Tea-picking robot autonomous navigation method based on tea ridge composite boundary identification and driving area adjustment

Abstract

The invention relates to a tea-picking robot autonomous navigation method based on tea ridge composite boundary identification and driving area adjustment, and belongs to the technical field of robot navigation. The method comprises the steps that tea ridge growth state information is evaluated, and a global path planning strategy is made; a root-ground intersection ground navigation datum line and lateral boundary constraints are established, the ground navigation datum line, the lateral boundary constraints and tea tree growth density and distribution characteristics are integrated, a tea ridge three-dimensional drivable road network model is established, and key geometric characteristics of the model are analyzed to establish an adaptive risk assessment system. Dividing the drivable area into areas with different risk levels; and establishing a multi-objective optimization function in combination with the risk region distribution diagram, fusing the multi-objective optimization function through an exponential weighted fusion mechanism, and generating an optimal navigation path under the risk gradient constraint. Guiding of intelligent navigation of the tea garden is achieved, the tea leaf picking efficiency and quality are improved, and technical support is provided for mechanical and intelligent picking of the tea garden.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of robot navigation, and relates to autonomous navigation of a tea-picking robot, in particular to a tea-picking robot autonomous navigation method based on tea ridge composite boundary identification and drivable area adjustment. BACKGROUND

[0002] As an important economic crop and traditional beverage in China, tea occupies an important position in the global market. Among them, the mountain tea garden can often produce high-quality tea due to its unique geographical and climatic conditions, but it also brings great challenges to the production and management of tea. Traditional tea garden management, especially picking, plant protection, and inspection, highly depends on manual labor, which is labor-intensive, low-efficiency, and the cost is rising year by year. Especially in the mountain tea garden with complex terrain and large slope changes, the difficulty and risk of manual operation are higher, which seriously restricts the scale and modernization development of the tea industry. Therefore, introducing intelligent and automated robot technology to replace or assist manual operation has become an urgent need for the sustainable development of mountain tea gardens.

[0003] In recent years, with the development of robot technology, artificial intelligence and sensor technology, agricultural robots have made some application progress in flat and regular farmland environment. However, when these technologies are directly applied to complex and unstructured mountain tea garden environment, there are many difficulties. First, the terrain of the mountain tea garden is undulating and the slope is different, which puts high requirements on the stable driving and path planning of the robot; second, tea trees are often irregularly planted, with large variation in row spacing and plant spacing, and the tea tree canopy and ground environment are complex and interlaced, making it difficult for traditional navigation methods based on GPS or simple vision / laser radar to accurately locate and perceive the drivable area, and easily causing collision or getting lost; third, existing navigation strategies often focus on obstacle avoidance and shortest path, and less consider the economic benefits (such as tea quality, yield distribution) of the task (such as picking) and the potential risks (such as steep slope, slippery) brought by complex terrain, resulting in low efficiency and safety of the task. At present, although some researches focus on robot navigation in agricultural environment, there is still a lack of a complete, robust and efficient autonomous navigation solution for the specific scene of mountain tea garden which combines complex three-dimensional terrain, unstructured crop environment and task value orientation. The existing methods still have deficiencies in environment perception, accurate modeling of drivable area, global path task planning and local path dynamic risk avoidance, and are difficult to meet the needs of safe, efficient and autonomous operation of intelligent tea-picking robots in real mountain tea garden environment. Therefore, it is urgent to develop an intelligent robot autonomous navigation method that can adapt to the complex environmental characteristics of mountain tea garden. SUMMARY

[0004] In view of this, the purpose of the present application is to provide a tea robot autonomous navigation method based on tea ridge composite boundary identification and drivable area adjustment, aiming to solve the problems of existing robot navigation technology in dealing with the complex terrain, unstructured environment and task requirements of mountain tea garden, such as inaccurate perception, low path planning efficiency, insufficient safety, etc., to realize the goal of safe, efficient and autonomous navigation of intelligent robots in such challenging environment, improve the environmental adaptability, navigation accuracy and operation safety of intelligent tea picking robots in complex environment of mountain tea garden, and provide effective technical support for realizing intelligent and unmanned management of mountain tea garden.

[0005] To achieve the above purpose, the present application provides the following technical scheme:

[0006] A tea robot autonomous navigation method based on tea ridge composite boundary identification and drivable area adjustment, the method comprising:

[0007] Comprehensively utilize the historical picking data obtained from the tea ridge historical database and the real-time evaluation of tea ridge growth state information through unmanned aerial vehicle multi-sensor fusion technology to develop a global path planning strategy;

[0008] Obtain the position of tea tree roots, perform semantic segmentation on the ground of tea ridge, establish ground navigation reference line, obtain tea crown boundary profile, establish lateral boundary constraint of robot driving channel, and comprehensively construct tea ridge three-dimensional drivable road network model based on ground navigation reference line, lateral boundary constraint and tea tree growth density and distribution characteristics;

[0009] Analyze the key geometric characteristics of the tea ridge three-dimensional drivable road network model, and design an adaptive risk assessment system based on the key geometric characteristics to divide the drivable area into areas of different risk levels;

[0010] Combined with the risk area distribution map, a multi-objective optimization function considering the length of each path, the risk level corresponding to the path and the energy consumption factor is established, and the multi-objective optimization function is fused through an exponential weighted fusion mechanism to generate the optimal navigation path under the risk gradient constraint.

