A method and system for extracting a cane field navigation reference line based on multi-source information fusion

Abstract

The application discloses a sugarcane field navigation reference line extraction method and system based on multi-source information fusion, and relates to the technical field of agricultural machinery navigation. The method first collects sugarcane planting GNSS historical trajectory data and unmanned aerial vehicle aerial photography data and pre-processes; extracts sugarcane crop row area through a semantic segmentation model, generates a curve navigation line through morphological optimization, skeleton extraction and feature point selection, and then converts the curve navigation line into a crop actual row line trajectory in the form of GNSS coordinates; adopts a sliding window mechanism, dynamically evaluates two trajectories based on multi-dimensional credibility, generates a fusion trajectory through weighted fusion and window smoothing processing; and finally compresses the trajectory through a Douglas-Pok algorithm and formats the output. The application fuses double-source information, solves the problem that traditional Beidou navigation is inaccurate in sugarcane field management and harvesting links, maintains stable navigation accuracy in complex scenes, supports flexible selection of models and algorithms, and adapts to different scale sugarcane field operation requirements.

Description

Technical Field

[0001] This invention belongs to the technical field of agricultural machinery, specifically relating to a method and system for extracting navigation baselines in sugarcane fields based on multi-source information fusion. Background Technology

[0002] Agricultural machinery navigation technology, based on the BeiDou satellite positioning system and combined with automatic steering control technology, can precisely guide agricultural machinery to autonomously travel and operate along a preset route. This technology not only significantly reduces the labor intensity of drivers and minimizes human error, but also significantly improves the uniformity, standardization, and overall efficiency of agricultural operations. It has been widely used in the planting of major crops such as wheat and rice in my country, becoming one of the key technologies driving agricultural modernization.

[0003] In the field of mechanized sugarcane production, BeiDou navigation technology has begun to play a role in the planting stage, guiding planting machinery to complete sowing operations by pre-setting parameters such as AB lines and operating width. However, it has not yet been systematically and on a large scale applied in key stages of the sugarcane's entire life cycle, including tillage, field management (intertillage, plant protection, etc.), and harvesting. The core reason for this application gap is that during field operations, sugarcane planting machinery is affected by various factors such as loose soil, undulating terrain, machinery bumps, and wheel slippage, which can easily cause deviations in the actual driving trajectory. As a result, the AB line and operating width information recorded by the BeiDou navigation system during the planting stage cannot accurately reconstruct the true growth position of the sugarcane crop rows.

[0004] When entering the field management or harvesting stages, if agricultural machinery still relies on the original navigation information recorded during the planting stage, it will be difficult to achieve precise alignment with the sugarcane crop rows, resulting in "inaccurate alignment." This leads to a series of problems such as missed tillage, missed pest control, crop crushing, or harvest losses, seriously affecting the quality of operations and production efficiency. This technical bottleneck not only restricts the improvement of efficiency and quality throughout the entire process of mechanized sugarcane production but also reduces the promotion value and cost-effectiveness of Beidou navigation technology products in the sugarcane industry, creating a significant gap between its application and the demands of modern agriculture for fully mechanized and precise operations.

[0005] Therefore, how to break through the limitations of traditional BeiDou navigation technology in sugarcane field management and harvesting, accurately reproduce the actual position of sugarcane crop rows, and construct a high-precision navigation baseline suitable for subsequent operations has become a key issue in promoting the in-depth application of BeiDou navigation technology in the entire process of sugarcane mechanized production. This is of great significance for improving the mechanization level of the sugarcane industry, reducing production costs, and ensuring yield. Summary of the Invention

[0006] To address the technical problems of existing technologies where BeiDou navigation planting AB lines are difficult to directly apply in mechanized field management and harvesting of sugarcane, and where row alignment is inaccurate, this invention provides a method and system for extracting navigation baselines for sugarcane fields based on multi-source information fusion. By integrating actual crop row information extracted from UAV aerial images with historical GNSS trajectory data from the planting stage, a high-precision and robust navigation baseline is generated, providing reliable support for BeiDou navigation of agricultural machinery in sugarcane field management and harvesting.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] In a first aspect, the present invention provides a method for extracting navigation baselines in sugarcane fields based on multi-source information fusion, comprising the following steps:

[0009] S1. When the sugarcane planting machinery is sowing using Beidou navigation, the GNSS trajectory of the sowing row is collected to obtain the sugarcane planting GNSS historical trajectory data. The planting GNSS historical trajectory data is preprocessed by coordinate transformation, outlier removal, smoothing filtering and row clustering segmentation.

