A vineyard quadruped robot inspection method and system

Abstract

This invention provides a method and system for inspecting vineyards using a quadruped robot, belonging to the field of agricultural robots. The method includes collecting meteorological data, soil condition data, and historical inspection anomaly point data from the vineyard, integrating them to obtain a basic inspection dataset; dividing the vineyard into inspection area units, performing risk assessments on each inspection area unit to obtain inspection priorities; controlling the quadruped robot to perform inspections according to the inspection path; when a new anomaly point is identified, updating the inspection priority, inserting the inspection area unit as a re-inspection node into the sequence of unexecuted inspection tasks, and partially rearranging the current inspection path; after the inspection is completed, summarizing the inspection results and updating the historical inspection anomaly point data. This invention generates priority paths and responds in real time to new anomalies during inspections, triggering online re-arrangement of tasks and paths. This enables the robot to prioritize and quickly re-inspect sudden high-risk points.

Description

Technical Field

[0001] This invention relates to the field of agricultural robot technology, specifically to a method and system for inspecting vineyards using a quadruped robot. Background Technology

[0002] Vineyards face complex environments with narrow row spacing, undulating terrain, and a tendency for low-lying, waterlogged areas. Traditional wheeled or tracked inspection robots have poor adaptability, often experiencing problems such as getting stuck, slipping, and insufficient obstacle-crossing ability, making it difficult to achieve full-coverage inspections in these complex field environments. In contrast, quadruped robots possess excellent terrain adaptability, obstacle-crossing ability, and flexible mobility, and are increasingly being applied to field operations.

[0003] Existing quadruped robot inspection solutions mostly employ fixed paths and task sequences for inspection. Even if new anomalies are identified in real time during the inspection process, the inspection logic cannot be flexibly adjusted, making it difficult to quickly conduct re-inspections of abnormal areas, resulting in delays in anomaly handling. This not only wastes the flexibility and maneuverability of quadruped robots but also prevents inspection resources from being allocated to high-risk areas, failing to fully utilize overall inspection efficiency and making it difficult to meet the refined and timely inspection management needs of vineyards. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for inspecting vineyards using a quadruped robot, which enables intelligent and dynamic inspection of vineyards, allows for rapid re-inspection when anomalies are detected, and improves the efficiency and precision of vineyard planting management.

[0005] To achieve the above objectives, this invention provides a method for inspecting vineyards using a quadruped robot, comprising: collecting meteorological data, soil condition data, and historical inspection anomaly point data of the vineyard, and integrating them to obtain an inspection base dataset; based on the inspection base dataset, dividing the vineyard into inspection area units, performing risk assessment on each inspection area unit, and obtaining inspection priorities; generating an inspection task sequence according to the inspection priorities, and generating an inspection path in conjunction with the vineyard row structure; controlling the quadruped robot to perform inspections according to the inspection path, and collecting inspection data during the inspection process to identify anomalies; when a new anomaly point is identified, updating the inspection priority based on the inspection area unit corresponding to the new anomaly point, inserting the inspection area unit as a re-inspection node into the unexecuted inspection task sequence, and partially rearranging the current inspection path; controlling the quadruped robot to perform inspections according to the reordered and re-inspection node-inserted inspection task sequence and the partially rearranged inspection path; and after the inspection is completed, summarizing the inspection results and updating the historical inspection anomaly point data.

[0006] Optionally, the step of risk assessment for each of the inspection area units to obtain inspection priorities includes: extracting risk-related features for each of the inspection area units; normalizing each of the risk-related features and calculating the comprehensive risk value of each of the inspection area units according to preset weights; acquiring multimodal perception data of each of the inspection area units and performing anomaly identification, fusing the results of each modality identification to obtain anomaly fusion confidence; and calculating the inspection priority of each of the inspection area units based on the comprehensive risk value and the anomaly fusion confidence, combined with the re-inspection demand coefficient, execution energy consumption, and path switching cost.

[0007] Optionally, the risk-related characteristics include frost risk factors, wet damage risk factors, disease transmission factors, historical abnormal residue factors, low-lying water accumulation factors, and accessibility factors.

[0008] Optionally, the step of collecting inspection data to identify abnormal locations during the inspection process includes: collecting multimodal perception data of the target area in real time; performing real-time analysis of the multimodal perception data using an edge-side recognition model deployed on the quadruped robot; identifying whether there are any abnormalities in the target area based on the real-time analysis results, wherein the abnormalities include at least one of abnormal plant growth, abnormal canopy environment, and abnormal soil condition; when an abnormality is identified, recording the spatial location, abnormality type, and abnormality fusion confidence of the abnormal location, and adding the abnormal location to the abnormal location database.

