0087] in , It is the average of N median points.

[0088] 2.2 Statistical Prior Initialization If RTK trajectories were recorded during planting, the controller can directly read the planting trajectory data, perform linear fitting on multiple planting trajectories, and calculate the standard deviation between the slope of each trajectory and the mean. The initial value.

[0089] If there is no historical data, the controller is set. The initial value is 0.025 cm / m (corresponding to a typical handover error of 5 cm over a 200m row length). It will be updated online based on actual observations in subsequent operations.

[0090] 2.3 Setting of Geometric Prior Parameters The maximum rate of change of curvature is set based on the turning performance of the seeder. The curvature is calculated using the difference in heading angle between adjacent sampling points: ,in The distance between sampling points is the arc length.

[0091] 2.4 Requirements for manual assisted marking When using a manual-assisted discrete point marking method to obtain prior information, there should be no less than 3 key points marked along the crop rows for each row.

[0092] 3. Online work process 3.1 Job Initialization When subsequent row operations begin, the driver aligns the wheat harvester with the starting point of the row of wheat to be harvested. After the starting point is aligned, the system automatically performs a tactile sensor validity check: it reads the contact status of the left and right tentacles, and if... = 0, the touchscreen displays "Start point offset, please fine-tune"; if only one side is in contact, it displays "Please adjust towards the contact side". After confirming valid contact, tap "Start Automatic Alignment" on the touchscreen. The controller will then adjust according to the reference slope. Generate a parallel navigation baseline using the starting coordinates:

[0093] 3.2 Real-time data acquisition The controller synchronously acquires the following data at a frequency of 10Hz: RTK positioning data: longitude, latitude, heading angle ,speed ; Haptic sensor data: Left tentacle angle Right antenna angle ; Front wheel angle sensor: actual front wheel steering angle ; Vehicle speed sensor: vehicle speed (Redundant with RTK speed).

[0094] 3.3 Calculation of tactile correction amount (1) Angle-offset conversion: The tentacle bending angle is converted into a lateral offset using a calibration function. in =0.25 cm / ° =0 (zero offset has been calibrated). Calculate similarly. .

[0095] (2) Overall lateral deviation:

[0096] (3) Low-pass filtering: for For a first-order low-pass filter with a cutoff frequency of 2Hz, the recursive formula is:

[0097] 3.4 Real-time calculation of confidence level (1) Soft-modified confidence : Contact detection signal , Judgment: When or If contact is detected, it is considered to be in contact (s=1); otherwise, there is no contact (s=0). Volatility Calculation: Take the most recent N=10 sampling points (corresponding to 1 second of data) Standard deviation, normalized; Volatility threshold (Normalized value, corresponding to approximately 10% of the signal variation); Substitute into equation (6) to calculate When the distance traveled At that time, calculations are not yet complete. Directly ordered (Initial trust prior); or use an approximation .

[0098] (2) Statistical prior confidence : Average slope deviation Calculation: Starting from the beginning of the row, maintain the cumulative amount. ,in This represents the current distance traveled (accumulated from the starting point). The sampling interval is approximately 0.25 m. Then... Simplified algorithm in practice: Update once per meter, using the current... The ratio to x is approximately equal.

[0099] Substitute into equation (7) to calculate

[0100]

[0101] (3) Geometric prior confidence : Real-time curvature calculation: ; Rate of change of curvature: ; like ,but ,otherwise .

[0102] 3.5 Generation of Fusion Navigation Lines The fused navigation line is generated based on the confidence level decision logic, and the specific rules are as follows: If tactile correction confidence The fused navigation line is taken as the parallel navigation baseline plus a tactile correction amount weighted by confidence level, i.e.

[0103] like = 0 and statistical prior confidence level >0.3: The system reverts to the sowing prior baseline, i.e. .

[0104] like = 0、 ≤ 0.3 and geometric prior confidence level = 1: Enable geometric extrapolation. First, calculate the extrapolation lateral deviation: in The step size is 0.25 m in this embodiment. The heading angle at the previous sampling time. The curvature at the previous sampling time. This represents the actual lateral deviation at the previous sampling time. This refers to the tangent direction of the navigation baseline at the current location.

[0105] The integrated navigation line is: .

[0106] If none of the above conditions are met (i.e.) = 0、 ≤ 0.3 and = 0), the system will issue an alarm and prompt manual intervention.

[0107] 3.6 Collision Detection and Long-Term Correction The controller performs conflict detection at the end of each line: (1) Calculation of cumulative conflict index (calculation of average slope deviation):

[0108] (2) Trend consistency calculation: Taking a window length W=20m, calculate the positive and negative proportions respectively: Pick .

[0109] (3) Continuous conflict determination: when and When a conflict occurs, the position range of that row is recorded. If five consecutive rows conflict within the same position range (error ≤ 10m), a correction is triggered.