[0010] S2. Two to three weeks after sugarcane seedlings emerge, aerial images of the sugarcane field during the seedling stage are collected by drone, and the aerial images are preprocessed for noise reduction and image enhancement.

[0011] S3. Use a semantic segmentation model to segment the preprocessed aerial image and output the binarized segmentation result to separate the sugarcane crop row area from the background.

[0012] S4. Perform morphological closing operation optimization on the binarized segmentation result, extract the sugarcane row skeleton through the edge detection algorithm, select the seedling sugarcane crop row feature points along the skeleton and record the pixel coordinates of each feature point;

[0013] S5. Based on the coordinate two-dimensional affine transformation matrix of aerial images, the affine transformation algorithm is used to establish the mapping relationship between the pixel coordinates of aerial images and the GNSS coordinates in the WGS-84 coordinate system, converting the pixel coordinates of the feature points of the sugarcane crop rows into the corresponding GNSS coordinates, and extracting the GNSS trajectory of the actual rows of sugarcane crops.

[0014] S6. A sliding window mechanism is adopted to calculate the local confidence of the actual crop row GNSS trajectory and the local confidence of the planting GNSS historical trajectory in each window. The fusion weight is dynamically allocated according to the two confidences. The two trajectories are weighted and averaged based on the multi-source information fusion method to obtain the fused trajectory. The fused trajectories of adjacent windows are smoothed.

[0015] S7. The Douglas-Puk algorithm is used to compress the discrete point set of the fused trajectory with a preset compression threshold. The compressed feature points are arranged in the order of operation and the navigation baseline is output in a preset data format.

[0016] As a preferred technical solution, in step S1, the planting GNSS historical trajectory data includes navigation AB points, operating width, three-dimensional position coordinates of the GNSS antenna / IMU module in the vehicle coordinate system, three-dimensional position coordinates of the sugarcane seed outlet in the vehicle coordinate system, latitude and longitude coordinates of the seeder's real-time GNSS positioning, real-time attitude angle of the seeder's IMU, navigation status flags and timestamps.

[0017] As a preferred technical solution, in step S2, the drone aerial image is a TIF format orthophoto image covering the entire sugarcane field. The orthophoto image saves image pixel information, geographical location information and projection information, and the ground sampling interval GSD is not less than 0.01m / pixel.

[0018] As a preferred technical solution, in step S3, the semantic segmentation model is selected as either the UNet model or the Deeplabv3+ model.

[0019] As a preferred technical solution, in step S4, the morphological closing operation is used to fill the holes inside the crop region in the binary image output by the sugarcane field semantic segmentation model, thereby enhancing the continuity of the region; the edge detection algorithm is used to extract the sugarcane row skeleton from the optimized binary image to obtain the central contour of the sugarcane crop row.

[0020] As a preferred technical solution, in step S6, the local reliability of the planted GNSS historical trajectory is determined by two intermediate indicators: GNSS quality score and dynamic stability score. The GNSS quality score is determined by the positioning identifier score and the positioning error score, and the dynamic stability score is determined by the three-axis angular velocity dynamic interference score and the three-axis acceleration dynamic interference score.

[0021] The local reliability of the actual row GNSS trajectory is determined by the mask hole rate and feature point dispersion. The mask hole rate is the ratio of the number of pixels belonging to the crop row and background categories within the bounded polygon of the mask region within the evaluation window. The feature point dispersion is the ratio of the standard deviation of the lateral distance error from each feature point to the local fitting curve to the maximum error within the evaluation window.

[0022] As a preferred technical solution, in step S6, the multi-source information fusion method dynamically determines the fusion weight within the evaluation window based on two locally calculated confidence levels, and performs a weighted average of the coordinates of the two trajectories to generate the fused trajectory of that segment.

[0023] The smoothing process employs either overlapping averaging or smoothing filtering.

[0024] As a preferred technical solution, in step S7, the compression threshold represents the maximum distance from any point on the compressed path to the line connecting the beginning and end.