[0009] Optionally, the step of collecting inspection data during the inspection process to identify abnormal points further includes: when the confidence level of the identified abnormal point is lower than a preset confidence threshold, marking the abnormal point as a node to be confirmed and generating a corresponding re-inspection task.

[0010] Optionally, updating the inspection priority based on the inspection area unit corresponding to the newly added anomaly point includes: increasing the re-inspection demand coefficient of the corresponding inspection area unit based on the newly added anomaly point; and recalculating the inspection priority of the inspection area unit based on the increased re-inspection demand coefficient and the anomaly fusion confidence of the newly added anomaly point.

[0011] Optionally, inserting the inspection area unit as a re-inspection node into the unexecuted inspection task sequence includes: calculating the incremental cost generated by inserting the re-inspection node into different positions in the unexecuted inspection task sequence; determining the target insertion position with the minimum incremental cost based on the calculation result; and inserting the re-inspection node into the target insertion position.

[0012] Optionally, calculating the incremental cost includes: obtaining the additional path length generated by inserting the re-inspection node into the candidate position; obtaining the terrain elevation change penalty value and obstacle density penalty value of the area where the re-inspection node is located; and, based on a preset adjustment coefficient, performing a weighted summation of the additional path length, the terrain elevation change penalty value, and the obstacle density penalty value to obtain the incremental cost.

[0013] Optionally, the partial rearrangement of the current inspection path includes: adding the re-inspection node as a new semantic node in the semantic topology graph of the vineyard rows; disconnecting the original association edges between the current node and the subsequent target node in the semantic topology graph; and establishing new association edges between the current node and the new semantic node, and between the new semantic node and the subsequent target node, to update the inspection path.

[0014] On the other hand, the present invention provides a vineyard quadruped robot inspection system for implementing a vineyard quadruped robot inspection method. The system includes a control module, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the computer program to implement the vineyard quadruped robot inspection method.

[0015] The aforementioned technical solution generates priority paths by dynamically assessing risks and responds in real time to new anomalies during inspections, triggering online reordering of tasks and paths. This enables the robot to prioritize and quickly re-inspect sudden high-risk points, improving response speed and inspection efficiency for problems such as defects, while avoiding the energy consumption and oscillations of global reordering.

[0016] Other features and advantages of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the following detailed description to explain the invention, but do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of the inspection method for a quadruped robot in a vineyard.

[0018] Figure 2 This is a flowchart of the dynamic re-inspection and path rearrangement of newly added anomalies. Detailed Implementation

[0019] The following is in conjunction with the appendix Figure 1 - Appendix Figure 2 The specific implementation methods of the embodiments of the present invention will be described in detail below. It should be understood that the specific implementation methods described herein are only for illustrating and explaining the embodiments of the present invention, and are not intended to limit the embodiments of the present invention.

[0020] It should be noted that the acquisition, transmission, storage, use, and processing of data in the technical solution of this application all comply with the relevant provisions of national laws and regulations. In the embodiments of this application, certain existing industry solutions such as software, components, and models may be mentioned. These should be considered exemplary, intended only to illustrate the feasibility of implementing the technical solution of this application, and do not imply that the applicant has already used or necessarily used such solutions.

[0021] In the process of realizing this invention, the inventors of this application discovered that the shortcomings of existing quadruped robot inspection schemes are that their inspection logic is static and preset, and they cannot dynamically adjust tasks and paths according to real-time anomalies during the inspection process, resulting in the inability to prioritize and promptly deal with sudden risks.

[0022] Example 1 Reference Figures 1-2 This is the first embodiment of the present invention, which provides a method for inspecting a vineyard using a quadruped robot, comprising: S100: Collect meteorological data, soil condition data, and historical inspection anomaly point data of the vineyard, and integrate them to obtain the basic inspection dataset.

[0023] In the embodiments of this application, meteorological data is collected through environmental monitoring stations deployed within the vineyard. This meteorological data includes at least ambient temperature, relative humidity, wind speed, rainfall, and frost warning information. Soil condition data is collected through a sensor network deployed in the vineyard soil. This soil condition data includes at least the soil volumetric water content, surface temperature, and humidity trends for each area. Historical inspection anomaly point data is retrieved from a stored historical inspection database. This data includes at least the anomaly coordinates, anomaly type, and anomaly occurrence time. The meteorological data, soil condition data, and historical inspection anomaly point data are timestamped and geospatially registered, and uniformly mapped to the same geographic coordinate system of the vineyard. The multi-source data after spatiotemporal alignment and coordinate mapping are integrated and standardized to obtain a standardized inspection base dataset.

