Mountainous area agricultural machine beidou precision operation method based on dynamic path planning

Abstract

The application discloses a mountainous area agricultural machine Beidou precision operation method of dynamic path planning, and relates to the technical field of intelligent agricultural automation. Terrain features are extracted based on boundary and elevation information, a three-dimensional field model is constructed, and a grid is divided. An initial coverage path is generated by using a quasi-inverse learning enhanced fog algorithm based on differential evolution and spiral search. The pose of the agricultural machine is solved in real time, and an effective operation range is dynamically locked. When a road condition anomaly is perceived, a local passable domain is reconstructed, a continuous safe alternative path segment is generated by using a path re-planning strategy, and a heading deviation is corrected in real time by Beidou, so that precision operation is achieved. The application improves the traditional static global path planning of the agricultural machine, takes into account safety and full coverage, effectively monitors the dynamic running state of the agricultural machine, and corrects in real time, so that precision operation of a field in a mountainous area is achieved.

Description

Technical Field

[0001] This invention relates to the field of intelligent agricultural automation technology, and in particular to a method for precise operation of agricultural machinery in mountainous areas using Beidou navigation based on dynamic path planning. Background Technology

[0002] Traditional agricultural machinery operation path planning relies heavily on manual experience and static global models. Based on flat terrain, it generates fixed operation trajectories using conventional path algorithms. This approach suffers from poor terrain adaptability, lack of dynamic obstacle avoidance capabilities, insufficient path smoothness, large positioning deviations, and incomplete operation coverage. It struggles to achieve safe operation in complex terrain while simultaneously ensuring real-time obstacle avoidance and full operation coverage. In the large-scale development of smart agriculture in mountainous areas, agricultural machinery operation planning needs to consider the complex operating environment resulting from the prevalent characteristics of mountain farmland, such as large terrain undulations, varying slopes, high surface roughness, and numerous scattered obstacles.

[0003] Existing agricultural machinery path planning systems have adopted electronic path generation combined with satellite navigation and positioning functions, but they still lack quantitative perception of complex three-dimensional terrain features in mountainous areas and the ability to fuse and optimize multiple constraints. They are difficult to efficiently match real-time ground conditions with agricultural machinery attitude and operation data. Path planning relies on static terrain data and general algorithm rules, lacking the ability to dynamically replan based on the real-time operating environment. They are slow to pay attention to and respond to sudden terrain changes and temporary obstacles in mountainous areas. Temporary path planning is slow and lacks accuracy. Path planning in dynamic environments is prone to getting trapped in local optima, posing safety hazards. Summary of the Invention

[0004] The technical problem solved by this invention is that the existing technology is not adaptable to the complex terrain of mountainous fields and is difficult to balance comprehensive operation coverage and safety assurance. In the static global mode, the real-time matching and monitoring of the path is insufficient, the dynamic obstacle avoidance capability is weak, and there are defects in the speed and accuracy of response to local emergencies and abnormal environments and replanning.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0006] A method for precise operation of agricultural machinery in mountainous areas using BeiDou navigation system with dynamic path planning includes the following steps:

[0007] Step S1: Collect latitude and longitude data of field boundaries, extract terrain features and establish a three-dimensional land parcel model by combining digital elevation information of mountainous areas;

[0008] Step S2: Based on the three-dimensional land parcel model, optimize the balance between coverage integrity and terrain safety to generate an initial coverage path;

[0009] Step S3: Combine the initial coverage path with the boundary model during the operation of the agricultural machinery to lock the effective operating range;

[0010] Step S4: Generate conflict areas based on the effective work area, calculate local passable domains and trigger dynamic path replanning to generate alternative path segments;

[0011] Step S5: Perform curvature continuity smoothing on the alternative path to obtain an accurate feasible route, and have the BeiDou navigation system correct the heading deviation in real time.

[0012] As a preferred embodiment of the dynamic path planning method for precise operation of agricultural machinery in mountainous areas using BeiDou navigation, as described in this invention, step S1 includes the following sub-steps:

[0013] Step S101: Based on the UAV aerial survey equipment, the latitude and longitude data of the field boundary and the digital elevation information of the mountain area are obtained by sampling using the Beidou RTK module.

[0014] Based on the image data collected by UAVs from multiple perspectives and their corresponding timestamps and geographic coordinates, combined with the BeiDou RTK module to sample the ground edge, calibrate the latitude and longitude data of the field boundaries, scan the surface structure, and obtain the original elevation information.

[0015] Step S102: Extract terrain features based on the digital elevation information of the mountainous area, including slope, aspect, elevation change value and surface roughness, and construct a three-dimensional land parcel model.

[0016] The original elevation data is standardized and preprocessed to achieve benchmark unification, timestamp synchronization and resolution unification. Noise is removed and errors are corrected to achieve fusion registration. Based on the generated regular grid digital elevation model, terrain features including slope, aspect, elevation abrupt change value and surface roughness are extracted. Based on these multi-dimensional attribute labels and obstacle markers, they are associated with the corresponding grid to construct a three-dimensional land parcel model.