[0011] Further, multi-dimensional information including tea quality score, yield record, growth state and historical picking record is extracted from the historical database of each tea ridge in the tea garden, and the appearance characteristics, physiological characteristics and moisture condition of each tea ridge are collected by the unmanned aerial vehicle equipped with multiple sensors; the collected multi-source heterogeneous data is processed by feature extraction and fusion to construct a tea ridge growth state evaluation model;

[0012] According to the tea quality, weights are given to each tea ridge, and a weighted sum is calculated to obtain a comprehensive quality score of each tea ridge. A tea ridge quality score model is constructed by combining the comprehensive quality score, yield score and distance cost of the tea ridge to determine the preliminary picking priority of each tea ridge.

[0013] According to the preliminary picking priority ranking, a preliminary visiting order of the tea ridges is determined. The path length and navigation difficulty between adjacent tea ridges are considered to optimize and adjust the preliminary visiting order, and a final visiting order of the tea ridges is determined to realize global path planning.

[0014] The tea ridge growth state evaluation model is represented as:

[0015] G i = k1·V i + k2·T i + k3·C i

[0016] In the formula, G i is the growth index of the i-th tea ridge, V i is the vegetation index of the i-th tea ridge, T i is the temperature index of the i-th tea ridge, C i is the canopy structure index of the i-th tea ridge; k1 represents the influence weight of the current vegetation growth condition, k2 represents the influence degree weight of temperature on tea quality, and k3 is the contribution weight coefficient of tea tree structure on quality.

[0017] The tea ridge quality score model is represented as:

[0018] P(i) = α·Q(i) + β·Y(i) - γ·D(i)

[0019]

[0020] In the formula, Q(i) represents the comprehensive quality score of tea ridge i, w i represents the weight coefficient of the i-th tea quality, and n represents the total number of evaluated tea ridges; P(i) represents the final priority score of tea ridge i, Y(i) represents the yield score of tea ridge i, D(i) represents the distance cost of visiting tea ridge i, and α, β, γ are the influence weight coefficients of quality, yield and distance, respectively.

[0021] The preliminary visiting order is optimized and adjusted considering the path length and navigation difficulty between adjacent tea ridges, as shown in the following formula:

[0022] C(i, m) = d(i, m)·(1 + η·T(i, m))

[0023]

[0024] In the formula, C(i, m) represents the comprehensive navigation cost from tea ridge i to tea ridge m, d(i, m) represents the physical distance between the two tea ridges, T(i, m) represents the complexity factor of the navigation path, and η is the complexity influence coefficient; L represents the total cost of the entire navigation sequence, and N is the total number of tea ridges to be visited; C(i, i+1) represents the actual navigation cost between the i th tea ridge and the i+1 th tea ridge in the to-be-visited sequence; by minimizing the total cost L, the final tea ridge visiting order is determined, and global path planning is realized.

[0025] Further, the robot-mounted camera collects tea tree images, adopts the EfficientDet recognition algorithm to process the collected images, recognizes the tea tree root position information, performs semantic segmentation on the tea ridge ground through the SegFormer segmentation network, classifies the ground area at the pixel level, and extracts the intersection area of the tea tree root and the ground; a radial basis function network model is established for fitting and identifying the spatial distribution of the root-ground intersection point:

[0026]

[0027] In the formula, F(x, y) represents the ground reference line height value at the plane coordinates (x, y), and p represents the plane position vector (x, y) of the query point; p j represents the plane coordinate vector of the j th root-ground intersection point sample; r=||p-p j || is the Euclidean distance between the query point and the j th sample point; w j is the weight coefficient corresponding to the j th sample point, φ(r) is a multi-quadratic radial basis function, c is a smoothing parameter, and J is the number of sample points; by solving the weight coefficient w j , the fitting error is minimized to obtain an accurate root-ground intersection point continuous curve, and a ground navigation reference line is generated;

[0028] The Mask R-CNN instance segmentation technology is used to identify the tea crown boundary contour, realize the description and boundary extraction of the tea tree crown layer structure, and establish a tea crown geometric model according to the extracted boundary data:

[0029]

[0030] B(x, y)=T(E(z))·(1-δ·H(z))

[0031] In the formula, E(c) represents the tea crown outer contour boundary function at position c; K(c, z j is a kernel function for realizing smooth interpolation and key feature extraction of the tea crown contour, z j is a sampling point position, a jwhere J is the number of sampling points; B(x, y) represents the projection function of the tea crown boundary to the ground; T(·) represents the projection transformation operation; H(z) represents the height influence factor; and δ is a safety boundary adjustment coefficient, which is used to define the safe operation boundary above the tea ridge.

