A method for farmland path recognition and planning based on aerial imagery

Abstract

A method for farmland path recognition and planning based on aerial imagery is proposed. First, a data set (DOM) is constructed, cropped, and scale-normalized to form a dataset for recognition and planning. Next, multi-scale identification of the inter-row crop passage area is performed to obtain passable areas. Then, path planning is performed based on the inter-row structure: the centerline of the passable area is extracted to obtain a skeleton; principal direction constraints and structural consistency analysis are applied to the skeleton; branch removal and break repair are performed to obtain continuous, directional, and consistent inter-row centerline paths. The paths are clustered, grouped, and sorted to automatically distinguish different ridges and determine the walking order within the plot. Based on the endpoint relationships of adjacent inter-row paths, cross-row connection paths are generated at the ridge ends, forming a global navigation path combining straight and turning routes, and determining the entrance and exit points. Finally, the results are converted into standard path data for output, providing navigation for ground-based equipment. This invention effectively improves the stability and consistency of planning.

Description

Technical Field

[0001] This invention belongs to the field of agricultural intelligence and agricultural information technology, and more specifically relates to a method for farmland path recognition and planning based on aerial imagery. It is mainly used to obtain the structural features of plots such as inter-row passages and drainage ditches in crop planting plots, and generate global driving paths for farmland operations. Background Technology

[0002] In large-scale open-field ridge-cultivated vegetable (such as broccoli, cabbage, and cauliflower) planting scenarios, key agronomical operations such as weeding and harvesting have long relied on manual labor, resulting in high labor intensity, limited operational efficiency, and difficulty in ensuring operational consistency. With the development of agricultural robots and autonomous navigation technology, using ground robots to replace manual labor in field operations has become an important trend. This requires the ability to stably and accurately acquire structural information of passable areas such as ridge-crossing paths, and based on this, to complete path planning and motion guidance, ensuring the safe and reliable operation of robots in complex farmland environments. Recent research indicates that passable area extraction and navigation line generation based on visual / semantic segmentation have become important technical approaches for ridge-cultivated farmland navigation. These methods can improve the perception of crop row structure to a certain extent and can be combined with curve fitting methods to generate navigation lines. Meanwhile, drones possess the advantages of rapidly covering large areas of land, acquiring high-resolution remote sensing imagery, and generating orthophotos (DOM / orthomosaic), and have been used to support map building and path planning related to row navigation. Existing work indicates that using drones to create land parcel maps can serve row navigation and coverage path planning (CPP), providing fundamental information for operational route optimization and obstacle avoidance. Furthermore, coverage operations are universally needed in agricultural scenarios, but different land parcel shapes, boundary constraints, and vehicle curvature constraints can significantly affect the quality of the final executable path.

[0003] However, existing technologies still have shortcomings in the end-to-end process of "aerial DOM → executable global path for ground robot": On the one hand, DOM-level images are often large in scale, complex in texture, and significantly affected by occlusion and illumination changes, making it challenging to accurately and stably extract inter-row passages and restore their topological continuity from large DOMs; on the other hand, many studies focus more on segmentation or local navigation line extraction, lacking systematic processing such as structured representation of centerlines for the entire plot, path grouping, break repair, and secondary smoothing, making it difficult to directly form an executable global path including "straight movement between rows - turning at the edge of the field - entering and exiting the plot". Recent related work also emphasizes that navigation line extraction needs to further obtain centerlines / center paths from the segmented contours and form a structured representation that can be used for planning, otherwise it is difficult to support subsequent path planning and control loop. At the same time, agricultural cover path planning has strong task dependence and application specificity, and the lack of a unified process and reproducible evaluation will also affect the implementation and promotion of algorithms on real plots. Therefore, there is an urgent need for a systematic path recognition and global path generation method that takes aerial DOM as input and is oriented towards the structural features of plots such as inter-row passages and drainage ditches. This method can automatically extract path structures from large-scale images and output continuous, stable, and directly executable global navigation paths, providing a reliable basis for harvesting robots to enter the ridges, travel between ridges, and turn around across ridges. Summary of the Invention

[0004] To overcome the problems of low automation, insufficient recognition accuracy, and difficulty in forming complete and executable paths in existing technologies for farmland path extraction, this invention proposes a farmland path recognition and planning method based on aerial imagery. This method uses large-scale farmland imagery acquired from an aerial source as input, and through multi-scale image processing and path structure analysis, it automatically identifies farmland path structures such as inter-row passages and drainage ditches within the planted area, and plans the overall movement path. This invention can accurately extract the geographic information and spatial distribution of inter-row passages from the digital orthophoto (DOM) imagery of the plot, and automatically generate the path centerline, ridge-end turning trajectory, and a global navigation path covering the entire plot area, ensuring good continuity and executability of the generated path results. This method effectively improves the stability and consistency of path planning results in farmland scenarios, providing a reliable path data foundation for subsequent autonomous operation of agricultural machinery.

[0005] The technical solution adopted by this invention to solve its technical problem is: A method for farmland path recognition and planning based on aerial imagery includes the following steps: Step 1, Dataset Preparation and Digital Orthophoto Construction: Based on the spatial extent and crop planting layout of the target farmland plot, aerial image data covering the target plot is acquired, and the original images are screened, registered, and scaled. On this basis, a digital orthophoto (DOM) of the target plot is generated through image stitching and geometric correction to eliminate image distortion and perspective differences, so that the distribution relationship of crops between rows in the farmland can be accurately expressed in a unified planar coordinate system.

[0006] The DOM is segmented, cropped, and scaled to form a dataset for subsequent path identification and planning, thereby providing a stable and continuous spatial data foundation for inter-row crop path planning.

[0007] Step 2, Multi-scale identification of inter-row crop passage areas: On the DOM dataset obtained in Step 1, farmland structure identification processing is performed to distinguish areas such as inter-row passages, crop ridges and drainage ditches, and to extract inter-row crop passage areas; the identification results are represented in the form of pixel-level masks to describe the spatial distribution range of inter-row passage areas in the plot.

[0008] The connectivity analysis and noise removal processes are performed on the access area mask to remove scattered areas and obvious non-accessible structures, resulting in an effective access area mask for path planning.

[0009] Step 3, Path planning based on inter-row crop structure: Based on the effective passage area mask obtained in Step 2, the centerline of the passage area is extracted to obtain the skeleton representation of the inter-row passage path; the main direction constraint and structural consistency analysis are performed on the skeleton, and branch removal and break repair are carried out to obtain a set of continuous and directional inter-row centerline paths.

[0010] The inter-row centerline paths are clustered, grouped, and sorted to automatically distinguish different inter-row paths and determine the walking order within the plot. Based on the spatial relationship between the endpoints of adjacent inter-row paths, cross-row connection paths are generated at the ridge ends to form a global path planning result that combines inter-row straight paths and ridge end turning paths. The plot entrance and exit paths are determined to obtain a global navigation path that covers the plot area, is continuous, and is executable.

[0011] Step 4, Path Result Output: Convert the global navigation path generated in Step 3 into standard path data format and output it as a reference path for ground operation equipment.

