A rhizoma alismatis harvesting inter-row rectification system

Abstract

The application discloses a reed harvesting inter-row deviation rectification system and relates to the technical field of agricultural machinery automation. The system comprises a sensing unit, a fusion unit and an execution unit. The sensing unit is used for collecting relative position information of a reed harvesting machine and a reed crop row and attitude information of the reed harvesting machine. The fusion unit is used for fusing the relative position information and the attitude information to obtain a comprehensive lateral deviation and a comprehensive heading deviation. The rectification unit is used for obtaining a turning radius based on the comprehensive lateral deviation and the comprehensive heading deviation, and obtaining a target speed difference of left and right side tracks of the reed harvesting machine based on the turning radius. The execution unit is used for controlling the rotational speed and the heading of the left and right side tracks of the reed harvesting machine based on the target speed difference, and controlling the position of a cutting device of the reed harvesting machine based on the comprehensive lateral deviation. The reed harvesting system can adapt to the reed harvesting operation environment, and the detection accuracy of the relative position of the reed harvesting machine and the crop row is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of agricultural machinery automation, in particular to a alisma orientale harvesting inter-row deviation rectification system. BACKGROUND

[0002] Alisma orientale is an aquatic Chinese medicinal material, and the soil moisture content in the field during harvesting is as high as 50%-60%. Therefore, a crawler-type harvester is often used to avoid the machine sinking and affecting the operation quality. The wider the crawler width, the smaller the ground pressure of the machine, and the less likely the machine is to sink into the mud. However, alisma orientale is a shallow-rooted medicinal material, with a growth depth of less than 10 cm and a row spacing of about 30 cm. If the crawler is too wide, the alisma orientale roots on both sides are easily crushed during walking, resulting in an increase in the damage rate of medicinal materials and causing economic losses. At the same time, if the harvester deviates laterally during walking, the plow of the harvester will deviate from the track, causing large-area injury or omission of alisma orientale root blocks, seriously damaging the integrity and quality of the medicinal materials, affecting the selling price, and causing great economic losses to the farmers.

[0003] At present, the traditional alisma orientale harvesting machine mainly relies on the experience of the driver and the single GPS or visual agricultural navigation technology, but the driver manually operates and rectifies the deviation, which is labor-intensive and difficult to ensure the accuracy under high-intensity operation. The single GPS positioning accuracy is easily disturbed in the farmland environment and cannot provide the direct deviation amount relative to the crop row. The single visual navigation has insufficient stability and reliability in the complex environment such as light change, water reflection, and soil covering in the muddy paddy field. SUMMARY

[0004] In order to solve the problem that the current alisma orientale harvesting system is difficult to adapt to the alisma orientale harvesting environment, resulting in low detection accuracy of the relative position of the harvester and the crop row, the present application provides an alisma orientale harvesting inter-row deviation rectification system, which comprises:

[0005] a perception unit for collecting the relative position information of the harvester and the alisma orientale crop row and the attitude information of the harvester;

[0006] a fusion unit for fusing the relative position information and the attitude information to obtain a comprehensive lateral deviation and a comprehensive heading deviation;

[0007] a deviation rectification unit for obtaining a turning radius based on the comprehensive lateral deviation and the comprehensive heading deviation, and obtaining a target speed difference of the left and right side crawlers of the harvester based on the turning radius;

[0008] an execution unit for controlling the rotation speed and the heading of the left and right side crawlers of the harvester based on the target speed difference, and controlling the position of the cutting device of the harvester based on the comprehensive lateral deviation.

[0009] The alisma plant row refers to a strip-shaped planting unit formed after alisma plants are planted in rows and columns in an alisma planting field according to a certain spacing and rule for the purpose of convenient management and optimal growth effect.

[0010] The system perceives the relative position information and the body attitude information of the harvester and the alisma plant row in real time through the fusion of visual (relative position), RTK-GPS (absolute position and heading), and IMU (high-frequency attitude) information, overcomes the limitations of a single sensor, and significantly improves the accuracy and anti-interference ability of inter-row detection in a complex paddy field environment; and through a closed-loop control algorithm, dynamically adjusts the speed difference of the two side tracks to correct the heading, reduces the track pressure on the two sides of the medicinal materials, and introduces a transverse active adjustment mechanism of the cutting device to minimize the root and stem injury rate caused by deviation; realizes intelligent control of the positions of the two side tracks and the cutting device, adapts to the changes in the traveling resistance caused by different soil conditions, realizes smooth and smooth correction action, reduces mechanical impact and additional soil disturbance caused by sharp turns, finally realizes accurate row-by-row operation of the harvester in narrow-row crops, effectively reduces the pressure and injury of medicinal materials, and realizes automatic operation, reduces the dependence on the operation experience of the driver, and improves the consistency of operation efficiency and quality.

[0011] Further, the perception unit specifically includes:

[0012] The machine vision unit is used to acquire a first image of the unharvested alisma plant row in front of the harvester, and obtain a crop row center line and alisma plant row features based on the first image, the alisma plant row features including spectral features, morphological features, and topological features;

[0013] The positioning unit is used to obtain the position coordinates and the heading angle of the harvester;

[0014] The inertial measurement unit is used to obtain three-dimensional acceleration and angular velocity of the harvester, and obtain attitude angle information based on the three-dimensional acceleration and the angular velocity, the attitude angle information including roll angle, pitch angle, and yaw angle;

[0015] The information unit is used to obtain the relative position information and the attitude information based on the crop row center line, the position coordinates, the heading angle, and the attitude angle information.

[0016] Further, the fusion unit specifically includes:

[0017] The visual deviation module is used to obtain a visual navigation reference based on the alisma plant row features, and obtain a relative deviation based on the crop row center line and the visual navigation reference, the relative deviation including a visual lateral deviation and a visual heading deviation;

[0018] The pose estimation module is configured to obtain a state prediction equation based on the three-dimensional acceleration and the angular velocity, obtain an observation value based on the position coordinates and the heading angle, and obtain a global pose based on the state prediction equation and the observation value.

[0019] The virtual path module is configured to obtain a plurality of continuous images based on the first image, obtain a reference center line of a plurality of crop rows based on the continuous images, obtain a virtual reference path based on the reference center line and the crop row center line, and obtain a virtual straight line equation of the virtual reference path.

[0020] The data fusion module is configured to construct a state vector and a measurement vector based on the virtual straight line equation, the relative deviation, and the global pose, obtain an adaptive weight matrix, and obtain the comprehensive lateral deviation and the comprehensive heading deviation based on the state vector, the measurement vector, and the adaptive weight matrix.

[0021] A hierarchical fusion strategy is adopted, and the IMU / GPS tight combination is first performed to solve the absolute position and high-frequency problems, and then the result is adaptively weighted and fused with vision, so that the advantages of various sensors are maximized, and the respective disadvantages are made up. In combination with the adaptive weighted Kalman filter, the relative position information of the crop row perceived by vision is used as a reference, the high-precision absolute position information is used for global correction, and the IMU data is used to compensate for high-frequency motion errors, so that a stable and reliable comprehensive navigation state quantity is finally output.