[0024] S200: Based on the basic inspection dataset, the inspection area units of the vineyard are divided, and the risk assessment of each inspection area unit is carried out to obtain the inspection priority.

[0025] In the embodiments of this application, risk assessment is performed on each inspection area unit to obtain the inspection priority, including: extracting risk-related features for each inspection area unit; the risk-related features include frost risk factors, wet damage risk factors, disease transmission factors, historical abnormal residue factors, low-lying water accumulation factors, and accessibility factors. Each risk-related feature is normalized, and a comprehensive risk value for each inspection area unit is calculated according to preset weights; multimodal perception data for each inspection area unit is acquired and anomaly identification is performed, and the anomaly fusion confidence score is obtained by fusing the results of each modality identification; based on the comprehensive risk value and the anomaly fusion confidence score, combined with the re-inspection demand coefficient, execution energy consumption, and path switching cost, the inspection priority for each inspection area unit is calculated.

[0026] It should be noted that, in the stage before generating the initial inspection task sequence, acquiring multimodal perception data refers to extracting historically relevant multimodal data records and synchronized environmental state data from the basic inspection dataset; anomaly identification is based on historical anomaly patterns and current environmental state data, using statistical inference models or rule-based reasoning models to predict the probability of anomalies occurring. The resulting anomaly fusion confidence score is a priori prediction of whether anomalies exist in the area before the inspection begins. For ease of distinction, it can be referred to as the prior anomaly confidence score at this stage.

[0027] In a preferred embodiment of this application, firstly, based on the unified geographic coordinates in the inspection base dataset, the vineyard is divided into multiple inspection area units by segmenting the vineyard rows or by regular rasterization, with each unit associated with a unique spatial location identifier.

[0028] Next, a risk assessment is performed on each inspection area unit to calculate its overall risk value. This process includes extracting and calculating multiple risk factors, including frost risk factors, wet damage risk factors, disease transmission factors, historical abnormal residue factors, low-lying water accumulation factors, and accessibility factors.

[0029] The frost risk factor is based on the ambient temperature, frost warning information and historical low temperature data of the currently assessed inspection area unit and its adjacent areas. It is calculated by using a temperature threshold judgment and duration weighted model to characterize the sensitivity of the area to frost. The wet damage risk factor is based on the soil volumetric water content, humidity change trend and canopy humidity estimate of the currently assessed inspection area unit. It is calculated by the probability of humidity exceeding the threshold and the trend slope to characterize the risk of excessive soil moisture or excessive canopy humidity. The disease transmission factor is calculated using a spatial decay function based on the historical anomaly types and spatial distances of the neighboring areas of the currently assessed inspection area unit in the inspection basic dataset. It is used to characterize the degree of influence of the disease spreading from the neighboring anomaly areas to this unit. The historical anomaly residual factor is based on the historical anomaly occurrence time, type and frequency of the currently evaluated inspection area unit in the inspection basic dataset. It is calculated by time decay function and frequency weighting and is used to characterize the persistent impact of historical anomaly problems. The low-lying water accumulation factor is calculated using topographic indices based on the digital elevation model data and rainfall information of the currently assessed inspection area unit and surrounding areas, and is used to characterize the water accumulation trend caused by topography. The accessibility factor is based on the terrain slope, surface roughness and obstacle distribution information of the currently evaluated inspection area unit. It is calculated by the robot kinematics model to form the accessibility difficulty coefficient, which is used to characterize the path cost and operational difficulty required for the quadruped robot to reach the unit.

[0030] After calculating the above risk factors, normalization is performed on each factor, and then a weighted sum is calculated based on the preset weight coefficients of each factor to obtain the comprehensive risk value of the inspection area unit.

[0031] It should be noted that the preset weight coefficients of each factor are determined by statistical analysis methods based on historical disaster data or by domain expert scoring methods, and are pre-configured in the system.

[0032] Furthermore, the inspection priority score for each inspection area unit is calculated. The inspection priority score is calculated based on the comprehensive risk value and incorporates the prior anomaly confidence level.

