Intelligent planting method, device and equipment for measuring offspring of forest trees

Abstract

This application discloses a method, device, and equipment for intelligent planting of forest trees for progeny testing, relating to the field of forest tree genetics and breeding. The method includes: acquiring planting design data and standardizing it into standardized data; constructing quantitative constraints for optimizing parental distribution, and iteratively optimizing the model according to the criterion of maximizing parental distance to obtain preliminary parental spatial configuration results; fine-tuning the preset basic model using knowledge from the forestry and grassland domain to obtain a fine-tuned model and generate structured feature text; based on the structured feature text, using a dual-pointer matching algorithm to match the parental configuration map in the preliminary parental spatial configuration results with the hillside pit map in the standardized data to obtain matching results; marking abnormal areas according to the matching results and performing visual fine-tuning and constraint verification to obtain fine-tuning results; and integrating the fine-tuning results and matching results into the target planting result. This application optimizes parental distribution and accurately adapts to terrain, improving the accuracy of breeding experiments.

Description

Technical Field

[0001] This application relates to the field of forest tree genetics and breeding technology, and in particular to a method, device and equipment for intelligent planting of forest trees with progeny determination. Background Technology

[0002] Progeny testing forests are the core carriers of forest tree breeding research, integrating knowledge from multiple disciplines such as dendrology, forest ecology, and biostatistics. They are a crucial link connecting the selection of breeding materials with the propagation and promotion of superior varieties. The scientific nature of the parental planting layout directly determines the accuracy of genetic gain estimation and the accuracy of selecting superior families, which is essential for shortening the breeding cycle and improving breeding efficiency. It is a key foundation for supporting the upgrading of forestry breeding technology and meeting the needs of high-quality industrial development. Currently, the demand for large-scale forestry breeding continues to rise, but progeny testing forests are mostly built in mountainous environments with complex terrain, featuring natural obstacles such as undulating slopes and rocky gullies, further exacerbating the difficulty of planting layout design. Therefore, in order to adapt to the trend of large-scale and precise breeding development, research on the planting of progeny testing forests is particularly important.

[0003] Currently, the existing method for planting parent trees for progeny determination in forests is based on manual operation. The parent trees to be determined are arranged by drawing lots, and then the planting map of the mountain area is drawn entirely by hand based on the arrangement of the parent trees. However, this method relies on manual experience to select parents, which makes it easy for the same parents to cluster together, resulting in a high adjacency rate of the same parents, which affects the accuracy of the evaluation of genetic traits. Furthermore, it has poor adaptability to complex mountainous terrain, low design efficiency, and the entire process of manual operation is prone to problems such as numbering errors and positional deviations, resulting in a high error rate. Summary of the Invention

[0004] The purpose of this application is to provide a method, device, and equipment for intelligent planting of forest trees with progeny determination, in order to solve the technical problems of clustering of the same parent trees, poor terrain adaptability, and high error rate in planting design.

[0005] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a method for intelligent planting of forests based on tree progeny determination, including: The planting design data of the progeny test forest of the forest tree were obtained and standardized to obtain standardized data; the planting design data included: parent information, block parameters and topographic data; Based on the standardized data, quantitative constraints for optimizing parental distribution are constructed. Based on these quantitative constraints, the optimization model is iteratively optimized according to the criterion of maximizing parental distance to obtain preliminary parental spatial configuration results. The optimization model is constructed using simulated annealing and Delaunay triangulation algorithms. The preset basic model is fine-tuned using knowledge from the forestry and grassland domain to obtain the fine-tuned model. The fine-tuned model is then used to process the standardized data to generate structured feature text. Based on the structured feature text, a two-pointer matching algorithm is used to match the parent configuration map in the preliminary parent spatial configuration result with the mountain and nest map in the standardized data to obtain the matching result; Based on the matching results, abnormal regions are marked, and the abnormal regions are visually fine-tuned and constraint-verified to obtain fine-tuning results. The fine-tuning results are then integrated with the matching results to generate target planting results.

[0006] Secondly, this application provides a smart planting device for forest progeny testing, the device comprising: The acquisition module is used to acquire planting design data of the progeny test forest of trees and perform standardization processing to obtain standardized data; the planting design data includes: parent information, block parameters and topographic data; The iterative optimization module is used to construct quantitative constraints for parental distribution optimization based on the standardized data, and based on the quantitative constraints, iteratively optimizes the optimization model according to the criterion of maximizing parental distance to obtain preliminary parental spatial configuration results; the optimization model is built using simulated annealing algorithm and Delaunay triangulation algorithm. The processing module is used to fine-tune the preset basic model with knowledge of forestry and grassland, to obtain the fine-tuned model, and to process the standardized data using the fine-tuned model to generate structured feature text. The matching module is used to match the parent configuration map in the preliminary parent spatial configuration result with the mountain and nest map in the standardized data based on the structured feature text using a two-pointer matching algorithm to obtain the matching result; The generation module is used to mark abnormal regions based on the matching results, perform visual fine-tuning and constraint verification on the abnormal regions, obtain fine-tuning results, and integrate the fine-tuning results with the matching results to generate target planting results.