[0022] Further, the virtual path module is specifically configured to:

[0023] obtain a plurality of continuous images based on the first image, obtain a reference center line of a plurality of crop rows based on the continuous images, obtain a reference straight line equation of the reference center line, convert the reference straight line equation into a three-dimensional world coordinate system with the harvester as the origin based on a pre-trained inverse perspective transformation model, and obtain a three-dimensional point set of each of the reference center lines;

[0024] fit the reference center line based on the three-dimensional point set, and obtain a parallel line cluster;

[0025] obtain the virtual reference path based on the relative deviation, the global pose, and the parallel line cluster, obtain a straight line equation of the virtual reference path, and obtain the virtual straight line equation.

[0026] Unlike preset absolute paths, this system dynamically generates a virtual path. Based on the center lines of multiple crop rows identified from the latest few frames of images, it fits an ideal reference path parallel to the current crop row in real time. The machine tracks an ideal line that is completely parallel to the current crop row, rather than a preset straight line. This better adapts to the natural curvature that may occur during crop sowing, reducing the inherent deviation between the absolute path and the relative crop row. It also reduces systematic deviations caused by uneven sowing or cumulative errors in global positioning, improving the system's practicality and robustness in real farmland environments. Furthermore, the virtual reference path is a relative path generated in real time based on biological features (crop rows). Even if there are minute centimeter-level drifts in RTK-GPS, or cumulative errors in IMU calculations due to tire slippage, the path tracked by the system is always relative to the crop row. This fundamentally eliminates the impact of absolute positioning errors on harvest quality, ensuring that the plow is always accurately aligned with the row.

[0027] Considering the unevenness of farmland bottoms, mechanical digging at a fixed depth either fails to remove all the tubers, leaving some behind, or digging too deep increases power consumption and stirs up a large amount of bottom debris (such as stones and hard mud clods), making subsequent cleaning and sorting difficult. Traditional harvesters control depth through depth-limiting wheels or mechanical contouring mechanisms, but these are prone to sinking in soft mud, have lagging contouring, and cannot guarantee accuracy, which can also lead to digging too deep or too shallow. This system solves the problem of digging depth control caused by uneven mud bottoms by using a multi-sensor information fusion system to perceive the relative position of the digging components to the mud bottom in real time and intelligently adjust the digging depth of the water plantain tubers.

[0028] Furthermore, the system also includes:

[0029] Depth Unit: Used to obliquely install a millimeter-wave radar on the front frame of the harvester based on a preset forward tilt angle. The beam of the millimeter-wave radar is directed towards the mud bottom area to be excavated in front of the harvester. A laser rangefinder is installed at an adjacent position of the millimeter-wave radar. The millimeter-wave radar has a built-in tilt sensor. Based on the millimeter-wave radar, the laser rangefinder, and the tilt sensor, the relative height between the digging device of the harvester and the mud bottom surface is obtained.

[0030] Positioning unit: used to install encoders to various rotating hinge points of the lifting device of the harvester, and to obtain the absolute height of the digging device and the frame of the harvester based on the encoders;

[0031] Control unit: Used to construct a fuzzy rule table, and obtain control parameters based on the fuzzy rule table, the relative height, and the absolute height;

[0032] Excavation unit: Used to drive the excavation device to excavate Alisma plantago-aquatica based on the control parameters.

[0033] The relative height is used as a feedforward control quantity to predict the terrain in front, and the absolute height is used as a feedback control quantity to represent the current actual position. The ranging is performed by a non-contact millimeter wave radar, and the feedforward-feedback compound control is used to perceive the terrain change in advance, so that the response is rapid, the control precision is high, the missed turnip roots are greatly reduced, and the over-digging is reduced.

[0034] Further, the depth unit is specifically used for:

[0035] The first slant range and the confidence score are obtained based on the millimeter wave radar, the second slant range and the signal-to-noise ratio are obtained based on the laser ranging sensor, and the radar attitude data of the millimeter wave radar is obtained based on the inclination sensor;

[0036] If the signal-to-noise ratio is greater than a first threshold value, a measurement deviation is obtained based on the first slant range and the second slant range, a persistent deviation is obtained based on the measurement deviation, the first slant range is compensated based on the persistent deviation to obtain a first calibrated slant range, a third slant range is obtained based on a first preset weight and the first calibrated slant range, and a first distance is obtained based on the third slant range;

[0037] If the signal-to-noise ratio is less than a second threshold value, the first distance is obtained based on the first slant range;

[0038] If the signal-to-noise ratio is greater than or equal to the second threshold value and less than or equal to the first threshold value, a second preset weight is obtained based on the signal-to-noise ratio, and the first distance is obtained based on the second preset weight and the first calibrated slant range;

[0039] The first distance is compensated based on the radar attitude data to obtain the relative height;

[0040] A first calculation formula of the relative height is:

[0041] ;

[0042] Wherein, 1 represents the relative height, represents the first distance, represents the installation inclination angle of the millimeter wave radar, represents the real-time change angle of the millimeter wave radar.

[0043] The millimeter wave radar has strong penetration and can effectively penetrate the mud and water mist raised during harvesting, sparse surface weeds and directly detect the real mud bottom surface without being limited by visible optical conditions. The FMCW (frequency-modulated continuous wave) type can also provide distance resolution and filter out false targets such as weeds near the ground. The radar is installed at a certain angle of forward inclination on the frame in front of the excavating shovel and above the mud surface, and the beam center is directed to the mud bottom at a certain distance in front of the excavating shovel operation track (pre-aiming point), which can provide terrain prediction time and solve the problem of system control lag. Combined with the laser ranging sensor, the laser has extremely high precision and resolution in a short distance, and is installed close to the millimeter wave radar with the beam pointing to the same pre-aiming point area. In non-turbid working conditions, the laser sensor provides ultra-high precision reference data, and through data fusion algorithm, the readings of the millimeter wave radar and the laser sensor are cross-checked. When the laser data is invalid due to extreme turbidity, the millimeter wave radar data is automatically trusted. When the environment is good, the laser data can be used to calibrate the millimeter wave radar online, compensate for its temperature drift and other errors, and realize the complementary advantages. The inclination sensor is integrated with the radar, which can monitor the pitch angle and roll angle changes of the radar sensor in real time. The real-time attitude data of the inclination sensor is used to compensate the slant range measured by the radar, and the real geometric vertical height between the radar and the mud bottom detection point in any shaking state is calculated. The three combined detection of the relative height of the excavating shovel blade and the mud bottom surface to be excavated solves the industry problem of accurate ranging of soft and irregular mud bottom surface in turbid mud water, weed shielding and machine vibration and other harsh working conditions.

