Cotton field robot fusion navigation path generation method

Abstract

The invention discloses a cotton field robot fusion navigation path generation method, and relates to the field of agricultural robot positioning navigation, and the method comprises the steps: 1, installing an RTK positioning module on a cotton field robot for position navigation; a deep learning visual camera is installed in front of the cotton field robot and used for recognizing and keeping lane lines, and visual navigation is achieved; a walking device, a turning device and an in-situ turning device are installed on the cotton field robot and used for achieving the control effect of walking, turning and in-situ turning. 2, constructing a visual navigation model based on semantic segmentation, and deploying the visual navigation model to an automatic driving system of the cotton field robot; step 3, dotting and marking land parcel information by the RTK dotting device, and importing the land parcel information into an automatic driving system of the cotton field robot; and step 4, starting the cotton field robot to walk and work in the sown cotton field, generating a fusion navigation path, and synchronously recording RTK position track coordinates. According to the invention, an operation path can be automatically planned on a sown standard cotton field.

Description

Technical Field

[0001] The present invention relates to the field of agricultural robot positioning and navigation, and specifically to a method for generating a fused navigation path for a cotton field robot. Background Art

[0002] Existing agricultural machinery path planning methods are mainly used in scenarios of farming and harvesting tractors and plant protection drones; the operation path planning of tractors, mainly used for farming or harvesting, is based on the premise that the plot is blank without crops or the crops are already mature, and there is no need to consider whether the operation path will cause damage to the crops; while the plant protection drone operates in the air, and the planning of its operation path also does not need to consider the crop information in the plot. In the above two cases, only the plot outline needs to be marked to carry out path planning.

[0003] Lane line detection can be divided into traditional visual lane line detection and deep learning-based lane line detection. Most traditional lane line detections are based on geometric methods and use some curve models to fit lane lines. In recent years, with the rapid development of convolutional neural networks, deep learning-based lane line detection methods have gradually become a research hotspot.

[0004] Especially the lane line recognition method based on semantic segmentation, by classifying each pixel of the scene picture to determine whether it belongs to the lane line or the background, realizes more accurate and detailed lane line recognition. This method uses a deep learning model to perform pixel-level classification tasks on images, and can effectively cope with complex cotton field environments and changing lighting conditions, providing reliable environmental perception capabilities for the agricultural machinery automatic driving system.

[0005] Cotton field lane line recognition is an important task for the environmental perception of agricultural machinery automatic driving. Accurately and quickly detecting the lane line is the premise for the safe and accurate operation of cotton field robots. When using cotton field robots for automated management on a sown standard cotton field, the operation path planning is based on the sown state in the plot. The cotton field robot needs to accurately walk on the blank tractor path, otherwise it will damage the seedlings. Therefore, it is impossible to use the path planning methods similar to traditional tractors and plant protection drones. The specific reasons are as follows:

[0006] 1) Since the standard cotton field plot is large, the workload of completely using RTK position marking for the operation path is large, and the more human participation, the more uncertainties will be generated, and it is impossible to ensure that the cotton field robot can operate safely according to the manually marked path.

[0007] 2) Based on RTK position navigation, due to the unfixed ridge spacing in cotton field planting, which is large or small, and the sowing operation path of the sowing tractor is not straight and the planting film is curved, the cotton field robot cannot simply use RTK position navigation to plan the operation path like traditional agricultural machinery.

[0008] 3) Based on visual navigation, when the cotton field robot changes rows during operation, it is unable to determine a reasonable U-turn and row-changing position, cannot identify the U-turn orientation, and there will be a blank area between the U-turn and row-changing position and the crops, where the lane lines cannot be recognized. Therefore, the cotton field robot cannot rely entirely on visual navigation to walk.

[0009] In summary, the path planning methods of traditional seeding and harvesting tractors and plant protection UAVs, as well as simply relying on RTK position navigation or visual navigation, cannot support the path planning of cotton field robots in the sown fields. Summary of the Invention

[0010] Object of the Invention: The object of the present invention is to provide a method for generating a fusion navigation path for a cotton field robot in view of the deficiencies of the prior art. The method of the present invention is a method for generating a fusion navigation path for a cotton field robot based on RTK positioning and deep learning vision in a sown cotton field.