[0017] As a preferred embodiment of the dynamic path planning method for precise operation of agricultural machinery in mountainous areas using BeiDou navigation, as described in this invention, step S2 includes the following sub-steps:

[0018] Step S201: Divide the three-dimensional land parcel model into rectangular grids. Each grid corresponds to a uniquely defined area in the model and includes obstacle attribute and operation attribute labels.

[0019] The continuous 3D model plots are divided into rectangular grids. The obstacle attributes and operational attributes of the grids are obtained based on the attribute information of the grids in the 3D plot model. This includes accessibility based on obstacle markers and elevation differences exceeding preset thresholds, and labels for average slope, average aspect, and average surface roughness based on the grids covered by each grid.

[0020] Step S202: Based on the constraint of the impact of terrain environment on operation safety and the consideration of coverage integrity, construct the objective function, and use the quasi-backward learning enhanced rime ice algorithm based on differential evolution and spiral search to plan the path and generate the initial coverage path.

[0021] Based on the pre-defined grid, accessibility is quantized into 0 and 1, with hard constraints set for reachable grid coverage reaching 95% and path nodes being adjacent grids. Paths are constructed from work points and encoded as a two-dimensional coordinate sequence; terrain safety and coverage integrity costs are introduced, and fitness values ​​are calculated considering path smoothness costs and boundary entry / exit penalties.

[0022] Diverse experimental paths are generated using differential evolution mutation. A global search for soft rime is conducted guided by terrain features, updating path point coordinates based on the current optimal individual position, random direction, and step size factor. A spiral search for hard rime is triggered probabilistically based on normalized fitness to complete local development, updating path point positions based on the difference between the optimal individual position and two random individuals, and the logarithmic spiral radius. Centroid boundary control is introduced to perform feasibility repair on the updated paths, replacing coordinates of deviated entry / exit points and internal points that cross boundaries or exceed thresholds. After each update, a greedy selection process is used to retain individuals with better fitness. In each iteration, the path with the best fitness is locally refined using quasi-oppositional learning. This iteration continues until the maximum number of iterations is reached, ultimately outputting the path with the best fitness, i.e., the initial coverage path that meets the conditions.

[0023] As a preferred embodiment of the dynamic path planning method for precise agricultural machinery operation in mountainous areas using BeiDou navigation, as described in this invention, step S3 includes the following sub-steps:

[0024] Step S301: Based on the Beidou RTK module mounted on the agricultural machinery, combined with the inertial measurement unit, lidar and vision sensor, the position and attitude information of the agricultural machinery is calculated in real time, including the position, heading and attitude data of the agricultural machinery.

[0025] The sensor spatial calibration and time synchronization are performed in advance. The position and speed of the agricultural machinery are calculated in real time based on the Beidou RTK module and the base station signal. The relative position increment, speed increment and attitude increment are recursively calculated based on the inertial measurement unit. The surrounding environment image and its core features are extracted based on the lidar and vision sensor, and the offset is calculated. The measurement data and recursion are combined to output the position and attitude of the agricultural machinery in real time.

[0026] Step S302: Perform ICP point cloud matching between the real-time pose of the agricultural machinery and the boundary of the three-dimensional plot model to dynamically lock the current effective operating range;

[0027] The current agricultural machinery is converted to the global coordinate system of the 3D plot model. The terrain ground points and boundary feature points are separated from the point cloud. First, coarse registration is used, and then ICP fine registration is used to match the 3D plot model in real time. After the validity is verified, the current local operation area is clipped globally.

[0028] Step S303: Simultaneously perceive the working environment through vehicle-mounted lidar and vision sensors to verify the consistency between the initial model and the actual terrain in the surrounding environment.

[0029] The registered point cloud is matched one by one with the corresponding grid in the 3D land parcel model. Based on the visual image, the extracted features are projected to the global coordinates for distance comparison to verify the consistency between the real-time environment and the model terrain. If the deviation exceeds the preset threshold, it is judged as inconsistent.

[0030] As a preferred embodiment of the dynamic path planning method for precise operation of agricultural machinery in mountainous areas using BeiDou navigation, as described in this invention, step S4 includes the following sub-steps:

[0031] Step S401: Based on the real-time perception of the vehicle-mounted sensors and the preset conditions of the three-dimensional land model, when it is determined that there is an abnormal terrain or temporary obstacle that conflicts with the preset conditions of the initial path, the sensors are triggered to perform a high-precision scan of the environment ahead.

[0032] When the terrain is determined to be inconsistent and temporary obstacles detected by the sensors conflict with the preset conditions of the initial path, and there is a risk of passage, the sensors are triggered to perform a high-precision scan, lock the spatial coordinate range of the conflict area, and use it as the target area for directional scanning.

[0033] Step S402: Reconstruct the local terrain model of the conflict area, update the terrain parameters and obstacle distribution attributes of the area, and upload and synchronize the global three-dimensional land parcel model to generate a local passable area.