[0044] Based on the laser signal-to-noise ratio, a quantifiable physical index directly reflecting the environmental state, a finite state machine including three states is constructed to realize smooth and intelligent transition of the control strategy. The dynamic weight distribution of smooth transition reduces the dramatic jump of output value near the critical point of the environment, ensuring the control stability of the system. The high-precision laser sensor is used as a ruler to calibrate the system error of the millimeter wave radar online and in real time when the working condition is good, solving the drift problem of the millimeter wave radar in long-term use, reducing the harsh requirements on the long-term absolute accuracy of the sensor itself, and improving the long-term reliability of the system.

[0045] Further, the position unit is specifically configured to:

[0046] Based on the encoder, a measured hinge point angle value between two adjacent rigid bodies is obtained;

[0047] A three-dimensional kinematic model of the harvester is constructed, and a first length between the rotary hinge points is obtained based on the three-dimensional kinematic model;

[0048] Based on the measured hinge point angle value, the first length and the three-dimensional kinematic model, the theoretical height is obtained.

[0049] installing an inertial measurement device to a frame of the harvester, obtaining a frame pitch angle of the frame based on the inertial measurement device, and obtaining the absolute height based on the frame pitch angle and the theoretical height;

[0050] A second calculation formula for obtaining the absolute height is:

[0051]

[0052] wherein, 2 represents the absolute height, represents an initial height of an origin of a machine coordinate system, represents the frame pitch angle, represents the theoretical height.

[0053] The precise kinematic model of the excavating mechanism is established, the super-high-precision absolute height of the excavating blade edge relative to the dynamic virtual reference surface of the frame is solved through forward kinematics, the problems of indirect measurement and large error of the traditional single sensor are overcome, the real-time compensation mechanism of the sensor is introduced, the deformation of the frame, a major error source ignored, is compensated, and the measurement result truly reflects the height relationship between the blade and the ground level.

[0054] Further, the position unit is further configured to:

[0055] installing a calibration sensor to a piston rod of a hydraulic cylinder of the harvester, and obtaining a hydraulic cylinder stroke value based on the calibration sensor;

[0056] obtaining a predicted hinge point angle value between two adjacent rigid bodies based on the hydraulic cylinder stroke value and an inverse kinematics algorithm;

[0057] obtaining a difference value between the measured hinge point angle value and the predicted hinge point angle value, and obtaining a warning result based on the difference value and a tolerance range.

[0058] Further, the control unit is specifically configured to:

[0059] obtaining a target height based on a preset height, the relative height and the absolute height;

[0060] The fuzzy rule table includes a plurality of fuzzy rules, each of the fuzzy rules corresponds to a fuzzy set, each of the fuzzy sets includes a plurality of language variables, and each of the language variables corresponds to a single-point value;

[0061] obtaining an error and an error rate based on the target height and the absolute height, converting the error and the error rate into a fuzzy language value, obtaining an activated rule based on the fuzzy language value and the fuzzy rule table, and obtaining an activated fuzzy set based on the activated rule.​

[0062] respectively, obtaining an activation strength of the active rule based on the first membership degree and the second membership degree, and obtaining an effective output value of the active rule based on the activation strength;

[0063] determining whether the activation strength is greater than a strength threshold value, if yes, obtaining a correction value of the language variable by weighted average of the effective output value and the activation strength, and if no, obtaining the correction value of the language variable based on a default value;

[0064] obtaining the control parameter based on the correction value and an initial parameter;

[0065] a third calculation formula of the target height is:

[0066] ;

[0067] wherein, represents the target height, represents a preset height.

[0068] A feedforward-feedback compound fuzzy PID control architecture is adopted, H1 (feedforward) provides foresight, so that the system can respond to terrain changes in advance, greatly reduces the system lag, and converts the target height into a target height trajectory that dynamically changes in the absolute coordinate system and follows the terrain; the fuzzy PID enables the controller parameters to be dynamically adjusted according to the current error state, quickly responds when the error is large, and finely adjusts when approaching the target, prevents overshoot oscillation, and perfectly adapts to nonlinear factors such as silt resistance changes; at the same time, a single-point output fuzzy set and a weighted average method are introduced, which simplifies the complex area calculation to extremely simple multiplication operation, so that the speed of the defuzzification process is extremely fast; an activation threshold and a default output mechanism are also introduced, so that the controller will not produce abnormal output when encountering unforeseen input combinations or slight sensor abnormalities, effectively solving the problem of unstable output or even meaningless control instructions of the traditional weighted average method in the case of low activation, greatly enhancing the robustness of the system in unknown or uncertain working conditions.

[0069] Further, the excavating unit is specifically configured to:

[0070] map the control parameter to a target PWM duty ratio;

[0071] construct a duty ratio-target current curve table, and obtain a target current value based on the duty ratio-target current curve table and the target PWM duty ratio;

[0072] An actual current value of the excavating device is acquired, a current error is obtained based on the target current value and the actual current value, the target PWM duty cycle is adjusted based on the current error to obtain a corrected PWM signal;

[0073] The corrected PWM signal is superimposed with a chattering signal to obtain a PWM command;

[0074] The excavating device is driven to excavate the reed based on the PWM command.

[0075] The curve is used for compensating the dead zone and nonlinearity of the proportional valve, the current negative feedback loop is introduced, the interference of factors such as coil resistance change with temperature, power supply voltage fluctuation, electromagnetic back electromotive force and the like is effectively overcome, and the high linearity and good repeatability between the force / flow supplied to the proportional valve and the control command are ensured; the chattering signal is superimposed, the static friction of the valve core is effectively overcome, the hysteresis phenomenon of the proportional valve is significantly reduced, the control is more linear, and the problems of poor switching control precision of the traditional electromagnetic valve and the slow action, shaking or inaccurate positioning of the excavating shovel caused by the influence of oil temperature and load on the control of the ordinary proportional valve are solved.

[0076] The one or more technical solutions provided by the present application have at least the following technical effects or advantages:

[0077] 1. The system overcomes the limitations of a single sensor by fusing visual (relative position), RTK-GPS (absolute position and heading) and IMU (high-frequency attitude) information through multi-sensor information fusion technology, and the perception unit perceives the relative position information and the body attitude information of the harvester and the reed crop row in real time, significantly improves the accuracy and anti-interference ability of inter-row detection in complex paddy field environment, and dynamically adjusts the speed difference of the two side tracks to correct the heading through a closed-loop control algorithm, reduces the pressure injury of the two side tracks, and introduces a cutting device transverse active adjustment mechanism to minimize the root and stem injury rate caused by deviation; the positions of the two side tracks and the cutting device are intelligently controlled, the changes in the travel resistance caused by different soil conditions are adapted, smooth and smooth correction actions are realized, mechanical impact and additional soil disturbance caused by sharp turns are reduced, and finally the harvester realizes accurate row operation between narrow-row crops, effectively reduces the pressure and cutting of medicinal materials, realizes automatic operation, reduces the dependence on the operation experience of the driver, and improves the consistency of operation efficiency and quality.