[0011] Technical Solution: A method for generating a fusion navigation path for a cotton field robot includes the following steps:

[0012] Step 1: Install an RTK positioning module on the cotton field robot for position navigation; install a deep learning vision camera in front of the cotton field robot for identifying and maintaining lane lines to achieve visual navigation; install a walking, turning, and U-turn device on the cotton field robot for achieving the control effects of walking, turning, and U-turn.

[0013] Step 2: Construct a visual navigation model based on semantic segmentation and deploy the visual navigation model to the autonomous driving system of the cotton field robot.

[0014] Step 3: Use an RTK marker to mark the plot information and import the plot information into the autonomous driving system of the cotton field robot.

[0015] Step 4: Start the cotton field robot to walk and operate in the sown cotton field, generate a fusion navigation path, and synchronously record the RTK position trajectory coordinates.

[0016] Furthermore, the specific operation of Step 2 is as follows:

[0017] 1) Data collection and preprocessing: Collect an image data set containing the cotton field environment and lane lines, and then perform image enhancement and data annotation operations on the data set.

[0018] 2) Model selection and establishment: According to the characteristics of the scene of walking along the tractor road inside the sown cotton field and the task of lane line recognition, select a suitable deep learning model architecture, and construct a visual navigation model for a lane line recognition model based on semantic segmentation, including defining the network structure and selecting the loss function.

[0019] 3) Dataset division and training preparation: Divide the dataset into a training set, a validation set, and a test set; perform image normalization and data augmentation operations on the dataset;

[0020] 4) Model training and tuning: Use the training set data to train the constructed visual navigation model, and perform model tuning and parameter optimization according to the performance on the validation set;

[0021] 5) Model evaluation and detection: Use the test set data to evaluate and detect the trained visual navigation model; analyze the performance and accuracy of the model based on the evaluation results, and further optimize and improve the model;

[0022] 6) Deployment and application: Deploy the visual navigation model after evaluation and detection to the cotton field autonomous driving system for real-time lane line detection and recognition, monitor the performance of the model, and update and optimize it.

[0023] Furthermore, in step three, on the plot to be operated, use an RTK dotting device to dot and mark the relevant GPS position coordinate information of the plot and save it; the steps for marking the plot information are as follows:

[0024] 1) Mark the starting point outside the top center of the first planting film;

[0025] 2) Mark two turning lines at both ends of the planting film where the cotton field robot can turn around and change rows. The turning lines must ensure that there are intersections with the midlines of all planting films;

[0026] 3) Mark a point every 50m in the middle of the first planting film at intervals from the starting point, mark two points. Based on the starting point, use the three-point connection to calculate the azimuth angle based on true north to represent the operation direction;

[0027] 4) After marking is completed, import the plot information from the dotting device into the autonomous driving system of the cotton field robot.

[0028] Furthermore, step four includes the following operations:

[0029] Step S1: Define the turning line at one end of the starting point as LFL1, and the turning line at the other end as LFL2; define the row labels of the planting films as CS1, CS2......CSn, the starting point of the row as CSnA, and the ending point of the row as CSnB; define the operation turning direction of the cotton field robot as D_LR; define the turning distance as D_LF; define the operation direction as WY;

[0030] Step S2: Start the cotton field robot, move the cotton field robot to the operation starting point, adjust the direction of the cotton field robot to face CS1, select the subsequent operation D_LR, set D_LF, start the task, and enter the fusion navigation path generation state;

[0031] Step S3: From the starting point position, draw a ray along WY to find the intersection point of the ray and LFL2, which is the point coordinate of CS1B.

[0032] Step S4: If the lane lines are not recognized by visual navigation, starting from CS1A, the cotton field robot travels towards CS1B relying on RTK position navigation.

[0033] Step S5: During the driving process, when the lane lines are recognized by visual navigation, it travels relying on visual navigation.

