Agricultural machinery automatic driving path planning method based on Beidou RTK

Abstract

The invention belongs to the technical field of path planning, and discloses an agricultural machinery automatic driving path planning method based on Beidou RTK. The method comprises the following steps: acquiring an original observation value of a Beidou satellite to obtain original observation data; the feature extraction module is used for calculating according to the original observation data to obtain a path combination observation value; calculating according to the path combination observation value to obtain path characteristics; analyzing data such as path characteristics, signal-to-noise ratio observation values and agricultural machine coordinates to obtain a dielectric anomaly index; according to the dielectric anomaly index, an obstacle probability distribution value is obtained through calculation; according to the obstacle probability distribution value, performing division to obtain a risk level; constructing a path total cost function according to the risk level, and calculating to obtain a minimum cost path; according to Beidou RTK underground obstacle detection, the dielectric anomaly index is calculated through the signal-to-noise ratio frequency domain characteristics to accurately recognize the underground obstacle, and the automatic driving safety and efficiency of agricultural machinery are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of path planning technology, and more specifically, to a path planning method for automated agricultural machinery based on BeiDou RTK. Background Technology

[0002] Existing methods for autonomous driving of agricultural machinery rely heavily on the identification of visible obstacles on the ground and simple shape analysis, lacking the ability to effectively detect underground obstacles and accurately assess underground risks. At the same time, traditional path planning only considers path length and basic obstacles, without comprehensively considering multi-dimensional factors such as underground risks, surface obstacles, soil characteristics, and terrain slope, resulting in insufficient safety and efficiency of the planned path, making it difficult to meet the needs of precision operations in modern agriculture.

[0003] Chinese patent application CN109240284A discloses an autonomous path planning method and device for unmanned agricultural machinery, relating to the field of intelligent agricultural machinery equipment. It employs a visual sensor and a BeiDou RTK positioning and orientation system to autonomously identify field ridges and optimize field boundary identification routes, reducing the computational load and extraction time of the boundary identification algorithm. The device includes a BeiDou RTK positioning and orientation unit, an information processing unit, and a data communication unit. The device first obtains field boundary information through a boundary identification algorithm. The user inputs the agricultural machinery's operating width through a human-machine interface. Then, based on the odd and even number of rows, it plans an operating path that meets agronomic requirements. The obtained field boundary information and path planning information are sent to a data service center for storage via a data transmission module, for automatic field matching and operational status monitoring of other agricultural implements.

[0004] While the above methods can meet the needs of most scenarios, research and practical application of these methods and existing technologies have revealed at least the following shortcomings:

[0005] The above methods rely on lidar and cameras to detect obstacles on the ground. However, lidar and cameras cannot penetrate soil and vegetation layers, making it impossible to effectively identify covered underground obstacles. This poses safety hazards such as equipment overturning and chassis damage during agricultural machinery operations, and can easily lead to work interruptions and equipment damage.

[0006] In view of this, the present invention proposes a path planning method for automatic agricultural machinery based on Beidou RTK to solve the above problems. Summary of the Invention

[0007] To overcome the aforementioned deficiencies of the prior art and to achieve the above objectives, the present invention provides the following technical solution: a method for path planning of automated agricultural machinery based on BeiDou RTK, comprising:

[0008] Collect raw observation data from BeiDou satellites; the raw observation data includes carrier phase observations, pseudorange observations, signal-to-noise ratio observations, agricultural machinery coordinates, satellite geometric elevation angle, and satellite geometric azimuth angle;

[0009] The path characteristics are calculated based on the carrier phase observations and pseudorange observations; the signal-to-noise ratio (SNR) observations are processed by short-time Fourier transform and analyzed to obtain the SNR frequency domain characteristics; the path characteristics, SNR frequency domain characteristics, agricultural machinery coordinates, satellite geometric elevation angle, and satellite geometric azimuth angle are used as inputs to the anomaly analysis model to obtain the dielectric anomaly index and its corresponding coordinates, and the presence of obstacles is determined based on the dielectric anomaly index.

[0010] When obstacles are present, Kriging interpolation is performed based on the dielectric anomaly index and the corresponding coordinates to obtain the obstacle probability distribution value of the farmland operation area; the obstacle probability distribution value of the farmland operation area is analyzed and clustered to obtain candidate areas; the risk level is determined based on the obstacle probability distribution value of the candidate areas.

[0011] Based on the risk level, a total path cost function is constructed, and the minimum cost path is calculated with the goal of minimizing the total path cost function; the minimum cost path is then used as the planning path for the automated driving of agricultural machinery.