[0078] 2. Unlike preset absolute paths, this system dynamically generates a virtual path. Based on the center lines of multiple crop rows identified from the latest few frames of images, it fits an ideal reference path parallel to the current crop row in real time. The machine tracks an ideal line that is completely parallel to the current crop row, rather than a preset straight line. This better adapts to the natural curvature that may occur during crop sowing, reducing the inherent deviation between the absolute path and the relative crop row. It also reduces systematic deviations caused by uneven sowing or cumulative errors in global positioning, improving the system's practicality and robustness in real farmland environments. Furthermore, the virtual reference path is a relative path generated in real time based on biological features (crop rows). Even if there are minute centimeter-level drifts in RTK-GPS, or cumulative errors in IMU calculations due to tire slippage, the path tracked by the system is always relative to the crop row. This fundamentally eliminates the impact of absolute positioning errors on harvest quality, ensuring that the plow is always accurately aligned with the row.

[0079] 3. This system uses relative height as the feedforward control variable to predict the terrain undulations ahead, and absolute height as the feedback control variable to represent the current actual position. It uses non-contact millimeter-wave radar for ranging, combined with feedforward-feedback composite control, to detect terrain changes in advance, respond quickly, and control with high precision, greatly reducing the omission of Alisma tubers and over-excavation. It adopts a fuzzy PID algorithm, which has strong robustness to disturbances such as changes in soil resistance and machine vibration, and is suitable for complex silt environments. It effectively reduces the digging up of bottom impurities (stones, hard mud), reduces the workload of subsequent cleaning and sorting, and improves the quality and purity of Alisma products.

[0080] 4. Based on the laser signal-to-noise ratio, a quantifiable physical indicator that directly reflects the environmental state, a finite state machine with three states was constructed to achieve smooth and intelligent transition of the control strategy. A smooth transition dynamic weight allocation was adopted to reduce drastic jumps in output values ​​near environmental critical points, ensuring the system's control stability. A high-precision laser sensor, used as a benchmark under good operating conditions, was used to calibrate the millimeter-wave radar's system error online and in real time, solving the drift problem of millimeter-wave radar during long-term use, reducing the stringent requirements for the sensor's long-term absolute accuracy, and improving the system's long-term reliability.

[0081] 5. A precise kinematic model of the excavating mechanism is established, and the ultra-high precision absolute height of the excavator blade relative to the dynamic virtual reference plane of the frame is calculated through forward kinematics. This overcomes the problems of indirect measurement and large error of traditional single sensor measurement. Furthermore, a real-time compensation mechanism for the sensor is introduced to compensate for the significant error source of frame deformation, which has been neglected, so that the measurement results truly reflect the height relationship between the blade and the ground level.

[0082] 6. A feedforward-feedback compound fuzzy PID control architecture is adopted, H1 (feedforward) provides foresight, so that the system can respond to terrain changes in advance, greatly reducing system lag, and converting the target height into a dynamically changing target height trajectory in the absolute coordinate system following the terrain; fuzzy PID enables the controller parameters to be dynamically adjusted according to the current error state, quickly responding when the error is large, and finely adjusting when approaching the target to prevent overshoot oscillation, perfectly adapting to nonlinear factors such as silt resistance changes; At the same time, single-point output fuzzy sets and weighted average method are introduced, which simplifies the complex area calculation to extremely simple multiplication operation, making the speed of the defuzzification process extremely fast; The activation threshold and default output mechanism are also introduced, so that the controller will not produce abnormal output when encountering unforeseen input combinations or slight sensor abnormalities, effectively solving the problem of unstable output or even meaningless control instructions of the traditional weighted average method in low activation conditions, greatly enhancing the robustness of the system in unknown or uncertain working conditions. BRIEF DESCRIPTION OF DRAWINGS

[0083] The accompanying drawings, which are included to provide a further understanding of the embodiments of the application and are incorporated in and constitute a part of this application, illustrate embodiments of the application and together with the description serve to explain the principles of the application.

[0084] Figure 1 is a process schematic diagram of an alisma harvesting inter-row deviation correction system in the application. DETAILED DESCRIPTION

[0085] In order to more clearly understand the above-mentioned purposes, features and advantages of the present application, the present application will be further described in detail below in combination with the drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0086] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application, however, the present application can also be implemented in other ways different from the scope described herein, therefore, the scope of protection of the present application is not limited by the specific embodiments disclosed below.

[0087] Embodiment one

[0088] Reference Figure 1 The present embodiment provides an alisma harvesting inter-row deviation correction system, which comprises:

[0089] A perception unit is configured to collect relative position information of the harvester and the alisma crop row and attitude information of the harvester;

[0090] A fusion unit is configured to fuse the relative position information and the attitude information to obtain a comprehensive lateral deviation and a comprehensive heading deviation;

[0091] The correction unit is used to obtain a steering radius based on the comprehensive lateral deviation and the comprehensive heading deviation, and obtain a target speed difference of the left and right side tracks of the harvester based on the steering radius; a fuzzy PID control or a pure pursuit model algorithm is used to take the comprehensive lateral deviation and the heading deviation as inputs, calculate a theoretical steering radius required to eliminate the deviation, and then calculate the target speed difference of the left and right side tracks.

[0092] The execution unit is used to control the rotation speed and heading of the left and right side tracks of the harvester based on the target speed difference, and control the position of the cutting device of the harvester based on the comprehensive lateral deviation.

[0093] If the left and right side tracks are controlled by a track driving control unit, the track driving control unit includes two independent electro-hydraulic proportional valves or servo motors, which are used to accurately control the rotation speed of the hydraulic motors or motors of the left and right tracks respectively, so as to realize differential steering and adjust the heading of the machine body. After receiving the target speed difference, the track driving control unit compares the target speed difference with the actual rotation speed value fed back by an encoder or a rotation speed sensor, calculates an error signal, processes the error signal through a PID control algorithm, generates an analog voltage signal (such as 0-10V) or a pulse width modulation (PWM) signal from the error signal, and transmits the analog voltage signal or the pulse width modulation (PWM) signal to an electro-hydraulic proportional valve (if the driving is hydraulic) or a servo motor driver (if the driving is electric). The electro-hydraulic proportional valve adjusts the flow to the hydraulic motor in proportion to the input voltage, so as to accurately control the rotation speed of the motor.