[0034] The terrain of the conflict area is remodeled based on local high-precision scanning. The terrain parameters and obstacle distribution of the area are reprojected and labeled in the 3D plot model based on lidar point cloud and visual sensor combined with the current pose of agricultural machinery. The local model is updated synchronously, and local passable domains are generated based on hard constraint marking.

[0035] Step S403: Dynamically replanning to generate alternative path segments based on the locally passable domain.

[0036] Based on the updated 3D land parcel model, feasible paths are planned for the updated local areas in real time, starting from the replanning starting point. The replanning endpoint is selected as the replanning map, extending to the boundary of the conflict area. Considering safety and current area coverage, continuous feasible alternative path segments are generated.

[0037] As a preferred embodiment of the dynamic path planning method for precise operation of agricultural machinery in mountainous areas using BeiDou navigation, as described in this invention, step S5 includes the following sub-steps:

[0038] Step S501: Perform curvature continuity detection and smoothing on the alternative path segment to ensure that the adjusted path meets the minimum turning radius constraint of agricultural machinery.

[0039] The curvature of the path points of the alternative path segment is calculated. If the rate of change of curvature of adjacent points exceeds the threshold, it is determined that the curvature is discontinuous. The maximum curvature is checked to see if it exceeds the upper limit of curvature corresponding to the minimum turning radius of the agricultural machinery. Points that exceed this value are smoothed by cubic B-spline curve.

[0040] Step S502: Update the adjusted path to the agricultural machinery driving control system, and correct the heading and position deviation of the agricultural machinery in real time based on the Beidou RTK system to carry out precise operations in mountainous fields.

[0041] Based on the adjusted precise route, the system updates the agricultural machinery driving control system in real time, matches the agricultural machinery's posture with the positioning in the preset route, and calculates and corrects deviations based on the Beidou RTK system to achieve precise operation in mountainous fields.

[0042] As a preferred embodiment of the Beidou-based precision operation method for agricultural machinery in mountainous areas based on dynamic path planning as described in this invention, during the operation, the operation path that has been advanced and matched is marked in real time based on the sensor perception of the environment and path verification process, the operation trajectory and travel parameters are recorded, and the path optimization log is uploaded.

[0043] As a preferred embodiment of the Beidou precision operation method for mountain agricultural machinery with dynamic path planning described in this invention, the dynamic replanning adopts a fusion path planning algorithm based on the improved A* algorithm and the dynamic window method optimized by the genetic algorithm. It introduces coverage integrity constraints and memory of already operated grid marks, and generates alternative path segments in the local passable domain based on safety and operation continuity.

[0044] Based on the marked completed grid, a penalty for repeated operations is added to the path planning cost. The path planning cost includes terrain safety, number of turns, and coverage. The replanning starting point of the path conflict extends to the replanning endpoint of the initial route at the boundary of the conflict area. A cost function is constructed based on the improved A* algorithm. A directional six-neighborhood search is used to obtain discrete polyline paths. Iterative folding is performed on each inflection point along the inner direction to eliminate redundant inflection points and smooth the discrete path. The DWA algorithm with offline weight optimization using a genetic algorithm is used to generate a continuous trajectory that conforms to agricultural machinery dynamics and replaces conflicting path segments.

[0045] As a preferred embodiment of the Beidou-based precision operation method for agricultural machinery in mountainous areas using dynamic path planning as described in this invention, the initial coverage route planning is generated offline before operation, while real-time pose matching and dynamic replanning are performed online during operation.

[0046] As a preferred embodiment of the Beidou precision operation method for mountain agricultural machinery with dynamic path planning described in this invention, in the abnormal working conditions of Beidou signal loss or path conflict detection, the agricultural machinery automatically reduces its operating speed, and when a dangerous situation is detected or the core sensor fails, the agricultural machinery automatically retreats to a safe area at low speed.

[0047] The beneficial effects of this invention are:

[0048] First, a high-precision terrain perception and multi-objective optimization framework is constructed. A 3D land parcel model is established based on the fusion of UAV aerial surveying and BeiDou RTK, terrain features are extracted, grids are divided, and accessibility is marked. A quasi-backward learning-enhanced rime ice algorithm based on differential evolution and spiral search is used to plan paths. The initial path satisfies a dynamic balance between coverage integrity and terrain safety, improving the terrain adaptability of path planning in complex mountainous terrain and forming an effective initial route.

[0049] Second, it achieves real-time pose matching and local dynamic replanning. The machine's pose is calculated in real-time, the effective operating range is locked, and high-precision scanning and reconstruction of the local passable domain are triggered promptly, meeting the real-time multi-source monitoring requirements of the machine's operation. In the replanning stage, an improved A* algorithm combined with a genetic algorithm-optimized DWA fusion strategy is adopted to reduce the complexity of online optimization planning and generate continuous and safe alternative path segments. Compared to static global planning, this approach satisfies the requirements for dynamic response to sudden operating conditions and real-time planning.