[0025] During compression, the entire point set is processed recursively in segments based on the compression threshold, eliminating redundant intermediate points and retaining only feature points that play a key role in the shape of the path.

[0026] As a preferred technical solution, in step S7, the preset data format is KML format, GPKG format or JSON format, and the output data includes path point sequence ID and preset operation speed extension information.

[0027] Secondly, the present invention provides a sugarcane field navigation baseline extraction system based on multi-source information fusion, which is applied to the sugarcane field navigation baseline extraction method based on multi-source information fusion, including a planting data acquisition module, an aerial image acquisition module, a semantic segmentation module, a feature extraction module, a coordinate transformation module, an information fusion module, and a compression module;

[0028] The planting data acquisition module is used to collect the GNSS trajectory of the sowing row when the sugarcane planting machinery is sowing with Beidou navigation, obtain the sugarcane planting GNSS historical trajectory data, and perform coordinate transformation, outlier removal, smoothing filtering and row clustering segmentation preprocessing on the planting GNSS historical trajectory data.

[0029] The aerial image acquisition module is used to collect drone aerial images of sugarcane fields during the seedling stage 2-3 weeks after sugarcane seedlings emerge, and to perform noise reduction and image enhancement preprocessing on the aerial images.

[0030] The semantic segmentation module is used to segment the preprocessed aerial image using a semantic segmentation model, output a binary segmentation result, and achieve the separation of the sugarcane crop row area from the background.

[0031] The feature extraction module is used to optimize the binarized segmentation result by performing morphological closing operation, extract the sugarcane row skeleton through the edge detection algorithm, select the sugarcane crop row feature points along the skeleton and record the pixel coordinates of each feature point;

[0032] The coordinate transformation module is based on the two-dimensional affine transformation matrix of the aerial image. It uses an affine transformation algorithm to establish the mapping relationship between the pixel coordinates of the aerial image and the GNSS coordinates in the WGS-84 coordinate system, converts the pixel coordinates of the feature points of the sugarcane crop row into the corresponding GNSS coordinates, and extracts the GNSS trajectory of the actual row line of the sugarcane crop.

[0033] The information fusion module is used to calculate the local confidence of the actual crop row GNSS trajectory and the local confidence of the planting historical GNSS trajectory in each window using a sliding window mechanism, dynamically allocate fusion weights according to the two confidences, perform a weighted average of the two trajectories based on a multi-source information fusion method to obtain the fused trajectory, and smooth the fused trajectories of adjacent windows.

[0034] The compression module is used to compress the discrete point set of the fused trajectory using the Douglas-Puk algorithm with a preset compression threshold, arrange the compressed feature points in the order of operation, and output the navigation baseline in a preset data format.

[0035] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0036] (1) This invention solves the core problem that traditional Beidou navigation cannot be applied to sugarcane field management and harvesting: it uses drone aerial photography to obtain images of sugarcane fields during the seedling stage, and extracts the actual position of sugarcane crop rows through semantic segmentation, skeleton extraction and curve fitting technology, providing high-precision navigation baseline information for agricultural machinery Beidou navigation, effectively avoiding operational losses caused by inaccurate row positioning.

[0037] (2) This invention improves navigation reliability in complex scenarios: By combining the sugarcane crop row positioning extracted from UAV aerial images with the GNSS navigation sowing row trajectory, and employing a multi-dimensional credibility assessment and adaptive weighted fusion method, the accurate extraction of the sugarcane crop row trajectory is achieved. Even in complex scenarios such as missing sugarcane seedlings, blurred row lines, and crooked crop rows, it can still provide accurate operating paths for agricultural machinery using Beidou navigation, thereby improving the quality and efficiency of operations.

[0038] (3) This invention is adaptable to different scales of operation conditions: it supports UNet and Deeplabv3+ model replacement and multiple fusion algorithm selection. It can flexibly adjust the technical solution according to the computing power configuration of the image processing and data fusion computer, taking into account both data processing efficiency and accuracy. It can meet the operation needs of sugarcane fields of different scales and has high promotion and application value. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is an overall flowchart of the sugarcane field navigation baseline extraction method based on multi-source information fusion of the present invention;

[0041] Figure 2 This is a block diagram of the sugarcane field navigation baseline extraction system based on multi-source information fusion, as described in this invention. Detailed Implementation

[0042] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present application without creative effort are within the scope of protection of the present application.