[0012] Furthermore, methods for obtaining path combination observations include:

[0013] Based on the carrier phase observations in the original observation data, the three-frequency geometric combination interpolation is calculated. The difference between the current three-frequency combination interpolation and the previous three-frequency combination interpolation is calculated to obtain the three-frequency combination difference. If the three-frequency combination difference is greater than the cycle slip detection threshold, it is determined to be a cycle slip. The carrier signal observations are repaired based on the repair window to obtain the repaired carrier phase sequence. Otherwise, the carrier phase observations are directly used as the repaired carrier phase sequence. The absolute value of the difference between the repaired carrier phase sequence and the pseudorange observations is calculated to obtain the path combination observations.

[0014] Furthermore, methods for obtaining signal-to-noise ratio frequency domain characteristics include:

[0015] The signal-to-noise ratio (SNR) observations are processed by short-time Fourier transform to obtain the spectrum matrix. The frequency with the largest amplitude in the spectrum matrix is ​​taken as the main oscillation frequency, and the amplitude corresponding to the main oscillation frequency is obtained. The main oscillation frequency and the corresponding amplitude are concatenated to obtain the SNR frequency domain characteristics.

[0016] Furthermore, methods for determining the presence of obstacles based on dielectric anomaly indices include:

[0017] When the dielectric anomaly index is greater than the obstacle probability threshold, an obstacle is determined to exist; otherwise, no obstacle is determined to exist.

[0018] Furthermore, methods for obtaining risk levels include:

[0019] The dielectric anomaly index is weighted by Kriging interpolation to obtain the initial barrier probability distribution value.

[0020] The initial obstacle probability distribution value is normalized to obtain the obstacle probability distribution value. The obstacle probability distribution value of each grid cell is compared with a preset probability threshold. If the obstacle probability distribution value is greater than the preset probability threshold, it is determined as a potential obstacle area; otherwise, it is determined as a safe area. The potential obstacle areas are clustered to obtain connected regions. The total number of grid cells in each connected region is counted and multiplied by the area of ​​a single grid cell to obtain a continuous area. The continuous area is compared with a preset minimum area, and regions with a continuous area greater than the preset minimum area are retained as candidate regions. The risk level is determined based on the obstacle probability distribution value of the candidate regions.

[0021] Furthermore, methods for obtaining the initial obstacle probability distribution values ​​include:

[0022] Using the boundary coordinates of the farmland operation area as the range, the farmland operation area is divided into 1m×1m grid cells. For each grid cell, the spatial correlation weight between the grid cell and the coordinates of the nearest H agricultural machinery is calculated. The spatial correlation weight is the reciprocal of the distance between the blank grid coordinates and the coordinates of the nearest H agricultural machinery. All spatial correlation weights are normalized to obtain the correlation weight. The dielectric anomaly index of all grid cells is weighted in combination with the correlation weight to obtain the initial obstacle probability distribution value.

[0023] Furthermore, methods for determining risk levels include:

[0024] When the obstacle probability distribution value of a candidate region is greater than the probability threshold but not greater than 1, it is determined to be a high-risk region; when the obstacle probability distribution value of a candidate region is greater than half of the probability threshold but not greater than the probability threshold, it is determined to be a medium-risk region; when the obstacle probability distribution value of a candidate region is not greater than half of the probability threshold, it is determined to be a low-risk region.

[0025] Furthermore, the total cost function of the path is obtained by adding the underground risk cost, the path length, and other cost items; the other cost items are obtained by adding the low surface obstacle cost, the soil compaction cost, and the slope cost.

[0026] Methods for obtaining underground risk costs include:

[0027] The underground risk cost is obtained by multiplying the obstacle probability distribution value and the risk coefficient corresponding to the risk level.

[0028] Methods for obtaining path length include:

[0029] The coordinates of each node on the path are obtained from the GPS positioning system, the distance between adjacent coordinates is calculated, and the path length is obtained.

[0030] Furthermore, methods for obtaining other cost items include:

[0031] Based on the coordinates of all points, calculate the mean and standard deviation of the average distance; collect the raw point cloud datasets from the LiDAR and cameras, and calculate the average distance from the current coordinates to the neighbor coordinates based on the raw point cloud datasets; set a filtering threshold based on the mean and standard deviation of the average distance, and remove the raw point clouds with an average distance greater than the filtering threshold to obtain the filtered point cloud; perform ground point cloud filtering on the filtered point cloud to obtain the obstacle point cloud set;

[0032] Obstacle feature vectors are calculated based on the obstacle point cloud set;

[0033] The obstacle feature vector is used as input to the type recognition model to obtain the obstacle type; the cost coefficient corresponding to the obstacle type is obtained, and the product of the cost coefficient and the coverage area is calculated to obtain the low surface obstacle cost.

[0034] Methods for obtaining soil compaction costs include:

[0035] The soil compaction cost is calculated by weighting the soil moisture content and soil density measured by sensors.

[0036] Methods for obtaining slope costs include:

[0037] Collect DEM data from airborne lidar and satellites, calculate the slope value, and then calculate the slope cost based on the slope value.