[0034] Step S6: During the driving process, when driving straight, record the point coordinates once every 10 m. If the driving trajectory deviates from the straight line, record the point coordinates whenever the lateral offset exceeds 3 cm. Fit the recorded point coordinates into a linear binary equation using the least squares method, and use this equation to find the intersection points of the current driving path line with LFL1 and LFL2, continuously correcting the point coordinates of CS1A and CS1B of CS1.

[0035] Step S7: When the cotton field robot travels along CS1 to a position where the lane lines cannot be recognized by visual navigation, starting from CS1A, it continues to travel towards CS1B relying on RTK position navigation until it reaches CS1B.

[0036] Step S8: From CS1B, add a D_LF along the lane change line LFL2 towards D_LR. If the obtained point is on LFL2, it is the point coordinate of CS2A.

[0037] Step S9: Based on CS2A, draw a ray along the reverse of WY to find the intersection point of the ray and LFL1, which is the point coordinate of CS2B.

[0038] Step S10: The cotton field robot makes a U-turn in place and turns towards CS2A.

[0039] Step S11: Starting from CS1B, continue to travel towards CS2A relying on RTK position navigation until it reaches CS2A.

[0040] Step S12: The cotton field robot makes a U-turn in place and turns towards CS2B. If the lane lines are not recognized by visual navigation at this time, starting from CS2A, the cotton field robot travels towards CS2B relying on RTK position navigation. During the driving process, when the lane lines are recognized by visual navigation, it travels relying on visual navigation.

[0041] Step S13: Repeat the above process until the point obtained by adding a D_LF along the lane change line towards D_LR is not on the lane change line, then end.

[0042] Step S14: During the above driving process, the trajectory coordinate points are continuously recorded to generate a path that can independently rely on RTK position navigation.

[0043] Further, in step S4, the RTK position navigation steps are as follows:

[0044] a), Obtain the real-time position coordinate information of the cotton field robot through the vehicle-mounted RTK positioning module, take CS1A as the starting point and CS1B as the ending point, and calculate the transverse difference in combination with the real-time position; the transverse difference refers to the yaw lateral distance;

[0045] b), Substitute the transverse difference into the PID control algorithm to output the expected turning angle, thereby controlling the driving trajectory of the cotton field robot.

[0046] Further, in step S5, the vision navigation steps are as follows:

[0047] a), Use the trained deep learning vision model to perform real-time recognition on the camera data;

[0048] b), When the lane lines are recognized, first find out the two lane lines in the middle of the planting film, and calculate the intercept on the x-axis of the intersection point formed with the points at the center of the lower edge of the image in the x, y coordinate system;

[0049] c), Determine the transverse difference based on the intercept;

[0050] d), Substitute the transverse difference into the PID control algorithm to output the expected turning angle, thereby controlling the driving trajectory of the cotton field robot.

[0051] Further, in step S10, the in-situ turning steps are as follows:

[0052] a), Take CS1B as the starting point and CS2A as the ending point, and calculate the target heading angle;

[0053] b), Obtain the real-time heading angle of the cotton field robot through the vehicle-mounted RTK positioning module;

[0054] c), Control the four wheels of the cotton field robot to change from a square distribution to a circular distribution;

[0055] d), Control the cotton field robot to turn in place, and stop after turning until the real-time heading angle is equal to the target heading angle;

[0056] e), Control the four wheels of the cotton field robot to change from a circular distribution to a square distribution, and the in-situ turning is completed.

[0057] Beneficial effects:

[0058] 1) The present invention can realize automatic planning of operation paths on a sown standard cotton field;

[0059] 2) The present invention realizes turning and changing rows by integrating RTK position navigation at the turning and changing row positions where vision navigation cannot recognize;

[0060] 3) In the scenario where the planting ridge spacing in the cotton field crop belt is inconsistent and the planting film is bent, the present invention combines RTK position navigation and visual navigation to achieve precise driving along the planting film;

[0061] 4) Since the visual system can no longer identify the lane lines after the cotton seedlings grow tall and form a continuous area, this method creates an operation path that only relies on RTK position navigation for subsequent re-operation;

[0062] 5) Reduce the workload of manual path marking, improve efficiency, and reduce uncertainty at the same time. Description of the Drawings

[0063] Figure 1 It is a flow chart of the method for generating a fusion navigation path of the cotton field robot of the present invention;

[0064] Figure 2 It is the route recognized on the cotton field planting film;

[0065] Figure 3 It is the result after smoothing the route recognized on the cotton field planting film (used to generate lane lines);

[0066] Figure 4 It is the identification of the plot information and the generated fusion navigation path in the embodiment of the present invention.