[0043] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this application can be combined with other embodiments.

[0044] like Figure 1 As shown in the figure, a sugarcane navigation baseline extraction technology based on a multi-source information fusion algorithm in this embodiment includes the following steps:

[0045] S1. Obtain historical GNSS trajectory data for sugarcane planting, specifically:

[0046] When sugarcane planting machinery uses Beidou navigation for sowing, the historical GNSS trajectory data of the sugarcane rows recorded during the planting stage is exported from the backend of the agricultural machinery's Beidou navigation system. The data should include key agronomic navigation information, such as navigation A and B points, operating width, three-dimensional position coordinates of the GNSS antenna / IMU module in the vehicle coordinate system, three-dimensional position coordinates of the sugarcane seed outlet in the vehicle coordinate system, real-time GNSS positioning latitude and longitude coordinates of the seeder, real-time attitude angle of the seeder's IMU, navigation status flags, and timestamps. All latitude and longitude coordinate data are uniformly based on the WGS-84 coordinate system.

[0047] Furthermore, the exported historical GNSS trajectory data for planting is subjected to coordinate transformation, smoothing filtering, and row clustering segmentation, and abnormal coordinate points are removed to lay the foundation for subsequent data fusion.

[0048] S2. Collect drone aerial images of sugarcane fields during the seedling stage, specifically:

[0049] A multi-rotor drone equipped with a high-precision GNSS positioning module was used to conduct full-coverage aerial photography of the target sugarcane field 2-3 weeks after the sugarcane seedlings emerged. The flight altitude was set at 10-20 meters to ensure that the acquired image resolution was no less than 1080P, and that each frame contained the corresponding GNSS coordinate information.

[0050] Furthermore, to enhance the model's generalization ability, data collection covered typical scenarios with different lighting conditions (such as sunny days and cloudy days) and different sugarcane growth states (such as normal growth and partial seedling loss), accumulating more than 1,000 image frames to construct the original image dataset.

[0051] Furthermore, the acquired aerial images are preprocessed, including using a Gaussian filtering algorithm for noise reduction and adjusting brightness and contrast to enhance the image and improve the distinction between sugarcane plants and the background.

[0052] After steps S1 and S2, multi-source data can be obtained, which is used for subsequent extraction of navigation baselines.

[0053] S3. Extraction of sugarcane crop row regions based on semantic segmentation, specifically:

[0054] S31. A semantic segmentation model is used to separate the crop region from the background in the preprocessed aerial image.

[0055] Optionally, the UNet model can be used as the base model, and it can be replaced with the Deeplabv3+ model to adapt to different computing power requirements.

[0056] S32. Model Training: Select 800 preprocessed aerial images as the training set and 200 frames as the test set. Manually label the sugarcane crop area and the background area. Set the model training iterations to 300 rounds, the learning rate to 0.001, and the batch size to 8.

[0057] S33. Image Segmentation: Input the preprocessed aerial image into the trained semantic segmentation model and output a binarized segmentation result (crop area is white, sugarcane field soil background is black) to achieve accurate extraction of sugarcane row area.

[0058] S4. Feature point extraction and curve navigation baseline generation, specifically:

[0059] S41. Binarization optimization: Perform morphological closing operation (structural element is a 5×5 rectangle) on the binary image output by semantic segmentation to fill the holes inside the crop region and enhance the continuity of the region.

[0060] S42. Skeleton Extraction: The edge detection algorithm (Canny operator, threshold range set to 50-150) is used to extract the skeleton of the sugarcane row from the optimized binary image, replacing the traditional point extraction method in the box, and accurately obtaining the center contour of the sugarcane row.

[0061] S43. Feature point selection: Select a feature point every 5-10 pixels along the midpoint of the extracted skeleton curve, and record the pixel coordinates of each feature point in the image.

[0062] S5. Pixel coordinate to GNSS coordinate conversion, specifically:

[0063] Based on the coordinate two-dimensional affine transformation matrix of TIFF aerial images, a mapping relationship between image pixel coordinates and WGS-84 GNSS coordinates is established. Using feature points with known GNSS coordinates in the image (such as sugarcane field boundary corners) for calibration, an affine transformation algorithm is employed to convert the pixel coordinates of the curved navigation line generated in step S3 into the corresponding GNSS coordinates, forming the GNSS trajectory of the actual crop row lines.

