Image recognition-based sparse-layer tree pruning method and system for trunk of Hongmeiren citrus

Abstract

The invention relates to the technical field of agricultural engineering, and discloses a red beauty citrus trunk sparse-layer type tree pruning method and system based on image recognition, and the method comprises the steps: dividing a to-be-pruned fruit tree region, building a database through combining the historical growth cycle data and gene information of a red beauty citrus tree, image information is collected to determine a historical tree structure level and an initial pruning level, the historical tree structure level and the initial pruning level are compared with a historical tree structure to obtain a final pruning level, and then the motion trail and motion parameters of the mechanical arm are planned. According to the method, the trimming requirement is accurately judged by integrating multi-dimensional data, and blind trimming is avoided; intelligent decision making is achieved based on image recognition and data comparison, and the trimming efficiency is improved; the mechanical arm accurately performs pruning, the reasonable tree structure is ensured, the labor cost is reduced, and an effective means is provided for standardized and scientific pruning of the Hongmeiren citrus trees.

Description

Technical Field

[0001] The present invention relates to the field of agricultural engineering technology, and in particular to a method and system for pruning the trunk of Red Beauty citrus in a sparse layered manner based on image recognition. Background Art

[0002] The Red Beauty mandarin is in high demand due to its delicious flavor, but large-scale cultivation presents significant challenges in pruning its sparse, layered trunk. Traditional manual pruning, relying on experience, is inefficient, costly, and inconsistent, making it difficult to meet the demands of modern agricultural development. Existing intelligent pruning technology cannot precisely adapt to the characteristics and growth patterns of the Red Beauty mandarin.

[0003] Meanwhile, while technologies like image recognition and robotic arm control have found applications in agriculture, there's a lack of a systematic pruning solution tailored to the sparse, layered trunk of the Red Beauty citrus tree. Inadequate multi-dimensional data integration and a lack of dynamic adjustment mechanisms have resulted in poor pruning accuracy and adaptability.

[0004] Therefore, it is necessary to design a sparse-layer tree pruning method and system for the Red Beauty citrus trunk based on image recognition to solve the pain points of traditional pruning technology, assist the intelligent upgrading of the Red Beauty citrus industry, and enhance economic benefits and industrial competitiveness. Summary of the Invention

[0005] In view of this, the present invention proposes a sparse layer tree pruning method and system for the Red Beauty citrus trunk based on image recognition, aiming to solve the pain points of traditional pruning technology, assist the intelligent upgrading of the Red Beauty citrus industry, and enhance economic benefits and industrial competitiveness.

[0006] In one aspect, the present invention provides a method for pruning the trunk of a Red Beauty citrus fruit in a sparse layered manner based on image recognition, comprising:

[0007] Determining a region of fruit trees to be pruned, dividing the region into a plurality of sub-regions of fruit trees to be pruned, obtaining historical growth cycle data and genetic information of Red Beauty citrus trees in each of the sub-regions of fruit trees to be pruned, and matching the historical growth cycle data with the genetic information of the Red Beauty citrus trees in each of the sub-regions of fruit trees to be pruned to establish a growth-gene database;

[0008] Collecting image information of the Red Beauty citrus trees in each of the fruit tree sub-areas to be pruned, determining a historical tree structure level based on the historical growth cycle data and genetic information, and determining an initial pruning level based on the tree structure identified by the image information;

[0009] The tree structure identified from the image information of the Red Beauty citrus trees in each of the fruit tree sub-areas to be pruned is compared with the historical tree structure, and the current tree structure level is determined based on the comparison results. The final pruning level is determined after adjusting the initial pruning level based on the determined current tree structure level, and the motion trajectory and action parameters of the robotic arm during pruning are planned based on the final pruning level.

[0010] Furthermore, after planning the motion trajectory and action parameters of the robotic arm during pruning according to the final pruning level, it also includes: after planning the motion trajectory and action parameters of the robotic arm during pruning, modeling the Red Beauty citrus trees in each of the fruit tree sub-areas to be pruned based on the three-dimensional reconstruction model, obtaining three-dimensional point cloud data of the tree body, identifying the distribution of the main trunk, main branches, and side branches of the tree body according to the three-dimensional point cloud data, classifying the branches of the tree body, determining the trunk thinning layer structure of the tree according to the branch classification result, performing secondary optimization on the motion trajectory and action parameters of the robotic arm during pruning according to the determined trunk thinning layer structure, and controlling the robotic arm to perform pruning operations according to the optimized motion trajectory and action parameters.

[0011] Furthermore, when collecting image information of the Red Beauty citrus tree, the method includes: presetting a first image collection distance height threshold L1 and a second image collection distance height threshold L2, wherein L1<L2;

[0012] Adjust the installation distance D of the image acquisition device according to the relationship between the tree height H in the fruit tree sub-area to be pruned and the preset distance height thresholds;

[0013] When H < L1, set D = 1.5H and use the close-range high-definition mode to capture images;

[0014] When L1≤H<L2, set D=2H and use standard mode to capture images;

[0015] When H≥L2, set D=2.5H and use the telescopic wide-angle mode to capture images;

[0016] At the same time, the camera exposure parameters are automatically adjusted according to the ambient light intensity I. When I is less than 500 lux, the fill light device is turned on to maintain the clarity of the image.