[0067] Among them, 1. Cotton field robot; 2. Planting film; 3. Tractor road; 4. Navigation operation route; 5. Route change line; 6. Starting point; 7. Lane line. Detailed Embodiment

[0068] The technical solution of the present invention will be described in detail below with reference to the drawings, but the protection scope of the present invention is not limited to the described embodiments.

[0069] As Figure 1 shown, a method for generating a fusion navigation path of a cotton field robot, the method generates a path based on RTK positioning and deep learning vision method; includes the following steps:

[0070] Step 1. Install an RTK positioning module on the cotton field robot to achieve centimeter-level positioning for position navigation; install a deep learning vision camera in front of the cotton field robot to identify and maintain lane lines for visual navigation; install walking, turning and in-situ turning devices on the cotton field robot to achieve walking, turning and in-situ turning control effects.

[0071] Prepare an RTK marker for marking the plot position coordinate information; RTK (Real-time kinematic) refers to real-time kinematic carrier phase differential technology; the RTK marker is a positioning device that relies on RTK technology to obtain centimeter-level accuracy longitude and latitude coordinate information.

[0072] Step 2: Construct a visual navigation model based on semantic segmentation and deploy the visual navigation model to the automatic driving system of the cotton field robot;

[0073] In the present invention, the visual navigation model is designed, established, and trained and detected based on a lane detection model for semantic segmentation. The specific construction and deployment application operations are as follows:

[0074] 1) Data collection and preprocessing: First, collect an image data set containing specific targets such as the cotton field environment and lane lines; then clean and preprocess the data set, including operations such as image enhancement and data annotation, to improve the data quality and the training effect of the model.

[0075] 2) Model selection and establishment: According to specific requirements and application scenarios (the scenario of walking along the tractor road in the sown cotton field), select a suitable deep learning model architecture, such as a convolutional neural network (CNN) or its variant. According to the task characteristics of lane line recognition, construct a visual navigation model for a lane line recognition model based on semantic segmentation, including defining the network structure, selecting a loss function, etc.

[0076] 3) Data set division and training preparation: Divide the data set into a training set, a validation set, and a test set. Further process the data set, such as operations like image normalization and data augmentation, to increase the diversity and quantity of the data and enhance the robustness of the model.

[0077] 4) Model training and tuning: Use the training set data to train the constructed visual navigation model, and perform model tuning and parameter optimization according to the performance of the validation set. Through repeated iterative training and validation, adjust the parameters and hyperparameters of the model to improve the generalization ability and accuracy of the model.

[0078] 5) Model evaluation and detection: Use the test set data to evaluate and detect the trained model. According to the evaluation results, analyze the performance and accuracy of the model and further optimize and improve the model.

[0079] 6) Deployment and application: Deploy the visual navigation model after model evaluation and detection to the actual cotton field automatic driving system for real-time lane line detection and recognition. Continuously monitor the performance of the model and update and optimize it to ensure the stability and reliability of the model in actual applications.

[0080] Semantic segmentation is an important image analysis technology in the field of computer vision, and its main goal is to classify each pixel in an image into one of the pre-defined categories.

[0081] Step 3: Mark the plot information by dotting and import the plot information into the automatic driving system of the cotton field robot;

[0082] On the plot to be operated, use an RTK dotter to dot and mark the relevant GPS position coordinate information of the plot and save it; as Figure 4 shown, the steps of marking the plot information are as follows:

[0083] 1) Mark the starting point 6 outside the top center of the first planting film 2;

[0084] The planting film refers to a film of cotton crops in a standard cotton field; the machine - tillage road 3 is between adjacent planting films 2;

[0085] 2) Mark two turning lines 5 at both ends of the planting film 2 where the cotton field robot 1 can turn around and change rows. The turning lines 5 must ensure that there are intersections with the mid - lines of all planting films 2;

[0086] The turning line (LF - line) refers to the passing route for the cotton field robot to turn around and change rows at the field head;

[0087] 3) Mark a point every 50 m from the starting point in the middle of the first planting film 2, mark two points. Based on the starting point, use the three - point connection to calculate the azimuth angle based on true north to represent the operation direction;

[0088] 4) After marking, import the plot information from the dotter into the automatic driving system of the cotton field robot.