[0064] S6. Multi-source information fusion algorithm extracts navigation baseline; To generate high-precision agricultural machinery operation navigation baselines, this step constructs a multi-source information fusion algorithm model, aiming to optimally fuse the "actual sugarcane crop row GNSS trajectory" with the "historical planting GNSS trajectory". The core processing logic is as follows:

[0065] S61. Set window sliding:

[0066] An evaluation window is formed by sliding along the work path using a fixed length (e.g., 20 meters) as the window.

[0067] S62. Reliability assessment of sugarcane crop row image trajectories:

[0068] The local reliability of the actual GNSS trajectory is determined by evaluating the mask void ratio and feature point dispersion; the mask void ratio R voidage Within the evaluation window, the ratio of the number of pixels belonging to the crop row and background categories within the bounded polygon of the mask region; feature point dispersion R deviation It is the ratio of the standard deviation of the lateral distance error from each feature point to the local fitted curve to the maximum error within the evaluation window.

[0069] Local confidence level C of actual GNSS trajectory vision for:

[0070] ;

[0071] S63. Local reliability assessment of sugarcane planting GNSS historical trajectories:

[0072] The reliability of historical GNSS trajectory points for sugarcane planting was evaluated using two intermediate indicators: GNSS quality score (Qgnss) and dynamic stability score (Qdynamic).

[0073] GNSS quality fraction Q gnss It is determined by the positioning identifier score Sfix and the positioning error score Serr.

[0074] Location Identifier Sfix: Assigned based on RTK location quality identifier. Fixed solution = 1.0, Floating solution = 0.6, Single-point solution = 0.3, Invalid = 0.

[0075] Positioning error fraction Serr:

[0076] ;

[0077] In the formula, err_std is the positioning standard deviation of the current positioning point, which can be obtained from the GST message in the NMEA 0183 protocol output by the GNSS module; th_err is the maximum value of the positioning standard deviation of the positioning point within the current sliding window.

[0078] GNSS quality fraction Q gnss It can be calculated using the following formula:

[0079] ;

[0080] The dynamic stability fraction Qdynamic is determined by the triaxial angular velocity dynamic disturbance fraction Dgyro and the triaxial acceleration dynamic disturbance fraction Dacc.

[0081] Three-axis angular velocity dynamic disturbance fraction D gyro The calculation formula is:

[0082] ;

[0083] In the formula, th_gyro is the empirical threshold for angular velocity, which should generally be greater than the maximum amplitude of the angular velocity of the three-axis gyroscope within the sliding window.

[0084] Three-axis acceleration dynamic disturbance fraction D acc The calculation formula is:

[0085] ;

[0086] In the formula, th_acc is the empirical threshold for acceleration, which should generally be greater than the maximum amplitude of the triaxial accelerometer within the sliding window.

[0087] The higher the dynamic stability score, the smoother the movement of the sugarcane planting machinery and the more stable the geometric relationship between the GNSS antenna and the seeding port.

[0088] ;

[0089] Reliability C of GNSS historical trajectory points for sugarcane planting gnss for:

[0090] ;

[0091] In the formula, the values ​​of α and β can be adjusted empirically.

[0092] S64, Trajectory Weighted Fusion;

[0093] Based on the two locally calculated confidence levels, the fusion weights within the window are dynamically determined. Within this window, the coordinates of the two trajectories are weighted and averaged to generate the fused trajectory for that segment.

[0094] Within a sliding window, let P be the GNSS point to be merged. gnss The corresponding image trajectory point P is obtained by searching based on the nearest neighbor principle. vison The fused coordinates P fused Calculated by the following formula:

[0095] ;

[0096] Among them, weight and Their credibility is dynamically determined:

[0097] , ;

[0098] S65, Smoothing between windows:

[0099] Overlap averaging or smoothing filtering is applied to the fusion results of adjacent windows to avoid abrupt changes in the navigation baseline at the window boundaries.