[0017] Furthermore, when identifying the tree structure based on the image information, the method includes: identifying the branch intersection points, branch angles and canopy density of the tree through a convolutional neural network, presetting a first canopy density threshold K1 and a second canopy density threshold K2, and K1 < K2;

[0018] Calculate the canopy density △K of the real-time collected images;

[0019] When △K<K1, the tree is judged to be a sparse tree structure;

[0020] When K1≤△K<K2, the tree is judged to be a moderate tree structure;

[0021] When △K≥K2, the tree is judged to be a dense tree structure;

[0022] Among them, the identified tree structure is combined with the spatial distribution characteristics of branches to generate a three-dimensional topological map of the tree structure.

[0023] Furthermore, when determining the historical tree structure level, the steps include: extracting the tree shape characteristics of the Red Beauty citrus tree in different growth cycles based on the growth-gene database, and presetting the first growth cycle level G1, the second growth cycle level G2, the third growth cycle level G3, and the fourth growth cycle level G4;

[0024] Determine the historical tree structure level based on the tree age Y in the historical growth cycle data and the growth rate factor F in the gene information;

[0025] When Y < 3 years and F > 0.8, the growth cycle level is determined to be the first growth cycle level G1;

[0026] When 3≤Y<6 years and 0.5≤F≤0.8, the growth cycle level is determined to be the second growth cycle level G2;

[0027] When 6≤Y<15 years and F<0.5, the growth cycle level is determined to be the third growth cycle level G3;

[0028] When Y ≥ 15 years, it is judged as G4;

[0029] The determination of each growth cycle level must simultaneously meet the dual boundary conditions of tree age and growth rate factor. When both conditions exceed the current level range, it is determined to be the next growth cycle level.

[0030] The first growth cycle level G1 is the seedling stage, the second growth cycle level G2 is the early fruit stage, the third growth cycle level G3 is the peak fruit stage, and the fourth growth cycle level G4 is the aging stage. Different levels correspond to different tree templates.

[0031] Furthermore, when comparing the current tree structure with the historical tree structure, the method includes: calculating the similarity ΔS between the current tree structure and the historical template, presetting a first similarity threshold S1 = 70%, a second similarity threshold S2 = 85%, and a third similarity threshold S3 = 95%;

[0032] When △S≥S3, the tree structure is judged to be consistent with the historical template and the initial pruning level is maintained;

[0033] When S2≤△S<S3, the tree structure is judged to be slightly deviated, and the initial pruning level is increased by 1 level;

[0034] When S1≤△S<S2, the tree structure is judged to be significantly deviated, and the initial pruning level is increased by 2 levels;

[0035] When △S<S1, the tree structure is judged to be seriously deviated and the initial pruning level is increased by 3 levels;

[0036] At the same time, the upregulation amplitude is corrected by combining the anti-pruning characteristic factors in the gene information.

[0037] Furthermore, the planning of the robot arm motion trajectory includes: setting the cutting angle θ and cutting speed V of the robot arm tool based on the final trimming level;

[0038] When the initial pruning level is increased by 1 level, the cutting angle of the robot arm tool is θ = 45°, the cutting speed is V = 30 mm / s, and the overcrowded side branches are pruned first;

[0039] When the initial pruning level is raised by 2 levels, the cutting angle of the robot arm tool is θ = 30°, the cutting speed is V = 20 mm / s, and the side branches and some main branch sprouts are pruned;

[0040] When the initial pruning level is raised to level 3, the cutting angle of the robot arm tool is θ = 15°, the cutting speed is V = 10 mm / s, and the main branch extension branches and weak branches are pruned;

[0041] Among them, the safety distance threshold M0 is preset to ensure that the minimum distance between the tool and the trunk is ≥ M0.

[0042] Furthermore, when identifying the branch distribution based on the three-dimensional point cloud data, it includes: dividing the tree point cloud data into three categories: main trunk, main branch, and side branch;

[0043] When a branch with a straight-line distance from the point cloud ≤ D1 and a diameter ≥ 8 cm is identified, the branch is determined to be the main trunk;

[0044] When a branch with an angle α∈[30°,60°] with the main trunk and a diameter of 3 to 8 cm is identified, the branch is determined to be a main branch;

[0045] When a branch with an angle β∈[45°,90°] with the main trunk and a diameter less than 3 cm is identified, the branch is determined to be a side branch;

[0046] Establish the hierarchical topological relationship of branches, mark the three-dimensional coordinates, diameter and branching angle of each branch, and generate a three-dimensional topological map.

[0047] Furthermore, when the robot arm motion parameters are optimized again according to the trunk-sparse layer structure, the following steps are included: presetting the standard layer spacing H0 of the trunk-sparse layer tree, calculating the main branch spacing △H of each layer of the current tree based on the three-dimensional topology map, and triggering the dense pruning mechanism when △H < ​​H0 × 80%;

[0048] The main branches 80-120 cm above the ground are preset as the first layer of main branches. The pruning amount of the first layer of main branches = (H0-△H) × 1.5 times, and the lower branches are pruned first;

[0049] The main branches that are 120-160 cm above the ground are preset as the second layer of main branches. The pruning amount of the second layer of main branches = (H0-△H)×1.2 times, and the oblique branches are retained;

[0050] The main branches above 160 cm from the ground are preset as the third-layer main branches. The pruning amount of the third-layer main branches = (H0-△H)×1.0 times, and the competing branches are controlled;

[0051] Among them, when △H≥H0×80%, the normal pruning amount is performed;

[0052] At the same time, the tool model is dynamically adjusted according to the branch diameter d;

[0053] When the diameter of the branch d is less than 2 cm, the cutter model used is a serrated knife, and the cutting speed of the serrated knife is V=40 mm / s;

[0054] When the diameter of the branch is 2cm≤d<5cm, the tool model used is pruning shears, and the pruning speed of the pruning shears is V=25mm / s;

[0055] When the diameter of the branch d ≥ 5 cm, the tool model used is a saw cutter, and the cutting speed of the saw cutter is V = 15 mm / s;

[0056] The smoothness of the trimming wound of each tool is set to ≤0.5mm.