[0038] Furthermore, methods for obtaining obstacle feature vectors include:

[0039] Based on the coordinates of the points in the obstacle point cloud set, calculate the difference between the maximum and minimum coordinate values ​​of each axis (X-axis, Y-axis, and Z-axis) to obtain the length, width, and height of the obstacle; calculate the projected area based on the convex hull polygon projected onto the horizontal plane, and calculate the ratio of the number of points in the obstacle point cloud set to the projected area to obtain the obstacle point density; calculate the mean Z-axis coordinate of all obstacle point clouds to obtain the mean obstacle height.

[0040] The obstacle feature vector is obtained by concatenating the length, width, height, point density, and average height of the obstacle.

[0041] The technical effects and advantages of the BeiDou RTK-based agricultural machinery automatic driving path planning method proposed in this invention are as follows:

[0042] This invention uses BeiDou RTK signal-to-noise ratio frequency domain characteristics to calculate the dielectric anomaly index, accurately identifying obstacles such as underground rocks and pipelines, effectively avoiding equipment overturning and chassis damage risks, and significantly improving the safety of agricultural machinery operations. Based on obstacle probability distribution and continuous area screening, it dynamically classifies risk levels, adjusts the cost coefficient of high-risk areas, and achieves proactive avoidance of high-risk areas, reducing the accident rate. It constructs a path cost model that integrates underground risks, surface obstacles, soil compaction, and slope costs, achieving optimal synergy between safety and efficiency, reducing operation interruption rates, and improving the efficiency of automated driving of agricultural machinery. Attached Figure Description

[0043] Figure 1 This is a schematic diagram of an agricultural machinery automatic driving path planning method based on Beidou RTK according to the present invention;

[0044] Figure 2 This is a flowchart of the method for obtaining path features according to the present invention;

[0045] Figure 3 This is a flowchart of the method for obtaining low surface obstacle costs according to the present invention;

[0046] Figure 4 This is a flowchart of the method for obtaining the minimum cost path according to the present invention. Detailed Implementation

[0047] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0048] Example 1:

[0049] Please see Figure 1 As shown in this embodiment, a method for automatic agricultural machinery path planning based on BeiDou RTK includes:

[0050] The system uses an onboard high-precision RTK receiver to collect raw observation data from BeiDou satellites; the raw observation data includes carrier phase observations. pseudorange observations Signal-to-noise ratio observations Agricultural machinery coordinates, satellite geometric elevation angle and satellite geometric azimuth The carrier phase observations in the original BeiDou RTK observation data are used to eliminate ionospheric errors through three-frequency geometric interpolation, improving observation accuracy. The pseudorange observations are combined with the repaired carrier phase sequence to calculate the combined observations, accurately characterizing the dielectric properties of the underground medium. The satellite geometric elevation angle and azimuth angle assist in correcting the signal propagation path and suppressing environmental interference. These data are fused with the signal-to-noise ratio observations to provide highly reliable input for the calculation of the dielectric anomaly index, enabling accurate detection of underground obstacles, significantly improving the underground obstacle identification capability of the risk assessment module, and providing a safety decision basis for the autonomous driving path planning of agricultural machinery.

[0051] It is important to emphasize that the above steps do not add any additional hardware, but only reuse the raw observation values ​​and signal-to-noise ratio data obtained by the Beidou RTK receiver already equipped on the agricultural machinery. Compared with the solution of using a dedicated underground detection sensor, this significantly reduces the system cost and integration complexity. To achieve this goal, this embodiment does not simply copy the existing soil moisture inversion method, but specially designs an algorithm chain consisting of three-frequency geometric combination, cycle slip repair, and path combination observation value calculation. This algorithm chain eliminates interference such as ionospheric delay and multipath effect to the greatest extent, and is specifically used to enhance the sensitivity of the signal to the abnormal dielectric properties of the underground medium, thereby providing a clean and reliable data foundation for subsequent underground obstacle identification.

[0052] Based on the original observation data, path combination observation values ​​are obtained; based on the path combination observation values, path characteristics are obtained; the path characteristics are analyzed with signal-to-noise ratio observation values, satellite geometric elevation angle, and satellite geometric azimuth angle to obtain the dielectric anomaly index.