[0094] If the cutting device is a plow, a plow controller (such as a stepper motor driver or a small servo driver) is arranged on the plow, which can make the plow finely adjust (such as ±5cm) in the lateral direction, and move compensatorily according to the calculated comprehensive lateral deviation, so as to ensure that the plow always aligns with the center of the crop row even if the machine body has a slight deviation. The plow controller converts the comprehensive lateral deviation into the number of pulses required for the stepper motor or servo motor to rotate according to the mechanical transmission parameters (such as the lead of the screw rod). After receiving the driving pulses, the stepper motor or servo motor starts to rotate. The rotation of the motor is transmitted to a set of precision ball screw pairs (or other mechanisms that convert rotary motion into linear motion, such as gear and rack) through a shaft coupling. The nut of the screw rod drives the slide or bracket fixedly connected with the plow to make accurate linear motion in the lateral direction on the linear guide rail, so as to realize real-time and high-precision lateral compensation of the position of the plow head, and ensure that the plow always aligns with the center of the crop row, regardless of whether the machine body is performing steering correction.

[0095] The perception unit specifically includes:

[0096] Machine vision unit: for collecting a first image of an unharvested alisma plant row in front of the harvester, obtaining a plant row center line and an alisma plant row feature based on the first image, the alisma plant row feature including spectral features, morphological features and topological features; such as mounting a front camera on the front end crossbeam of the harvester to collect image information of the unharvested alisma row in front, identifying the plant row center line through image processing algorithm, using image segmentation algorithm based on color features (green vegetation and mud water background) and morphological processing to extract alisma plant row features.

[0097] Positioning unit: for installing an existing RTK-GPS on the top of the harvester to obtain the position coordinates and heading angle of the harvester;

[0098] Inertial measurement unit: for installing an inertial measurement unit (IMU) near the center of gravity of the harvester to obtain three-dimensional acceleration and angular velocity of the harvester, obtaining attitude angle information based on the three-dimensional acceleration and the angular velocity, the attitude angle information including roll angle, pitch angle and yaw angle;

[0099] Information unit: for obtaining the relative position information and the attitude information based on the plant row center line, the position coordinates, the heading angle and the attitude angle information.

[0100] Among them, the fusion unit specifically includes:

[0101] Visual deviation module: for obtaining a visual navigation reference based on the alisma plant row feature, obtaining a relative deviation based on the plant row center line and the visual navigation reference, the relative deviation including visual lateral deviation (i.e. the lateral distance between the current center line of the machine and the center line of the plant row) and visual heading deviation (i.e. the angle difference between the current driving direction of the machine and the ideal navigation reference direction calculated based on the visual sensor); such as using Hough Transform or least squares method to fit the visual navigation reference line in the current image and calculating a visual lateral deviation and a visual heading deviation.

[0102] Pose estimation module: for obtaining a state prediction equation based on the three-dimensional acceleration and the angular velocity, obtaining an observation value based on the position coordinates and the heading angle, obtaining a global pose based on the state prediction equation and the observation value; such as using Kalman filter to tightly combine the position coordinates, the heading angle, the three-dimensional acceleration and the angular velocity: using the three-dimensional acceleration and the angular velocity as the state prediction equation, and using the position coordinates and the heading angle as the observation value for updating, which can smooth the jump noise of RTK-GPS, and use IMU for short-time high-precision calculation when GPS signal is temporarily locked (such as passing through a shelter), finally output a high-frequency, smooth fused global pose.

[0103] a virtual path module configured to: obtain a plurality of continuous images based on the first image, obtain a plurality of reference center lines of the plurality of Alisma orientale crop rows based on the continuous images, and obtain a virtual reference path based on the reference center lines and the crop row center lines, and obtain a virtual straight line equation of the virtual reference path;

[0104] a data fusion module configured to: construct a state vector and a measurement vector based on the virtual straight line equation, the relative deviation, and the global pose, obtain an adaptive weight matrix based on the state vector, the measurement vector, and the adaptive weight matrix, and obtain the integrated lateral deviation and the integrated heading deviation based on the state vector, the measurement vector, and the adaptive weight matrix; wherein the global lateral deviation is calculated according to the virtual reference path and the global pose, the straight line equation of the virtual reference path is converted to the global geodetic coordinate system based on the global pose, the straight line equation is ( denotes a slope, denotes a slope distance), the current pose of the machine is regarded as a point , , the distance between the point and the straight line equation is calculated , the global lateral deviation is obtained based on the distance and the straight line equation, and the distance of the formula is signed, if , the machine deviates to the right and needs to be corrected to the left, if , the machine deviates to the left and needs to be corrected to the right, and if it is 0, no correction is needed.

[0105] a Kalman filter or an extended Kalman filter algorithm is used to construct a state vector based on the integrated lateral deviation, a differential (change rate) of the integrated lateral deviation, and the integrated heading deviation, construct a state transition equation based on the state vector, construct a measurement vector based on the relative deviation, the global pose, and the global lateral deviation, construct an observation equation based on the measurement vector, obtain an existing adaptive weight matrix, and based on the Kalman filter recursive algorithm, at each time, perform the prediction and update steps of the standard Kalman filter to finally obtain the optimal estimated integrated lateral deviation and integrated heading deviation.

[0106] The virtual path module is specifically configured to:

[0107] obtain a plurality of continuous images based on the first image, obtain a plurality of reference center lines of the plurality of Alisma orientale crop rows based on the continuous images, and obtain a virtual reference path based on the reference center lines and the crop row center lines, and obtain a virtual straight line equation of the virtual reference path; such as using a computer vision algorithm (such as a semantic segmentation model based on deep learning or an optimized traditional image processing process) to simultaneously recognize a plurality of complete Alisma orientale crop rows in the current field of view, for each recognized crop row, a straight line equation of the center line thereof is fitted using a least squares method;

[0108] Based on the pre-trained inverse perspective transformation model, the reference straight line equation is converted into a three-dimensional world coordinate system with the harvester as the origin, and a three-dimensional point set of each reference center line is obtained; an inverse perspective transformation model of a pre-calibrated image pixel coordinate system to a harvester body coordinate system is established, the model is applied to each center line equation, and the center line equation is converted from a two-dimensional image coordinate to a three-dimensional world coordinate (a body coordinate system) with the harvester as the origin. After conversion, each crop row center line in the body coordinate system can be expressed as a set of points, that is, a three-dimensional point set.

[0109] The reference center line is fitted based on the three-dimensional point set, and a parallel line cluster is obtained; based on the three-dimensional point set, a least squares method or a random sample consensus (RANSAC) algorithm is used to fit all the crop row center lines as a whole, and a parallel line cluster is established. The parallel line cluster represents a geometric model of the crop row, and is used to estimate the overall trend (that is, a direction vector) and an average row spacing of all the crop rows in the current working area.

[0110] Based on the relative deviation, the global pose and the parallel line cluster, the virtual reference path is obtained, a straight line equation of the virtual reference path is obtained, and the virtual straight line equation is obtained. According to the current position of the harvester (from the RTK-GPS / IMU fusion result) and the agricultural requirements (for example, the requirement that the plow is aligned with the middle of two crop rows), a virtual reference path is dynamically specified from the parallel line cluster, for example, if it is determined that the machine is currently working between the first row and the second row, the virtual reference path is set to be a straight line that is parallel to the center lines of the two rows and is located exactly in the middle of the two rows, and a straight line equation of the virtual reference path in the body coordinate system is determined.