[0012] Furthermore, step 1 includes the following sub-steps: Step 1.1, Digital Orthophoto DOM Construction: Based on the image data taken by the UAV, the target farmland plot is image stitched and orthorectified to generate a digital orthophoto DOM covering the entire plot area, so as to ensure the true proportional relationship of the structure such as inter-row crops, inter-row passages and drainage ditches in the farmland in the plane coordinate system. Step 1.2, DOM segmentation and scale normalization: The digital orthophoto is segmented and cropped to divide the large DOM into several sub-images of preset sizes, and scale unification and format normalization are performed on each sub-image to form a dataset suitable for subsequent path recognition and planning.

[0013] Preferably, before step 2, based on the digital orthophoto dataset constructed in step 1, samples are labeled for areas such as inter-row passages, crop ridges, and drainage ditches in farmland plots, and the labeled samples are used to train a path structure recognition model, so that the model has the ability to recognize farmland path structures at the pixel level; the trained model is used for the identification of inter-row passage areas in subsequent steps.

[0014] Furthermore, step 2 includes the following sub-steps: Step 2.1, Initial identification of inter-row crop passage areas: On the DOM dataset obtained in Step 1, load the passage area image segmentation model trained in Step 1, perform semantic segmentation processing on the plot image, distinguish areas such as inter-row passages and crop ridge drainage ditches, and generate pixel-level masks of inter-row crop passage areas to describe the spatial distribution range of the passage areas in the plot; the identification results are represented in the form of pixel-level masks to describe the spatial distribution range of inter-row crop passage areas in the plot; Step 2.2, Connectivity analysis and noise removal of the passage area mask: Perform connectivity analysis on the passage area mask obtained in Step 2.1 to identify continuous passage area structures; based on area threshold, spatial location and morphological characteristics, remove scattered areas, discontinuous structures and areas that obviously do not have passage significance to obtain an effective inter-row crop passage area mask for path planning.

[0015] Step 3 includes the following sub-steps: Step 3.1, Extraction of centerline and determination of main direction of inter-row crop passage area: Based on the effective inter-row crop passage area mask obtained in step 2, centerline extraction and direction analysis are performed on the passage area to obtain the inter-row crop centerline path for path planning; Step 3.2, Branch Removal and Fragment Repair of Centerline Paths: Based on the set of inter-row crop centerline paths obtained in Step 3.1, the centerline paths undergo structural optimization to eliminate branch structures caused by noise, occlusion, or local misidentification, and to repair any broken areas in the centerlines, resulting in continuous and structurally sound inter-row crop travel paths. Note that the directional consistency constraint in Step 3.1 is used for initial screening of centerline segments, while the directional consistency constraint in Step 3.2 performs topology-level optimization and fragment repair on the remaining paths based on the graph structure. Step 3.3, Grouping and sorting of inter-row centerline paths: Based on the set of continuous, branchless inter-row crop centerline paths obtained in Step 3.2, each centerline path is grouped and sorted to distinguish different inter-row crop rows and determine the driving order of agricultural machinery within the plot, providing an ordered path structure for subsequent global path generation. Step 3.4, ridge end connection and global path generation: Based on the grouping and sorting results of the inter-ridge centerline paths obtained in Step 3.3, the adjacent inter-ridge paths are connected at the ridge end positions, and a global driving path for inter-ridge crops covering the entire plot is generated to meet the actual needs of continuous agricultural machinery operation.

[0016] Preferably, in step 4, the ground-side application outputs standard path data as a reference path for the ground-side equipment. The ground-side equipment then uses the path data to complete inter-row driving and cross-row operations, thereby realizing farmland operations based on the aerial path planning results.

[0017] Step 4 includes the following sub-steps: Step 4.1, Coordinate Mapping and Representation of Global Path Data: The global path planning results obtained in Step 3.4 are mapped from the image coordinate system to the geographic coordinate system or the farmland operation coordinate system, so that each path point in the path has clear spatial location information; the path is represented in the form of an ordered sequence of path points, used to describe the continuous driving trajectory of agricultural machinery within the farmland plot. The coordinate mapping can be completed based on the spatial reference information of digital orthophotos or a ground-based positioning system. Step 4.2, Formatted output of path data: The mapped path data is formatted to generate a standard path data file. The path data file contains at least path point coordinate information and path sequence information, which are used to characterize the driving route and operation sequence of agricultural machinery in the inter-row crop environment. Step 4.3, Ground-based loading and retrieval of path results: Before operation, the ground-based operating equipment reads the path data file generated in step 4.2 and loads the path data as a navigation reference path into the ground-based control system. This guides the agricultural machinery to enter the field, travel along the inter-row path, and complete cross-row turning operations. Step 4.4, Ground-end constrained driving based on global path: During the actual driving of the agricultural machinery, the ground-end control system will match its own position information acquired in real time with the global path, and complete the inter-row driving and ridge-end turning operations under the constraints of the global path, thereby ensuring that the agricultural machinery drives stably in the farmland according to the path planned by the aerial end.

[0018] Step 4.5, Output Results: By outputting the above path and the ground application process, the following effects are achieved: The global path for inter-row crops generated in the air can be directly read and executed by ground-based equipment. The travel paths of agricultural machinery within farmland plots exhibit good continuity and consistency; Reduce the workload of manual path planning and guidance, and improve the efficiency and stability of farmland operations.

[0019] This invention uses large-scale farmland image data acquired from an aerial terminal as input, and through multi-scale path structure analysis and processing, it realizes automatic identification and global path planning of path structures such as inter-row passages and drainage ditches in farmland plots.

[0020] The beneficial effects of this invention are mainly reflected in: 1. Based on aerial remote sensing images of farmland, this invention realizes automated inspection and path structure identification of vegetable planting plots. It can accurately extract key plot structure features such as inter-row passages and drainage ditches from large-scale digital orthophotos acquired by UAVs, effectively replacing manual labor and improving the efficiency and consistency of path acquisition. 2. This invention uses multi-scale image processing and path structure analysis methods to transform the scattered and complex inter-row passage areas in farmland into continuous and clearly structured centerline paths. Based on this, it automatically generates global path planning results covering the entire plot. The generated paths have good continuity, stability and executability. 3. The global path planning results generated by this invention can be directly used as a reference path for the driving of ground agricultural machinery, which significantly improves the autonomous navigation stability and path tracking accuracy of agricultural machinery in the inter-row crop environment, and reduces the operational risks caused by path discontinuity during ridge-end turning and cross-ridge driving. 4. This invention effectively supports the unmanned operation requirements of intelligent agricultural machinery in crop planting scenarios, and provides a reliable data foundation and path basis for path planning, navigation control and operation organization in the field operation process. It has good engineering practical value and application prospects. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the overall process of a multi-scale path recognition and planning method for farmland based on aerial imagery.