[0007] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the intelligent planting method for determining the progeny of trees as described above.

[0008] According to the specific embodiments provided in this application, the following technical effects are disclosed: This application provides a method, device, and equipment for intelligent planting of trees based on progeny generation. Compared with existing technologies, this solution acquires and standardizes planting design data, providing more comprehensive data guidance for subsequent planting treatments. It eliminates the reliance on experience in traditional manual operations, avoiding the numbering errors and positional deviations caused by manual lot drawing and hand-drawn maps. Based on standardized data, quantitative constraints are constructed, and an optimization model integrating simulated annealing and Delaunay triangulation algorithms is iteratively optimized according to the parental distance maximization criterion to obtain preliminary configuration results. This achieves a scientifically dispersed layout of parents at the algorithmic level, effectively reducing the clustering of identical parents and improving the accuracy of genetic trait evaluation. Furthermore, the basic model is fine-tuned using forestry and grassland knowledge, allowing the fine-tuned model to accurately analyze terrain and pit features, laying a precise data foundation for subsequent matching. Finally, structured feature text is generated based on the fine-tuned model to solve... This approach addresses the problem of poor adaptability of traditional methods to complex mountainous terrain. A dual-pointer matching algorithm is then used to accurately match the parent plant configuration map with the mountain pit map, ensuring the initial configuration results closely align with the actual terrain and improving the terrain adaptability and efficiency of the design. Finally, abnormal areas in the matching results are marked, and visual fine-tuning and constraint verification are performed. The fine-tuning and matching results are integrated to generate the target planting results. This approach significantly improves planting design efficiency through automation while ensuring compliance and operability through manual fine-tuning. Overall, it solves many problems associated with traditional manual planting methods, such as unreasonable parent plant distribution, poor terrain adaptability, low design efficiency, and high error rates. It achieves intelligent, standardized, and precise parent plantation for forest progeny testing, significantly improving the scientific rigor of planting schemes. Compared to traditional manual methods, this application improves design efficiency by 8 times, reduces the adjacent rate of the same parent plants from 28.7% to 0%, and reduces the manual error rate from 15.3% to 1.2%. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a structural schematic diagram of the application environment of a method for intelligent planting of forest progeny testing in an embodiment of this application; Figure 2 A schematic flowchart of a method for intelligent planting of forests based on progeny determination provided in an embodiment of this application; Figure 3 A flowchart illustrating a method for fine-tuning a preset basic model using knowledge from the forestry and grassland domain, as provided in an embodiment of this application, to obtain a fine-tuned model. Figure 4This is a flowchart illustrating how a matching result is obtained by matching according to a pre-matching list and preset matching rules, as provided in an embodiment of this application. Figure 5 This application provides a schematic diagram of the functional modules of an intelligent planting device for determining the progeny of trees in an embodiment of the present application. Figure 6 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0011] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0012] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0013] Existing techniques for designing the planting of parent trees for progeny testing in forests are mainly divided into two categories: traditional manual techniques and preliminary AI application techniques. One traditional technique relies on manual operation, arranging the proposed parent tree numbers by drawing lots, and then manually drawing a planting map of the mountain area based on the parent tree arrangement results. However, this approach has the following problems: First, the clustering of identical parents leads to larger experimental errors. Traditional designs rely on manual experience to select parents without scientific optimization strategies. The adjacent rate of identical parents reaches 28%-35%, exceeding the pollen dispersal interference range of most tree species. This results in pollen cross-interference in offspring traits, directly reducing the objectivity of genetic trait evaluation, the accuracy of genetic gain estimation, and the reliability of screening superior families.

[0014] Secondly, the design is poorly adapted to complex mountainous terrain, resulting in low efficiency. Most progeny testing forests are established in mountainous areas, facing obstacles such as undulating slopes and rocky gullies. Traditional methods require repeated on-site surveys to adjust the matching relationship between the design drawings and the actual planting sites. According to data from the Guangxi Gaofeng Forest Farm, the average time for planting design on 1 hectare of mountainous land is 3 working days. This inefficiency cannot meet the time-sensitive requirements of large-scale breeding.

[0015] Third, the high error rate of manual operation leads to a surge in rectification costs. The entire process of parent stock configuration, numbering and arrangement, and image conversion is done manually, which easily results in problems such as incorrect numbering and positional deviations. The error rate generally exceeds 15%, leading to an increase of more than 30% in subsequent construction and rectification costs, significantly increasing the economic cost of breeding.