[0111] Embodiment Two

[0112] In this embodiment, the system further includes:

[0113] The depth unit is used for installing a millimeter wave radar on the front frame of the harvester at a preset front inclination angle, the beam of the millimeter wave radar points to a mud bottom area to be excavated in front of the harvester (that is, a pre-look point), a laser ranging sensor is installed at a position adjacent to the millimeter wave radar, an inclination sensor (such as an inertial measurement unit, which is internally provided with a three-axis gyroscope and a three-axis accelerometer and can be directly installed inside the shell of the millimeter wave radar sensor) is built in the millimeter wave radar, and based on the millimeter wave radar, the laser ranging sensor and the inclination sensor, the relative height of the excavating device of the harvester and the mud bottom surface is obtained. The millimeter wave radar has strong penetration and is not affected by water vapor and splashing mud, can accurately measure the slant distance between the sensor and the mud bottom point, and can calculate the vertical height through the inclination sensor data.

[0114] A position unit is configured to install an encoder (such as an existing multi-turn absolute value encoder) to each rotary articulation point (such as the articulation point between the boom and the frame, the articulation point between the arm and the boom) of the lifting device of the harvester, and obtain the absolute height of the excavating device and the frame of the harvester based on the encoder; for example, the encoder is installed at the lifting hydraulic cylinder of the excavating bucket or the articulation point of the lifting linkage, and the stroke of the cylinder or the angle of the articulation point is directly measured to convert the current absolute height of the cutting edge of the excavating bucket relative to the dynamic reference frame of the frame.

[0115] A control unit is configured to construct a fuzzy rule table, and obtain a control parameter based on the fuzzy rule table, the relative height and the absolute height.

[0116] An excavating unit is configured to drive the excavating device to excavate the reed based on the control parameter.

[0117] The depth unit is specifically configured to:

[0118] The first slant range and the confidence score are obtained based on the millimeter wave radar, the score can be obtained based on the target intensity provided by the internal DSP (digital signal processing technology) processing of the radar, the echo quality index, or by calculating the variance of the latest group of radar data, and a small variance indicates stable data, and thus a high confidence; the second slant range and the signal-to-noise ratio are obtained based on the laser ranging sensor, the signal-to-noise ratio is a key index for judging whether the laser sensor is working or not, and the radar attitude data of the millimeter wave radar is obtained based on the tilt sensor.

[0119] If the signal-to-noise ratio is greater than a first threshold value, it indicates that the environment is in an excellent state, the water body is clear, there is no obstruction, and the laser sensor is working in an optimal state, then the measurement deviation is obtained based on the first slant range and the second slant range, the measurement deviation = the first slant range - the second slant range, the continuous deviation is obtained based on the measurement deviation, for example, the measurement deviation is input into an existing first-order low-pass filter or a sliding average window to form a smooth and continuous deviation estimate, the first calibrated slant range is obtained by compensating the first slant range based on the continuous deviation, the first calibrated slant range = the first slant range - the continuous deviation, the third slant range is obtained based on the first preset weight and the first calibrated slant range, the third slant range = the first preset weight * the second slant range + (1 - the first preset weight) * the first calibrated slant range, and the first distance is obtained based on the third slant range.

[0120] If the signal-to-noise ratio is less than a second threshold value, it indicates that the environment is in a poor state, the water body is extremely turbid, and the laser is completely ineffective or the data is not reliable, then the first distance is obtained based on the first slant range, and the output of the millimeter wave radar is used as the reference;

[0121] If the signal-to-noise ratio is greater than or equal to the second threshold value and less than or equal to the first threshold value, it indicates that the environment is in an environment transition state, the water body is slightly turbid, the laser data quality is reduced but not completely invalid, and then a second preset weight is obtained based on the signal-to-noise ratio, for example, the second preset weight is linearly interpolated or exponentially attenuated between 0 and 0.9, the worse the laser data quality, the lower the weight, and the weight of the millimeter wave radar is correspondingly increased, and the first distance is obtained based on the second preset weight and the first calibrated slant range, at this time, the first distance = second preset weight * second slant range + (1-second preset weight) * first calibrated slant range;

[0122] The first distance is compensated based on the radar attitude data to obtain the relative height;

[0123] The first calculation formula for obtaining the relative height is:

[0124] ;

[0125] Wherein, 1 represents the relative height, represents the first distance, represents the installation inclination angle of the millimeter wave radar, represents the real-time pitch change angle of the millimeter wave radar.

[0126] The position unit is specifically configured to:

[0127] Based on the encoder, a measured hinge point angle value between two adjacent rigid bodies is obtained;

[0128] A three-dimensional kinematics model of the harvester is constructed, the harvester is regarded as a mechanical arm rather than a simple four-bar linkage mechanism, and a first length between the rotary hinges is obtained based on the three-dimensional kinematics model;

[0129] Based on the measured hinge point angle value, the first length and the three-dimensional kinematics model, forward kinematics calculation is performed according to geometric principles (such as the cosine theorem, D-H parameter method, etc.), and the accurate three-dimensional coordinates (X, Y, Z) of the digging blade edge relative to the mechanical coordinate system origin defined on the frame under the current attitude are directly and real-timely solved, wherein the Z coordinate is the required absolute height value, i.e. the theoretical height; this method directly obtains the spatial position of the blade edge, and reduces the cumulative error and precision loss caused by measuring the cylinder stroke and then indirectly converting through complex trigonometric functions.

[0130] An inertial measurement device is installed to the key point of the frame of the harvester, the frame pitch angle of the frame is obtained based on the inertial measurement device, and the absolute height is obtained based on the frame pitch angle and the theoretical height;

[0131] A second calculation formula for obtaining the absolute height is:

[0132]

[0133] wherein, 2 represents the absolute height, represents the initial height of the origin of the mechanical coordinate system, represents the rack pitch angle, represents the theoretical height.

[0134] Considering that the rack itself is not absolutely rigid and will be twisted and deformed when driving or excavating in rough mud, the absolute coordinate system established on the rack is actually also dynamically changing. An auxiliary IMU (inertial measurement unit) is installed at a key position of the rack (such as near the root of the excavating mechanism) to monitor the pitch and roll changes of the rack itself. Reading the pitch angle of the IMU, the final compensated absolute height of the excavating blade edge = theoretical height + the height compensation amount calculated according to the rack pitch angle and the position of the origin of the mechanical coordinate system, which ensures that the height measurement is relative to a horizontal virtual reference, not a physical reference that shakes with the vehicle body, and the precision reaches the level of industrial robots.

[0135] In this embodiment, the rigid body refers to each main component of the alisma harvesting machine excavating device, which is idealized as a solid object that does not elastically deform or bend during movement.