[0022] Figure 2 This diagram illustrates the planning of agricultural machinery's travel path within a farmland plot. Taking a regular ridge-cultivated farmland as an example, it uses a top-down view to show the spatial relationships between crop ridges, inter-ridge passageways, irrigation canal structures, and the agricultural machinery's travel path. The diagonally filled areas represent crop ridges, the black areas represent the agricultural machinery's travel path, the blank rectangular areas represent irrigation canal structures, and the arrows indicate the agricultural machinery's travel direction and centerline in the inter-ridge passageways. The agricultural machinery travels straight along the centerline of the inter-ridge passageway. Upon reaching the plot boundary or ridge end, it uses a pre-planned U-shaped turning path to cross ridges and continue traveling into the adjacent inter-ridge passageway, thus forming a continuous travel path covering the entire plot. This illustrates the overall path planning concept of the invention.

[0023] Figure 3 The image shows an example of a farm machinery driving path in an aerial photograph of a farmland plot. The aerial photograph of the farmland plot, acquired by an inspection drone, is overlaid with the driving path generated based on the method of this invention. It can be seen that the generated driving path maintains good consistency with the actual crop ridge structure in the farmland. The path is continuously distributed within the passage area of ​​each ridge and forms a smooth turning connection at the ridge end, verifying the applicability and feasibility of the path planning method in real farmland scenarios.

[0024] Figure 4This is an example of the entire process of the farmland path identification and planning method of the present invention. The typical area of ​​the plot is selected in the figure, and the processing stages are as follows: (a) is the initial segmentation result of the inter-row passage area. This figure is extracted from the original orthophoto by the semantic segmentation method. The white area in the figure represents the identified inter-row passage area. Due to the influence of crop shading, uneven lighting and noise interference, there are still local breaks, burr noise and adhesion phenomena in the initial segmentation result. (b) is the result after performing morphological purification processing on the initial passage area. Through connected component analysis and morphological operations, small fragmented noise areas and lateral structures that do not have passage significance are removed, making the overall structure of the inter-row passage area more prominent, but still retaining its original spatial distribution characteristics. (c) is the initial result of the centerline obtained by skeletonization processing based on the purification result. The passage area is compressed into a centerline structure with a single pixel width through skeletonization operation to express the topological skeleton of the inter-row passage area. However, this result still contains branches and redundant line segments caused by noise and local structural irregularities. (d) This is the centerline diagram after applying directional consistency constraints and branch removal to the skeleton results. Through main direction analysis and length constraints, branch structures inconsistent with the main direction of the ridge were removed, retaining continuous and directionally consistent effective centerlines, providing a stable geometric basis for subsequent path generation. (e) This is the result diagram after grouping and clustering the centerlines. Based on spatial location and directional consistency, the centerlines of different inter-ridge travel areas are automatically divided into several independent paths, realizing the differentiation and organization of multiple ridge lines. (f) This is the inter-ridge straight path diagram obtained by performing curve smoothing on the grouped centerlines. Curve fitting is used to smooth the centerlines, so that the path has good continuity and geometric stability while maintaining the original spatial constraints, suitable for actual ground robot driving. (g) This is the result diagram after secondary smoothing and continuity enhancement based on the first smoothing, further suppressing local jitter and residual bending, and smoothly connecting possible minor breaks, making the path shape more regular and more suitable for actual tracking control. (h) is the final generated driving path planning result map. This path includes the driving center line path and the ridge end U-shaped turn path, forming a continuous driving trajectory covering the entire plot, which can be directly used as the global navigation path input for the harvesting robot. Detailed Implementation

[0025] The present invention will now be further described with reference to the accompanying drawings.

[0026] Reference Figure 1 A method for farmland path recognition and planning based on aerial imagery includes the following steps: Step 1, Dataset Preparation and Digital Orthophoto Construction: Based on the spatial extent and crop planting layout of the target farmland plot, aerial image data covering the target plot is acquired, and the original images are screened, registered, and scaled. On this basis, a digital orthophoto (DOM) of the target plot is generated through image stitching and geometric correction to eliminate image distortion and perspective differences, so that the distribution relationship of crops between rows in the farmland can be accurately expressed in a unified planar coordinate system.

[0027] The DOM is segmented, cropped, and scaled to form a dataset for subsequent path identification and planning, thereby providing a stable and continuous spatial data foundation for inter-row crop path planning.

[0028] Step 1 includes the following sub-steps: Step 1.1, Digital Orthophoto DOM Construction: Based on the image data taken by the UAV, the target farmland plot is image stitched and orthorectified to generate a digital orthophoto DOM covering the entire plot area, so as to ensure the true proportional relationship of the structure such as inter-row crops, inter-row passages and drainage ditches in the farmland in the plane coordinate system. Step 1.2, DOM segmentation and scale normalization: The digital orthophoto is segmented and cropped to divide the large DOM into several sub-images of preset sizes, and scale unification and format normalization are performed on each sub-image to form a dataset suitable for subsequent path recognition and planning.

[0029] Before proceeding to step 2, based on the digital orthophoto dataset constructed in step 1, samples are labeled for areas such as inter-row passages, crop ridges, and drainage ditches in farmland plots. The labeled samples are then used to train a path structure recognition model, enabling the model to perform pixel-level recognition of farmland path structures. The trained model is then used for inter-row passage area recognition in subsequent steps.

[0030] Step 2, Multi-scale identification of inter-row crop passage areas: On the DOM dataset obtained in Step 1, farmland structure identification processing is performed to distinguish areas such as inter-row passages, crop ridges and drainage ditches, and to extract inter-row crop passage areas; the identification results are represented in the form of pixel-level masks to describe the spatial distribution range of inter-row passage areas in the plot.

[0031] The connectivity analysis and noise removal processes are performed on the access area mask to remove scattered areas and obvious non-accessible structures, resulting in an effective access area mask for path planning.

[0032] Step 2 includes the following sub-steps: Step 2.1, Initial identification of inter-row crop passage areas: On the DOM dataset obtained in Step 1, load the passage area image segmentation model trained in Step 1, perform semantic segmentation processing on the plot image, distinguish areas such as inter-row passages and crop ridge drainage ditches, and generate pixel-level masks of inter-row crop passage areas to describe the spatial distribution range of the passage areas in the plot; the identification results are represented in the form of pixel-level masks to describe the spatial distribution range of inter-row crop passage areas in the plot; Step 2.2, Connectivity analysis and noise removal of the passage area mask: Perform connectivity analysis on the passage area mask obtained in Step 2.1 to identify continuous passage area structures; based on area threshold, spatial location and morphological characteristics, remove scattered areas, discontinuous structures and areas that obviously do not have passage significance to obtain an effective inter-row crop passage area mask for path planning.

[0033] Step 3, Path planning based on inter-row crop structure: Based on the effective passage area mask obtained in Step 2, the centerline of the passage area is extracted to obtain the skeleton representation of the inter-row passage path; the main direction constraint and structural consistency analysis are performed on the skeleton, and branch removal and break repair are carried out to obtain a set of continuous and directional inter-row centerline paths.

[0034] The inter-row centerline paths are clustered, grouped, and sorted to automatically distinguish different inter-row paths and determine the walking order within the plot. Based on the spatial relationship between the endpoints of adjacent inter-row paths, cross-row connection paths are generated at the ridge ends to form a global path planning result that combines inter-row straight paths and ridge end turning paths. The plot entrance and exit paths are determined to obtain a global navigation path that covers the plot area, is continuous, and is executable.