[0032] Further, the ground navigation reference line and the tea crown boundary projection are taken as constraint conditions to construct a three-dimensional drivable road network model of the tea ridge:

[0033]

[0034] where H k (z) represents the feasibility measurement function of the three-dimensional space of the tea ridge; φ k (x, y) is the horizontal plane radial basis function; H k (z) is the elevation weight function; W k is the constraint weight coefficient; and K is the total number of basis functions; P(t) represents the three-dimensional drivable road parameter equation; c l is the control point; B l (t) is the basis function; L is the number of control points; and t is the curve parameter.

[0035] Further, the key geometric characteristics of the three-dimensional drivable road network model of the tea ridge are analyzed, including the slope change, the radius of curvature and the width constraint, and an adaptive risk assessment system is established:

[0036] R(t) = R s (t) + R w (t) + δ·R e (t)

[0037] P'(t) = P0(t) + k·R(t) + ε·N(t)

[0038] where R(t) represents the comprehensive risk value of the three-dimensional drivable road of the tea ridge at the parameter t; t is the parameterization variable of the path, representing the current position of the robot on the three-dimensional drivable road of the tea ridge; R s (t) represents the slope and bending risk; R w (t) represents the width risk; R e (t) represents the environmental factor risk; δ is the environmental factor weight; P'(t) represents the optimal navigation path after risk adjustment; P0(t) is the reference navigation trajectory; N(t) is the path normal correction term; k is the gradient adjustment coefficient; and ε is the normal adjustment coefficient.

[0039] According to the comprehensive risk value R(t), the drivable area is identified and divided into areas of different risk levels, wherein the risk levels include high-risk areas, medium-risk areas and safe areas.

[0040] Furthermore, combined with the risk area distribution map, a multi-objective optimization function is established that comprehensively considers the length of each path segment, the corresponding risk level of the path segment, and energy consumption factors:

[0041] F(t)=ω1·R(t)+ω2·S(t)+ω3·E(t)

[0042] Where F(t) represents the multi-objective optimization function, t is the path parameterization variable, S(t) represents the smoothness function; E(t) represents the energy efficiency function, which is used to evaluate the energy consumption of the robot traveling on path t; ω1 is the risk weight coefficient, which controls the importance of safety in path planning; ω2 is the smoothness weight coefficient, which is used to adjust the continuity and turning smoothness of the path; ω3 is the energy efficiency weight coefficient, which balances the relationship between energy consumption and path length.

[0043] The multi-objective optimization functions are fused through the exponential weighted fusion mechanism to generate the optimal navigation path under the risk gradient constraint:

[0044] N(t)=N0(t)·e -λF(t) +β·[1-e -λF(t) ]

[0045] Where N(t) represents the final generated navigation path, N0(t) is the initial navigation trajectory, β is the safety correction coefficient, and λ is the adjustment factor.

[0046] The beneficial effects of the present invention are:

[0047] 1) By integrating information from a historical tea ridge database with real-time assessment results from drone multi-sensors, this paper proposes a global path planning strategy based on tea quality weights. This overcomes the limitations of traditional navigation methods that only focus on the shortest path or simple coverage, and can prioritize tea ridges with higher picking value (combining quality and yield). This significantly improves the overall economic benefits and task orientation of intelligent tea picking operations, making navigation decisions more in line with actual production needs.

[0048] 2) Compared with navigation methods that rely on single sensor information or simple environmental models, this paper proposes a method for identifying the composite boundaries of tea ridges and calculating the feasible domain. By comprehensively applying multiple advanced visual recognition algorithms (EfficientDet, SegFormer, Mask R-CNN), it accurately captures the roots, ground, and canopy boundaries of tea ridges. Through spatial mapping, geometric analysis, and triangulation technology, it constructs an accurate three-dimensional drivable road network model with multi-dimensional constraints such as height, width, and inclination. This greatly improves the perception accuracy and depth of understanding of the complex and unstructured environment of mountain tea gardens, provides robots with more reliable and detailed drivable space information, and effectively avoids collision risks and path planning deviations caused by inaccurate environmental modeling.

[0049] 3) The application further proposes a risk gradient division method based on the three-dimensional drivable road of tea ridge, which performs hierarchical risk assessment on the environment according to the constructed three-dimensional drivable road model (such as identifying high-risk areas such as steep slopes, narrow roads, sharp turns, etc.), and combines path length, safety factor, energy consumption and other multi-objectives to optimize the path, so that the finally generated navigation path not only considers efficiency, but also significantly enhances the driving safety, stability and environmental adaptability of the robot in complex and variable mountainous terrain, solving the problem of insufficient reliability of traditional path planning in the face of potential terrain risks.