[0053] Methods for obtaining path combination observations include:

[0054] Based on the carrier phase observations in the original observation data, the three-frequency non-geometric combination interpolation is calculated, such as the three-frequency non-geometric combination interpolation. ,in The carrier phase observation value at frequency B1. The carrier phase observation value at frequency B2 is used. Based on the absence of geometric interpolation for the three frequencies, the difference between the current three-frequency combination interpolation value and the previous three-frequency combination interpolation value is calculated to obtain the three-frequency combination difference. If the three-frequency combination difference is greater than the cycle slip detection threshold, it is determined to be a cycle slip. The cycle slip detection threshold is set according to the actual situation, preferably 0.7-0.9 cycles. A repair window of size 'a' and step size 'b' is set, where 'a' is preferably 5-10 observation times, and 'b' is preferably the time interval between adjacent interpolation points, matching the observation frequency. The carrier phase observation values ​​in the repair window are linearly interpolated to obtain the repaired carrier phase sequence, such as the satellite... Repaired carrier phase sequence ,in The starting observation time of the window. The end of the observation window. Satellite at the start of the observation window The corresponding carrier phase observation value, Satellite at the end of the observation window The corresponding carrier phase observation value is used; otherwise, the carrier phase observation value is directly used as the repaired carrier phase sequence. The absolute value of the difference between the repaired carrier phase sequence and the pseudorange observation value is calculated to obtain the path combination observation value. Under normal conditions without underground obstacles, the path combination observation value should be within a stable range. When underground obstacles exist, changes in dielectric properties will cause the propagation path deviation of the carrier phase sequence to increase, while the pseudorange observation value is less affected. At this time, the difference will significantly exceed the normal range, which is the physical basis for capturing underground medium anomalies. The path combination observation value is filtered with a filter window of c and a step size of d to obtain path characteristics. Among them, c and d match the observation frequency, and c is preferably 5-10 times the observation frequency; d is preferably 1-2 times the observation frequency. Figure 2 The invention eliminates ionospheric errors through three-frequency geometric-free interpolation, accurately characterizing the interference characteristics of underground media on satellite signals and providing high-precision input for signal-to-noise ratio (SNR) frequency domain feature calculation. After filtering, path features effectively suppress noise fluctuations, extracting stable signal propagation characteristics and ensuring improved sensitivity of the dielectric anomaly index to underground obstacles. These intermediate data form the core foundation for accurate underground risk detection, enabling the fusion of path combination observations and SNR frequency domain features to reliably distinguish between underground obstacles and environmental interference, providing highly reliable input for risk level classification and path planning.

[0055] Methods for obtaining the frequency domain characteristics of the signal-to-noise ratio include:

[0056] The signal-to-noise ratio (SNR) observations are processed by short-time Fourier transform to obtain the spectrum matrix. The frequency with the largest amplitude in the spectrum matrix is ​​taken as the main oscillation frequency, and the amplitude corresponding to the main oscillation frequency is obtained. The main oscillation frequency and the corresponding amplitude are concatenated to obtain the SNR frequency domain characteristics.

[0057] The path characteristics, signal-to-noise ratio frequency domain characteristics, agricultural machinery coordinates, satellite geometric elevation angle, and satellite geometric azimuth angle are used as inputs to the anomaly analysis model to obtain the dielectric anomaly index and its corresponding coordinates. When the dielectric anomaly index is greater than the obstacle probability threshold, it is determined that there is an obstacle; otherwise, it is determined that there is no obstacle and normal agricultural operations can be carried out. The obstacle probability threshold is set according to empirical values, and the dielectric anomaly index value of normal soil is stable in the range of 0.2-0.4, preferably 0.5-0.6.

[0058] Training methods for anomaly analysis models include:

[0059] Pre-collected B1 group of abnormal training data, including path features, signal-to-noise ratio frequency domain features, agricultural machinery coordinates, satellite geometric elevation angle and satellite geometric azimuth angle, as well as the corresponding dielectric anomaly index and corresponding coordinates.

[0060] The path features, signal-to-noise ratio frequency domain features, agricultural machinery coordinates, satellite geometric elevation angle, and satellite geometric azimuth angle are used as inputs to the anomaly analysis model. The corresponding dielectric anomaly index and its corresponding coordinates are used as outputs. The network parameters of the anomaly analysis model are iteratively optimized to minimize the error between the output dielectric anomaly index and its corresponding coordinates and the actual dielectric anomaly index and its corresponding coordinates. The network parameters that minimize the error between the output dielectric anomaly index and its corresponding coordinates and the actual dielectric anomaly index and its corresponding coordinates are obtained. The anomaly analysis model constructed with the corresponding network parameters is used as the trained anomaly analysis model.

[0061] High-precision detection of underground obstacles is achieved through frequency domain feature analysis of BeiDou RTK signal-to-noise ratio (SNR) observations. The system utilizes the dominant oscillation frequency and amplitude features extracted from the SNR observations via short-time Fourier transform, combined with path features and satellite geometric parameters to construct a multi-dimensional feature vector. This vector is then used to output a dielectric anomaly index through an anomaly analysis model, enabling accurate identification of underground obstacles. The dielectric anomaly index, as a core quantitative indicator, allows the system to accurately detect underground obstacles without additional sensors. This provides a highly reliable risk input for agricultural machinery autonomous driving path planning, effectively reducing accident rates and chassis damage, and significantly improving operational safety and efficiency.