[0050] Third, ensuring operational accuracy and safety. The alternative path undergoes curvature continuity detection and cubic B-spline smoothing, coupled with real-time correction of heading deviations via BeiDou, guaranteeing accurate operation along a precise route. When BeiDou signal is lost or danger is detected, the system automatically reduces speed or retreats to a safe area, ensuring both safety and operational efficiency. Attached Figure Description

[0051] Figure 1 The flowchart illustrates the steps of a method for precise operation of agricultural machinery in mountainous areas using BeiDou navigation based on dynamic path planning, as provided in one embodiment of the present invention.

[0052] Figure 2 This is a flowchart of the quasi-backward learning-enhanced rime ice algorithm based on differential evolution and spiral search used in this invention. Detailed Implementation

[0053] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0054] Combination Figure 1 and Figure 2 This invention provides a method for precise operation of agricultural machinery in mountainous areas using BeiDou navigation based on dynamic path planning, comprising the following specific steps:

[0055] Step S1: Collect latitude and longitude data of field boundaries, extract terrain features and establish a three-dimensional land parcel model by combining digital elevation information of mountainous areas;

[0056] Step S2: Based on the three-dimensional land parcel model, optimize the balance between coverage integrity and terrain safety to generate an initial coverage path;

[0057] Step S3: Combine the initial coverage path with the boundary model during the operation of the agricultural machinery to lock the effective operating range;

[0058] Step S4: Generate conflict areas based on the effective work area, calculate local passable domains and trigger dynamic path replanning to generate alternative path segments;

[0059] Step S5: Perform curvature continuity smoothing on the alternative path to obtain an accurate feasible route, and have the BeiDou navigation system correct the heading deviation in real time.

[0060] The steps of acquiring data and 3D modeling include:

[0061] Step S101: Based on the UAV aerial survey equipment, the latitude and longitude data of the field boundary and the digital elevation information of the mountainous area are obtained by sampling using the Beidou RTK module;

[0062] Data is collected by multi-view aerial photography of UAVs equipped with lidar and high-resolution cameras. The system combines ground-based BeiDou reference stations and airborne RTK modules to achieve positioning and signal transmission, sample and calibrate field latitude and longitude data, and obtain raw surface elevation data.

[0063] Step S102: Extract terrain features based on the digital elevation information of the mountainous area, including slope, aspect, elevation change value and surface roughness, and construct a three-dimensional land parcel model.

[0064] BeiDou RTK reference stations and airborne RTK systems output data using the 2000 National Geodetic Coordinate System and the 1985 National Height Datum. Raw point clouds from lidar and aerial photographic POS data are also converted to the 2000 National Geodetic Coordinate System, achieving unified datum alignment for multi-source data. This completes the standardization of datum, resolution, and time alignment. Bundle adjustment is used for error calibration to obtain preprocessed data. Gaussian filtering, progressive morphological filtering for separation, and redundancy removal are applied to the lidar point clouds. Influence point clouds are generated based on aerial triangulation results. ICP precise registration and point cloud fusion are then employed to obtain clean discrete point clouds.

[0065] The preprocessed data is interpolated using inverse distance weighted interpolation to generate a fixed-resolution regular grid digital elevation model, with each grid storing a unique average absolute elevation value. A global orthophoto is generated based on aerial imagery, and georegistration is used to overlay the digital elevation model and the digital orthophoto to obtain a three-dimensional geometric model combining elevation and texture.

[0066] Preferably, the resolution of the digital elevation model is set to 0.5 meters.

[0067] Based on field boundary vector data, spatial indexing is superimposed onto a three-dimensional geometric model to accurately delineate the boundary range of agricultural machinery operations.

[0068] Based on the digital elevation model, the 3×3 window difference method is used to calculate the terrain features such as slope, aspect, elevation abrupt change value, and surface roughness for each grid cell. The acquired attributes are mapped to a unique ID corresponding to each grid cell and stored. Special cases such as elevation abrupt changes are marked.

[0069] Obstacle information is obtained by clustering non-ground point clouds obtained through filtering and separation. The size of the obstacle is preset. Obstacles larger than the preset size are judged as potentially affecting the operation and are marked by associating them with the corresponding coordinate grid.

[0070] The three-dimensional geometric model, which has been mapped by terrain feature attributes and marked with obstacles, is stored in a hierarchical architecture to form a high-precision three-dimensional land parcel model.

[0071] The step of obtaining the initial coverage path using the frost algorithm includes:

[0072] Step S201: Divide the three-dimensional land parcel model into rectangular grids. Each grid corresponds to a uniquely defined area in the model and includes obstacle attribute and operation attribute labels.

[0073] Based on a continuous three-dimensional plot model, a rectangular grid network matching the characteristics of agricultural machinery operations is obtained. The grid row and column directions are parallel to the grid row and column directions of the three-dimensional plot model. Each grid corresponds to a uniquely determined geographic spatial area in the plot. The grid resolution is determined according to the actual operating width of the agricultural machinery.

[0074] Preferably, the grid resolution is set to be equal to the width of the agricultural machinery's working area.

[0075] Raster accessibility is determined based on markers such as obstacles and elevation changes, and grid accessibility is marked to indicate whether or not grids need to be avoided.