[0050] Other advantages, objects, and features of the application will be set forth in part in the description which follows, and in part will become apparent to those skilled in the art upon examination of the following specification or can be learned by practice of the application. The objects and other advantages of the application can be realized and attained by the methods and processes particularly pointed out in the description. BRIEF DESCRIPTION OF DRAWINGS

[0051] In order to make the objects, technical solutions and advantages of the application clearer, the preferred detailed description of the application will be combined with the drawings below, in which:

[0052] Figure 1 The flowchart of the autonomous navigation method of the tea-picking robot based on tea ridge composite boundary recognition and drivable area adjustment provided by an embodiment of the application is shown. DETAILED DESCRIPTION

[0053] The embodiments of the application are described below through specific examples, and those skilled in the art can easily understand other advantages and effects of the application from the disclosure of the specification. The application can also be implemented or applied by different specific embodiments, and the details in the specification can be modified or changed based on different views and applications without departing from the spirit of the application. It should be noted that the diagrams provided in the following examples only illustrate the basic concept of the application in a schematic manner, and the following examples and features in the examples can be combined with each other without conflict.

[0054] The drawings are only used for illustrative description, and the representation is only a schematic diagram, not a physical diagram, and cannot be understood as a limitation of the application; in order to better illustrate the embodiments of the application, some components of the drawings are omitted, enlarged or reduced, and do not represent the size of the actual product; for those skilled in the art, it can be understood that some well-known structures and their descriptions in the drawings can be omitted.

[0055] The same or similar reference numerals in the drawings of the embodiments of the present application correspond to the same or similar components; in the description of the present application, it is understood that if the orientations or positional relationships indicated by the terms "upper", "lower", "left", "right", "front", "back" and the like are based on the orientations or positional relationships shown in the drawings, they are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, therefore the terms describing the positional relationship in the drawings are only used for exemplary illustration, and cannot be understood as a limitation on the present application, for those skilled in the art, the specific meanings of the above terms can be understood according to the specific circumstances.

[0056] The present application provides a tea-picking robot autonomous navigation method applied to mountain tea gardens, which integrates global task planning based on historical data and real-time perception, fine environment modeling of multi-modal sensor fusion, and local path dynamic optimization considering terrain risk. Specifically, the method first uses tea ridge historical database information and unmanned aerial vehicle multi-sensor fusion evaluation results to develop a globally optimal navigation sequence considering tea quality and yield; secondly, by combining advanced deep learning vision algorithms (such as EfficientDet, SegFormer, Mask R-CNN) for composite boundary identification of tea ridge roots, ground and canopy, and using spatial mapping and geometric analysis techniques to construct an accurate three-dimensional drivable area model; finally, based on the generated three-dimensional environment model, a three-dimensional guided risk area division strategy is proposed, and combined with a multi-objective optimization function (considering path length, safety factor, energy consumption, etc.), a safe, smooth and efficient final navigation path is generated to adapt to complex terrain. Through the method proposed in the present application, the environmental adaptability, navigation accuracy and operation safety of the intelligent tea-picking robot in the complex environment of the mountain tea garden can be significantly improved, providing effective technical support for realizing intelligent and unmanned management and operation of the mountain tea garden.

[0057] Based on this, an embodiment of the present application provides a tea-picking robot autonomous navigation method based on tea ridge composite boundary identification and drivable area adjustment, as shown in Figure 1 The method is as follows:

[0058] I. First, the historical picking data obtained from the tea ridge historical database and the real-time evaluation of the tea ridge growth state information obtained by the multi-sensor fusion technology of the unmanned aerial vehicle are comprehensively utilized to formulate a global path planning strategy. According to these input information, a tea ridge quality weight evaluation and global path planning strategy is proposed: in the first stage, a tea ridge quality evaluation model is established, combining tea quality indicators and yield data to calculate the revenue contribution and picking value of each tea ridge, and determine the preliminary picking priority of each tea ridge; in the second stage, based on the work efficiency principle and considering the distance and terrain characteristics between adjacent tea ridges, the preliminary priority based on the weight coefficient is dynamically adjusted; finally, according to the optimized navigation sequence, a global navigation path with optimal task value is output, including the following steps:

[0059] 1: Obtain the historical quality database of each tea ridge in the tea garden, including tea quality score, yield record, growth state and historical picking record, etc. The sensor system carried by the unmanned aerial vehicle includes RGB camera, multispectral camera and thermal imaging sensor, which respectively collect the appearance characteristics, physiological characteristics and water status of the tea ridge; the collected multi-source heterogeneous data are processed by feature extraction and fusion, and a tea ridge growth state evaluation model is constructed, as shown in formula (1); finally, the historical data and real-time sensor data are constructed into a complete tea ridge quality evaluation basic data set, providing basic data support for subsequent priority evaluation.