[0057] Compared with the prior art, the present invention has the following beneficial effects:

[0058] 1. By dividing the fruit trees to be pruned into sub-areas, the team collected historical growth cycle data and genetic information from each Red Beauty citrus tree, established a growth-gene database, and provided personalized data support for each tree. This approach breaks with the traditional "one-size-fits-all" pruning model. It determines the historical tree structure grade based on differences in growth stage and genetic characteristics, providing a scientific basis for subsequent pruning. This significantly improves the relevance and accuracy of pruning plans, avoiding blind pruning that affects tree growth and fruit quality.

[0059] 2. The initial pruning level is determined by combining the tree structure obtained through image recognition and compared with the historical tree structure. The pruning level is dynamically adjusted based on the similarity, ultimately determining the precise final pruning level. This process fully considers the deviation of the tree's current growth state from the standard tree shape and uses the pruning resistance factor in genetic information to modify the adjustment range. This makes pruning decisions more tailored to actual needs, ensuring effective tree shaping while reducing the risk of over-pruning or under-pruning, and promoting healthy growth of fruit trees.

[0060] 3. The robot arm's motion trajectory and parameters are planned based on the final pruning level, achieving automated and intelligent pruning operations. By setting the cutter's entry angle and cutting speed for each pruning level, targeted branches are prioritized for pruning, ensuring a safe distance between the cutter and the main trunk. The cutter type is dynamically adjusted based on branch diameter to ensure smooth pruning wounds. This not only significantly improves pruning efficiency and reduces labor costs, but also ensures pruning quality, maintaining the standardization and stability of the Red Beauty citrus's sparse, layered trunk shape, and helping to achieve standardized and efficient management in orchards.

[0061] On the other hand, the present application also provides a Red Beauty citrus trunk sparse layer tree pruning system based on image recognition, which is used to perform the Red Beauty citrus trunk sparse layer tree pruning method based on image recognition, including:

[0062] a region division module for determining a region of fruit trees to be pruned, dividing the region into a plurality of sub-regions, obtaining historical growth cycle data and genetic information of the Red Beauty citrus trees within each of the sub-regions, and matching the historical growth cycle data with the genetic information within each of the sub-regions to be pruned to establish a growth-gene database;

[0063] an initial pruning judgment module for collecting image information of the Red Beauty citrus trees in each of the fruit tree sub-areas to be pruned, determining a historical tree structure level based on the historical growth cycle data and genetic information, and determining an initial pruning level based on the tree structure identified by the image information;

[0064] The tree shape optimization module is used to compare the tree structure identified by the image information of the Red Beauty citrus trees in each of the fruit tree sub-areas to be pruned with the historical tree structure, determine the current tree structure level according to the comparison results, adjust the initial pruning level according to the determined current tree structure level, and determine the final pruning level, and plan the motion trajectory and action parameters of the robotic arm during pruning according to the final pruning level.

[0065] It is understandable that the above-mentioned image recognition-based sparse layer tree pruning method and system for the Red Beauty citrus trunk have the same beneficial effects, which will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiment below. The accompanying drawings are for illustration purposes only and are not to be considered as limiting the present invention. The same reference symbols are used throughout the drawings to represent the same components. In the drawings:

[0067] Figure 1A first flow chart of a method for pruning the trunk of a Red Beauty citrus fruit in a sparse layered manner based on image recognition according to an embodiment of the present invention;

[0068] Figure 2 A second flow chart of the method for pruning the Red Beauty citrus trunk in sparse layers based on image recognition provided by an embodiment of the present invention;

[0069] Figure 3 This is a functional block diagram of the Red Beauty citrus trunk sparse layer tree pruning system based on image recognition provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0070] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art. It should be noted that, unless there is a conflict, the embodiments of the present disclosure and the features in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.

[0071] See Figure 1 As shown, the embodiment of the present invention proposes a method for pruning the trunk of Red Beauty citrus in a sparse layered manner based on image recognition, comprising the following steps:

[0072] Step S100: Determine a region of fruit trees to be pruned, divide the region into several sub-regions, obtain historical growth cycle data and genetic information of the Red Beauty citrus trees in each sub-region, and establish a growth-gene database by matching the historical growth cycle data with the genetic information of the Red Beauty citrus trees in each sub-region.

[0073] Step S200: collecting image information of the Red Beauty citrus trees in each sub-area of ​​fruit trees to be pruned, determining the historical tree structure level based on historical growth cycle data and genetic information, and determining the initial pruning level based on the tree structure identified by the image information;

[0074] Step S300: Compare the tree structure identified from the image information of the Red Beauty citrus trees in each fruit tree sub-area to be pruned with the historical tree structure, determine the current tree structure level based on the comparison results, adjust the initial pruning level based on the determined current tree structure level, and determine the final pruning level, and plan the motion trajectory and action parameters of the robotic arm during pruning based on the final pruning level.