[0076] The area-weighted method is used to calculate all terrain attributes of the operational raster. The formula is as follows:

[0077] ;

[0078] in, For the ( For each of the following raster cells, calculate the average elevation, average slope, average aspect, and standard deviation of elevation. M×N represents the total number of grid cells intersecting with this raster cell. For the ( The actual area of ​​the grid covered by the operation grid. For the ( The original attribute values ​​of each grid cell.

[0079] For edge grids, a grid is considered valid if its area is greater than or equal to half the area of ​​a single standard grid. Its grid attributes are calculated and recorded. If its area is less than half the area of ​​a single standard grid, it is considered an invalid grid and is directly marked as an impassable area.

[0080] Slope limits are set based on the slope limits of the type of agricultural machinery used, and paths exceeding the slope limits are marked as unreachable.

[0081] Preferably, the slope limit is set to 25° based on industry standard specifications.

[0082] Step S202: Based on the constraint of the impact of terrain environment on operation safety and the consideration of coverage integrity, construct the objective function, and use the quasi-backward learning enhanced rime ice algorithm based on differential evolution and spiral search to plan the path and generate the initial coverage path.

[0083] Build the full path :

[0084] ;

[0085] in, These are internal work path points, and each point has its corresponding grid coordinates.

[0086] The objective is to minimize the fitness value. The formula is:

[0087] ;

[0088] in, To cover the cost of integrity, For path security costs, For path smoothness cost, Cost of elevation difference fluctuation, The weights are set based on the optimization objective and constraints. .

[0089] Coverage integrity cost: A quantitative metric used to measure the extent to which the generated path covers the target work area.

[0090] The formula for the cost of coverage integrity is:

[0091] ;

[0092] in, For the raster set that needs to be covered, This is the set of all grid cells passed through within the path. This represents the number of times grid g is traversed by the path. The penalty coefficient for repeated coverage is set based on the requirement for the shortest path.

[0093] Preferably, in this embodiment, the primary objective is to ensure coverage integrity, and the following settings are made: It is 0.2.

[0094] Path safety cost: Used to measure the danger level of each path segment between grids in the path, based on the slope of the grid and the angle between the driving direction and the slope aspect to obtain the instantaneous risk of the path segment.

[0095] The formula for path security cost is:

[0096] ;

[0097] in, This is the comprehensive slope penalty value for the k-th path segment.

[0098] Comprehensive slope penalty value: The slope is decomposed into lateral slope and longitudinal slope, and obtained by weighted summation of piecewise function square penalties with different thresholds applied to the angles corresponding to the lateral slope and longitudinal slope.

[0099] Preferably, since lateral slope is the main factor affecting rollover, the threshold is set to 5°-15°. Within this range, a squared penalty is applied, with below 5° considered safe and above 15° considered impassable. The longitudinal slope threshold is set to 10°-25°, with below 10° considered safe and above 25° considered impassable.

[0100] Preferably, the longitudinal slope has a smaller impact on safety events such as rollover compared to the lateral slope, with the lateral slope being the dominant risk. The penalty function weights for the lateral slope and the longitudinal slope are set to 0.7 and 0.3, respectively.

[0101] Path smoothness cost: Used to measure the number of sharp turns on the path, assessing operational efficiency and wear. The calculation uses a normalized average of the squared steering angles, with a normalization factor set based on the minimum turning radius of the agricultural machinery, and treats excessive steering angles as hard constraints.

[0102] Elevation difference fluctuation cost: A quantitative indicator that measures the degree of vertical undulation of a path. It is calculated by normalizing the sum of the squares of the elevation differences between adjacent grid cells along the path.

[0103] Preferably, since the main objectives of this embodiment are coverage integrity and terrain security, these two items are given high weights. The weight parameters are screened and optimized based on the orthogonal experiment method, and the weight factor level is set with a step size of 0.05.

[0104] Table 1

[0105] choose Orthogonal arrays are used to calculate the mean of the objective function at each level based on the algorithm in this embodiment. The optimal weight combinations are obtained by considering the range R and the range R.

[0106] Table 2

[0107] The initial population size is set to 30, the number of iterations T is 100, and the dimension is the number of nodes + 2. The number of nodes is set based on the actual field area.

[0108] Initial path point coordinates are generated based on Tent chaotic mapping, projected onto the feasible region, and then mapped to the actual coordinates.

[0109] Differential evolution mutation and crossover are used to generate experimental vectors for each individual.

[0110] ;

[0111] in, , This is the index of an individual that is randomly selected from the population and is different from i.

[0112] Cross dimensions are generated for dimension i based on a cross probability of 0.9.

[0113] Global search based on probability A soft frost search strategy is adopted.