[0089] Step 4: Start the cotton field robot to walk and operate in the sown cotton field, generate a fused navigation path, and synchronously record the RTK position trajectory coordinates; the specific steps are as follows:

[0090] Step S1: Parameter definition: Define the turning line at one end of the starting point as LFL1, and the turning line at the other end as LFL2; define the row labels of the planting films as CS1, CS2......CSn, the starting point of the row as CSnA, and the ending point of the row as CSnB; define the row - changing direction of the cotton field robot operation as D_LR (left shift / right shift); define the row - changing distance as D_LF; define the operation direction as WY;

[0091] Step S2: Start the cotton field robot, move the cotton field robot to the operation starting point, adjust the direction of the cotton field robot to face CS1, select the subsequent operation D_LR, set D_LF, start the task, and enter the fused navigation path generation state;

[0092] Step S3: From the starting point position, make a ray along WY, and find the intersection point of the ray and LFL2, that is, the point coordinates of CS1B;

[0093] Step S4: If the lane lines are not recognized by visual navigation at this time, starting from CS1A, the cotton field robot travels towards CS1B relying on RTK position navigation. The RTK position navigation steps are as follows:

[0094] a) Obtain the real-time position coordinate information of the cotton field robot through the on-vehicle RTK positioning module. Starting from CS1A and ending at CS1B, calculate the cross difference in combination with the real-time position. The cross difference refers to the lateral distance of yaw.

[0095] b) Substitute the cross difference into the PID control algorithm to output the expected turning angle, thereby controlling the driving trajectory of the cotton field robot.

[0096] Step S5: During the driving process, when the lane lines are recognized by visual navigation, rely on visual navigation to drive. The visual navigation steps are as follows:

[0097] a) Use the trained deep learning visual model to perform real-time recognition on the camera data.

[0098] b) When the lane lines are recognized, first find the two lane lines in the middle of the planting film, and calculate the intercept on the x-axis of the intersection point formed with the point at the center of the lower edge of the image in the x, y coordinate system.

[0099] c) Determine the cross difference based on the intercept.

[0100] d) Substitute the cross difference into the PID control algorithm to output the expected turning angle, thereby controlling the driving trajectory of the cotton field robot.

[0101] Step S6: During the driving process, when driving straight, record the point coordinates once every 10 m. If the driving trajectory deviates from the straight line, record the point coordinates whenever the lateral offset exceeds 3 cm. Fit the recorded point coordinates into a linear binary equation using the least squares method, and use this equation to find the intersection points of the current driving path line with LFL1 and LFL2, and continuously correct the point coordinates of CS1A and CS1B of CS1.

[0102] Step S7: When the cotton field robot travels along CS1 to a position where the lane lines cannot be recognized by visual navigation, starting from CS1A, continue to rely on RTK position navigation to travel towards CS1B until reaching CS1B.

[0103] Step S8: Along the switching route LFL2 from CS1B, add a D_LF towards D_LR. If the obtained point is on LFL2, it is the point coordinate of CS2A.

[0104] Step S9: Based on CS2A, make a ray along the reverse of WY, and find the intersection point of the ray and LFL1, which is the point coordinate of CS2B.

[0105] Step S10: The cotton field robot turns around in place and steers towards CS2A. The steps for turning around in place are as follows:

[0106] a) Calculate the target heading angle with CS1B as the starting point and CS2A as the ending point.

[0107] b) Obtain the real-time heading angle of the cotton field robot through the on-vehicle RTK positioning module.

[0108] c) Control the four wheels of the cotton field robot to change from a square distribution to a circular distribution.

[0109] d) Control the cotton field robot to turn in place and stop after the real-time heading angle is equal to the target heading angle.