[0060] G i = k1·V i + k2·T i + k3·C i (1)

[0061] Where G i is the growth index of the i-th tea ridge, V i is the vegetation index of the tea ridge, T i is the temperature index, and C i is the canopy structure index; k1 represents the influence weight of the current vegetation growth condition, k2 represents the influence degree weight of temperature on tea quality, and k3 is the contribution weight coefficient of tea tree structure to quality. The current growth state of the tea ridge is evaluated by this model to provide data support for path planning.

[0062] 2: A priority decision method based on tea ridge quality evaluation is proposed. By analyzing the information in the tea ridge historical quality database, a tea ridge quality score model is constructed, as shown in (2) and (3):

[0063]

[0064] P(i) = a·Q(i) + b·Y(i) - g·D(i) (3)

[0065] wherein Q(i) represents the quality comprehensive score of tea ridge i, w i represents the weight coefficient of the quality of the ith tea ridge, n represents the total number of tea ridges evaluated; P(i) represents the final priority score of tea ridge i, Y(i) represents the yield score of tea ridge i, D(i) represents the distance cost of visiting tea ridge i, and a, β, γ are the influence weight coefficients of quality, yield, and distance, respectively.

[0066] 3: Establish a tea ridge navigation sequence planning strategy based on priority scores. First, sort the calculated priority scores P(i) of each tea ridge to determine the preliminary visiting order of each tea ridge; then, consider the path length and navigation difficulty between adjacent tea ridges, and optimize and adjust the preliminary visiting order through the following equations (4) and (5):

[0067] C(i, m) = d(i, m) · (1 + η · T(i, m)) (4)

[0068]

[0069] wherein C(i, m) represents the comprehensive navigation cost from tea ridge i to tea ridge m, d(i, m) represents the physical distance between the two tea ridges, T(i, m) represents the complexity factor of the navigation path, and η is the complexity influence coefficient; L represents the total cost of the entire navigation sequence, and N is the total number of tea ridges to be visited. C(i, i+1) represents the actual navigation cost between the ith tea ridge and the i+1th tea ridge in the sequence to be visited. By minimizing the total cost L, the final tea ridge visiting order is determined, and efficient global path planning is achieved.

[0070] Second, tea ridge composite boundary identification and feasible region calculation: accurately identify the complex boundary of the tea ridge and calculate the three-dimensional drivable space of the robot. On the one hand, to determine the bottom boundary of the tea ridge and the navigation reference, EfficientDet is used to identify the root position of the tea tree, and SegFormer is used for semantic segmentation of the tea ridge ground. Then, the root-ground intersection information is processed by the Radial Basis Function Space Mapping algorithm, and finally the accurate ground navigation reference line is established. On the other hand, to determine the lateral constraint of the driving channel, MaskR-CNN is used to identify the crown boundary profile of the tea tree, and based on the geometric analysis of the tea crown and the projection relationship with the ground, a model is established to finally calculate the lateral boundary constraint of the robot driving channel, overcoming the path planning deviation caused by single boundary identification. Finally, by integrating the ground navigation reference line, the lateral boundary constraint, and the growth density and distribution characteristics of the tea tree, a three-dimensional drivable road network model of the tea ridge is constructed using the triangulation algorithm, which includes height, width, and inclination constraints. The accurate driving feasible region information is output, and the steps include:

[0071] 1: Multimodal tea ridge boundary detection. First, the EfficientDet recognition algorithm is used to process the collected images to identify the location information of the tea tree root. This algorithm uses a single-stage target detection framework to achieve high-precision positioning of the root area by setting appropriate anchor box sizes and positions. Second, the SegFormer segmentation network is deployed to perform semantic segmentation on the tea ridge ground. This network combines a visual transformer encoder and a lightweight decoder to perform pixel-level classification on the ground area, accurately distinguishing between tea ridge, walking path, and obstacle areas. Finally, the Mask R-CNN instance segmentation technology is used to identify the tea crown boundary contour. This technology uses a two-stage detection process to first generate region proposals, then classify each region and generate a pixel-level mask, achieving accurate description and boundary extraction of the tea tree canopy structure.