[0035] Step 3 includes the following sub-steps: Step 3.1, Centerline Extraction and Main Direction Determination of Inter-row Crop Travel Area: Based on the effective inter-row crop travel area mask obtained in Step 2, centerline extraction and direction analysis are performed on the travel area to obtain the inter-row crop centerline path for path planning; the process is as follows: 3.1.1) Skeletonization of the passage area: Assume the effective passage area mask output from step 2 is a binary image: ; in, = 1 represents a pixel. Located within the area where crops pass between ridges; Skeletonization is performed on the binary access region mask, converting the region representation into a set of central skeletons with a width of one pixel. ; in, , representing the set of pixel points along the center line of the inter-ridge passage area; By using skeletonization, the geometric width information of the passage area is compressed while maintaining the overall topology of the passage area. Only the skeleton structure that reflects the geometric center distribution of the passage area is retained, providing a basis for subsequent path structure analysis and direction determination.

[0036] 3.1.2) Segmented representation based on the centerline of connected components: for the skeleton set Connectivity analysis is performed, and the skeleton is divided into several candidate center line segments based on the 8-neighbor connectivity of each pixel: ,in, Indicates the first The set of pixels corresponding to a connected center line segment; Each centerline segment This represents a potential inter-row crop passage path, but it may contain branching structures caused by noise, shading, or segmentation errors.

[0037] 3.1.3) Principal Direction Analysis (PCA): For each candidate centerline segment... Represent its pixel coordinates as a two-dimensional point set: ; Calculate the mean vector of the point set: ; And construct the covariance matrix: ; Perform eigenvalue decomposition on the covariance matrix and extract the eigenvector corresponding to the largest eigenvalue: ; The feature vector Indicates the first The main direction vector of the center line segment is used to describe the overall extension direction of the crop arrangement between ridges.

[0038] 3.1.4) Centerline Direction Consistency Constraint and Initial Screening: Based on the prior characteristic that crops between ridges in farmland have an overall consistent direction, direction consistency constraint analysis is performed on the principal direction vectors of all centerline segments; Let the principal direction vectors of any two centerline segments be respectively and Calculate the included angle: ; When the included angle When the value exceeds the preset threshold, the corresponding center line segment is identified as an abnormal direction line segment and removed, thereby eliminating non-row-together directional structures caused by noise, local occlusion, or misidentification. Among them, the abnormal direction line segments include short branches caused by local occlusion, lateral or oblique skeleton structures caused by noise or missegmentation, and non-passage structures that do not conform to the overall arrangement direction of crops between rows; By using directional consistency constraints, interference segments with abnormal branch structures and directions can be effectively removed, leaving only the center line segments that are consistent with the main direction of the inter-row crops.

[0039] 3.1.5) Output Results: Through the above-described skeletonization, connected component segmentation, principal direction analysis, and direction consistency constraint processing, the final set of inter-row crop centerline paths that satisfy the following conditions is obtained: Located within the valid passage area; It has a continuous geometric structure; The main direction is consistent with the direction of the inter-row crop arrangement; Abnormal line segments caused by noise, occlusion, or missegmentation were removed; The set of centerline paths will serve as the input basis for subsequent centerline correction, path grouping, and global path planning.

[0040] Step 3.2, Branch Removal and Fragment Repair of Centerline Paths: Based on the set of inter-row crop centerline paths obtained in Step 3.1, structural optimization is performed on the centerline paths to eliminate branch structures caused by noise, occlusion, or local misidentification, and to repair any broken areas in the centerlines, resulting in continuous and structurally sound inter-row crop travel paths. Note that the directional consistency constraint in Step 3.1 is used for initial screening of centerline segments, while the directional consistency constraint in Step 3.2 performs topology-level optimization and fragment repair on the remaining paths based on the graph structure; the process is as follows: 3.2.1) Graph Theory Modeling of the Centerline Path: The pixel set of the centerline path obtained in step 3.1 is represented as an undirected graph structure: ; in, This represents the set of pixel nodes along the center line, with each node corresponding to the coordinates of a single pixel. , This represents the connection relationship between pixel nodes. If the node and If the skeleton image satisfies the 8-neighborhood connectivity relationship, then an undirected edge is established between the two. In a graph structure, the degree of a node is defined as: ; when When 1 is true, the node is a path endpoint; when At time 2, the node is a normal path node; when At that time, the node is a path branch node.

[0041] By using graph theory modeling, the geometric structure of the centerline can be explicitly represented as the topological relationship between nodes and edges, thus providing a unified structural analysis framework for the identification of branch structures and the detection of path breaks.

[0042] 3.2.2) Identification and determination of branch structure: In graph G, starting from the branch node, traverse all the path branches connected to it until an endpoint node or the next branch node is encountered, thus obtaining several candidate branch paths: ; in, This represents the first branch starting from a certain branch node. Branch paths; For each candidate branch path The corresponding pixel coordinates are represented as ; 3.2.3) Distance-angle joint constraint for branch culling: For each candidate branch path Calculate their geometric lengths respectively: ; in ; Simultaneously, based on the inter-row principal direction vector calculated in step 3.1... Calculate the overall direction vector of the branch path. And calculate the angle between them: ; When candidate branch path A branch structure is identified and removed if it meets all of the following conditions: ; in, This is the preset minimum effective path length threshold. An angle threshold that is inconsistent with the main direction between rows; By using the above distance-angle joint constraints, short, off-direction branch paths caused by local noise, weed obstruction, or segmentation errors can be effectively removed, while preserving the true main driving path between rows.

[0043] 3.2.3) Centerline Path Break Detection Based on Rotated Bounding Box: Based on the spatial characteristics of the lateral separation and stable spacing of crop rows in farmland, a rotated bounding box is introduced to limit the break repair range. After branch removal, the endpoint nodes of each path in the centerline path set are extracted. Let the endpoint set be: ; To avoid erroneous connections between different inter-row paths, a rotational bounding box constraint based on the main crop direction between rows is introduced to spatially limit break detection, including: 3.2.3.1 Principal Direction Estimation: For each centerline path or path combination, the principal component analysis method is used to calculate the principal direction angle based on the second-order statistics of its pixels, which is used as the direction of the rotating bounding box; 3.2.3.2 Constructing the Initial Rotating Strip: Using the geometric center of the centerline path as the frame center, construct a rotating rectangle in the main direction with a length greater than the actual path length. Set its width to the preset inter-row coverage width to cover any possible breaks in the same inter-row path. Note that the rotating strips are generated in descending order of the length of the centerline path. Once any centerline is assigned to any strip, it will no longer generate a separate strip. When the previously generated strip contains more than 90% of the pixels of other centerlines, the centerline can be assigned to that strip. At this time, the centerline paths within the same strip are considered as the same group of centerlines. The original strip must be discarded, and a new strip is generated using this group of centerlines to continue the above process. This method can automatically calibrate the strip angle during the strip generation process. 3.2.3.3 Filtering endpoints within the bounding box: Only filter endpoints of different centerline paths within the same rotating bounding box. Perform fracture detection; For each pair of endpoints that satisfy the rotation bounding box constraint, obtain its pixel coordinates. And calculate the Euclidean distance: ; Simultaneously, the local direction vector at the endpoint is calculated based on the adjacent path pixels within the endpoint neighborhood. And calculate the included angle of direction: ; 3.2.4) Determination and connection of centerline path breakage repair: When any pair of endpoints located within the same rotating bounding box... and If both conditions are met simultaneously, they are determined to be broken sections of the same inter-row path, and connection repair is performed: ;

[0044] in, This is the maximum allowed connection distance threshold; The directional continuity angle threshold; Under the condition that the above conditions are met, the two centerline paths are connected by interpolating between the endpoints to generate a connection path; the connection path can be generated by straight line interpolation or smooth curve interpolation to restore the continuity of the centerline path in spatial position and direction.