[0114] ;

[0115] in Let t be the coordinate of the j-th dimension of the current best individual in the population, and t be the current iteration number. A positive or negative random factor, ranging from -1 to 1, is used to determine the search direction. To control the probability threshold of whether the soft frost search is triggered, the range is 0~1, and h is the adhesion coefficient, which simulates the degree of random adhesion between frost and the environment during frost growth, and is uniformly distributed in the range of 0~1; To determine the angle coefficient of the cosine term in the soft rime search, it increases linearly with the number of iterations, used to simulate the natural phenomenon of the wind direction slowly rotating over time. express; The step size decay factor is used to simulate the rapid to slow growth of rime ice. The expression is... .

[0116] The local search employs a hard frost puncture mechanism combined with a spiral search to obtain the optimal individual.

[0117] ;

[0118] in, , This is the index of an individual that is randomly selected from the population and is different from i. Let be the normalized fitness value of the i-th individual. This is a spiral search term.

[0119] All of the above updates introduce centroid boundary control to perform feasibility repair on the midpoint coordinates of individuals. Based on all feasible individuals in the population, the centroid coordinates are obtained by averaging the horizontal and vertical coordinates. For individuals that exceed the limits, the coordinate values ​​are corrected by using the nearest boundary coordinate value of the exceeding point and the mean centroid coordinates.

[0120] The population is updated based on greedy selection, retaining individuals with better fitness. After each iteration, quasi-opposite learning is applied to the best individual, calculating the quasi-opposite point for each path point and comparing it with the optimal solution with smaller fitness.

[0121] The step of real-time calculation of pose matching and recording path includes the following sub-steps:

[0122] Step S301: Based on the Beidou RTK module mounted on the agricultural machinery, combined with the inertial measurement unit, lidar and vision sensor, the position and attitude information of the agricultural machinery is calculated in real time, including the position, heading and attitude data of the agricultural machinery.

[0123] The external parameters of each sensor are calibrated in advance, the three-dimensional offset and installation attitude of the sensor relative to the coordinate system of the agricultural machinery are measured, and the rigid body transformation matrix is ​​established. At the same time, the camera intrinsic parameter calibration and the lidar zero-bias calibration are completed. The PPS second pulse based on Beidou RTK is used as the global time reference, and all data are bound to a unified Beidou time stamp. Linear interpolation is used to align the data timestamps for sensors with different sampling frequencies.

[0124] The BeiDou RTK system, combined with differential correction data transmitted from farmland reference stations, calculates position and velocity in real time. Based on inertial measurement unit (IMU) pre-integration, it recursively calculates the relative position, velocity, and attitude increments between two RTK observation times. Terrain features are extracted using lidar and visual sensors to calculate lateral offset and heading deviation. Based on inertial navigation mechanics, the system recursively calculates the agricultural machinery's position, velocity, and attitude angles in real time. Combined with measurement data, it corrects the IMU's cumulative drift error using Kalman gain, and employs point cloud registration for pose and visual heading deviation as auxiliary methods to accurately output the agricultural machinery's pose data.

[0125] Step S302: Perform ICP point cloud matching between the real-time pose of the agricultural machinery and the boundary of the three-dimensional plot model to dynamically lock the current effective operating range;

[0126] Based on the output agricultural machinery pose, ground point segmentation is performed on the point cloud to separate terrain ground points and boundary feature points, and non-ground obstacle points are filtered out. The point cloud is coarsely registered based on the sample consistency initial registration algorithm, and then finely registered based on the point-to-surface iterative nearest point algorithm. The current local working area is cropped with the registered agricultural machinery position as the center and the effective sensing radius of the sensor.

[0127] Step S303: Simultaneously perceive the working environment through vehicle-mounted lidar and vision sensors to verify the consistency between the initial model and the actual terrain in the surrounding environment.

[0128] After registration, the point cloud is matched one by one with the corresponding grid cells of the 3D land parcel model according to the grid resolution. The deviation between the real-time point cloud elevation and the model elevation of each grid cell is counted, and the elevation deviation of a single grid cell and the average deviation of a local area are calculated. In this embodiment, if the absolute value of the elevation deviation of a single grid cell is >0.3m, it is marked as an elevation inconsistency grid cell; if the deviation of more than 5 consecutive grid cells exceeds the standard, it is determined that the terrain of a local area is inconsistent.

[0129] Extract the outline of ground features from the current frame visual image, project the boundary features in the image onto the global coordinate system, and compare the distance with the boundary lines of the plots in the initial model. In this embodiment, if the boundary line offset distance is >0.5m, the area boundary is determined to be inconsistent.

[0130] For terrain areas with consistent assessments, the already operated grid is marked based on the route already traveled by the agricultural machinery.

[0131] The step of local replanning to obtain alternative path segments includes the following sub-steps:

[0132] Step S401: Based on the real-time perception of the vehicle-mounted sensors and the preset conditions of the three-dimensional land model, when it is determined that there are abnormal terrain or temporary obstacles that conflict with the preset conditions of the initial path, the sensors are triggered to perform a high-precision scan of the environment ahead.