[0062] Based on the dielectric anomaly index, the probability distribution value of the obstacle is calculated; based on the probability distribution value of the obstacle, the risk level is defined.

[0063] Methods for obtaining obstacle probability distribution values ​​include:

[0064] The dielectric anomaly index is weighted by Kriging interpolation to obtain the initial barrier probability distribution value.

[0065] Methods for obtaining the initial obstacle probability distribution values ​​include:

[0066] Using the boundary coordinates of the farmland operation area as the range, the farmland operation area is divided into 1m×1m grid cells. For each grid cell, the spatial correlation weight between the grid cell and the coordinates of the nearest H agricultural machinery is calculated. The spatial correlation weight is the reciprocal of the distance between the blank grid coordinates and the coordinates of the nearest H agricultural machinery. All spatial correlation weights are normalized to obtain the correlation weight. The dielectric anomaly index of all grid cells is weighted in combination with the correlation weight to obtain the initial obstacle probability distribution value.

[0067] The initial obstacle probability distribution value is normalized to obtain the obstacle probability distribution value. The obstacle probability distribution value of each grid cell is compared with a preset probability threshold. If the obstacle probability distribution value is greater than the preset probability threshold, it is determined to be a potential obstacle area; otherwise, it is determined to be a safe area. The preset probability threshold is obtained by combining historical data statistical analysis. Potential obstacle areas are clustered to obtain connected regions. The total number of grid cells in each connected region is counted and multiplied by the area of ​​a single grid cell to obtain a continuous area. The continuous area is compared with a preset minimum area. Regions with a continuous area greater than the preset minimum area are retained as candidate regions. The preset minimum area is obtained by combining historical data statistical analysis. The risk level is determined based on the obstacle probability distribution value of the candidate regions. When the obstacle probability distribution value of the candidate region... When the probability distribution value is greater than the probability threshold but not greater than 1, it is determined to be a high-risk area; when the obstacle probability distribution value of the candidate area is greater than half of the probability threshold but not greater than the probability threshold, it is determined to be a medium-risk area; when the obstacle probability distribution value of the candidate area is not greater than half of the probability threshold, it is determined to be a low-risk area. The probability threshold is set based on empirical values. The discrete dielectric anomaly index is fused into a continuous obstacle probability distribution through Kriging interpolation. After normalization, the probability value is ensured to be within the range of zero to one. Combined with the probability threshold and the minimum area dual screening, small-scale interference is eliminated, and high-risk, medium-risk and low-risk areas are accurately divided. This provides intelligent guidance for path planning, realizes forced avoidance of high-risk areas, optimized passage of medium-risk areas, and safe passage of low-risk areas, and significantly improves the safety and efficiency of agricultural machinery automatic driving operation.

[0068] The dielectric anomaly index, probability distribution, area screening, risk level, and differentiated cost coefficient proposed in this embodiment constitute a complete closed-loop quantitative strategy. It transforms the complex and continuous underground physical property inversion results into a finite number of discrete parameters that path planning algorithms can directly understand and process. This allows the system to achieve proactive and powerful avoidance in high-risk areas and a balance between efficiency and safety in medium- and low-risk areas without running complex multi-objective optimization algorithms, greatly improving the engineering practicality and computational efficiency of the entire method.

[0069] The total cost function of the path is constructed based on the risk level, and the minimum cost path is calculated with the goal of minimizing the total cost function. The total cost function of the path is obtained by adding the underground risk cost, the path length and other cost items. The other cost items are obtained by adding the low surface obstacle cost, soil compaction cost and slope cost.

[0070] Methods for constructing the total path cost function include:

[0071] The underground risk cost is obtained by multiplying the obstacle probability distribution value and the risk coefficient corresponding to the risk level. When the risk level is high, the corresponding risk coefficient is preferably 8.0, which can be adjusted according to the actual situation, but it is higher than the corresponding risk coefficient when the risk level is medium, to ensure that the route planning prioritizes avoiding high-risk areas. When the risk level is medium, the corresponding risk coefficient is preferably 4.0, which can be adjusted according to the actual situation. When the risk level is low, the corresponding risk coefficient is 0. The coordinates of each node on the route are obtained from the GPS positioning system, the distance between adjacent coordinates is calculated, and the route length is obtained.