[0100] S7. Formatted output of navigation baseline; To transform the fused high-density discrete point set into a preset path that can be efficiently resolved by the agricultural machinery Beidou navigation system, feature-preserving compression and formatted output are required; the specific steps are as follows:

[0101] S71. Trajectory Compression: The dense discrete point set generated in step S65 is compressed using the Douglas-Puk algorithm.

[0102] Set a compression threshold: Based on the steering control accuracy and operational requirements of the agricultural machinery, set a maximum compression error (e.g., 0.1 meters). This threshold means that the maximum distance from any point on the compressed path to the line connecting the beginning and end does not exceed 0.1 meters.

[0103] The compression algorithm is then applied: using this threshold as a standard, the entire point set is processed in segments, eliminating redundant intermediate points and retaining only feature points that are crucial to the path shape. This step significantly reduces the number of points while perfectly preserving the path's curvature characteristics.

[0104] S72. Path point sequence generation: The feature points retained after compression are arranged in order from the start point to the end point of the operation to form an ordered path point sequence with a reduced data volume.

[0105] S73. Define the output data format: Encapsulate and output the final path point sequence and its attributes using a data format commonly used in agricultural machinery Beidou navigation systems (such as KML, GPKG, or a custom JSON format). In addition to coordinates, the output data packet may also include extended information such as point sequence IDs and preset operating speeds.

[0106] It should be noted that, for the sake of simplicity, the aforementioned method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, because according to the present invention, some steps can be performed in other orders or simultaneously.

[0107] Based on the same idea as the sugarcane field navigation baseline extraction method based on multi-source information fusion in the above embodiments, the present invention also provides a sugarcane field navigation baseline extraction system based on multi-source information fusion. This system can be used to execute the above-described sugarcane field navigation baseline extraction method based on multi-source information fusion. For ease of explanation, the structural schematic diagram of the embodiment of the sugarcane field navigation baseline extraction system based on multi-source information fusion only shows the parts related to the embodiments of the present invention. Those skilled in the art will understand that the illustrated structure does not constitute a limitation on the device, and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0108] Please see Figure 2 In another embodiment of this application, a sugarcane field navigation baseline extraction system 100 based on multi-source information fusion is provided. The system includes a planting data acquisition module 101, an aerial image acquisition module 102, a semantic segmentation module 103, a feature extraction module 104, a coordinate transformation module 105, an information fusion module 106, and a compression module 107.

[0109] The planting data acquisition module 101 is used to collect the GNSS trajectory of the sowing row when the sugarcane planting machinery is sowing with Beidou navigation, obtain the sugarcane planting GNSS historical trajectory data, and perform coordinate transformation, outlier removal, smoothing filtering and row clustering segmentation preprocessing on the planting GNSS historical trajectory data.

[0110] The aerial image acquisition module 102 is used to collect aerial images of sugarcane fields taken by drones during the seedling stage 2-3 weeks after sugarcane seedlings emerge, and to perform noise reduction and image enhancement preprocessing on the aerial images.

[0111] The semantic segmentation module 103 is used to segment the preprocessed aerial image using a semantic segmentation model, output a binarized segmentation result, and achieve the separation of the sugarcane crop row area from the background.

[0112] The feature extraction module 104 is used to perform morphological closing operation optimization on the binarized segmentation result, extract the sugarcane row skeleton through the edge detection algorithm, select seedling sugarcane crop row feature points along the skeleton and record the pixel coordinates of each feature point.

[0113] The coordinate transformation module 105 generally uses the coordinate two-dimensional affine transformation matrix of the aerial image to establish the mapping relationship between the pixel coordinates of the aerial image and the GNSS coordinates in the WGS-84 coordinate system, based on the affine transformation algorithm, to convert the pixel coordinates of the feature points of the sugarcane crop row into the corresponding GNSS coordinates and extract the GNSS trajectory of the actual row line of the sugarcane crop.

[0114] The information fusion module 106 is used to employ a sliding window mechanism to calculate the local confidence of the actual crop row GNSS trajectory and the local confidence of the planting historical GNSS trajectory in each window, dynamically allocate fusion weights based on the two confidences, perform a weighted average of the two trajectories based on a multi-source information fusion method to obtain the fused trajectory, and smooth the fused trajectories of adjacent windows.

[0115] The compression module 107 is used to compress the discrete point set of the fused trajectory using the Douglas-Puk algorithm with a preset compression threshold, arrange the compressed feature points in the order of operation, and output the navigation baseline in a preset data format.