[0045] By introducing a rotational bounding box constraint, it can be ensured that fracture repair only occurs within the same row of crops, avoiding accidental cross-row connections.

[0046] 3.2.5) Output Results: Through branch removal and break repair, the final set of inter-row crop centerline paths that meet the following conditions is obtained: The path structure is continuous, with no obvious breaks; It does not include branch structures that deviate from the main direction of the inter-row crops; The geometric shape is consistent with the direction of the inter-row crop arrangement; The optimized centerline path set will serve as the input basis for subsequent inter-row path grouping, sorting, and global path generation.

[0047] Step 3.3, Grouping and Sorting of Inter-row Centerline Paths: Based on the set of continuous, branchless inter-row crop centerline paths obtained in Step 3.2, the centerline paths are grouped and sorted to distinguish different inter-row crop rows and determine the driving order of agricultural machinery within the plot, providing an ordered path structure for subsequent global path generation; the process is as follows: 3.3.1) Unified overall direction of centerline paths: Let the set of inter-row centerline paths output in step 3.2 be: ; Among them, each centerline path Composed of a set of ordered pixels: ; Based on the main direction vector of inter-row crops obtained in step 3.1 The direction of each point along the centerline path is unified so that the direction from the start to the end point is consistent with the direction of the centerline path. Maintaining consistency ensures that the direction of subsequent sorting is consistent.

[0048] 3.3.2) Calculation of the projected distance of the centerline path between rows: In order to group different inter-row paths, an inter-row normal direction vector is introduced: ; in, Main direction of inter-row crops Vertical, used to characterize the lateral distribution direction between rows.

[0049] For each centerline path Calculate its centroid coordinates: ; Project the centroid onto the normal direction Above, the projected distance is obtained: ; The projection distance Characterized the first The relative position of the centerline path in the transverse direction between rows.

[0050] 3.3.3) Grouping and row numbering of centerline paths between ridges: based on the projected distance of each centerline path. For the set of centerline paths Sort: ; Then, assign row numbers to the sorted centerline paths in sequence: ; in, Indicates the row number of crops between rows, used to distinguish different paths between rows; In cases where there is local path density or fragmentation, the difference in projected distance between adjacent centerline paths can be further considered: ; when When the value is less than the preset threshold, the corresponding centerline path is determined to be a different segment of the same inter-row path, and its row number identifier is merged.

[0051] 3.3.4) Sequencing of driving order for the centerline path between ridges: After completing the ridge numbering, the driving order of the paths between ridges is sorted according to the actual driving needs of agricultural machinery operations; Preferably, the rows are sorted in ascending or descending order of their row numbers to form a basic driving sequence. Based on this, the alternating or round-trip sequence can be adopted by considering the positional relationship of the endpoints of adjacent row paths to reduce the number of times the rows turn around and improve overall operational efficiency.

[0052] 3.3.5) Output Results: Through the above grouping and sorting processes, the following results are obtained: Each centerline path between rows corresponds to a unique row number; All inter-row paths are grouped according to their lateral spatial distribution. The sequence of agricultural machinery movement between rows within the plot was determined; The grouping and sorting results will serve as input conditions for subsequent monopoly connection and global path generation.

[0053] Step 3.4, Ridge-end Connection and Global Path Generation: Based on the grouping and sorting results of the inter-ridge centerline paths obtained in Step 3.3, adjacent inter-ridge paths are connected at the ridge-end positions, and a global inter-ridge crop travel path covering the entire plot is generated to meet the actual needs of continuous agricultural machinery operation; the process is as follows: 3.4.1) Inter-row centerline path endpoint extraction: Let the k-th inter-row centerline path obtained in step 3.3 be: ; Extract its starting point and ending point as path endpoints respectively: ; The path endpoints are used to describe the spatial location of the inter-row crop travel path in the ridge end area. 3.4.2) Matching determination of endpoints of adjacent ridge paths: Based on the ridge number and driving sequence determined in step 3.3, select the centerline paths of two adjacent ridges. and Calculate the Euclidean distance between their corresponding endpoints: ; And combine the local direction vectors of the two paths at their endpoints , Calculate the included angle of its direction: ; When the distance between endpoints Less than the preset distance threshold, and the directional angle When the continuity constraint is satisfied, it is determined that two inter-row paths can be connected at the ridge end. 3.4.3) Ridge end turning path generation: Under the condition of satisfying the endpoint connection, ridge end turning paths are generated between the endpoints of the path to connect the centerline paths between adjacent ridges; Preferably, the turning path at the ridge end is represented by a circular arc curve or a smooth parametric curve, and its geometric expression can be represented as follows: ; The curve's starting and ending points are connected to the endpoints of adjacent ridge paths, satisfying positional and directional continuity constraints, thus ensuring that agricultural machinery can smoothly complete turning operations in the ridge-end area. 3.4.4) Construction of the global path for inter-row crops: Following the inter-row driving sequence determined in step 3.3, connect each inter-row straight path with its corresponding ridge-end turning path in sequence to form a continuous driving path covering the entire plot: ; in, Indicates the first Straight paths between rows Indicates the connection of the first Article and No. The ridge-end turning path of the inter-row path; 3.4.5) Determination of plot entrance and exit paths: During the global path construction process, based on the spatial distribution relationship between plot boundaries and inter-row paths, the endpoints of inter-row paths located at the edge of the plot are selected as plot entrance and exit locations, and connected to the beginning and end of the global path to form a complete plot entry and exit path. 3.4.5) Output Results: Through the above ridge-end connection and path construction processing, the global path planning results for inter-ridge crops that meet the following conditions are obtained: The path is continuous and unbroken within the inter-row area; The ridge-end turning path is smooth, meeting the turning requirements of agricultural machinery; The global path covers all inter-row crop rows in the plot; The path results are executable and can be directly used for agricultural machinery operation; The global path planning result serves as the input data for the path output in step 4.

[0054] Step 4, Path Result Output: Convert the global navigation path generated in Step 3 into standard path data format and output it as a reference path for ground operation equipment.

[0055] In step 4, the ground-side application outputs standard path data as a reference path for the ground-side equipment. The ground-side equipment then uses this path data to complete inter-row travel and cross-row operations, thereby realizing farmland operations based on the aerial path planning results.