[0110] (4) Correction execution: Collect the most recent 5 rows Data points (sampling interval 0.25m) are calculated according to formula (12). Update according to formula (13) Update according to formula (14) .

[0111] 3.7 Path Tracking Control The controller performs path tracing at a frequency of 10Hz: Calculate the lateral deviation d and the heading deviation ; Calculate the aiming distance using formula (15) ; Calculate the target front wheel steering angle according to formula (16). .

[0112] 3.8 Closed-loop control of front wheel steering angle The controller performs corner closed-loop control at a frequency of 50Hz: Read the actual front wheel steering angle ; Calculate the angle deviation ; PWM duty cycle calculation uses proportional control plus dead-time compensation, as shown in the following formula: like (Turn right): , ; like <0 (Turn left): , ; like = 0: .in , (No action is taken when the duty cycle is below 35%).

[0113] 4. Data Storage and Logs The controller's built-in memory records the following data in a loop: The start and end times of each job line; Reference slope for each line of work and statistical error ; Average absolute deviation and maximum absolute deviation for each line of work; The time, location, and slope correction amount for each correction ; The timing and cause of abnormal events (alarms, manual takeover).

[0114] Data can be exported via touchscreen for subsequent analysis and algorithm optimization.

[0115] 5. Parameter Tuning Instructions Those skilled in the art can adjust the following parameters according to actual operating conditions:

[0116] 6. Exception Handling 6.1 RTK Loss Handling When the RTK positioning module outputs a lock-out flag or the positioning accuracy exceeds 5cm, the controller reduces the operating speed to 3km / h and increases the tactile correction confidence level. Weight, if If the value is 0, geometric prior extrapolation is enabled. If the lock is lost continuously for more than 30 seconds, an alarm will be issued and a manual takeover will be requested.

[0117] 6.2 Tactile sensor malfunction When the strobe signal remains unchanged for an extended period or exceeds a reasonable range, the controller is set... ， Increase statistical prior confidence The system will display a fault message indicating a weighting issue, suggesting that the sensor be checked.

[0118] 6.3 Dual Failure Handling when and and When the value is 0, the controller will issue an audible and visual alarm, automatically decelerate to a stop, and prompt manual intervention.

[0119] 7. Human-computer interaction interface Touchscreen displays offer the following features: Real-time display: current speed, lateral deviation, heading deviation, and confidence levels; Track display: Shows historical tracks and the current navigation line; Parameter settings: Allows adjustment of various control parameters; Job statistics: Displays the number of rows harvested, average deviation, and maximum deviation; Alarm information: Displays the abnormal status and handling suggestions.

[0120] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An automatic row alignment method for an intercropped wheat harvester, characterized in that, Includes the following steps: Acquire prior information for sowing, including complete trajectory data of the sowing process, real-time recorded data of the first row of manual driving, a baseline trajectory formed by manual-assisted discrete point marking, as well as prior information on the structure of the sowing machine, prior information on the statistical aspects of sowing operations, and prior information on the geometric aspects of the sowing trajectory. Obtain tactile correction amount; Calculate the haptic correction confidence, statistical prior confidence, and geometric prior confidence; Based on tactile correction information, seeding prior information, tactile correction confidence, statistical prior confidence, and geometric prior confidence, a progressive fusion decision logic is formed, which determines the parallel benchmark through structural prior, quantifies the handover line error through statistical prior, constrains the trajectory shape through geometric prior, and compensates for local deviations through tactile correction, ultimately forming a navigation path.

2. The automatic row alignment method for intercropped wheat harvesters as described in claim 1, characterized in that, The aforementioned tactile correction setting The calculation is based on the offset detected by the flexible tendril sensors symmetrically arranged on both sides of the front header of the harvester. The offsets detected by the left and right sensors are defined as follows: and Then define the tactile correction amount. : 。 3. The automatic row alignment method for intercropped wheat harvesters as described in claim 1, characterized in that, The haptic correction confidence level The calculation process is as follows: Let the contact detection signals of the left and right tentacles be respectively and s when in contact L =1、s R =1, otherwise s L =0、s R =0; simultaneously define the volatility of the whisker signal. Let N be the standard deviation of the signals from the most recent N sampling points; then the formula for calculating the tactile correction confidence level is: , when =1 and =1 and , when( =1 or =1) and , when =0 and =0 in The volatility threshold is calibrated based on sensor characteristics.

4. The automatic row alignment method for intercropped wheat harvesters as described in claim 1, characterized in that, The statistical prior confidence level The calculation process is as follows: The initial confidence level of the statistical prior for sowing operations is 1.0, and it is dynamically adjusted based on the observed average slope deviation. Let the average slope deviation from the start of the row to the current position be: in, ξ represents the instantaneous lateral deviation, ξ represents the distance traveled, and x represents the distance traveled from the current starting point. The statistical prior confidence level is then: Where λ is the attenuation coefficient. The standard deviation of the slope error for the handover line represents the typical inter-line deviation magnitude; when the average slope deviation... Less than hour, Approaching 1; when Much larger hour, It decreases and approaches 0.