[0072] 2: Root-ground intersection navigation reference line generation method based on radial basis space mapping. First, collect point cloud data of the tea ridge area and extract the intersection area between the tea tree root and the ground. Then, a radial basis function network model is established to fit and identify the spatial distribution of the root-ground intersection points, as shown in equations (6) and (7):

[0073]

[0074] where F(x, y) represents the ground reference line height value at the plane coordinates (x, y), and p represents the plane position vector (x, y) of the query point. p j represents the plane coordinate vector of the jth root-ground intersection sample; r = ||p-p j || is the Euclidean distance between the query point and the jth sample point, used to measure spatial correlation. w j is the weight coefficient corresponding to the jth sample point, φ(r) is a multi-quadratic radial basis function, c is a smoothing parameter, and J is the number of sample points. By solving the weight coefficient w j to minimize the fitting error, an accurate continuous curve of the root-ground intersection points is obtained, providing a reliable basis for the generation of tea ridge navigation reference lines.

[0075] 3: A driving boundary determination method based on tea crown geometry analysis and ground projection is proposed. First, collect the three-dimensional shape data of the tea crown of the tea ridge and use an adaptive segmentation algorithm to extract the outer contour boundary of the tea crown. Then, for the extracted boundary data, a tea crown geometry model is established, as shown in equations (8) and (9):

[0076]

[0077] B(x, y) = T(E(z)) · (1 - δ · H(z)) (9)

[0078] where E(c) denotes the tea crown outer contour boundary function at position c; K(c, z j ) is a kernel function for realizing smooth interpolation and key feature extraction of the tea crown contour, z j is the sampling point position, a j is the corresponding coefficient, and J is the number of sampling points; B(x, y) denotes the projection function of the tea crown boundary to the ground, T(·) denotes the projection transformation operation, H(z) denotes the height influence factor, and δ is a safety boundary adjustment coefficient, which can accurately define the safe operation boundary above the tea ridge to ensure that the machine will not damage the tea crown during operation and provide lateral constraints for subsequent path planning.

[0079] 4: A tea ridge three-dimensional drivable road construction method based on triangular subdivision constraints is proposed. The root-ground intersection navigation reference line and the tea crown boundary projection are taken as constraint conditions to construct the tea ridge three-dimensional drivable road, as shown in equations (10) and (11):

[0080]

[0081] where H k (z) denotes the feasibility measurement function of the tea ridge in three-dimensional space, φ k (x, y) is the horizontal plane radial basis function, H k (z) is the elevation weight function, W k is the constraint weight coefficient, and K is the total number of basis functions; P(t) denotes the parameter equation of the three-dimensional drivable road, c l is the control point, B l (t) is the basis function, L is the number of control points, and t is the curve parameter. Through this method, a tea ridge three-dimensional drivable road network model is obtained, which not only satisfies the constraints of tea tree root and crown boundary but also has appropriate geometric characteristics.

[0082] III. Risk gradient division method based on tea ridge three-dimensional drivable road: Based on the three-dimensional drivable road network model constructed in step two, the navigation path is evaluated and optimized in detail to adapt to complex terrain. First, the key geometric characteristics of the network model (such as slope change, curvature radius, width constraint) are analyzed, and an adaptive risk assessment system is designed based on this to adapt to different slopes, bends, and width changes. The drivable area is identified and divided into different risk level areas such as high-risk area, medium-risk area, and safe area; then, combined with the risk area distribution map, a multi-objective optimization function is established considering the length of each path, the risk level corresponding to the path, and the energy consumption factor. Through an exponential weighted fusion mechanism, the multi-objective optimization function is fused to generate an optimal navigation path under the risk gradient constraint, and finally an optimized navigation path is output, which is smooth, safe, and efficient and can adapt to the complex environment of mountain tea garden. The steps include:

[0083] 1: A risk gradient division method based on the three-dimensional drivable road of tea ridge is proposed. The constructed three-dimensional drivable road network model is analyzed, the key geometric characteristics are extracted and the corresponding traffic risk is evaluated, as shown in equations (12) and (13):

[0084] R(t) = R s (t) + R w (t) + δ·R e (t) (12)

[0085] P'(t) = P0(t) + k·R(t) + ε·N(t) (13)

[0086] Where R(t) represents the comprehensive risk value of the three-dimensional drivable road of tea ridge at parameter t; t is the path parameter variable, representing the current position of the robot on the three-dimensional drivable road of tea ridge; R s (t) represents the slope and curvature risk, R w (t) represents the width risk, R e (t) represents the environmental factor risk, δ is the environmental factor weight; P'(t) represents the optimal navigation path after risk adjustment, P0(t) is the baseline navigation trajectory, N(t) is the path normal correction term, k is the gradient adjustment coefficient, ε is the normal adjustment coefficient. Through the guidance and adjustment of risk gradient, the dynamic optimization and safe navigation of the path are realized.