[0110] e) Control the four wheels of the cotton field robot to change from a circular distribution to a square distribution, and the in-place turning is completed.

[0111] Step S11: Starting from CS1B, continue to drive towards CS2A relying on RTK position navigation until reaching CS2A.

[0112] Step S12: The cotton field robot turns around in place and steers towards CS2B. If the lane lines are not recognized by visual navigation at this time, starting from CS2A, the cotton field robot drives towards CS2B relying on RTK position navigation. During the driving process, when the lane lines are recognized by visual navigation, it drives relying on visual navigation.

[0113] Step S13: Repeat the above process until the point obtained by adding a D_LF to D_LR along the lane change line is not on the lane change line, then end.

[0114] Step S14: During the above driving process, the trajectory coordinate points are continuously recorded to generate a path that can independently rely on RTK position navigation.

[0115] As described above, although the present invention has been shown and described with reference to specific preferred embodiments, it should not be construed as a limitation of the present invention itself. Various changes can be made in its form and details without departing from the spirit and scope of the present invention defined by the appended claims.

Claims

1. A method for generating a fusion navigation path of a cotton field robot, characterized in that, It includes the following steps: Step 1: Install an RTK positioning module on the cotton field robot for position navigation; install a deep learning vision camera in front of the cotton field robot for identifying and maintaining the lane line to achieve visual navigation; install a walking, turning and in-place turning device on the cotton field robot for achieving walking, turning and in-place turning control effects; Step 2: Construct a visual navigation model based on semantic segmentation and deploy the visual navigation model into the autonomous driving system of the cotton field robot; Step 3: Use an RTK dotter to mark the plot information and import the plot information into the autonomous driving system of the cotton field robot; Step 4: Start the cotton field robot to walk and operate in the sown cotton field, generate a fused navigation path, and synchronously record the RTK position trajectory coordinates.

2. The method for generating a fusion navigation path of a cotton field robot according to claim 1, wherein, The specific operations of the said Step 2 are as follows: 1) Data collection and preprocessing: Collect an image data set containing the cotton field environment and lane lines, and then perform image enhancement and data annotation operations on the data set; 2) Model selection and establishment: According to the scene of walking along the tractor road in the sown cotton field and the task characteristics of lane line recognition, select a suitable deep learning model architecture, and construct a visual navigation model of a lane line recognition model based on semantic segmentation, including defining the network structure and selecting the loss function; 3) Data set division and training preparation: Divide the data set into a training set, a validation set and a test set; Perform image normalization and data augmentation operations on the data set; 4) Model training and tuning: Use the training set data to train the constructed visual navigation model, and perform model tuning and parameter optimization according to the performance of the validation set; 5) Model evaluation and detection: Use the test set data to evaluate and detect the trained visual navigation model; according to the evaluation results, analyze the performance and accuracy of the model, and further optimize and improve the model; 6) Deployment and application: Deploy the visual navigation model after model evaluation and detection into the cotton field autonomous driving system for real-time lane line detection and recognition, monitor the performance of the model and perform updates and optimizations.

3. A method for generating a fused navigation path of a cotton field robot according to claim 1, characterized in that, In the said Step 3, on the plot to be operated, use an RTK dotter to mark and save the relevant GPS position coordinate information of the plot; the steps of plot information marking are as follows: 1) Mark the starting point outside the top center of the first planting film; 2) Mark two turning lines at the two ends of the planting film where the cotton field robot can turn around and change rows. The turning lines must ensure that there are intersections with the midlines of all planting films; 3) Mark one point every 50 m at the middle of the first planting film at intervals from the starting point, mark two points. Based on the starting point, calculate the azimuth angle based on true north with the three-point connection line to represent the operation direction; 4) After marking is completed, import the plot information from the dotter into the autonomous driving system of the cotton field robot.