[0033] It should be noted that the anomaly fusion confidence score is updated in real time during the actual execution phase of the inspection task (S400) through the following process: When the quadruped robot arrives at the target inspection area unit according to the inspection path, it collects real-time multimodal perception data of the unit through its onboard visible light, thermal infrared, and depth sensors. Subsequently, the multimodal perception data is analyzed in real time by an edge-side recognition model deployed on the robot body, which uses existing image recognition and pattern recognition technologies, to obtain preliminary anomaly judgment results corresponding to each of the visible light, thermal infrared, and depth modalities. By performing a weighted fusion calculation on the preliminary anomaly judgment results of each modality, a value used to quantify the current anomaly probability of the target unit is generated, namely the real-time anomaly fusion confidence score, calculated as follows:

[0034] in, The expression represents the anomaly fusion confidence level of the i-th inspection area unit, and M represents the total number of modalities in the multimodal sensing data. This represents the initial confidence level of the m-th modality in determining the anomaly of the i-th region unit (the result directly output by the edge-side recognition model deployed on the quadruped robot after real-time analysis of the single-modal perception data). This represents the preset weight coefficient corresponding to the m-th mode (assigned based on prior knowledge).

[0035] The above formula can be used to incorporate leaf discoloration in visible light images, local temperature differences in thermal infrared images, and canopy morphology anomalies in depth information into the anomaly confirmation process, thus avoiding misjudgment based on a single mode.

[0036] In the embodiments of this application, the inspection priority score is calculated using a preset priority calculation model employing existing weighted scoring or decision model techniques. During the initial planning phase, the inspection priority score is calculated based on the aforementioned comprehensive risk value and prior anomaly confidence level. During the dynamic reordering phase (S500), it is calculated based on the comprehensive risk value and real-time anomaly fusion confidence level, combined with the re-inspection demand coefficient configured for the inspection area unit, the energy consumption cost expected to execute the inspection task, and the path switching cost from the previous inspection unit to the current unit. Through comprehensive calculation, an inspection priority score is obtained to guide inspection path planning and dynamic scheduling. The formula for calculating the inspection priority score is as follows:

[0037] in, This represents the inspection priority score of the i-th inspection area unit. This represents the overall risk value of the i-th inspection area unit. This represents the anomaly fusion confidence level of the i-th inspection area unit. This represents the re-inspection requirement coefficient for the i-th inspection area unit. This represents the energy consumption for executing the inspection task of the i-th inspection area unit. This represents the path switching cost to the i-th inspection area unit. , , , All of these represent preset adjustment coefficients.

[0038] It should be noted that the re-inspection demand coefficient These are binary status parameters set based on historical inspection records. If the i-th inspection area unit is marked as having an anomaly or requiring re-inspection in the historical inspection database, then... Assign a value to the first preset value (e.g., 1.0); if the unit in the inspection area has no abnormalities and does not require re-inspection, then The value is assigned to 0. In the initial risk assessment phase, The parameters are set based on historical inspection records. If the i-th inspection area unit has been marked as having an anomaly in the historical inspection database, then... A baseline value UB greater than 0 (e.g., 1.0) is assigned to reflect the baseline level of concern required for this area due to historical issues; if the unit has no historical anomaly records, then... The value is assigned to 0. The re-inspection demand coefficient is used to quantitatively characterize whether the corresponding area unit needs to be prioritized for re-inspection. Execution energy consumption. The energy consumption prediction is calculated using an existing robot energy consumption prediction model. This model takes the geospatial attributes of the i-th inspection area unit, the preset inspection task requirements, and the basic motion power consumption parameters of the quadruped robot as input, and outputs the result after comprehensive estimation. This value characterizes the expected energy consumption required to complete a full inspection of the area unit. Path switching cost. This is calculated in real time using path planning algorithms (such as A algorithm, Dijkstra's algorithm, etc.). The calculation is based on the real-time pose of the quadruped robot, the center coordinates of the i-th target inspection area unit, and the terrain elevation data and obstacle distribution information stored in the electronic map as inputs. The output is... The value represents the displacement cost required for the robot to move from its previous work position to the target inspection area unit.

[0039] In one specific embodiment, if an inspection area unit is located in a low-lying area, experiences hot and humid conditions, and has had lesions appear in historical inspections, then the comprehensive risk value of that area unit is... With re-inspection demand coefficient The risk level of a certain area unit will increase synchronously, causing it to automatically move forward in the initial inspection task sorting. If the risk value of an inspection area unit is average, but the cost of switching the current path is extremely high, the inspection priority of that area unit will be appropriately suppressed, thereby avoiding unnecessary frequent jumps in the task sequence.

[0040] The aforementioned solution divides the vineyard into refined inspection areas, extracts multi-dimensional risk factors, normalizes and weights them to calculate a comprehensive risk value, and integrates prior anomaly confidence levels. It also determines inspection priorities based on re-inspection needs, execution energy consumption, and path switching costs. This achieves quantitative risk assessment of inspection areas, scientifically distinguishes between high- and low-risk areas, and solves the problems of traditional inspections lacking focus and unreasonable resource allocation. It improves the efficiency of inspection resource utilization and the accuracy of risk prediction.