[0136] The control unit is specifically configured to:

[0137] obtain a target height based on the preset height, the relative height and the absolute height;

[0138] The fuzzy rule table includes a plurality of fuzzy rules, such as IF error is PB (positive big, i.e., digging shallow) AND error rate is ZO (error change rate zero), THEN proportional gain is PB (greatly increase proportional gain), integral gain is PM (moderately increase integral gain), and differential gain is NS (smallly decrease differential gain).

[0139] Each fuzzy rule corresponds to a fuzzy set, and each fuzzy set includes a plurality of language variables, and each language variable corresponds to a single-point value.

[0140] Based on the target height and the absolute height, an error and an error rate are obtained, error = target height - absolute height, and the error rate can be the difference between the current error and the last error divided by the period time.

[0141] ​​The PID parameters are dynamically set using fuzzy logic, the error and the error rate are converted into fuzzy language values, an active rule is obtained based on the fuzzy language values and the fuzzy rule table, and an active fuzzy set is obtained based on the active rule;

[0142] The first membership and the second membership of the error and the error rate in the active fuzzy set are respectively obtained by using an existing algorithm, the active strength of the active rule is obtained based on the first membership and the second membership, that is, a minimum operation is performed in the first membership and the second membership, the effective output value of the active rule is obtained based on the active strength, and the effective output value = the active strength * the single-point value;

[0143] It is judged whether the active strength is greater than a strength threshold value, if yes, it is judged that the fuzzy reasoning result of the current state is reliable, only rules with the active strength greater than the strength threshold value are aggregated, the effective output value and the active strength are weighted and averaged to obtain a correction value of the language variable, such as a proportional gain = total effective output value / total active strength, if not, it is judged that the current state is an undefined unknown state, the correction value of the language variable is obtained based on a default value, which is an intelligent decision of no action or maintaining the status quo, preventing the system from oscillating or losing stability due to random parameter adjustment in an uncertain state.

[0144] The control parameter is obtained based on the correction value and an initial parameter, the correction value is added to the initial value of the PID parameter to obtain the parameter finally used in the current control period, and then a standard PID operation is performed to generate a basic control amount and obtain the control parameter.

[0145] A third calculation formula of the target height is:

[0146] ;

[0147] wherein, the target height is represented by Htarget, the preset height is represented by Hpre. According to a geometric relationship, approximately equals the absolute height of the mud bottom at the front preview point, which is a feedforward amount, and informs the controller in advance that the front terrain will change how, and is subtracted from Therefore, the final position of the blade is lower than the mud bottom of the preview point by a target depth, and this formula ingeniously converts the relative depth target into a target height trajectory that dynamically changes in the absolute coordinate system and follows the terrain.

[0148] The excavating unit is specifically configured to:

[0149] The control parameter is mapped into a target PWM (pulse width modulation) duty cycle based on an existing algorithm;

[0150] constructing a duty cycle-target current curve table, obtaining a target current value based on the duty cycle-target current curve table and the target PWM duty cycle;

[0151] obtaining an actual current value of the excavating device, such as by a high-precision, low-drift Hall current sensor, real-time detecting an actual current value flowing through the proportional valve coil, obtaining a current error (target current value-actual current value) based on the target current value and the actual current value, adjusting the target PWM duty cycle based on the current error to obtain a modified PWM signal, such as immediately increasing the duty cycle of the output PWM to increase the current when the actual current is low;

[0152] superimposing a high-frequency, low-amplitude jitter signal on the modified PWM signal to obtain a PWM command; the jitter signal can be digitally synthesized by a software algorithm (such as a timing interrupt service program) in a micro control unit (MCU) in combination with a built-in PWM generator hardware module.

[0153] driving the excavating device to excavate the sedge based on the PWM command, such as using an electro-hydraulic proportional valve connected with a hydraulic cylinder controlling the lifting of the excavating shovel, controlling the opening of the electro-hydraulic proportional valve based on the PWM command, the opening degree of the valve core being in precise proportion to the current size, thereby precisely controlling the flow and direction of the hydraulic oil to the hydraulic cylinder, the controlled hydraulic oil driving the hydraulic cylinder piston rod to extend or retract, through a mechanical linkage mechanism, ultimately driving the excavating shovel to precisely and smoothly lift, completing the adaptive adjustment of the depth.

[0154] The position unit is further configured to:

[0155] installing a calibration sensor (such as an existing magnetostrictive linear displacement sensor) to the hydraulic cylinder piston rod of the harvester, obtaining a hydraulic cylinder stroke value based on the calibration sensor;

[0156] obtaining a predicted hinge point angle value between two adjacent rigid bodies based on the hydraulic cylinder stroke value and inverse kinematics algorithm;

[0157] obtaining a difference value of the measured hinge point angle value and the predicted hinge point angle value, obtaining a warning result based on the difference value and a tolerance range, such as determining that a sensor failure, abnormal deformation of the mechanical structure, or severe skidding may occur when the difference value exceeds the tolerance range.

[0158] Although the preferred embodiments of the present application have been described, those skilled in the art can make further changes and modifications to the embodiments once they know the basic inventive concept. Therefore, the appended claims are intended to be interpreted as including all changes and modifications falling within the scope of the present application.

[0159] It will be apparent to those skilled in the art that various modifications and variations can be made to the present application without departing from the spirit or scope of the application. Thus, it is intended that the present application cover modifications and variations of this application provided they come within the scope of the appended claims and their equivalents.

Claims

1. An in-row correction system for alisma harvesting, characterized in that, The system comprises: a perception unit configured to acquire relative position information of the harvester and the reed crop row and attitude information of the harvester; a fusion unit configured to fuse the relative position information and the attitude information to obtain a comprehensive lateral deviation and a comprehensive heading deviation; a correction unit configured to obtain a turning radius based on the comprehensive lateral deviation and the comprehensive heading deviation, and obtain a target speed difference of left and right side tracks of the harvester based on the turning radius; an execution unit configured to control rotation speeds and headings of the left and right side tracks of the harvester based on the target speed difference, and control a position of a cutting device of the harvester based on the comprehensive lateral deviation; The system further comprises: a depth unit configured to install a millimeter wave radar on a front frame of the harvester at a preset front inclination angle, a beam of the millimeter wave radar being directed to a mud bottom area to be excavated in front of the harvester, a laser ranging sensor being installed at a position adjacent to the millimeter wave radar, and an inclination sensor being built-in in the millimeter wave radar, the relative height of an excavating device of the harvester and a mud bottom surface being acquired based on the millimeter wave radar, the laser ranging sensor and the inclination sensor; a position unit configured to install an encoder to each rotating hinge point of a lifting device of the harvester, and to acquire an absolute height of the excavating device and the frame of the harvester based on the encoder; a control unit configured to construct a fuzzy rule table, and to obtain a control parameter based on the fuzzy rule table, the relative height and the absolute height; an excavating unit configured to drive the excavating device to excavate reed based on the control parameter; The depth unit is specifically configured to: acquire a first slant distance and a confidence score based on the millimeter wave radar, acquire a second slant distance and a signal-to-noise ratio based on the laser ranging sensor, and acquire radar attitude data of the millimeter wave radar based on the inclination sensor; if the signal-to-noise ratio is greater than a first threshold value, acquire a measurement deviation based on the first slant distance and the second slant distance, acquire a persistent deviation based on the measurement deviation, compensate the first slant distance based on the persistent deviation to obtain a first calibrated slant distance, acquire a third slant distance based on a first preset weight and the first calibrated slant distance, and acquire a first distance based on the third slant distance; if the signal-to-noise ratio is less than a second threshold value, acquire the first distance based on the first slant distance; if the signal-to-noise ratio is greater than or equal to the second threshold value and less than or equal to the first threshold value, acquire a second preset weight based on the signal-to-noise ratio, and acquire the first distance based on the second preset weight and the first calibrated slant distance; compensate the first distance based on the radar attitude data to obtain the relative height; a first calculation formula of the relative height is: ； wherein, 1 represents a relative height, represents a first distance, represents an installation inclination angle of the millimeter wave radar, represents a real-time pitch change angle of the millimeter wave radar.