[0133] When the actual terrain does not match the initial model, and the sensor detects unmarked obstacles in the model, a high-precision scanning trigger signal is output when there is a risk of passage due to a conflict with the preset conditions of the initial path. The environment around the agricultural machinery coordinates is locked as the conflict area. In subsequent directional scanning of the target area, the scanning accuracy is improved by increasing the radar sampling frame rate and line density, and switching the vision sensor to high-definition acquisition.

[0134] Step S402: Reconstruct the local terrain model of the conflict area, update the terrain parameters and obstacle distribution attributes of the area, and upload the synchronized global 3D land parcel model to generate a local passable area.

[0135] Based on LiDAR point cloud and visual sensor image acquisition, the data is synchronized, denoised, filtered and extracted. Combined with the agricultural machinery pose, the data is re-superimposed on the basic geometric model to obtain a new three-dimensional land parcel model. Local new terrain feature markers are extracted and stored to generate local passable domains.

[0136] Step S403: Dynamically replanning to generate alternative path segments based on the locally passable domain.

[0137] A fusion path planning algorithm based on an improved A* algorithm and a genetic algorithm-optimized dynamic window method is adopted. Coverage integrity constraints and memory of previously completed raster markers are introduced. The replanning starts at the conflict point, extends to the boundary of the coverage conflict area, and connects to the initial route as the replanning endpoint, forming the replanning map. Raster terrain features within the local area are extracted, and the initial path is used as a sub-objective sequence to closely approximate the original route. A cost function is set based on the A* algorithm.

[0138] ;

[0139] in, To calculate the cumulative cost across multiple objectives, the same calculation method as minimizing the fitness value is adopted, and constraints are imposed on safety and integrity. A conservative heuristic estimate of the cost from the current node to the destination.

[0140] The formula is:

[0141] ;

[0142] in, This represents the local obstacle density in the region from the current node to the endpoint. It is calculated as the ratio of the number of obstacle grids in the remaining region to the number of remaining local grids.

[0143] The path with minimum cost is retrieved using a directional six-neighborhood search until the replanning endpoint is retrieved and added to the path. Inflection points of the path are identified, and inflection points to be smoothed are marked. The inner direction of the corner is obtained, and the inflection point position is moved inward along the angle bisector direction. After each movement, it is checked whether the two newly generated path segments pass through obstacles or exceed the operation boundary. If they meet the rules, the iteration continues; if they do not meet the rules, the folding is terminated, and the path is backed up to the previous feasible position to obtain a smooth path with simplified inflection points. Tangent circular arcs are used to connect adjacent discrete line segments based on circular arc interpolation.

[0144] Guided by a local reference path, feasible speed sets are generated by sampling speed combinations within a dynamic window, resulting in a real-time driving trajectory that conforms to agricultural machinery dynamics. A comprehensive evaluation function is then optimized based on a genetic algorithm. Obtain the optimal trajectory:

[0145] ;

[0146] in, The speed term is obtained by normalizing the travel speed based on the maximum speed, which is used to ensure operational efficiency. The heading term is obtained by normalizing the heading angle and trajectory pose based on π, which is used to reduce the deviation between the heading at the end of the trajectory and the direction of the sub-target. The obstacle distance term is obtained by normalizing the nearest Euclidean distance from the path point to the point cloud of all obstacles in the environment, and is used to ensure that the path stays away from obstacles. , , For each weight, We use a genetic algorithm to optimize the weights offline.

[0147] In this embodiment, the three weight values ​​obtained are 0.09, 0.33, and 0.58.

[0148] Replace the conflicting path segments of the original path with the obtained optimal trajectory.

[0149] The precise route correction operation includes the following sub-steps:

[0150] Step S501: The curvature continuity of the alternative path segment is detected and smoothed to ensure that the adjusted path meets the minimum turning radius constraint of the agricultural machinery.

[0151] The curvature of the path points of the alternative path segment is calculated. If the rate of change of curvature of adjacent points exceeds the threshold, it is determined that the curvature is discontinuous. The maximum curvature is checked to see if it exceeds the upper limit of curvature corresponding to the minimum turning radius of the agricultural machinery. Points that exceed this value are smoothed by cubic B-spline curve.

[0152] Preferably, the threshold for curvature change between adjacent points is set to 0.1m based on the terrain. -2 ~0.2m -2 In this embodiment, the intermediate value of 0.15m is used. -2 .

[0153] Step S502: Update the adjusted path to the agricultural machinery driving control system, and correct the heading and position deviation of the agricultural machinery in real time based on the Beidou RTK system to carry out precise operations in mountainous fields.

[0154] Based on the adjusted precise route, the system updates the agricultural machinery driving control system in real time, matches the agricultural machinery's posture with the positioning in the preset route, and calculates and corrects deviations based on the Beidou RTK system to achieve precise operation in mountainous fields.

[0155] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the protection scope of the present invention.