[0041] S300: Generates an inspection task sequence based on inspection priority and generates an inspection path in combination with the vineyard row structure.

[0042] In the embodiments of this application, based on the calculated inspection priority scores of each inspection area unit, all inspection area units are sorted from high to low scores to generate an initial inspection task set. Combining the quadruped robot's single-run endurance, preset inspection duration, and row / ridge passage restrictions, this initial set is scheduled and optimized based on multiple constraints. During the scheduling and optimization process, if the estimated total energy consumption of the task sequence exceeds the endurance, low-priority tasks are removed from the end of the sequence until the requirements are met. If the estimated total duration exceeds the preset duration, tasks are processed in batches or removed. Simultaneously, the task order is adjusted to conform to the passage logic of the row / ridge topology, forming an ordered inspection task sequence.

[0043] Based on the vineyard spatial structure data contained in the inspection basic dataset, the abnormal point records in the historical inspection database, the real-time collected environmental data, and the preset operation point information, the start and end points and feature points of the vineyard row centerline, the inter-row turning areas, the historical and newly added abnormal points, the low-lying areas identified by terrain analysis, and the preset supply points are defined and instantiated as nodes in the vineyard row semantic topology map.

[0044] Each target inspection area unit in the inspection task sequence is associated with the nearest semantic node with the lowest passage cost based on its spatial location in the semantic topology map of the vineyard rows, thus generating an ordered sequence of target nodes composed of semantic nodes. This ordered sequence of target nodes serves as the critical path points that must be visited sequentially, and the node connection relationships and passage rules defined in the semantic topology map of the vineyard rows form the search space. A path planning algorithm is used to plan the global inspection path for the quadruped robot. Under the premise of satisfying the constraint of sequentially visiting the target node sequence, this planning process uses the cost of edges in the semantic topology map as the optimization basis, comprehensively incorporating the passage cost transformed by terrain elevation changes and obstacle distribution, to generate a global inspection path with optimal or near-optimal total passage cost. The core cost components (i.e., path length, terrain difficulty, and obstacle penalty) upon which the passage cost value of each edge in the semantic topology map of the vineyard rows is calculated are based on the same principles as the path switching cost defined in S200 and the incremental cost in S500.

[0045] The above solution generates a task sequence based on inspection priority, constructs a semantic topology graph based on the vineyard row structure, and plans a global inspection path. It also incorporates constraints such as robot endurance and inspection duration to optimize task scheduling and generate the inspection path with the optimal total travel cost. This optimal inspection path not only conforms to the vineyard row passage logic but also avoids unnecessary detours and task overload, ensuring the orderly execution of robot inspection tasks and improving the rationality and practicality of global path planning.

[0046] S400: Controls the quadruped robot to perform inspections according to the inspection path and collects inspection data during the inspection process to identify abnormal points.

[0047] In the embodiments of this application, multimodal perception data of the target area is collected in real time; the multimodal perception data is analyzed in real time by an edge-side recognition model deployed on a quadruped robot; based on the real-time analysis results, it is identified whether there are any anomalies in the target area, including at least one of abnormal plant growth, abnormal canopy environment, and abnormal soil condition; when an anomaly is identified, the spatial location, anomaly type, and real-time anomaly fusion confidence of the anomaly point are recorded, and the anomaly point is added to the anomaly point database.

[0048] In the embodiments of this application, when the confidence level of the anomaly fusion of the identified abnormal points is lower than the preset confidence threshold, the abnormal points are marked as nodes to be confirmed, and a corresponding re-inspection task is generated.

[0049] In a preferred embodiment of this application, the quadruped robot is controlled to autonomously travel along a global inspection path, sequentially arriving at each inspection area unit and performing in-situ inspection operations. Upon arriving at each target inspection area unit, the robot simultaneously collects plant images, canopy temperature information, and terrain depth information in that area using visible light sensors, thermal infrared sensors, and depth sensors, forming real-time multimodal perception data.

[0050] By deploying an edge-side recognition model on the quadruped robot body, real-time multimodal perception data is analyzed and anomaly detection is performed to identify whether there are abnormal plant growth, abnormal canopy environment, or abnormal soil conditions in the target area.

[0051] When an anomaly is identified, the real-time spatial location information of the anomaly point is obtained, its geographical coordinates, anomaly type and corresponding real-time anomaly fusion confidence level are recorded, and this record is synchronously written into the anomaly point database.