4. A method for generating a fused navigation path of a cotton field robot according to claim 1, characterized in that, The said Step 4 includes the following operations: Step S1: Define the lane-changing route at one end of the starting point as LFL1 and the lane-changing route at the other end as LFL2; define the row labels of the planting film as CS1, CS2......CSn, the starting point of the row as CSnA, and the ending point of the row as CSnB; define the lane-changing direction of the cotton field robot operation as D_LR; define the lane-changing distance as D_LF; define the operation direction as WY. Step S2: Start the cotton field robot, move the cotton field robot to the operation starting point, adjust the direction of the cotton field robot to face CS1, select the subsequent operation D_LR, set D_LF, start the task, and enter the state of generating a fused navigation path. Step S3: From the starting point position, draw a ray along WY to obtain the intersection point of the ray and LFL2, that is, the point coordinates of CS1B. Step S4: If the lane lines are not recognized by visual navigation, starting from CS1A, the cotton field robot relies on RTK position navigation to drive towards CS1B. Step S5: During the driving process, when the lane lines are recognized by visual navigation, rely on visual navigation to drive. Step S6: During the driving process, when driving straight, record the point coordinates once every 10m. If the driving trajectory deviates from the straight line, record the point coordinates whenever the lateral deviation exceeds 3cm; use the least squares method to fit the recorded point coordinates into a linear binary equation, and use this equation to find the intersection points of the current driving path line and LFL1 and LFL2, and continuously correct the point coordinates of CS1A and CS1B of CS1. Step S7: When the cotton field robot drives along CS1 to a position where the lane lines cannot be recognized by visual navigation, starting from CS1A, continue to rely on RTK position navigation to drive towards CS1B until reaching CS1B. Step S8: From CS1B, increase by one D_LF along the lane-changing route LFL2 in the direction of D_LR. If the obtained point is on LFL2, it is the point coordinates of CS2A. Step S9: Based on CS2A, draw a ray in the opposite direction along WY to obtain the intersection point of the ray and LFL1, that is, the point coordinates of CS2B. Step S10: The cotton field robot turns around in place and turns to face CS2A. Step S11: Starting from CS1B, continue to rely on RTK position navigation to drive towards CS2A until reaching CS2A. Step S12: The cotton field robot turns around in place and turns to face CS2B. If the lane lines are not recognized by visual navigation at this time, starting from CS2A, the cotton field robot relies on RTK position navigation to drive towards CS2B; during the driving process, when the lane lines are recognized by visual navigation, rely on visual navigation to drive. Step S13: Repeat the above process until the point obtained by increasing by one D_LF along the lane-changing route in the direction of D_LR is not on the lane-changing line, then end. Step S14: During the above driving process, continuously record the trajectory coordinate points to generate a path that can rely independently on RTK position navigation.

5. A method for generating a fused navigation path of a cotton field robot according to claim 4, characterized in that, In the step S4, the RTK position navigation steps are as follows: a) Obtain the real-time position coordinate information of the cotton field robot through the vehicle-mounted RTK positioning module. Starting from CS1A and ending at CS1B, calculate the transverse difference in combination with the real-time position; the transverse difference refers to the lateral distance of yaw. b), Substitute the lateral deviation into the PID control algorithm to output the desired steering angle, thereby controlling the driving trajectory of the cotton field robot.

6. A method for generating a fusion navigation path of a cotton field robot according to claim 4, characterized in that In the step S5, the visual navigation steps are as follows: a), Use the trained deep learning visual model to perform real-time recognition on the camera data; b), When the lane lines are recognized, first find the two lane lines in the middle of the planting film, and calculate the intercept on the x-axis of the intersection point formed with the points at the center of the lower edge of the image in the x, y coordinate system; c), Determine the lateral deviation based on the intercept; d), Substitute the lateral deviation into the PID control algorithm to output the desired steering angle, thereby controlling the driving trajectory of the cotton field robot.

7. A method for generating a fused navigation path of a cotton field robot according to claim 4, characterized in that, In the step S10, the in-place turning steps are as follows: a), Take CS1B as the starting point and CS2A as the ending point to calculate the target heading angle; b), Obtain the real-time heading angle of the cotton field robot through the vehicle-mounted RTK positioning module; c), Control the four wheels of the cotton field robot to change from a square distribution to a circular distribution; d), Control the cotton field robot to turn in place, and stop after the real-time heading angle is equal to the target heading angle; e), Control the four wheels of the cotton field robot to change from a circular distribution to a square distribution, and the in-place turning is completed.

Images