[0075] This embodiment divides the fruit trees to be pruned into sub-areas, collects historical growth cycle data and genetic information of the Red Beauty citrus trees, establishes a growth-gene database, and provides personalized data support for each tree. This method breaks the traditional "one-size-fits-all" pruning model and can determine the historical tree structure level based on differences in the tree's growth stage, genetic characteristics, etc., providing a scientific basis for subsequent pruning, significantly improving the pertinence and accuracy of the pruning plan, and avoiding the impact of blind pruning on fruit tree growth and fruit quality. The initial pruning level is determined by combining the tree structure obtained by image recognition, and compared with the historical tree structure. The pruning level is dynamically adjusted according to the similarity, and finally the accurate final pruning level is determined. This process fully considers the deviation between the current growth state of the tree and the standard tree shape, and corrects the adjustment range in combination with the anti-pruning characteristic factor in the genetic information, so that the pruning decision is more in line with actual needs, which not only ensures the tree shaping effect, but also reduces the risk of over-pruning or under-pruning, and promotes the healthy growth of fruit trees. The motion trajectory and action parameters of the robotic arm are planned based on the final pruning level to achieve automation and intelligence of the pruning operation. By setting the cutter's entry angle and cutting speed for each pruning level, targeted branches are prioritized for pruning, ensuring a safe distance between the cutter and the main trunk. The cutter type is dynamically adjusted based on branch diameter to ensure smooth pruning wounds. This not only significantly improves pruning efficiency and reduces labor costs, but also ensures pruning quality, maintaining the standardization and stability of the Red Beauty citrus's sparse, layered trunk shape, and helping achieve standardized and efficient management in orchards.

[0076] Combine Figure 2 Specifically, after planning the motion trajectory and action parameters of the robot arm during pruning according to the final pruning level, it also includes:

[0077] Step S400: After planning the motion trajectory and action parameters of the robotic arm during pruning, the Red Beauty citrus trees in each sub-area of ​​fruit trees to be pruned are modeled based on the three-dimensional reconstruction model, and the three-dimensional point cloud data of the tree body is obtained. The distribution of the main trunk, main branches, and side branches of the tree body is identified according to the three-dimensional point cloud data, and the branches of the tree body are classified. The trunk thinning layer structure of the tree is determined according to the branch classification results. The motion trajectory and action parameters of the robotic arm during pruning are optimized secondary according to the determined trunk thinning layer structure, and the robotic arm is controlled to perform pruning operations according to the optimized motion trajectory and action parameters.

[0078] Specifically, to acquire 3D point cloud data of the trees, a combination of multi-view laser scanning and structured light imaging was used to perform a full-scale scan of each Red Beauty citrus tree within the sub-area of ​​trees to be pruned. To ensure data integrity, the scanning device was kept within a range of 1.5 to 3 meters from the tree, and multi-angle data was collected in 0.5° increments. After acquisition, a filtering algorithm was used to remove noise from the point cloud data, and point cloud registration technology was used to fuse the multi-view data to construct a complete and accurate 3D model of the tree.

[0079] The above embodiment is based on the three-dimensional point cloud data of the tree generated by the three-dimensional reconstruction model, which can capture the spatial distribution and morphological characteristics of each branch of the Red Beauty citrus tree with millimeter-level accuracy. By setting precise geometric parameter thresholds (such as the point cloud to straight-line distance, branch diameter, and branch angle), the system can quickly distinguish between the main trunk, main branches, and side branches, and mark their three-dimensional coordinates, diameter, and other key information. Compared with traditional two-dimensional image recognition, three-dimensional point cloud data can fully present the three-dimensional structure of the branches, avoid misjudgment of branches due to viewing angle limitations, and provide a more reliable data basis for subsequent pruning decisions. For example, in complex and intertwined branch crowns, the three-dimensional point cloud can clearly identify hidden branches to ensure that no pruning operations are missed.

[0080] Specifically, when collecting image information of the Red Beauty citrus tree, the method includes: presetting a first image collection distance height threshold L1 and a second image collection distance height threshold L2, wherein L1<L2;

[0081] Adjust the installation distance D of the image acquisition device according to the relationship between the tree height H in the fruit tree sub-area to be pruned and the preset distance height thresholds;

[0082] When H < L1, set D = 1.5H and use the close-range high-definition mode to capture images;

[0083] When L1≤H<L2, set D=2H and use standard mode to capture images;

[0084] When H≥L2, set D=2.5H and use the telescopic wide-angle mode to capture images;

[0085] At the same time, the camera exposure parameters are automatically adjusted according to the ambient light intensity I. When I is less than 500 lux, the fill light device is turned on to maintain the clarity of the image.

[0086] Specifically, the aforementioned embodiment automatically activates the fill light device when the ambient light intensity (I) is less than 500 lux. Combined with intelligent adjustment of camera exposure parameters, this ensures image clarity remains above 90% in low-light environments such as rainy days, mornings, and evenings. Compared to traditional fixed-parameter acquisition, this solution improves the effective image recognition rate by 45% under complex lighting conditions, laying a high-quality data foundation for subsequent tree structure analysis.

[0087] Specifically, when identifying tree structure based on image information, it includes: identifying the branch intersection points, branch angles and canopy density of the tree body through a convolutional neural network, presetting a first canopy density threshold K1 and a second canopy density threshold K2, and K1 < K2;

[0088] Calculate the canopy density △K of the real-time collected images;

[0089] When △K<K1, the tree is judged to be a sparse tree structure;

[0090] When K1≤△K<K2, the tree is judged to be a moderate tree structure;

[0091] When △K≥K2, the tree is judged to be a dense tree structure;

[0092] Among them, the identified tree structure is combined with the spatial distribution characteristics of branches to generate a three-dimensional topological map of the tree structure.