[0116] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0117] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0118] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. A method for extracting navigation baselines in sugarcane fields based on multi-source information fusion, characterized in that, Includes the following steps: S1. When the sugarcane planting machinery is sowing using Beidou navigation, the GNSS trajectory of the sowing row is collected to obtain the sugarcane planting GNSS historical trajectory data. The planting GNSS historical trajectory data is preprocessed by coordinate transformation, outlier removal, smoothing filtering and row clustering segmentation. S2. Two to three weeks after sugarcane seedlings emerge, aerial images of the sugarcane field during the seedling stage are collected by drone, and the aerial images are preprocessed for noise reduction and image enhancement. S3. Use a semantic segmentation model to segment the preprocessed aerial image and output the binarized segmentation result to separate the sugarcane crop row area from the background. S4. Perform morphological closing operation optimization on the binarized segmentation result, extract the sugarcane row skeleton through the edge detection algorithm, select the seedling sugarcane crop row feature points along the skeleton and record the pixel coordinates of each feature point; S5. Based on the coordinate two-dimensional affine transformation matrix of aerial images, the affine transformation algorithm is used to establish the mapping relationship between the pixel coordinates of aerial images and the GNSS coordinates in the WGS-84 coordinate system, converting the pixel coordinates of the feature points of the sugarcane crop rows into the corresponding GNSS coordinates, and extracting the GNSS trajectory of the actual rows of sugarcane crops. S6. A sliding window mechanism is adopted to calculate the local confidence of the actual crop row GNSS trajectory and the local confidence of the planting GNSS historical trajectory in each window. The fusion weight is dynamically allocated according to the two confidences. The two trajectories are weighted and averaged based on the multi-source information fusion method to obtain the fused trajectory. The fused trajectories of adjacent windows are smoothed. Local reliability C of planted GNSS history trajectory gnss is: ； In the formula, the values ​​of α and β are adjusted based on experience; GNSS quality score Q gnss The positioning identification score Sfix and the positioning error score Serr are determined, and the calculation formula is as follows: ； Among them, the positioning identifier score Sfix is ​​assigned a value based on the RTK positioning quality identifier; The positioning error fraction Serr is expressed as: ； Where err_std is the standard deviation of the current location point, and th_err is the maximum value of the standard deviation of the current location point within the current sliding window; Dynamic stability fraction Q dynamic Dynamic disturbance fraction D from triaxial angular velocity gyro Dynamic disturbance fraction D from triaxial angular velocity acc Decide; Three-axis angular velocity dynamic disturbance fraction D gyro The calculation formula is: ； In the formula, th_gyro is the empirical threshold for angular velocity, which is greater than the maximum amplitude of the angular velocity of the three-axis gyroscope within the sliding window; Three-axis acceleration dynamic disturbance fraction D acc The calculation formula is: ； In the formula, th_acc is the empirical acceleration threshold, which is greater than the maximum acceleration amplitude of the triaxial accelerometer within the sliding window; The higher the dynamic stability score, the smoother the movement of the sugarcane planting machinery and the more stable the geometric relationship between the GNSS antenna and the seeding port. ； The local reliability of the actual row GNSS trajectory is determined by the mask hole rate and feature point dispersion. The mask hole rate is the ratio of the number of pixels belonging to the crop row and background categories within the bounding polygon of the mask region within the evaluation window. The feature point dispersion is the ratio of the standard deviation of the lateral distance error from each feature point to the local fitting curve to the maximum error within the evaluation window. S7. The Douglas-Puk algorithm is used to compress the discrete point set of the fused trajectory with a preset compression threshold. The compressed feature points are arranged in the order of operation and the navigation baseline is output in a preset data format.

2. The method for extracting navigation baselines in sugarcane fields based on multi-source information fusion according to claim 1, characterized in that, In step S1, the planting GNSS historical trajectory data includes navigation AB points, operating swath width, three-dimensional position coordinates of GNSS antenna and IMU module in the vehicle coordinate system, three-dimensional position coordinates of sugarcane seed outlet in the vehicle coordinate system, real-time GNSS positioning latitude and longitude coordinates of seeder, real-time attitude angle of seeder IMU, navigation status flag and timestamp.