[0056] Step 4 includes the following sub-steps: Step 4.1, Coordinate Mapping and Representation of Global Path Data: The global path planning results obtained in Step 3.4 are mapped from the image coordinate system to the geographic coordinate system or the farmland operation coordinate system, so that each path point in the path has clear spatial location information; the path is represented in the form of an ordered sequence of path points, used to describe the continuous driving trajectory of agricultural machinery within the farmland plot. The coordinate mapping can be completed based on the spatial reference information of digital orthophotos or a ground-based positioning system. Step 4.2, Formatted output of path data: The mapped path data is formatted to generate a standard path data file. The path data file contains at least path point coordinate information and path sequence information, which are used to characterize the driving route and operation sequence of agricultural machinery in the inter-row crop environment. Step 4.3, Ground-based loading and retrieval of path results: Before operation, the ground-based operating equipment reads the path data file generated in step 4.2 and loads the path data as a navigation reference path into the ground-based control system. This guides the agricultural machinery to enter the field, travel along the inter-row path, and complete cross-row turning operations. Step 4.4, Ground-end constrained driving based on global path: During the actual driving of the agricultural machinery, the ground-end control system will match its own position information acquired in real time with the global path, and complete the inter-row driving and ridge-end turning operations under the constraints of the global path, thereby ensuring that the agricultural machinery drives stably in the farmland according to the path planned by the aerial end.

[0057] Step 4.5, Output Results: By outputting the above path and the ground application process, the following effects are achieved: The global path for inter-row crops generated in the air can be directly read and executed by ground-based equipment. The travel paths of agricultural machinery within farmland plots exhibit good continuity and consistency; Reduce the workload of manual path planning and guidance, and improve the efficiency and stability of farmland operations.

[0058] This embodiment uses large-scale farmland image data acquired from an aerial terminal as input, and through multi-scale path structure analysis and processing, it realizes automatic identification and global path planning of path structures such as inter-row passages and drainage ditches in farmland plots.

[0059] This embodiment uses remote sensing images of farmland acquired by inspection drones as input. The ground station performs multi-scale processing and path structure analysis on the image data to automatically identify the passageway structure such as the passage between rows and drainage ditches in the farmland, and further generates a global driving path that can be directly executed by the ground mobile robot.

[0060] like Figure 1 As shown, the overall process of the method of the present invention includes steps such as aerial image acquisition, ground station image processing and path recognition, and path planning result output, forming a complete processing framework from image input to path generation.

[0061] like Figure 2 and Figure 3As shown, the agricultural machinery travel path generated based on the method of this invention presents a regular and continuous covering trajectory within the farmland. The agricultural machinery travels straight along the centerline of each ridge-to-ridge passage area, and upon reaching the ridge end, it turns around across the ridge according to a pre-planned turning path, then enters the adjacent ridge-to-ridge passage area to continue traveling, thus achieving continuous coverage of the entire plot. Through the above path planning method, the agricultural machinery's travel path within the plot maintains good continuity and consistency, effectively avoiding problems such as repeated travel and missed work caused by inaccurate manual path planning or ad-hoc manual judgment, significantly reducing the workload of manual guidance and path planning, and improving the operational efficiency and stability during farmland operations.

[0062] Figure 4 This is a diagram illustrating the entire process of farmland path identification and planning methods. (Example:) Figure 3 As shown, the intermediate results of the inter-row passage area at different processing stages are as follows: First, the initial inter-row passage area mask is obtained by semantic segmentation, which includes the fracture, noise and local adhesion structures caused by factors such as crop shading and light changes; then, through connectivity analysis and morphological processing, the fragmented noise area and non-major passage structure are removed, and the continuity of the inter-row passage area is enhanced. The cleaned passage area undergoes skeletonization to extract its initial path shape. Subsequently, through directional consistency constraints and abnormal branch removal, the bending, jittering, and redundant branches in the centerline are optimized. Then, based on the slope and positional relationship of the resulting fragmented line segments, a clustering operation is performed. Fragmented line segments within the same box (of the same type) are then connected, and after two smoothing algorithms, a continuous, smooth, and independent path result is obtained. Finally, an executable vehicle path map (i.e., the vehicle's center path map) is generated using an algorithm. Figure 4 As can be seen from the full process shown, this invention can effectively address the difficulties in path recognition caused by factors such as crop occlusion and regional adhesion in farmland scenarios. While maintaining the overall consistency of the inter-row structure, it can achieve stable path extraction, providing a reliable path foundation for subsequent global path planning and autonomous robot driving.

[0063] Through the above-mentioned multi-scale path recognition and planning methods, this invention realizes the automated generation of global driving paths for ground mobile robots from aerial images, providing reliable path data support for autonomous operations in farmland scenarios.

[0064] The ground robot in the air-ground cooperative path recognition and global path planning system of this embodiment is described as follows: The ground robot is a mobile platform with autonomous walking capabilities. The mobile platform adopts a four-wheel steering and four-wheel drive structure, which can stably travel along the inter-row passage in the crop planting plot and complete the cross-row turning operation.

[0065] The ground robot includes a vehicle body, a walking mechanism, a control unit, and a positioning module. The walking mechanism drives the robot to move within the site; the control unit receives global path planning results generated by the ground station and executes path tracking control; the positioning module includes an RTK positioning system for acquiring the robot's real-time position and heading information within the site.

[0066] During operation, the ground robot matches the real-time location information obtained by its own RTK system with the global path point sequence generated by the method of this invention, thereby achieving autonomous driving along the center line of the inter-row passage; when it reaches the end of the ridge, the robot completes the cross-ridge movement according to the planned turning path and enters the adjacent inter-row passage to continue driving.

[0067] The above-described ground robot structure is used to illustrate the execution carrier of the path recognition and global path planning method of the present invention, and its specific structural form does not constitute a limitation on the scope of protection of the present invention.

[0068] The air-ground cooperative path identification and global path planning method in this embodiment, such as Figure 1 As shown, the system mainly consists of three parts: an inspection drone, a ground station, and a ground mobile robot. The method uses the inspection drone to acquire high-throughput aerial images of the land parcels. The ground station performs multi-scale processing on the images and extracts path structure features such as inter-row passages and drainage ditches, thereby generating a global path planning result that can be executed by the ground mobile robot.

[0069] Compared to traditional methods that rely on human experience for field navigation planning, this invention can automatically identify and model the path structure of a plot, generating continuous and regular driving paths. This effectively improves the autonomous navigation stability and path execution efficiency of ground robots in complex farmland environments. This method provides reliable technical support for the automated inspection and operation path planning of intelligent agricultural machinery in crop-growing plots.

[0070] The embodiments described in this specification are merely examples of implementations of the inventive concept and are for illustrative purposes only. The scope of protection of this invention should not be considered limited to the specific forms described in these embodiments; rather, it extends to equivalent technical means conceived by those skilled in the art based on the inventive concept.