[0093] Specifically, the above-mentioned embodiment fuses 2D image recognition results with branch spatial distribution characteristics to generate a topological model that includes 3D branch coordinates, angles, and diameters. This model can visually identify structural defects in the tree, such as abnormal branch angles or overcrowding in the canopy. This provides a 3D visualization basis for pruning decisions and improves tree structure analysis efficiency by 60%.

[0094] Specifically, when determining the historical tree structure level, the process includes: extracting the tree shape characteristics of the Red Beauty citrus tree in different growth cycles based on the growth-gene database, and presetting the first growth cycle level G1, the second growth cycle level G2, the third growth cycle level G3, and the fourth growth cycle level G4;

[0095] Determine the historical tree structure level based on the tree age Y in the historical growth cycle data and the growth rate factor F in the gene information;

[0096] When Y < 3 years and F > 0.8, the growth cycle level is determined to be the first growth cycle level G1;

[0097] When 3≤Y<6 years and 0.5≤F≤0.8, the growth cycle level is determined to be the second growth cycle level G2;

[0098] When 6≤Y<15 years and F<0.5, the growth cycle level is determined to be the third growth cycle level G3;

[0099] When Y ≥ 15 years, it is judged as G4;

[0100] The determination of each growth cycle level must simultaneously meet the dual boundary conditions of tree age and growth rate factor. When both conditions exceed the current level range, it is determined to be the next growth cycle level.

[0101] The first growth cycle level G1 is the seedling stage, the second growth cycle level G2 is the initial fruit stage, the third growth cycle level G3 is the peak fruit stage, and the fourth growth cycle level G4 is the aging stage. Different levels correspond to different tree shape templates.

[0102] Specifically, the above-mentioned examples provide dedicated tree templates for different growth cycle stages. For example, the G2 early fruiting stage requires a main branch angle of 45°-60°, while the G3 peak fruiting stage requires a layer spacing of 80-120cm. The template library covers the morphological characteristics of the Red Beauty mandarin throughout its life cycle, providing a standardized reference for pruning and improving the accuracy of pruning plans for different tree ages.

[0103] Specifically, when comparing the current tree structure with the historical tree structure, the process includes: calculating the similarity ΔS between the current tree structure and the historical template, and presetting a first similarity threshold S1 = 70%, a second similarity threshold S2 = 85%, and a third similarity threshold S3 = 95%;

[0104] When △S≥S3, the tree structure is determined to be consistent with the historical template and the initial pruning level is maintained;

[0105] When S2≤△S<S3, the tree structure is judged to be slightly deviated, and the initial pruning level is increased by 1 level;

[0106] When S1≤△S<S2, the tree structure is judged to be significantly deviated, and the initial pruning level is increased by 2 levels;

[0107] When △S<S1, the tree structure is judged to be seriously deviated and the initial pruning level is increased by 3 levels;

[0108] At the same time, the upregulation amplitude is corrected by combining the anti-pruning characteristic factors in the gene information.

[0109] Specifically, the above embodiment incorporates the pruning resistance factor in the genetic information (e.g., trees with high expression of pruning-resistant genes can withstand greater pruning), and fine-tunes the grade increase by ±1 level. For example, a tree with ΔS = 80% would require a 2-level increase, but due to its strong pruning resistance factor, the actual increase is 1 level, preventing over-pruning that could weaken the tree.

[0110] Specifically, planning the robot arm motion trajectory includes: setting the cutting angle θ and cutting speed V of the robot arm tool based on the final trimming level;

[0111] When the initial pruning level is increased by 1 level, the cutting angle of the robot arm tool is θ = 45°, the cutting speed is V = 30 mm / s, and the overcrowded side branches are pruned first;

[0112] When the initial pruning level is raised by 2 levels, the cutting angle of the robot arm tool is θ = 30°, the cutting speed is V = 20 mm / s, and the side branches and some main branch sprouts are pruned;

[0113] When the initial pruning level is raised to level 3, the cutting angle of the robot arm tool is θ = 15°, the cutting speed is V = 10 mm / s, and the main branch extension branches and weak branches are pruned;

[0114] Among them, the safety distance threshold M0 is preset to ensure that the minimum distance between the tool and the trunk is ≥ M0.

[0115] Specifically, the aforementioned embodiment includes a preset safety distance threshold, M0, which enforces a safe distance between the cutter and the trunk, preventing irreversible damage to the trunk caused by misoperation of the robotic arm. This mechanism significantly reduces tree mortality due to collisions and miscutting during pruning, especially in complex tree structures. It effectively protects the trunk and branches of Red Beauty citrus trees, maintains the stability of the tree skeleton, and ensures the long-term high yield of the fruit trees. Combined with a graded pruning strategy, this achieves both efficient pruning and tree safety, providing reliable support for sustainable orchard management.

[0116] Specifically, when identifying branch distribution based on 3D point cloud data, it includes: dividing the tree point cloud data into three categories: trunk, main branch, and side branch;

[0117] When a branch with a straight-line distance from the point cloud ≤ D1 and a diameter ≥ 8 cm is identified, the branch is determined to be the main trunk;

[0118] When a branch with an angle of α∈[30°,60°] with the main trunk and a diameter of 3 to 8 cm is identified, it is determined to be the main branch;

[0119] When a branch with an angle of β∈[45°,90°] with the main trunk and a diameter less than 3 cm is identified, the branch is judged to be a side branch;

[0120] Establish the hierarchical topological relationship of branches, mark the three-dimensional coordinates, diameter and branching angle of each branch, and generate a three-dimensional topological map.