5. The automatic row alignment method for intercropped wheat harvesters as described in claim 1, characterized in that, Geometric prior confidence The calculation method is as follows: When the rate of change of curvature satisfies the equation hour, ;otherwise ; Where s is the current arc length position, κ is the curvature, and κ(s) is the curvature of the trajectory at arc length s. The step size is the arc length. This represents the maximum rate of change of curvature.

6. The automatic row alignment method for intercropped wheat harvesters as described in claim 1, characterized in that, The navigation path as follows: in: To calculate the prior confidence level; Adjust confidence level for tactile feedback; For tactile correction; Geometric prior confidence; Parallel navigation baseline; For the preset threshold, Lateral deviation for geometric extrapolation: The specific integration strategy is as follows: (1) That is, when the tactile information is reliable, the fusion weight is determined according to the confidence level; when At that time, fully trust haptic correction and adjust the path in real time; when In some cases, a compromise should be made, taking into account both prior knowledge and modifications; (2) When tactile information is unreliable, if at this time That is, if the prior information is reliable, we can revert to the prior baseline for sowing; (3) When and That is, when both touch and the a priori of sowing are unreliable, if Then, use the geometric prior of the sowing trajectory for smooth extrapolation; if This indicates that the actual trajectory does not satisfy the smoothness assumption, and the system should not rely on geometric extrapolation. At this point, all information is unavailable, and an alarm is triggered to prompt manual intervention.

7. The automatic row alignment method for intercropped wheat harvesters as described in claim 1, characterized in that, At the end of each row operation, the coordinates of the current row's endpoint and heading are recorded. Based on the seeder's prior structural knowledge, the predicted position of the next row's starting point is automatically calculated: the harvester is laterally shifted by a fixed row distance from the endpoint, while maintaining the heading. After automatically navigating to the predicted starting point, a tactile search is performed. Proceed at low speed while making small, alternating turns to the left and right, monitoring the contact status of the left and right tentacles, and searching for a stable contact position on both sides; the search distance does not exceed a preset threshold and the time does not exceed 10 seconds. If a stable connection is successfully found, the system will enter normal fusion navigation mode; if the search fails, an alarm will be triggered to prompt manual intervention.

8. The automatic row alignment method for intercropped wheat harvesters as described in claim 1, characterized in that, When the statistical prior for sowing operations and the tactile correction information continue to conflict, it indicates that the statistical prior itself needs to be updated. The specific steps are as follows: (1) Conflict detection The average slope deviation up to the present is used as the conflict evaluation index. A conflict is considered persistent when the following conditions are met simultaneously: A. The average slope deviation exceeds the threshold, i.e. β is a preset coefficient; This represents the average slope deviation from the starting point of the trip to the current position. The standard deviation of the slope error at the handover point; B. Within the preset window length, the tactile correction amount The proportion of identical numbers exceeds a preset threshold; C. The conflicting states A and B occur repeatedly in the same position interval across multiple consecutive rows; (2) Correction execution Once persistent conflict is detected, the accumulated [data / information] is utilized. For the data points, the slope difference is estimated using linear regression through the origin: This represents the distance traveled from the starting point of the current row to the i-th sampling point. This refers to the tactile lateral deviation at the corresponding sampling point; Before performing the origin return, the controller automatically resets the tactile deviation at the line start point to zero, thus ensuring... ; Update reference slope: K new The updated slope; K ref For reference slope; Simultaneously update the statistical error: in The slope difference estimated from the most recent N rows. i The value ranges from 1 to N, and std is the sample standard deviation.

9. The automatic row alignment method for intercropped wheat harvesters as described in claim 1, characterized in that, The path tracking control is performed using a speed-adaptive pure tracking algorithm, as detailed below: Pre-aiming distance Dynamically adjust based on speed: in As the lower limit of speed, Minimum aiming distance; v For real-time speed; Derive the target front wheel steering angle from geometric relationships: in For the heading deviation, in the formula: For heading angle; This refers to the wheelbase.

10. An automatic row alignment system for intercropped wheat harvesters, characterized in that, include: The sowing prior information acquisition module is configured to acquire sowing prior information, including sowing machine structure prior, sowing operation statistical prior, and sowing trajectory geometric prior; The tactile correction amount acquisition module is configured to acquire tactile correction amounts; The confidence calculation module is configured to calculate the haptic correction confidence, statistical prior confidence, and geometric prior confidence. The navigation path generation module is configured to form a progressive fusion decision logic based on tactile correction information, seeding prior information, tactile correction confidence, statistical prior confidence, and geometric prior confidence. This logic involves determining the parallel benchmark using structural priors, quantifying the handover line error using statistical priors, constraining the trajectory shape using geometric priors, and compensating for local deviations using tactile corrections. Finally, a navigation path is generated.

Images