[0072] Methods for obtaining other cost items include:

[0073] refer to Figure 3 Methods to achieve low surface obstacle costs include:

[0074] Based on the coordinates of all points, calculate the mean and standard deviation of the average distance; collect raw point cloud datasets from LiDAR and cameras, and calculate the average distance from the current coordinate to its neighboring coordinates based on the raw point cloud datasets; set a filtering threshold based on the mean and standard deviation of the average distance, and remove raw point clouds with average distances greater than the filtering threshold to obtain filtered point clouds; the filtering threshold is set according to the actual situation; perform ground point cloud filtering on the filtered point clouds: randomly select three filtered point clouds from the filtered point clouds to fit a temporary plane equation, calculate the point cloud distance from all filtered point clouds to the temporary plane equation, if the point cloud distance is small... If the initial threshold is met, the point is identified as an interior point; otherwise, it is identified as an exterior point. If the ratio of the number of interior points to the total number of filtered point clouds exceeds a preset ratio, the temporary plane equation is considered valid; otherwise, three filtered point clouds are selected to fit a new temporary plane equation. Using all points in the interior point set corresponding to the valid temporary plane equation, the plane equation is obtained by fitting using the least squares method. The plane distance between all exterior points corresponding to the valid temporary plane equation and the plane corresponding to the plane equation is calculated, and filtered point clouds with plane distances not less than a preset plane distance threshold are retained to obtain the obstacle point cloud set. The initial threshold is set based on empirical values, such as 0.3 meters.

[0075] Obstacle feature vectors are calculated based on the obstacle point cloud set.

[0076] Methods for obtaining obstacle feature vectors include:

[0077] Based on the coordinates of the points in the obstacle point cloud set, calculate the difference between the maximum and minimum coordinate values ​​on each of the X, Y, and Z axes to obtain the length, width, and height of the obstacle; based on the convex hull polygon projected onto the horizontal plane, calculate the projected area, such as the projected area... Where u is the number of vertices of the convex hull polygon. For the kth vertex coordinate, Let be the y-coordinate of the k-th vertex. For the (k+1)th vertex coordinate, Let y be the y-coordinate of the (k+1)th vertex; calculate the ratio of the number of points in the obstacle point cloud set to the projected area to obtain the obstacle point density; calculate the mean Z-axis coordinate of all obstacle point clouds to obtain the mean obstacle height.

[0078] The obstacle feature vector is obtained by concatenating the length, width, height, obstacle point density, and average height of obstacles. A total path cost function is constructed through multi-dimensional cost comprehensive calculation, enabling precise planning and safety optimization of the agricultural machinery autonomous driving path. Low-surface obstacle cost calculation employs point cloud data processing technology, filtering outliers using the mean and standard deviation of average distances. A ground separation algorithm distinguishes between ground and non-ground point clouds, accurately extracting the obstacle point cloud set and calculating the coverage area, effectively identifying surface obstacles such as shrubs, rocks, and garbage dumps. Underground risk costs are dynamically adjusted based on obstacle probability distribution values ​​and risk levels, applying a high-risk coefficient to high-risk areas and a medium-risk coefficient to medium-risk areas, achieving precise avoidance of underground obstacles. Path length is precisely calculated using GPS coordinates to determine the distance between nodes. Simultaneously, soil compaction costs and slope costs are integrated, comprehensively considering soil characteristics and terrain factors. This multi-dimensional cost calculation method enables path planning to intelligently identify and avoid high-risk underground obstacles and high-density surface obstacles, optimize path length, balance safety and efficiency, significantly improve the safety and efficiency of automated agricultural machinery operations, avoid equipment damage and operation interruptions, and realize intelligent and efficient precision agricultural operations.

[0079] The obstacle feature vector is used as input to the type recognition model to obtain the obstacle type; the cost coefficient corresponding to the obstacle type is obtained, and the product of the cost coefficient and the coverage area is calculated to obtain the low surface obstacle cost. Among them, when the obstacle type is a vegetation obstacle, the cost coefficient is preferably 1.0, and when the obstacle type is a geological obstacle, the cost coefficient is preferably 2.5; otherwise, the cost coefficient is 0.

[0080] Training methods for type recognition models include:

[0081] Type identification data for group B2 was collected in advance. The type identification data includes obstacle feature vectors and corresponding obstacle types.

[0082] The obstacle feature vector is used as the input of the type recognition model, and the corresponding obstacle type is used as the output of the type recognition model. With the goal of minimizing the error between the output obstacle type and the actual obstacle type, the network parameters of the type recognition model are iteratively optimized to obtain the network parameters that minimize the error between the output obstacle type and the actual obstacle type. The type recognition model constructed with the corresponding network parameters is used as the trained type recognition model.

[0083] Methods for obtaining soil compaction costs include:

[0084] The soil compaction cost is calculated by weighting the soil moisture content and soil density measured by the sensor. The weighting coefficients are set based on empirical values, with the soil weight coefficient preferably being 0.8 and the soil moisture content weight coefficient preferably being 0.2.