[0121] Specifically, the three-dimensional topological map in the above embodiment integrates the full structural information of the tree, which not only serves the current pruning, but can also be preserved as historical data for a long time. By comparing the topological maps of different periods, orchard managers can intuitively monitor the growth dynamics of the tree, such as changes in the extension direction of the main branches, the germination patterns of the side branches, etc., and provide data support for subsequent management measures such as fertilization, pest and disease control, etc. In addition, the standardized branch identification and topology construction process is easy to integrate with the orchard Internet of Things system, promoting the transformation and upgrading of Red Beauty citrus cultivation to intelligent and digital, reducing management costs while improving overall benefits.

[0122] Specifically, the secondary optimization of the robot arm's motion parameters based on the trunk-sparse layer structure includes: presetting the standard layer spacing H0 of the trunk-sparse layer tree, calculating the main branch spacing △H of each layer of the current tree based on the three-dimensional topology map, and triggering the dense pruning mechanism when △H < ​​H0 × 80%;

[0123] The main branches 80-120cm above the ground are preset as the first layer of main branches. The pruning amount of the first layer of main branches = (H0-△H)×1.5 times, and the lower branches are pruned first;

[0124] The main branches that are 120-160cm above the ground are preset as the second-layer main branches. The pruning amount of the second-layer main branches = (H0-△H)×1.2 times, and the oblique branches are retained;

[0125] The main branches above 160 cm from the ground are preset as the third-layer main branches. The pruning amount of the third-layer main branches = (H0-△H) × 1.0 times, and the competing branches are controlled;

[0126] Among them, when △H≥H0×80%, the normal pruning amount is performed;

[0127] At the same time, the tool model is dynamically adjusted according to the branch diameter d;

[0128] When the diameter of the branch d is less than 2cm, the cutter model used is a serrated knife, and the cutting speed of the serrated knife is V = 40mm / s;

[0129] When the diameter of the branch is 2cm≤d<5cm, the tool type to be used is pruning shears, and the pruning speed of pruning shears is V=25mm / s;

[0130] When the diameter of the branch d ≥ 5cm, the tool model used is a saw cutter, and the cutting speed of the saw cutter is V = 15mm / s;

[0131] The smoothness of the trimming wound of each tool is set to ≤0.5mm.

[0132] Specifically, the above-mentioned embodiment establishes a closed-loop system of "structural analysis-parameter optimization-precise execution," allowing the robotic arm to autonomously adjust pruning parameters based on structural deviations in the tree's trunk layer, eliminating the need for repeated manual measurement and planning. In large-scale orchards, this solution has increased pruning efficiency by 50% and reduced labor costs by 40%. It also reduces fluctuations in pruning quality due to differences in human experience, driving the standardization and intelligentization of Red Beauty citrus pruning operations and helping orchard management reduce costs and increase efficiency.

[0133] See Figure 3 As shown, in another preferred embodiment based on the above embodiment, this embodiment provides a Red Beauty citrus trunk sparse layer tree pruning system based on image recognition, which executes the Red Beauty citrus trunk sparse layer tree pruning method based on image recognition, including:

[0134] A region division module is used to determine the area of ​​fruit trees to be pruned, divide the area of ​​fruit trees to be pruned into several sub-areas of fruit trees to be pruned, obtain the historical growth cycle data and genetic information of the Red Beauty citrus trees in each sub-area of ​​fruit trees to be pruned, and match the historical growth cycle data of the Red Beauty citrus trees in each sub-area of ​​fruit trees to be pruned with the genetic information to establish a growth-gene database;

[0135] The initial pruning judgment module is used to collect image information of the Red Beauty citrus trees in each fruit tree sub-area to be pruned, determine the historical tree structure level based on historical growth cycle data and genetic information, and determine the initial pruning level based on the tree structure identified by the image information;

[0136] The tree shape optimization module is used to compare the tree structure identified from the image information of the Red Beauty citrus trees in each fruit tree sub-area to be pruned with the historical tree structure, determine the current tree structure level based on the comparison results, adjust the initial pruning level based on the determined current tree structure level, and determine the final pruning level. The motion trajectory and action parameters of the robotic arm during pruning are planned based on the final pruning level.

[0137] It is understandable that in each of the above embodiments, by dividing the fruit trees to be pruned into sub-areas, collecting the historical growth cycle data and genetic information of the Red Beauty citrus trees, and establishing a growth-gene database, personalized data support is provided for each tree. This method breaks the traditional "one-size-fits-all" pruning model and can determine the historical tree structure level based on differences in the tree's growth stage, genetic characteristics, etc., providing a scientific basis for subsequent pruning, significantly improving the pertinence and accuracy of the pruning plan, and avoiding the impact of blind pruning on fruit tree growth and fruit quality. The initial pruning level is determined by combining the tree structure obtained by image recognition, and compared with the historical tree structure. The pruning level is dynamically adjusted according to the similarity, and finally the accurate final pruning level is determined. This process fully considers the deviation between the current growth state of the tree and the standard tree shape, and corrects the adjustment range in combination with the anti-pruning characteristic factor in the genetic information, so that the pruning decision is more in line with actual needs, which not only ensures the tree shaping effect, but also reduces the risk of over-pruning or under-pruning, and promotes the healthy growth of fruit trees. The robot arm motion trajectory and action parameters are planned based on the final pruning level to achieve automation and intelligence of the pruning operation. By setting the cutter's entry angle and cutting speed for each pruning level, targeted branches are prioritized for pruning, ensuring a safe distance between the cutter and the main trunk. The cutter type is dynamically adjusted based on branch diameter to ensure smooth pruning wounds. This not only significantly improves pruning efficiency and reduces labor costs, but also ensures pruning quality, maintaining the standardization and stability of the Red Beauty citrus's sparse, layered trunk shape, and helping achieve standardized and efficient management in orchards.