Claims

1. A method for farmland path recognition and planning based on aerial imagery, characterized in that, The method includes the following steps: Step 1, Dataset Preparation and Digital Orthophoto Construction: Based on the spatial range and crop planting layout of the target farmland plot, aerial image data covering the target plot is acquired, and the original images are screened, registered, and scaled. A digital orthophoto DOM of the target plot is generated through image stitching and geometric correction. The DOM is then segmented, cropped, and scaled to form a dataset for subsequent path recognition and planning. Step 2, Multi-scale identification of inter-row crop passage areas: On the DOM dataset, farmland structure identification processing is performed to distinguish between inter-row passages, crop ridges and drainage ditch areas, and inter-row crop passage areas are extracted. The identification results are represented in the form of pixel-level masks to describe the spatial distribution range of inter-row passage areas in the plot. Connectivity analysis and noise removal processing are performed on the passage area masks to obtain effective passage area masks for path planning. Step 3, Path planning based on inter-row crop structure: Extract the centerline of the passage area to obtain the skeleton representation of the inter-row passage path; perform main direction constraint and structural consistency analysis on the skeleton, perform branch removal and break repair to obtain a set of continuous and directional inter-row centerline paths; Clustering and sorting of the centerline paths between ridges, automatically distinguishing different inter-ridge paths and determining the walking order within the plot; based on the spatial relationship of the endpoints of adjacent inter-ridge paths, generating cross-ridge connection paths at the ridge ends, forming a global path planning result that combines inter-ridge straight paths and ridge end turning paths, and determining the plot entrance and exit paths, thereby obtaining a global navigation path that covers the plot area, is continuous and executable. Step 4, Path Result Output: Convert the global navigation path generated in Step 3 into standard path data format and output it as a reference path for ground operation equipment.

2. The method for farmland path recognition and planning based on aerial imagery as described in claim 1, characterized in that, Step 1 includes the following sub-steps: Step 1.1, Digital Orthophoto DOM Construction: Based on the image data taken by the UAV, the target farmland plot is image stitched and orthorectified to generate a digital orthophoto DOM covering the entire plot area, so as to ensure the true proportional relationship of the inter-row crops, inter-row passages and drainage ditch structures in the farmland in the plane coordinate system. Step 1.2, DOM segmentation and scale normalization: The digital orthophoto is segmented and cropped to divide the large DOM into several sub-images of preset sizes, and scale unification and format normalization are performed on each sub-image to form a dataset suitable for subsequent path recognition and planning.

3. The method for farmland path recognition and planning based on aerial imagery as described in claim 2, characterized in that, Based on the constructed digital orthophoto dataset, samples of inter-row passages, crop ridges, and drainage ditch areas in farmland plots are labeled, and a path structure recognition model is trained using the labeled samples, enabling the model to perform pixel-level recognition of farmland path structures; the trained model is used for inter-row passage area recognition in subsequent steps.

4. A method for farmland path recognition and planning based on aerial imagery as described in any one of claims 1 to 3, characterized in that, Step 2 includes the following sub-steps: Step 2.1, Initial identification of inter-row crop passage areas: On the DOM dataset obtained in Step 1, load the passage area image segmentation model trained in Step 1, perform semantic segmentation processing on the plot image, distinguish between inter-row passages and crop ridge drainage ditch areas, and generate pixel-level masks of inter-row crop passage areas to describe the spatial distribution range of the passage areas in the plot; the identification results are represented in the form of pixel-level masks to describe the spatial distribution range of inter-row crop passage areas in the plot; Step 2.2, Connectivity analysis and noise removal of the passage area mask: Perform connectivity analysis on the passage area mask obtained in Step 2.1 to identify continuous passage area structures; based on area threshold, spatial location and morphological characteristics, remove scattered areas, discontinuous structures and areas that obviously do not have passage significance to obtain an effective inter-row crop passage area mask for path planning.

5. A method for farmland path recognition and planning based on aerial imagery as described in any one of claims 1 to 3, characterized in that, Step 3 includes the following sub-steps: Step 3.1, Extraction of centerline and determination of main direction of inter-row crop passage area: Based on the effective inter-row crop passage area mask obtained in step 2, centerline extraction and direction analysis are performed on the passage area to obtain the inter-row crop centerline path for path planning; Step 3.2, Branch Removal and Break Repair of Centerline Path: Based on the set of inter-row crop centerline paths obtained in Step 3.1, the centerline paths are structurally optimized to eliminate branch structures caused by noise, occlusion, or local misidentification, and the broken areas in the centerline are repaired to obtain a continuous and structurally reasonable inter-row crop travel path. Step 3.3, Grouping and sorting of inter-row centerline paths: Based on the set of continuous, branchless inter-row crop centerline paths obtained in Step 3.2, each centerline path is grouped and sorted to distinguish different inter-row crop rows and determine the driving order of agricultural machinery within the plot, providing an ordered path structure for subsequent global path generation. Step 3.4, ridge end connection and global path generation: Based on the grouping and sorting results of the inter-ridge centerline paths obtained in Step 3.3, the adjacent inter-ridge paths are connected at the ridge end positions, and a global driving path for inter-ridge crops covering the entire plot is generated to meet the actual needs of continuous agricultural machinery operation.

6. The method for farmland path recognition and planning based on aerial imagery as described in claim 5, characterized in that, The process of step 3.1 is as follows: 3.1.1) Skeletonization of Passage Area: Assume the effective passage area mask output in step 2 is a binary image; perform skeletonization on the binary passage area mask to convert the region representation into a set of central skeletons with a width of one pixel; 3.1.2) Segmented representation based on the centerline of connected components: for the skeleton set Connectivity analysis is performed, and based on the 8-neighbor connectivity of each pixel, the skeleton is divided into several candidate centerline segments. This represents a potential inter-row crop passage path, but it may contain branching structures caused by noise, shading, or segmentation errors. 3.1.3) Centerline Main Direction Analysis: For each candidate centerline segment The pixel coordinates are represented as a two-dimensional point set; the mean vector of the point set is calculated and a covariance matrix is ​​constructed; the covariance matrix is ​​subjected to eigenvalue decomposition, and the eigenvector corresponding to the largest eigenvalue is taken; the eigenvector... Indicates the first The main direction vector of the center line segment is used to describe the overall extension direction of the crop arrangement between ridges; 3.1.4) Centerline Direction Consistency Constraint and Initial Screening: Based on the prior knowledge that crops between ridges in farmland have an overall consistent direction, direction consistency constraint analysis is performed on the principal direction vectors of all centerline segments; let the principal direction vectors of any two centerline segments be respectively... and Calculate the included angle When the included angle When the value exceeds the preset threshold, the corresponding center line segment is identified as an abnormal direction line segment and removed, thereby eliminating non-row-together directional structures caused by noise, local occlusion, or misidentification. 3.1.5) Output Results: Through the above-described skeletonization, connected component segmentation, principal direction analysis, and direction consistency constraint processing, the final set of inter-row crop centerline paths that satisfy the following conditions is obtained: Located within the valid passage area; It has a continuous geometric structure; The main direction is consistent with the direction of the inter-row crop arrangement; Abnormal line segments caused by noise, occlusion, or missegmentation were removed.

7. The method for farmland path recognition and planning based on aerial imagery as described in claim 6, characterized in that, The process of step 3.2 is as follows: 3.2.1) Graph modeling of the centerline path: The pixel set of the centerline path obtained in step 3.1 is represented as an undirected graph structure; in the graph structure, the degree of the nodes is defined. Through graph modeling, the geometric structure of the centerline can be explicitly represented as the topological relationship between nodes and edges, thus providing a unified structural analysis framework for the identification of branch structures and the detection of path breaks. 3.2.2) Identification and determination of branch structure: In graph G, starting from the branch node, traverse all its connected path branches until an endpoint node or the next branch node is encountered, thus obtaining several candidate branch paths; for each candidate branch path Give the corresponding pixel coordinates; 3.2.3) Distance-angle joint constraint for branch culling: For each candidate branch path Calculate their geometric lengths respectively, and simultaneously, based on the inter-row principal direction vectors calculated in step 3.