3. The method for extracting navigation baselines in sugarcane fields based on multi-source information fusion according to claim 1, characterized in that, In step S2, the drone aerial image is a TIF format orthophoto covering the entire sugarcane field. The orthophoto saves image pixel information, geographical location information and projection information, and the ground sampling interval GSD is not less than 0.01m / pixel.

4. The method for extracting navigation baselines in sugarcane fields based on multi-source information fusion according to claim 1, characterized in that, In step S3, the semantic segmentation model is selected as either the UNet model or the Deeplabv3+ model.

5. The method for extracting navigation baselines in sugarcane fields based on multi-source information fusion according to claim 1, characterized in that, In step S4, the morphological closing operation is used to fill the holes inside the crop region in the binary image output by the sugarcane field semantic segmentation model, thereby enhancing the continuity of the region; the edge detection algorithm is used to extract the sugarcane row skeleton from the optimized binary image and obtain the central contour of the sugarcane crop row.

6. The method for extracting navigation baselines in sugarcane fields based on multi-source information fusion according to claim 1, characterized in that, In step S6, the multi-source information fusion method dynamically determines the fusion weight within the evaluation window based on two locally calculated confidence levels, and performs a weighted average of the coordinates of the two trajectories to generate the fused trajectory for that segment. The smoothing process employs either overlapping averaging or smoothing filtering.

7. The method for extracting navigation baselines in sugarcane fields based on multi-source information fusion according to claim 1, characterized in that, In step S7, the compression threshold represents the maximum distance from any point on the compressed path to the line connecting the beginning and end. During compression, the entire point set is processed recursively in segments based on the compression threshold, eliminating redundant intermediate points and retaining only feature points that play a key role in the shape of the path.

8. The method for extracting navigation baselines in sugarcane fields based on multi-source information fusion according to claim 1, characterized in that, In step S7, the preset data format is KML, GPKG, or JSON, and the output data includes path point sequence IDs and preset job speed extension information.

9. A sugarcane field navigation baseline extraction system based on multi-source information fusion, characterized in that, The sugarcane field navigation baseline extraction method based on multi-source information fusion, applied to any one of claims 1-8, includes a planting data acquisition module, an aerial image acquisition module, a semantic segmentation module, a feature extraction module, a coordinate transformation module, an information fusion module, and a compression module; The planting data acquisition module is used to collect the GNSS trajectory of the sowing row when the sugarcane planting machinery is sowing with Beidou navigation, obtain the sugarcane planting GNSS historical trajectory data, and perform coordinate transformation, outlier removal, smoothing filtering and row clustering segmentation preprocessing on the planting GNSS historical trajectory data. The aerial image acquisition module is used to collect drone aerial images of sugarcane fields during the seedling stage 2-3 weeks after sugarcane seedlings emerge, and to perform noise reduction and image enhancement preprocessing on the aerial images. The semantic segmentation module is used to segment the preprocessed aerial image using a semantic segmentation model, output a binary segmentation result, and achieve the separation of the sugarcane crop row area from the background. The feature extraction module is used to optimize the binarized segmentation result by performing morphological closing operation, extract the sugarcane row skeleton through the edge detection algorithm, select the sugarcane crop row feature points along the skeleton and record the pixel coordinates of each feature point; The coordinate transformation module is based on the two-dimensional affine transformation matrix of the aerial image. It uses an affine transformation algorithm to establish the mapping relationship between the pixel coordinates of the aerial image and the GNSS coordinates in the WGS-84 coordinate system, converts the pixel coordinates of the feature points of the sugarcane crop row into the corresponding GNSS coordinates, and extracts the GNSS trajectory of the actual row line of the sugarcane crop. The information fusion module is used to calculate the local confidence of the actual crop row GNSS trajectory and the local confidence of the planting historical GNSS trajectory in each window using a sliding window mechanism, dynamically allocate fusion weights according to the two confidences, perform a weighted average of the two trajectories based on a multi-source information fusion method to obtain the fused trajectory, and smooth the fused trajectories of adjacent windows. The compression module is used to compress the discrete point set of the fused trajectory using the Douglas-Puk algorithm with a preset compression threshold, arrange the compressed feature points in the order of operation, and output the navigation baseline in a preset data format.

Images