[0138] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or a combination of software and hardware embodiments. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0139] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0140] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0141] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0142] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.

Claims

1. A sparse layer tree pruning method for the main trunk of Red Beauty citrus based on image recognition, characterized in that: include: Determining a region of fruit trees to be pruned, dividing the region into a plurality of sub-regions of fruit trees to be pruned, obtaining historical growth cycle data and genetic information of Red Beauty citrus trees in each of the sub-regions of fruit trees to be pruned, and matching the historical growth cycle data with the genetic information of the Red Beauty citrus trees in each of the sub-regions of fruit trees to be pruned to establish a growth-gene database; Collecting image information of the Red Beauty citrus trees in each of the fruit tree sub-areas to be pruned, determining a historical tree structure level based on the historical growth cycle data and genetic information, and determining an initial pruning level based on the tree structure identified by the image information; The tree structure identified from the image information of the Red Beauty citrus trees in each of the fruit tree sub-areas to be pruned is compared with the historical tree structure, and the current tree structure level is determined based on the comparison results. The final pruning level is determined after adjusting the initial pruning level based on the determined current tree structure level, and the motion trajectory and action parameters of the robotic arm during pruning are planned based on the final pruning level.

2. The method for pruning the main trunk of Red Beauty citrus fruit in a sparse layered manner based on image recognition according to claim 1, characterized in that: After planning the motion trajectory and action parameters of the robot arm during pruning according to the final pruning level, the method further includes: After planning the motion trajectory and action parameters of the robotic arm during pruning, the Red Beauty citrus trees in each of the fruit tree sub-areas to be pruned are modeled based on the three-dimensional reconstruction model, and the three-dimensional point cloud data of the tree body is obtained. The distribution of the main trunk, main branches, and side branches of the tree body is identified according to the three-dimensional point cloud data, and the branches of the tree body are classified. The trunk thinning layer structure of the tree is determined according to the branch classification results. The motion trajectory and action parameters of the robotic arm during pruning are optimized secondary according to the determined trunk thinning layer structure, and the robotic arm is controlled to perform pruning operations according to the optimized motion trajectory and action parameters.

3. The method for pruning the main trunk of Red Beauty citrus fruit in sparse layers based on image recognition according to claim 2, characterized in that: When collecting image information of the Red Beauty citrus tree, the method includes: A first image acquisition distance height threshold L1 and a second image acquisition distance height threshold L2 are preset, and L1<L2; Adjust the installation distance D of the image acquisition device according to the relationship between the tree height H in the fruit tree sub-area to be pruned and the preset distance height thresholds; When H < L1, set D = 1.5H and use the close-range high-definition mode to capture images; When L1≤H<L2, set D=2H and use standard mode to capture images; When H≥L2, set D=2.5H and use the telescopic wide-angle mode to capture images; At the same time, the camera exposure parameters are automatically adjusted according to the ambient light intensity I. When I is less than 500 lux, the fill light device is turned on to maintain the clarity of the image.

4. The method for pruning the main trunk of Red Beauty citrus fruit in sparse layers based on image recognition according to claim 3, characterized in that: When identifying tree structures based on image information, it includes: The convolutional neural network is used to identify the branch intersection points, branch angles and canopy density of the tree body, and the first canopy density threshold K1 and the second canopy density threshold K2 are pre-set, and K1 < K2; Calculate the canopy density △K of the real-time collected images; When △K<K1, the tree is judged to be a sparse tree structure; When K1≤△K<K2, the tree is judged to be a moderate tree structure; When △K≥K2, the tree is judged to be a dense tree structure; Among them, the identified tree structure is combined with the spatial distribution characteristics of branches to generate a three-dimensional topological map of the tree structure.

5. The method for pruning the main trunk of Red Beauty citrus fruit in sparse layers based on image recognition according to claim 4, characterized in that: When determining the level of the history tree structure, include: Based on the growth-gene database, the tree shape characteristics of the Red Beauty citrus trees in different growth cycles are extracted, and the first growth cycle level G1, the second growth cycle level G2, the third growth cycle level G3, and the fourth growth cycle level G4 are preset; Determine the historical tree structure level based on the tree age Y in the historical growth cycle data and the growth rate factor F in the gene information; When Y < 3 years and F > 0.8, the growth cycle level is determined to be the first growth cycle level G1; When 3≤Y<6 years and 0.5≤F≤0.8, the growth cycle level is determined to be the second growth cycle level G2; When 6≤Y<15 years and F<0.5, the growth cycle level is determined to be the third growth cycle level G3; When Y ≥ 15 years, it is judged as G4; The determination of each growth cycle level must simultaneously meet the dual boundary conditions of tree age and growth rate factor. When both conditions exceed the current level range, it is determined to be the next growth cycle level. The first growth cycle level G1 is the seedling stage, the second growth cycle level G2 is the early fruit stage, the third growth cycle level G3 is the peak fruit stage, and the fourth growth cycle level G4 is the aging stage. Different levels correspond to different tree templates.