[0087] 2: A navigation line generation method based on risk gradient is proposed. According to the risk evaluation result R(t), a navigation line generation model considering multiple factors is constructed, as shown in equations (14) and (15):

[0088] F(t) = ω1·R(t) + ω2·S(t) + ω3·E(t) (14)

[0089] N(t) = N0(t)·e -λF(t) + β·[1-e －λF(t) ] (15)

[0090] Wherein, F(t) represents a comprehensive objective function of navigation line optimization, t is a path parameter variable, S(t) represents a smoothness function; E(t) represents an energy efficiency function, used to evaluate the energy consumption of the robot driving on the path t, considering factors such as slope change and steering amplitude; ω1 is a risk weight coefficient, controlling the importance of safety in path planning; ω2 is a smoothness weight coefficient, used to adjust the continuity and turning softness of the path, avoiding sharp turns and mutations; ω3 is an energy efficiency weight coefficient, balancing the relationship between energy consumption and path length, ensuring the optimal energy efficiency under the requirements of safety and smoothness; N(t) represents the finally generated navigation line, N0(t) is the initial navigation trajectory, β is a safety correction coefficient, and λ is an adjustment factor. Through the exponential weighted fusion mechanism, the optimal navigation path under the risk gradient constraint is generated, providing a safe and efficient driving reference path for the tea-picking robot.

[0091] Finally, it should be pointed out that the above embodiments are only used to illustrate the technical solutions of the present application and are not limiting. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced equivalently without departing from the purpose and scope of the technical solutions, which should be covered in the scope of the claims of the present application.

Claims

1. An autonomous navigation method for a tea-picking robot based on tea ridge composite boundary recognition and drivable area adjustment, characterized in that: The method includes: Comprehensively utilize historical picking data obtained from the tea ridge history database and use drone multi-sensor fusion technology to evaluate the growth status of tea ridges in real time to formulate a global path planning strategy; The root locations of tea trees are obtained, the ground of the tea ridges is semantically segmented, and a ground navigation baseline is established. The boundary contours of the tea crowns are obtained and lateral boundary constraints are established for the robot's driving path. The ground navigation baseline, lateral boundary constraints, and the growth density and distribution characteristics of the tea trees are integrated to construct a three-dimensional drivable road network model for the tea ridges. The key geometric characteristics of the three-dimensional drivable road network model of the tea ridge are analyzed, and an adaptive risk assessment system is designed based on the key geometric characteristics to divide the drivable area into areas with different risk levels. Combined with the risk area distribution map, a multi-objective optimization function is established that comprehensively considers the length of each path segment, the corresponding risk level of the path segment, and energy consumption factors. The multi-objective optimization function is fused through the exponential weighted fusion mechanism to generate the optimal navigation path under the risk gradient constraint.

2. The method according to claim 1, characterized in that Multi-dimensional information, including tea quality scores, yield records, growth status, and historical picking records, was extracted from the historical database of each tea ridge in the tea garden. A drone equipped with multiple sensors collected the appearance characteristics, physiological properties, and moisture status of each tea ridge. Feature extraction and fusion processing were performed on the collected multi-source heterogeneous data to construct a tea ridge growth status assessment model. Weights were assigned to each tea ridge based on tea quality, and the weighted sum was used to calculate the comprehensive quality score of each tea ridge. A tea ridge quality scoring model was constructed by combining the comprehensive quality score, yield score, and distance cost of the tea ridge to determine the initial picking priority of each tea ridge. The preliminary visiting order of tea ridges is determined according to the preliminary picking priority ranking. The preliminary visiting order is optimized and adjusted considering the path length and navigation difficulty between adjacent tea ridges. The final tea ridge visiting order is determined to achieve global path planning.

3. The method according to claim 2, characterized in that The tea ridge growth state evaluation model is expressed as: G i =k1·V i +k2·T i +k3·C i Where G i is the growth index of the i-th tea ridge, V i is the vegetation index of the i-th tea ridge, T i is the temperature index of the i-th tea ridge, C i is the canopy structure index of the i-th tea ridge; k1 represents the influence weight of the current vegetation growth status, k2 represents the influence weight of temperature on tea quality, and k3 is the contribution weight coefficient of tea tree structure to quality; The tea ridge quality scoring model is expressed as: P(i)=α·Q(i)+β·Y(i)-γ·D(i) Where Q(i) represents the comprehensive quality score of tea ridge i, w i represents the weight coefficient of the quality of the i-th tea ridge, n represents the total number of tea ridges evaluated; P(i) represents the final priority score of tea ridge i, Y(i) represents the yield score of tea ridge i, D(i) represents the distance cost of visiting tea ridge i, α, β, γ are the influence weight coefficients of quality, yield and distance respectively; Considering the path length and navigation difficulty between adjacent tea ridges, the initial visit order is optimized and adjusted as shown in the following formula: C(i,m)=d(i,m)·(1+η·T(i,m)) where C(i,m) represents the comprehensive navigation cost from tea ridge i to tea ridge m, d(i,m) represents the physical distance between two tea ridges, T(i,m) represents the complexity factor of the navigation path, and η is the complexity influence coefficient; L represents the total cost of the entire navigation sequence, N is the total number of tea ridges to be visited; C(i,i+1) represents the actual navigation cost from the i-th tea ridge to the i+1-th tea ridge in the sequence to be visited; by minimizing the total cost L, the final tea ridge visiting order is determined, and global path planning is achieved.