[0052] If the real-time anomaly fusion confidence level of the identified abnormal point is lower than the preset confidence level threshold, the point is marked as a node to be confirmed, and a re-inspection task corresponding to the node is generated to be inserted into the subsequent inspection task sequence for priority verification.

[0053] The above-described scheme controls a robot to perform inspections along a planned path. It collects data in real time using multimodal sensors, identifies anomalies in plants, canopy, and soil based on edge models, records anomaly information, and marks low-confidence anomalies as nodes requiring confirmation, achieving real-time online anomaly identification and location. The fusion of multimodal data avoids misjudgments based on a single modality. This scheme not only generates timely re-inspection tasks but also effectively improves anomaly detection accuracy, providing real-time data for dynamic path adjustments.

[0054] S500: When a new abnormal point is identified, the inspection priority is updated based on the inspection area unit corresponding to the new abnormal point, the inspection area unit is inserted as a re-inspection node into the unexecuted inspection task sequence, and the current inspection path is partially rearranged.

[0055] In the embodiments of this application, when a new abnormal point is identified during the inspection process, the inspection area unit to which the new abnormal point belongs is first determined, and the inspection area unit is marked as a priority re-inspection object. Based on the new abnormal point, the re-inspection demand coefficient of the corresponding inspection area unit is increased; based on the increased re-inspection demand coefficient, the real-time anomaly fusion confidence level corresponding to the new abnormal point, and the original comprehensive risk value, execution energy consumption, and path switching cost of the inspection area unit, the inspection priority of the inspection area unit is recalculated, thereby realizing the dynamic upward adjustment of the inspection priority of the inspection area unit.

[0056] In an embodiment of this application, inserting the inspection area unit as a re-inspection node into an unexecuted inspection task sequence includes: calculating the incremental cost generated by inserting the re-inspection node into different positions in the unexecuted inspection task sequence; determining the target insertion position with the minimum incremental cost based on the calculation results; and inserting the re-inspection node into the target insertion position.

[0057] In embodiments of this application, calculating the incremental cost includes: obtaining the additional path length generated by inserting the re-inspection node into the candidate position; obtaining the terrain elevation change penalty value and obstacle density penalty value of the area where the re-inspection node is located; and, based on a preset adjustment coefficient, performing a weighted summation of the additional path length, terrain elevation change penalty value, and obstacle density penalty value to obtain the incremental cost. The formula for calculating the incremental cost is as follows:

[0058] in, This represents the incremental cost of inserting the re-inspection node into the candidate position. This represents the difference between the length of the newly planned path and the original path length after inserting the re-inspection node into the candidate position. This represents the penalty value for terrain elevation changes in the area where the re-inspection node is located. It is obtained by converting the elevation difference and terrain slope within the area based on the digital elevation model data of the area where the re-inspection node is located. This represents the obstacle density penalty value in the area where the re-inspection node is located. It is obtained by converting the number of obstacles and their distribution density based on the obstacle distribution information in the area where the re-inspection node is located. , , All represent preset weighting coefficients greater than zero, used to balance the impact of various costs.

[0059] In the embodiments of this application, the current inspection path is partially rearranged, including: adding the re-inspection node as a new semantic node in the semantic topology graph of the vineyard rows; in the semantic topology graph, disconnecting the original association edges between the current node and the subsequent target node; and establishing new association edges between the current node and the new semantic node, and between the new semantic node and the subsequent target node, to update the inspection path.

[0060] In a preferred embodiment of this application, firstly, the spatial location information corresponding to the re-inspection node and its associated inspection area unit attributes are added as a new semantic node to the data structure of the vineyard row semantic topology graph. A unique identifier is assigned to this new node, and attribute fields including geographic coordinates, the latest inspection priority score, and the re-inspection demand coefficient are assigned values.

[0061] Then, in the semantic topology graph of the vineyard rows, based on the real-time positioning coordinates of the quadruped robot, it is mapped to the nearest semantic node in the graph, and this node is defined as the current node. At the same time, the next target node planned to be visited immediately in the original global inspection path is located. The graph data structure operation interface is called to delete the original directed association edge connecting the two nodes, thereby removing the original planning constraints of this path segment.