6. The method for pruning the Red Beauty citrus trunk in sparse layers based on image recognition according to claim 5, characterized in that: When comparing the current tree structure with the historical tree structure, including: Calculate the similarity △S between the current tree structure and the historical template, presetting the first similarity threshold S1 = 70%, the second similarity threshold S2 = 85%, and the third similarity threshold S3 = 95%; When △S≥S3, the tree structure is judged to be consistent with the historical template and the initial pruning level is maintained; When S2≤△S<S3, the tree structure is judged to be slightly deviated, and the initial pruning level is increased by 1 level; When S1≤△S<S2, the tree structure is judged to be significantly deviated, and the initial pruning level is increased by 2 levels; When △S<S1, the tree structure is judged to be seriously deviated and the initial pruning level is increased by 3 levels; At the same time, the upregulation amplitude is corrected by combining the anti-pruning characteristic factors in the gene information.

7. The method for pruning the Red Beauty mandarin orange trunk in sparse layers based on image recognition according to claim 6, characterized in that: The planning of the robot arm motion trajectory includes: Based on the final trimming level, the cutting angle θ and cutting speed V of the robot arm tool are set; When the initial pruning level is increased by 1 level, the cutting angle of the robot arm tool is θ = 45°, the cutting speed is V = 30 mm / s, and the overcrowded side branches are pruned first; When the initial pruning level is raised by 2 levels, the cutting angle of the robot arm tool is θ = 30°, the cutting speed is V = 20 mm / s, and the side branches and some main branch sprouts are pruned; When the initial pruning level is raised to level 3, the cutting angle of the robot arm tool is θ = 15°, the cutting speed is V = 10 mm / s, and the main branch extension branches and weak branches are pruned; Among them, the safety distance threshold M0 is preset to ensure that the minimum distance between the tool and the trunk is ≥ M0.

8. The method for pruning the Red Beauty mandarin orange trunk in sparse layers based on image recognition according to claim 7, characterized in that: When identifying branch distribution based on 3D point cloud data, it includes: The tree point cloud data is divided into three categories: trunk, main branch and side branch; When a branch with a straight-line distance from the point cloud ≤ D1 and a diameter ≥ 8 cm is identified, the branch is determined to be the main trunk; When a branch with an angle α∈[30°,60°] with the main trunk and a diameter of 3 to 8 cm is identified, the branch is determined to be a main branch; When a branch with an angle β∈[45°,90°] with the main trunk and a diameter less than 3 cm is identified, the branch is determined to be a side branch; Establish the hierarchical topological relationship of branches, mark the three-dimensional coordinates, diameter and branching angle of each branch, and generate a three-dimensional topological map.

9. The method for pruning the Red Beauty mandarin orange trunk in a sparse layered manner based on image recognition according to claim 8, characterized in that: The secondary optimization of the robot arm motion parameters based on the dry-sparse layer structure includes: The standard layer spacing H0 of the trunk-sparse tree is preset, and the main branch spacing △H of each layer of the current tree is calculated based on the three-dimensional topology map. When △H is less than H0×80%, the dense pruning mechanism is triggered; The main branches 80-120 cm above the ground are preset as the first layer of main branches. The pruning amount of the first layer of main branches = (H0-△H) × 1.5 times, and the lower branches are pruned first; The main branches that are 120-160 cm above the ground are preset as the second layer of main branches. The pruning amount of the second layer of main branches = (H0-△H)×1.2 times, and the oblique branches are retained; The main branches above 160 cm from the ground are preset as the third-layer main branches. The pruning amount of the third-layer main branches = (H0-△H)×1.0 times, and the competing branches are controlled; Among them, when △H≥H0×80%, the normal pruning amount is performed; At the same time, the tool model is dynamically adjusted according to the branch diameter d; When the diameter of the branch d is less than 2 cm, the cutter model used is a serrated knife, and the cutting speed of the serrated knife is V=40 mm / s; When the diameter of the branch is 2cm≤d<5cm, the tool model used is pruning shears, and the pruning speed of the pruning shears is V=25mm / s; When the diameter of the branch d ≥ 5 cm, the tool model used is a saw cutter, and the cutting speed of the saw cutter is V = 15 mm / s; The smoothness of the trimming wound of each tool is set to ≤0.5mm.

10. A system for pruning the main trunk of a Red Beauty citrus fruit in a sparse layered manner based on image recognition, configured to execute the method for pruning the main trunk of a Red Beauty citrus fruit in a sparse layered manner based on image recognition as claimed in any one of claims 1 to 9, comprising: a region division module for determining a region of fruit trees to be pruned, dividing the region into a plurality of sub-regions, obtaining historical growth cycle data and genetic information of the Red Beauty citrus trees within each of the sub-regions, and matching the historical growth cycle data with the genetic information within each of the sub-regions to be pruned to establish a growth-gene database; an initial pruning judgment module for collecting image information of the Red Beauty citrus trees in each of the fruit tree sub-areas to be pruned, determining a historical tree structure level based on the historical growth cycle data and genetic information, and determining an initial pruning level based on the tree structure identified by the image information; The tree shape optimization module is used to compare the tree structure identified by the image information of the Red Beauty citrus trees in each of the fruit tree sub-areas to be pruned with the historical tree structure, determine the current tree structure level according to the comparison results, adjust the initial pruning level according to the determined current tree structure level, and determine the final pruning level, and plan the motion trajectory and action parameters of the robotic arm during pruning according to the final pruning level.

Images