Claims

1. A method for precise operation of agricultural machinery in mountainous areas using BeiDou navigation based on dynamic path planning, characterized in that: Includes the following steps: Step S1: Collect latitude and longitude data of field boundaries, extract terrain features and establish a three-dimensional land parcel model by combining digital elevation information of mountainous areas; Step S2: Based on the three-dimensional land parcel model, optimize the balance between coverage integrity and terrain safety to generate an initial coverage path; Step S3: Combine the initial coverage path with the boundary model during the operation of the agricultural machinery to lock the effective operating range; Step S4: Generate conflict areas based on the effective work area, calculate local passable domains and trigger dynamic path replanning to generate alternative path segments; Step S5: Perform curvature continuity smoothing on the alternative path to obtain an accurate feasible route, and have the BeiDou navigation system correct the heading deviation in real time.

2. The method for precise operation of agricultural machinery in mountainous areas using BeiDou navigation based on dynamic path planning as described in claim 1, characterized in that: Step S1 includes the following sub-steps: Step S101: Based on the UAV aerial survey equipment, the latitude and longitude data of the field boundary and the digital elevation information of the mountainous area are obtained by sampling using the Beidou RTK module; Step S102: Extract terrain features based on the digital elevation information of the mountainous area, including slope, aspect, elevation change value and surface roughness, and construct a three-dimensional land parcel model.

3. The method for precise operation of agricultural machinery in mountainous areas using BeiDou navigation based on dynamic path planning as described in claim 2, characterized in that, Step S2 includes the following sub-steps: Step S201: Divide the three-dimensional land parcel model into rectangular grids. Each grid corresponds to a uniquely defined area in the model and includes obstacle attribute and operation attribute labels. Step S202: Based on the constraint of the impact of terrain environment on operation safety and the consideration of coverage integrity, construct the objective function, and use the quasi-backward learning enhanced rime ice algorithm based on differential evolution and spiral search to plan the path and generate the initial coverage path.

4. The method for precise operation of agricultural machinery in mountainous areas using BeiDou navigation based on dynamic path planning as described in claim 3, characterized in that, Step S3 includes the following sub-steps: Step S301: Based on the Beidou RTK module mounted on the agricultural machinery, combined with the inertial measurement unit, lidar and vision sensor, the position and attitude information of the agricultural machinery is calculated in real time, including the position, heading and attitude data of the agricultural machinery. Step S302: Perform ICP point cloud matching between the real-time pose of the agricultural machinery and the boundary of the three-dimensional plot model to dynamically lock the current effective operating range; Step S303: Simultaneously perceive the working environment through vehicle-mounted lidar and vision sensors to verify the consistency between the initial model and the actual terrain in the surrounding environment.

5. The method for precise operation of agricultural machinery in mountainous areas using BeiDou navigation based on dynamic path planning as described in claim 4, characterized in that, Step S4 includes the following sub-steps: Step S401: Based on the real-time perception of the vehicle-mounted sensors and the preset conditions of the three-dimensional land model, when it is determined that there is an abnormal terrain or temporary obstacle that conflicts with the preset conditions of the initial path, the sensors are triggered to perform a high-precision scan of the environment ahead. Step S402: Reconstruct the local terrain model of the conflict area, update the terrain parameters and obstacle distribution attributes of the area, and upload and synchronize the global three-dimensional land parcel model to generate a local passable area. Step S403: Dynamically replanning to generate alternative path segments based on the locally passable domain.

6. The method for precise operation of agricultural machinery in mountainous areas using BeiDou navigation based on dynamic path planning as described in claim 5, characterized in that, Step S5 includes the following sub-steps: Step S501: Perform curvature continuity detection and smoothing on the alternative path segment to ensure that the adjusted path meets the minimum turning radius constraint of agricultural machinery. Step S502: Update the adjusted path to the agricultural machinery driving control system, and correct the heading and position deviation of the agricultural machinery in real time based on the Beidou RTK system to carry out precise operations in mountainous fields.

7. The method for precise operation of agricultural machinery in mountainous areas using BeiDou navigation based on dynamic path planning as described in claim 6, characterized in that, During the operation, based on the sensor perception of the environment and the verification of the path, the grid of the operation path that has been advanced and matched is marked in real time, the operation trajectory and travel parameters are recorded, and the path optimization log is uploaded.

8. The method for precise operation of agricultural machinery in mountainous areas using BeiDou navigation based on dynamic path planning as described in claim 7, characterized in that, The dynamic replanning adopts a fusion path planning algorithm based on the improved A* algorithm and the dynamic window method optimized by the genetic algorithm. It introduces coverage integrity constraints and memory of already completed grid marks, and generates alternative path segments in the local traversable domain based on safety and operation continuity.

9. The method for precise operation of agricultural machinery in mountainous areas using BeiDou navigation based on dynamic path planning as described in claim 8, characterized in that, Initial coverage route planning is generated offline before the operation, while real-time pose matching and dynamic replanning are performed online during the operation.

10. The method for precise operation of agricultural machinery in mountainous areas using BeiDou navigation based on dynamic path planning as described in claim 9, characterized in that... When the BeiDou signal is lost or an abnormal working condition such as a path conflict is detected, the agricultural machinery will automatically reduce its operating speed. If a dangerous situation is detected or a core sensor fails, the agricultural machinery will automatically retreat to a safe area at low speed.

Images