[0062] Next, two new directed edges are established in the semantic topology graph to integrate the newly added node. The first edge establishes a connection from the current node to the newly added semantic node, and the second edge establishes a connection from the newly added semantic node to the original subsequent target node. The calculation of the travel cost values ​​of the two edges follows the same principles as the calculation of path switching cost and incremental cost, specifically: For the first edge, its passage cost is calculated based on the basic path length derived from the spatial relationship between the two nodes, and then weighted and summed by the terrain elevation change penalty value and obstacle density penalty value of the region where the newly added semantic node (i.e., the re-inspection node) is located. After the calculation is completed, this cost is assigned as the weight attribute of this edge.

[0063] For the second edge, its passage cost is calculated based on the basic path length derived from the spatial relationship between the two nodes, and then weighted and summed by incorporating the terrain elevation change penalty and obstacle density penalty values ​​of the original subsequent target node's region. After calculation, this cost is assigned as the weight of this edge.

[0064] Based on the updated semantic topology of the vineyard rows (which now includes new nodes, edges, and edge weights), the current position of the quadruped robot is used as the starting point for path planning. Using the same path planning algorithm as in S300, a global inspection path is recalculated that sequentially visits all incomplete target nodes (including semantic nodes mapped by newly inserted re-inspection nodes) and has the optimal total travel cost. This path naturally includes travel segments adjusted to prioritize access to newly added re-inspection nodes, thus completing a local rearrangement of the inspection path.

[0065] The above solution dynamically increases the inspection priority of the corresponding area for newly added anomaly points, inserts re-inspection nodes into the task sequence with minimal incremental cost, and rearranges the paths locally based on the semantic topology graph. This approach can quickly respond to on-site anomalies, prioritize the verification of high-risk re-inspection points, and minimize the energy consumption and time loss caused by path changes. It enables dynamic adaptive adjustment of inspection tasks and paths, improving the flexibility of the inspection process and emergency response capabilities.

[0066] S600: Controls the quadruped robot to perform inspections based on the inspection task sequence after reordering and inserting re-inspection nodes, as well as the inspection path after partial rearrangement.

[0067] In the embodiments of this application, the quadruped robot is controlled to continue autonomously following the updated inspection task sequence and the partially rearranged inspection path, sequentially traversing all target locations in the task sequence that include regular inspection units and re-inspection nodes. Upon reaching each target location, real-time data acquisition and anomaly identification are performed using multimodal sensors. A secondary check is performed on re-inspection nodes marked as pending confirmation, and the anomaly fusion confidence of the corresponding points is updated. During the inspection process, the quadruped robot's pose information, completed inspection units, re-inspection results, and current remaining battery life are recorded in real time.

[0068] The above solution controls the robot to continue inspection according to the updated task sequence and rearranged path, completes routine inspection and re-inspection tasks, performs secondary verification of nodes to be confirmed and updates the confidence level of anomalies, and records the robot's pose, battery life and task completion status in real time, ensuring the continuous execution of inspection tasks and accurately confirming abnormal states.

[0069] S700: After the inspection is completed, summarize the inspection results and update the historical inspection abnormality data.

[0070] In the embodiments of this application, after the entire global inspection task is completed, the multimodal perception data collected during the inspection process, the identified abnormal points, the confirmation results of the re-inspection nodes, the anomaly fusion confidence level and spatial location information are uniformly summarized and organized.

[0071] Based on the compiled inspection results, the historical database of abnormal locations is updated, and newly added abnormal locations are stored in the database. For existing abnormal locations, the latest occurrence time, anomaly type, and confidence level are updated; for locations where anomalies have been eliminated, their status is corrected or their markings are cleared. Finally, the complete inspection record is archived and saved to provide data support for subsequent risk assessments, priority calculations, and route planning for the inspection area.

[0072] The present invention also provides a vineyard quadruped robot inspection system for implementing a vineyard quadruped robot inspection method. The system includes a control module, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the computer program to implement the vineyard quadruped robot inspection method.

[0073] This invention provides a storage medium storing a program that, when executed by a processor, implements a vineyard quadruped robot inspection method.

[0074] This invention provides a processor for running a program, wherein the program executes a vineyard quadruped robot inspection method during runtime.

[0075] This invention provides a device including a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it implements a vineyard quadruped robot inspection method. The device described herein can be a server, PC, tablet, mobile phone, etc.

[0076] This application also provides a computer program product that, when executed on a data processing device, is suitable for performing a vineyard quadruped robot inspection method.