4. The method according to claim 1, wherein The robot's onboard camera captures images of tea trees. The EfficientDet recognition algorithm is used to process these images and identify the location of tea tree roots. The SegFormer segmentation network performs semantic segmentation on the tea ridge ground, classifies the ground area at the pixel level, and extracts the intersection area between the tea tree roots and the ground. A radial basis function network model is then established to fit and identify the spatial distribution of root-ground intersection points. Where F(x,y) represents the ground reference line height value at the plane coordinate (x,y), p represents the plane position vector (x,y) of the query point; p j represents the plane coordinate vector of the jth root-ground intersection sample; r = || pp j || is the Euclidean distance between the query point and the jth sample point; w j is the weight coefficient corresponding to the jth sample point, φ(r) is the multi-quadratic radial basis function, c is the smoothing parameter, and J is the number of sampling points; by solving the weight coefficient w j Minimize the fitting error, obtain an accurate continuous curve of the root-ground intersection, and generate the ground navigation reference line; The Mask R-CNN instance segmentation technique is used to identify the boundary contours of the tea canopy, describe the tea tree canopy structure and extract its boundaries. Based on the extracted boundary data, a tea canopy geometric model is established: B(x,y)=T(E(z))·(1-δ·H(z)) Where, E(c) represents the outer contour boundary function of the tea crown at position c; K(c, z j ) is the kernel function used to achieve smooth interpolation of tea crown contour and key feature extraction, z j is the sampling point location, a j is the corresponding coefficient, J is the number of sampling points; B(x, y) represents the projection function of the tea crown boundary to the ground, T(·) represents the projection transformation operation, H(z) represents the height influence factor, and δ is the safety boundary adjustment coefficient, which is used to define the safe operation boundary above the tea ridge.

5. The method according to claim 4, characterized in that Taking the ground navigation baseline and the projection of the tea crown boundary as constraints, a three-dimensional drivable road network model of the tea ridge is constructed: Where H k (z) represents the feasibility measurement function of the three-dimensional space of tea ridges, φ k (x,y) is the radial basis function of the horizontal plane, H k (z) is the elevation weight function, W k is the constraint weight coefficient, K is the total number of basis functions; P(t) represents the three-dimensional drivable road parameter equation, c l is the control point, B l (t) is the basis function, L is the number of control points, and t is the curve parameter.

6. The method according to claim 1, characterized in that The key geometric characteristics of the 3D drivable road network model of the tea ridge, including slope variation, curvature radius, and width constraints, were analyzed to establish an adaptive risk assessment system: R(t)=R s (t)+R w (t)+δ·R e (t) P'(t)=P0(t)+k·R(t)+ε·N(t) Where R(t) represents the comprehensive risk value of the three-dimensional drivable road of the tea ridge at parameter t; t is the path parameterized variable, which represents the current position of the robot on the three-dimensional drivable road of the tea ridge; R s (t) represents the slope and curvature risk, R w (t) represents the width risk, R e (t) represents the environmental factor risk, δ is the environmental factor weight; P′(t) represents the optimal navigation path after risk adjustment, P0(t) is the reference navigation trajectory, N(t) is the path normal correction term, k is the gradient adjustment coefficient, and ε is the normal adjustment coefficient; According to the comprehensive risk value R(t), the drivable area is identified and divided into areas with different risk levels, where the risk levels include high-risk area, medium-risk area and safe area.

7. The method according to claim 6, characterized in that Combined with the risk area distribution map, a multi-objective optimization function is established that comprehensively considers the length of each path segment, the corresponding risk level of the path segment, and energy consumption factors: F(t)=ω1·R(t)+ω2·S(t)+ω3·E(t) Where F(t) represents the multi-objective optimization function, t is the path parameterization variable, S(t) represents the smoothness function; E(t) represents the energy efficiency function, which is used to evaluate the energy consumption of the robot traveling on path t; ω1 is the risk weight coefficient, which controls the importance of safety in path planning; ω2 is the smoothness weight coefficient, which is used to adjust the continuity of the path and the smoothness of the turn; ω3 is the energy efficiency weight coefficient, which balances the relationship between energy consumption and path length; The multi-objective optimization functions are fused through the exponential weighted fusion mechanism to generate the optimal navigation path under the risk gradient constraint: N(t)=N0(t)·e -λF(t) +β·[1-e -λF(t) ] Where N(t) represents the final generated navigation path, N0(t) is the initial navigation trajectory, β is the safety correction coefficient, and λ is the adjustment factor.

Images