[0085] Methods for obtaining slope costs include:

[0086] Collect DEM data from airborne lidar and satellites to calculate slope values, such as slope values. ,in It is the arctangent function. For elevation difference, , The horizontal distance is used as the basis for calculation. Slope cost is calculated based on the slope value. If the slope value is below a slope threshold, the slope cost is 0. Otherwise, the slope difference between the slope value and the slope threshold is calculated, and then the product of the slope difference and the slope coefficient is calculated to obtain the slope cost. If this is the case, the slope has no impact on the cost; otherwise, a cost is incurred. The slope threshold is set based on empirical values, preferably 15°. The costs of low surface obstacles, soil compaction, and slope are added together to obtain other cost items. The total path cost function is obtained by weighting the underground risk cost, path length, and other cost items. Based on the total path cost function, the minimum cost path is calculated with the goal of minimizing the total path cost function. (Reference) Figure 4 The system uses the minimum cost path as the optimal route planning; it accurately identifies obstacle types using a random forest classifier and dynamically calculates the cost of low-surface obstacles based on coverage area; it calculates soil compaction costs based on soil moisture content and density, assesses slope costs based on slope values, and integrates these cost items with underground risk costs and path length into a total path cost function; it intelligently calculates the minimum cost path, effectively avoiding high-risk underground obstacles, high-density surface obstacles, and steep slope areas, optimizing path length and operational efficiency, significantly improving the safety, accuracy, and overall efficiency of agricultural machinery autopilot, avoiding equipment damage and operational interruptions, and achieving intelligent, safe, and efficient agricultural operations.

[0087] This embodiment, for the first time, achieves systematic and coordinated optimization of four key factors in agricultural machinery autonomous driving path planning: underground risk fields, surface obstacles, soil compaction characteristics, and terrain slope. Existing technologies mostly focus on single types, especially the avoidance of surface obstacles. This embodiment, however, achieves a paradigm shift from localized, single-source obstacle avoidance to global, multi-source risk-coordinated planning by constructing a unified, multi-dimensional total path cost function. This integrated planning can simultaneously prevent equipment from overturning due to hidden underground obstacles, avoid collisions with surface objects, reduce damage to crops from excessive soil compaction, and control the risks of operating on steep slopes, thereby significantly improving the safety, efficiency, and agronomic rationality of operations at the global level.

[0088] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0089] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for path planning of automated agricultural machinery based on BeiDou RTK, characterized in that, include: Collect raw observation data from BeiDou satellites; The raw observation data includes carrier phase observations, pseudorange observations, signal-to-noise ratio observations, agricultural machinery coordinates, satellite geometric elevation angle, and satellite geometric azimuth angle; Path characteristics are calculated based on carrier phase observations and pseudorange observations; The signal-to-noise ratio (SNR) observations are processed by short-time Fourier transform, and the SNR frequency domain characteristics are obtained through analysis and calculation. The path characteristics, SNR frequency domain characteristics, agricultural machinery coordinates, satellite geometric elevation angle, and satellite geometric azimuth angle are used as inputs to the anomaly analysis model to obtain the dielectric anomaly index and its corresponding coordinates. The presence of obstacles is determined based on the dielectric anomaly index. When obstacles are present, Kriging interpolation is performed based on the dielectric anomaly index and the corresponding coordinates to obtain the obstacle probability distribution value of the farmland operation area; the obstacle probability distribution value of the farmland operation area is analyzed and clustered to obtain candidate regions. The risk level is determined based on the obstacle probability distribution values ​​of the candidate region; Construct a total path cost function based on the risk level, and calculate the minimum cost path with the goal of minimizing the total path cost function. The minimum cost path is used as the planning path for the automatic driving of agricultural machinery.

2. The method for agricultural machinery automatic driving path planning based on Beidou RTK according to claim 1, characterized in that, Methods for obtaining path combination observations include: Based on the carrier phase observations in the original observation data, the three-frequency geometric combination interpolation is calculated. The difference between the current three-frequency combination interpolation and the previous three-frequency combination interpolation is calculated to obtain the three-frequency combination difference. If the three-frequency combination difference is greater than the cycle slip detection threshold, it is determined to be a cycle slip. The carrier signal observations are repaired based on the repair window to obtain the repaired carrier phase sequence. Otherwise, the carrier phase observations are directly used as the repaired carrier phase sequence. The absolute value of the difference between the repaired carrier phase sequence and the pseudorange observations is calculated to obtain the path combination observations.

3. The method for agricultural machinery automatic driving path planning based on Beidou RTK according to claim 1, characterized in that, Methods for obtaining the frequency domain characteristics of the signal-to-noise ratio include: The signal-to-noise ratio (SNR) observations are processed by short-time Fourier transform to obtain the spectrum matrix. The frequency with the largest amplitude in the spectrum matrix is ​​taken as the main oscillation frequency, and the amplitude corresponding to the main oscillation frequency is obtained. The main oscillation frequency and the corresponding amplitude are concatenated to obtain the SNR frequency domain characteristics.