2. A system for correcting deviation of alisma harvesting inter-row according to claim 1, characterized in that, The perception unit specifically comprises: a machine vision unit configured to acquire a first image of an unharvested reed crop row in front of the harvester, and to acquire a crop row center line and reed crop row features based on the first image, the reed crop row features including spectral features, morphological features and topological features; a positioning unit configured to acquire a position coordinate and a heading angle of the harvester; an inertial measurement unit, configured to acquire three-dimensional acceleration and angular velocity of the harvester, obtain attitude angle information based on the three-dimensional acceleration and the angular velocity, the attitude angle information comprising a roll angle, a pitch angle and a yaw angle; an information unit, configured to obtain the relative position information and the attitude information based on the crop row center line, the position coordinates, the heading angle and the attitude angle information.

3. A system for correcting deviation of a reed harvester according to claim 2, wherein The fusion unit specifically comprises: a visual deviation module, configured to obtain a visual navigation reference based on the Alisma orientale crop row features, and obtain a relative deviation based on the crop row center line and the visual navigation reference, the relative deviation comprising a visual lateral deviation and a visual heading deviation; a pose estimation module, configured to obtain a state prediction equation based on the three-dimensional acceleration and the angular velocity, obtain an observation value based on the position coordinates and the heading angle, and obtain a global pose based on the state prediction equation and the observation value; a virtual path module, configured to acquire a plurality of continuous images based on the first image, obtain reference center lines of a plurality of Alisma orientale crop rows based on the continuous images, obtain a virtual reference path based on the reference center lines and the crop row center line, and acquire a virtual straight line equation of the virtual reference path; a data fusion module, configured to construct a state vector and a measurement vector based on the virtual straight line equation, the relative deviation and the global pose, acquire an adaptive weight matrix, and obtain the comprehensive lateral deviation and the comprehensive heading deviation based on the state vector, the measurement vector and the adaptive weight matrix.

4. A system for correcting deviation of reed canary grass between rows of harvest according to claim 3, characterized in that, The virtual path module is specifically configured to: acquire a plurality of continuous images based on the first image, obtain reference center lines of a plurality of Alisma orientale crop rows based on the continuous images, acquire a reference straight line equation of the reference center lines, convert the reference straight line equation into a three-dimensional world coordinate system with the harvester as the origin based on a pre-trained inverse perspective transformation model, and obtain a three-dimensional point set of each of the reference center lines; fit the reference center lines based on the three-dimensional point set, and obtain a parallel line cluster; obtain the virtual reference path based on the relative deviation, the global pose and the parallel line cluster, acquire a straight line equation of the virtual reference path, and obtain the virtual straight line equation.

5. The in-row correction system for alisma harvesting according to claim 1, wherein, The position unit is specifically configured to: obtain a measured hinge point angle value between two adjacent rigid bodies based on the encoder; construct a three-dimensional kinematics model of the harvester, and obtain a first length between the rotary hinges based on the three-dimensional kinematics model; obtain a theoretical height based on the measured hinge point angle value, the first length and the three-dimensional kinematics model; install an inertial measurement device to a key point of a frame of the harvester, obtain a frame pitch angle of the frame based on the inertial measurement device, and obtain the absolute height based on the frame pitch angle and the theoretical height; a second calculation formula of the absolute height is: ； wherein, 2 represents an absolute height, represents an initial height of the origin of the machine coordinate system, represents a machine tilt angle, represents a theoretical height.

6. A system for correcting deviation of a reed harvester according to claim 5, wherein The position unit is further configured to: install a calibration sensor to a piston rod of a hydraulic cylinder of the harvester, and obtain a hydraulic cylinder stroke value based on the calibration sensor; obtain a presumed hinge point angle value between two adjacent rigid bodies based on the hydraulic cylinder stroke value and an inverse kinematics algorithm; A difference between the measured hinge point angle value and the inferred hinge point angle value is obtained, and based on the difference and a tolerance range, a warning result is obtained.

7. A system for correcting deviation of alisma harvesting inter-row according to claim 6, characterized in that, The control unit is specifically configured to: A preset height is obtained, and based on the preset height, the relative height, and the absolute height, a target height is obtained. The fuzzy rule table includes a plurality of fuzzy rules, each of which corresponds to a fuzzy set, each of which includes a plurality of linguistic variables, each of which corresponds to a single-point value. Based on the target height and the absolute height, an error and an error rate are obtained, the error and the error rate are converted into fuzzy language values, based on the fuzzy language values and the fuzzy rule table, an active rule is obtained, and based on the active rule, an active fuzzy set is obtained. The first membership degree and the second membership degree of the error and the error rate in the active fuzzy set are obtained respectively, the activation strength of the active rule is obtained based on the first membership degree and the second membership degree, and the effective output value of the active rule is obtained based on the activation strength. It is judged whether the activation strength is greater than a strength threshold value, if yes, the effective output value and the activation strength are weighted and averaged to obtain a correction value of the language variable; if not, the correction value of the language variable is obtained based on a default value. The control parameter is obtained based on the correction value and an initial parameter. A third calculation formula of the target height is: ； wherein, represents a target height, represents a preset height.

8. The in-row correction system for alisma harvesting according to claim 1, wherein, The mining unit is specifically configured to: The control parameter is mapped to a target PWM duty cycle; A duty cycle-target current curve table is constructed, and based on the duty cycle-target current curve table and the target PWM duty cycle, a target current value is obtained; An actual current value of the mining device is obtained, based on the target current value and the actual current value, a current error is obtained, based on the current error, the target PWM duty cycle is adjusted, and a corrected PWM signal is obtained; The corrected PWM signal is superimposed with a chattering signal to obtain a PWM command; Based on the PWM command, the mining device is driven to mine rehmannia glutinosa.

Images