1. Calculate the overall direction vector of the branch path. And calculate the angle between the two; when the candidate branch path If the following conditions are met simultaneously, it will be identified as a branch structure and removed; 3.2.3) Centerline path break detection based on rotating bounding box: Based on the spatial characteristics of the crop rows between ridges in farmland being separated from each other in the lateral direction and having stable spacing, a rotating bounding box is introduced to limit the break repair range. After the branches are removed, the endpoint nodes of each path in the centerline path set are extracted. 3.2.4) Determination and connection of centerline path breakage repair: When any pair of endpoints located within the same rotating bounding box... and When the following conditions are met simultaneously, it is determined that the two are broken parts of the same inter-row path, and connection repair is performed; when the above conditions are met, a connection path is generated by interpolation between the endpoints to connect the two centerline paths; the connection path can be generated by straight line interpolation or smooth curve interpolation to restore the continuity of the centerline path in spatial position and direction. 3.2.5) Output Results: Through branch removal and break repair, the final set of inter-row crop centerline paths that meet the following conditions is obtained: The path structure is continuous, with no obvious breaks; It does not include branch structures that deviate from the main direction of the inter-row crops; The geometric shape is consistent with the direction of the inter-row crop arrangement; The optimized centerline path set will serve as the input basis for subsequent inter-row path grouping, sorting, and global path generation.

8. The method for farmland path recognition and planning based on aerial imagery as described in claim 7, characterized in that, In section 3.2.3, to avoid erroneous connections between different inter-row paths, a rotational bounding box constraint based on the main direction of the inter-row crops is introduced to spatially limit the break detection, including: 3.2.3.1 Principal Direction Estimation: For each centerline path or path combination, the principal component analysis method is used to calculate the principal direction angle based on the second-order statistics of its pixels, which is used as the direction of the rotating bounding box; 3.2.3.2 Constructing the Initial Rotating Strip: Using the geometric center of the centerline path as the frame center, construct a rotating rectangle in the main direction with a length greater than the actual path length. Set its width to the preset inter-row coverage width to cover any possible breaks in the same inter-row path. Note that the rotating strips are generated in descending order of the length of the centerline path. Once any centerline is assigned to any strip, it will no longer generate a separate strip. When the previously generated strip contains more than 90% of the pixels of other centerlines, the centerline can be assigned to that strip. At this time, the centerline paths within the same strip are considered as the same group of centerlines. The original strip must be discarded, and a new strip is generated using this group of centerlines to continue the above process. This method can automatically calibrate the strip angle during the strip generation process. 3.2.3.3 Filtering endpoints within the bounding box: Only filter endpoints of different centerline paths within the same rotating bounding box. Perform break detection; for endpoint pairs that satisfy the rotated bounding box constraint, obtain their pixel coordinates respectively. The Euclidean distance is calculated, and the local direction vector at the endpoint is calculated based on the adjacent path pixels within the endpoint neighborhood. And calculate the included angle of direction.

9. The method for farmland path recognition and planning based on aerial imagery as described in claim 7, characterized in that, The process of step 3.3 is as follows: 3.3.1) Unified overall direction of centerline paths: Let the set of inter-row centerline paths output in step 3.2 be defined, and each centerline path... It consists of a set of ordered pixels; based on the main direction vector of the inter-row crops obtained in step 3.

1. The direction of each point along the centerline path is unified so that the direction from the start to the end point is consistent with the direction of the centerline path. Maintaining consistency ensures that the direction of subsequent sorting is consistent; 3.3.2) Calculation of the projected distance of the centerline path between rows: In order to group different inter-row paths, an inter-row normal direction vector is introduced. For each centerline path... Calculate its centroid coordinates; and project the centroid onto the normal direction. The projected distance is obtained from the above. 3.3.3) Grouping and row numbering of centerline paths between ridges: based on the projected distance of each centerline path. For the set of centerline paths The paths are sorted; and row numbers are assigned to the sorted centerline paths in sequence; in cases where there are dense or fragmented local paths, the difference in projection distance between adjacent centerline paths can be further considered. ;when When the value is less than the preset threshold, the corresponding centerline path is determined to be different segments of the same inter-row path, and its row number identifier is merged. 3.3.4) Sequencing of driving order for the centerline path between ridges: After completing the ridge numbering, the driving order of the paths between ridges is sorted according to the actual driving needs of agricultural machinery operations; 3.3.5) Output Results: Through the above grouping and sorting processes, the following results are obtained: Each centerline path between rows corresponds to a unique row number; All inter-row paths are grouped according to their lateral spatial distribution. The sequence of agricultural machinery movement between rows within the plot was determined; The grouping and sorting results will serve as input conditions for subsequent monopoly connection and global path generation.

10. The method for farmland path recognition and planning based on aerial imagery as described in claim 9, characterized in that, The process of step 3.4 is as follows: 3.4.1) Extraction of the endpoints of the inter-row centerline path: Let the k-th inter-row centerline path obtained in step 3.3 be extracted; its starting point and ending point are extracted as the path endpoints, which are used to describe the spatial position of the inter-row crop travel path in the ridge end area; 3.4.2) Matching determination of endpoints of adjacent ridge paths: Based on the ridge number and driving sequence determined in step 3.3, select the centerline paths of two adjacent ridges. and Calculate the Euclidean distance between the corresponding endpoints; and combine this with the local direction vectors of the two paths at the endpoints. , Calculate the included angle of its direction, when the distance between the endpoints Less than the preset distance threshold, and the directional angle When the continuity constraint is satisfied, it is determined that two inter-row paths can be connected at the ridge end. 3.4.3) Ridge end turning path generation: Under the condition of satisfying the endpoint connection, ridge end turning paths are generated between the endpoints of the path to connect the centerline paths between adjacent ridges; 3.4.4) Construction of global path for inter-row crops: According to the inter-row driving sequence determined in step 3.3, connect each inter-row straight path with the corresponding ridge end turning path in sequence to form a continuous driving path covering the entire plot area; 3.4.5) Determination of plot entrance and exit paths: During the global path construction process, based on the spatial distribution relationship between plot boundaries and inter-row paths, the endpoints of inter-row paths located at the edge of the plot are selected as plot entrance and exit locations, and connected to the beginning and end of the global path to form a complete plot entry and exit path. 3.4.5) Output Results: Through the above ridge-end connection and path construction processing, the global path planning results for inter-ridge crops that meet the following conditions are obtained: The path is continuous and unbroken within the inter-row area; The ridge-end turning path is smooth, meeting the turning requirements of agricultural machinery; The global path covers all inter-row crop rows in the plot; The path results are executable and can be directly used for agricultural machinery operation; The global path planning result serves as the input data for the path output in step 4.

Images