4. The method for agricultural machinery automatic driving path planning based on Beidou RTK according to claim 1, characterized in that, Methods for determining the presence of obstacles based on dielectric anomaly indices include: When the dielectric anomaly index is greater than the obstacle probability threshold, an obstacle is determined to exist; otherwise, no obstacle is determined to exist.

5. The method for agricultural machinery automatic driving path planning based on Beidou RTK according to claim 1, characterized in that, Methods for obtaining a risk level include: The dielectric anomaly index is weighted by Kriging interpolation to obtain the initial barrier probability distribution value. The initial obstacle probability distribution value is normalized to obtain the obstacle probability distribution value. The obstacle probability distribution value of each grid cell is compared with a preset probability threshold. If the obstacle probability distribution value is greater than the preset probability threshold, it is determined as a potential obstacle area; otherwise, it is determined as a safe area. The potential obstacle areas are clustered to obtain connected regions. The total number of grid cells in each connected region is counted and multiplied by the area of ​​a single grid cell to obtain a continuous area. The continuous area is compared with a preset minimum area, and regions with a continuous area greater than the preset minimum area are retained as candidate regions. The risk level is determined based on the obstacle probability distribution value of the candidate regions.

6. The method for agricultural machinery automatic driving path planning based on Beidou RTK according to claim 5, characterized in that, Methods for obtaining the initial obstacle probability distribution values ​​include: Using the boundary coordinates of the farmland operation area as the range, the farmland operation area is divided into 1m×1m grid cells. For each grid cell, the spatial correlation weight between the grid cell and the coordinates of the nearest H agricultural machinery is calculated. The spatial correlation weight is the reciprocal of the distance between the blank grid coordinates and the coordinates of the nearest H agricultural machinery. All spatial correlation weights are normalized to obtain the correlation weight. The dielectric anomaly index of all grid cells is weighted in combination with the correlation weight to obtain the initial obstacle probability distribution value.

7. The method for agricultural machinery automatic driving path planning based on Beidou RTK according to claim 6, characterized in that, Methods for determining risk levels include: When the obstacle probability distribution value of a candidate region is greater than the probability threshold but not greater than 1, it is determined to be a high-risk region; when the obstacle probability distribution value of a candidate region is greater than half of the probability threshold but not greater than the probability threshold, it is determined to be a medium-risk region; when the obstacle probability distribution value of a candidate region is not greater than half of the probability threshold, it is determined to be a low-risk region.

8. The method for agricultural machinery automatic driving path planning based on Beidou RTK according to claim 1, characterized in that, The total cost function of the path is obtained by adding the underground risk cost, the path length, and other cost items; The other cost items are obtained by adding the costs of low surface obstacles, soil compaction, and slope. Methods for obtaining underground risk costs include: The underground risk cost is obtained by multiplying the obstacle probability distribution value and the risk coefficient corresponding to the risk level. Methods for obtaining path length include: The coordinates of each node on the path are obtained from the GPS positioning system, the distance between adjacent coordinates is calculated, and the path length is obtained.

9. A method for agricultural machinery automatic driving path planning based on Beidou RTK according to claim 8, characterized in that, Methods for obtaining other cost items include: Based on the coordinates of all points, calculate the mean and standard deviation of the average distance; collect the raw point cloud datasets from the LiDAR and cameras, and calculate the average distance from the current coordinates to the neighbor coordinates based on the raw point cloud datasets; set a filtering threshold based on the mean and standard deviation of the average distance, and remove the raw point clouds with an average distance greater than the filtering threshold to obtain the filtered point cloud; perform ground point cloud filtering on the filtered point cloud to obtain the obstacle point cloud set; Obstacle feature vectors are calculated based on the obstacle point cloud set; The obstacle feature vector is used as input to the type recognition model to obtain the obstacle type; the cost coefficient corresponding to the obstacle type is obtained, and the product of the cost coefficient and the coverage area is calculated to obtain the low surface obstacle cost. Methods for obtaining soil compaction costs include: The soil compaction cost is calculated by weighting the soil moisture content and soil density measured by sensors. Methods for obtaining slope costs include: Collect DEM data from airborne lidar and satellites, calculate the slope value, and then calculate the slope cost based on the slope value.

10. A method for agricultural machinery automatic driving path planning based on Beidou RTK according to claim 9, characterized in that, Methods for obtaining obstacle feature vectors include: Based on the coordinates of the points in the obstacle point cloud set, calculate the difference between the maximum and minimum coordinate values ​​of each axis (X-axis, Y-axis, and Z-axis) to obtain the length, width, and height of the obstacle; calculate the projected area based on the convex hull polygon projected onto the horizontal plane, and calculate the ratio of the number of points in the obstacle point cloud set to the projected area to obtain the obstacle point density; calculate the mean Z-axis coordinate of all obstacle point clouds to obtain the mean obstacle height. The obstacle feature vector is obtained by concatenating the length, width, height, point density, and average height of the obstacle.

Images