[0077] Those skilled in the art will understand that embodiments of this application can provide methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0078] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0079] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0080] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0081] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0082] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0083] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0084] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0085] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for inspecting vineyards using a quadruped robot, characterized in that, include: Meteorological data, soil condition data, and historical inspection anomaly data of the vineyards were collected and integrated to obtain the basic inspection dataset; Based on the aforementioned inspection dataset, the vineyard is divided into inspection area units, and a risk assessment is performed on each inspection area unit to obtain the inspection priority. An inspection task sequence is generated based on the inspection priority, and an inspection path is generated in combination with the vineyard row structure. Control the quadruped robot to perform inspections according to the inspection path, and collect inspection data during the inspection process to identify abnormal points; When a new abnormal point is identified, the inspection priority is updated based on the inspection area unit corresponding to the new abnormal point, the inspection area unit is inserted as a re-inspection node into the unexecuted inspection task sequence, and the current inspection path is partially rearranged. The quadruped robot is controlled to perform inspections based on the inspection task sequence after reordering and inserting re-inspection nodes, and the inspection path after partial rearrangement. After the inspection is completed, the inspection results are summarized and the historical inspection anomaly point data is updated.

2. The vineyard quadruped robot inspection method according to claim 1, characterized in that, The step of conducting risk assessments on each of the aforementioned inspection area units to obtain inspection priorities includes: For each of the aforementioned inspection area units, extract risk-related features; The risk-related features are normalized, and the comprehensive risk value of each inspection area unit is calculated according to the preset weights. Multimodal perception data of each inspection area unit is acquired and anomaly identification is performed. The anomaly fusion confidence score is obtained by fusing the results of each modality identification. Based on the comprehensive risk value and the anomaly fusion confidence level, combined with the re-inspection demand coefficient, execution energy consumption, and path switching cost, the inspection priority of each inspection area unit is calculated.

3. The vineyard quadruped robot inspection method according to claim 2, characterized in that, The risk-related characteristics include frost risk factors, wet damage risk factors, disease transmission factors, historical abnormal residue factors, low-lying water accumulation factors, and accessibility factors.

4. The vineyard quadruped robot inspection method according to claim 1, characterized in that, The process of collecting inspection data during the inspection to identify abnormal locations includes: Real-time acquisition of multimodal sensing data of the target area; The multimodal perception data is analyzed in real time using an edge-side recognition model deployed on the quadruped robot. Based on real-time analysis results, identify whether there are any anomalies in the target area. The anomalies include at least one of the following: abnormal plant growth, abnormal canopy environment, and abnormal soil condition. When an anomaly is identified, the spatial location, anomaly type, and anomaly fusion confidence level of the anomaly point are recorded, and the anomaly point is added to the anomaly point database.

5. The vineyard quadruped robot inspection method according to claim 2, characterized in that, The process of collecting inspection data during the inspection to identify abnormal points also includes: when the confidence level of the identified abnormal point is lower than a preset confidence threshold, marking the abnormal point as a node to be confirmed and generating a corresponding re-inspection task.

6. The vineyard quadruped robot inspection method according to claim 1, characterized in that, The step of updating the inspection priority based on the inspection area unit corresponding to the newly added abnormal point includes: Based on the newly added abnormal points, the re-inspection requirement coefficient of the corresponding inspection area unit is increased; Based on the improved re-inspection requirement coefficient and the anomaly fusion confidence of the newly added abnormal points, the inspection priority of the inspection area unit is recalculated.

7. The vineyard quadruped robot inspection method according to claim 1, characterized in that, The step of inserting the inspection area unit as a re-inspection node into the unexecuted inspection task sequence includes: Calculate the incremental cost of inserting the re-inspection node into different positions in the sequence of unexecuted inspection tasks; Based on the calculation results, determine the target insertion position with the minimum incremental cost; Insert the re-inspection node into the target insertion position.

8. The vineyard quadruped robot inspection method according to claim 7, characterized in that, Calculating the incremental cost includes: Obtain the additional path length generated by inserting the re-inspection node into the candidate position; Obtain the terrain elevation change penalty value and obstacle density penalty value of the area where the re-inspection node is located; Based on a preset adjustment coefficient, the additional path length, the terrain elevation change penalty value, and the obstacle density penalty value are weighted and summed to obtain the incremental cost.

9. The vineyard quadruped robot inspection method according to claim 1, characterized in that, The partial rearrangement of the current inspection path includes: Add the re-inspection node as a new semantic node in the semantic topology graph of the vineyard rows; In the semantic topology graph, disconnect the original associated edges between the current node and the subsequent target node; Establish new association edges between the current node and the newly added semantic node, and between the newly added semantic node and the subsequent target node, to update the inspection path.

10. A quadruped robot inspection system for vineyards, characterized in that, The system includes a control module, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the computer program to implement the vineyard quadruped robot inspection method according to any one of claims 1-9.

Images