[0016] Another approach is to adopt preliminary AI application solutions, such as the automated tree placement solution based on Arcpy. However, this solution can only achieve simple arrangement of planting units and does not incorporate parental distance constraints and terrain adaptation rules, resulting in the same parental adjacency rate still exceeding 5%, which cannot meet the requirements of breeding experiments. For example, the enhanced monarch genetic algorithm is not adapted to the parental layout scenario, and ChatGPT, Lin Longda model and other models have not been implemented in this scenario. Moreover, existing technologies either only focus on parental genetic distance analysis and lack spatial optimization solutions, or are like Excel breeding tools with limited functions and no ability to adapt to mountainous terrain, which cannot meet the needs of complex scenarios.

[0017] In summary, current technologies still suffer from core defects such as unreasonable parent distribution, poor terrain adaptability, low efficiency, high error, and insufficient AI technology adaptation. There is an urgent need to develop technical solutions that take into account parent distribution optimization, adaptation to complex terrain, and precise and efficient design, in order to break through traditional bottlenecks and meet the needs of modern large-scale forestry breeding.

[0018] To address the aforementioned shortcomings, this application provides an intelligent planting method for forestry based on progeny generation determination. Compared to existing technologies, this method acquires and standardizes planting design data, providing more comprehensive data guidance for subsequent planting processes. It eliminates the reliance on experience inherent in traditional manual operations, avoiding issues like incorrect numbering and positional deviations caused by manual lotteries and hand-drawn maps. Based on standardized data, quantitative constraints are constructed, and an optimization model integrating simulated annealing and triangulation algorithms is iteratively optimized according to the parental distance maximization criterion to obtain preliminary configuration results. This algorithmically achieves a scientifically dispersed layout of parents, effectively reducing the clustering of identical parents and improving the accuracy of genetic trait evaluation. Furthermore, the basic model is fine-tuned using forestry and grassland knowledge, allowing the fine-tuned model to accurately analyze terrain and pit features, laying a precise data foundation for subsequent matching. Based on the fine-tuned model... This method generates structured feature text to address the poor adaptability of traditional methods to complex mountainous terrain. A dual-pointer matching algorithm then accurately matches the parent plant configuration map with the mountain pit map, ensuring the initial configuration results align with the actual terrain and improving the terrain adaptability and efficiency of the design. Finally, abnormal areas in the matching results are marked, and visual fine-tuning and constraint verification are performed. The fine-tuning and matching results are integrated to generate the target planting results. This approach significantly improves planting design efficiency through automation while ensuring compliance and operability through manual fine-tuning. Overall, it solves many problems associated with traditional manual planting methods, such as unreasonable parent plant distribution, poor terrain adaptability, low design efficiency, and high error rates. This enables intelligent, standardized, and precise planting of parent plants for progeny testing in forests, significantly improving the scientific rigor of planting plans.

[0019] This application provides an embodiment of a method for intelligent planting of forests based on progeny determination, which can be applied to, for example... Figure 1The application environment of the intelligent planting method for progeny-tested forests shown is illustrated. This application environment includes a terminal 102, a server 104, and a data storage system. The terminal 102 communicates with the server 104 via a network. The data storage system can store planting design data. The data storage system can be set up independently, integrated into the server 104, or located in the cloud or on another server. The terminal 102 can send the acquired planting design data to the server 104. After receiving the planting design data, the server 104 performs intelligent planting processing to generate the target planting result. Furthermore, in some embodiments, the intelligent planting method for progeny-tested forests can also be implemented independently by the server 104 or the terminal 102. For example, the terminal 102 can directly generate the target planting result by performing intelligent planting processing on the planting design data.

[0020] The terminal 102 can be, but is not limited to, various desktop computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, and smart in-vehicle devices. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted devices. The server 104 can be implemented using a standalone server or a server cluster composed of multiple servers, or it can be a cloud server.

[0021] In one exemplary embodiment, such as Figure 2 As shown, a method for intelligent planting of trees for progeny determination is provided. This method is executed by computer equipment, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method is applied to... Figure 1 Taking the server in the example, the explanation includes the following steps S201 to S205. Wherein: Step S201: Obtain the planting design data of the progeny test forest and perform standardization processing to obtain standardized data; the planting design data includes: parent information, block parameters and topographic data.

[0022] It should be noted that the aforementioned parental information is the core parental object for progeny testing and planting in forests, and is the foundational object for optimizing parental configuration. Parental information may include parental numbers, genetic priorities, etc., and is stored in JSON format. Among these, genetic priority determines the allocation order of parents in high-quality planting units, directly affecting the efficiency of the transmission and screening of superior genetic traits in breeding experiments, and is a crucial basis for the column pointer allocation of planting holes in the subsequent two-pointer algorithm.

[0023] The aforementioned block parameters are the core parameters defining the spatial layout framework for progeny testing of forest trees. They are stored in JSON format, and key information includes: the number of blocks, the size of the planting unit matrix (rows × columns), and the side length of the planting unit. A block is the basic replication unit of a breeding experiment, and a planting unit is the planting carrier for a single parent, such as a square with a side length of 2 meters. By clearly defining the quantitative specifications of blocks and planting units, a unified spatial benchmark is provided for the spatial location calibration of parents and the calculation of distance constraints.

[0024] The aforementioned topographic data is the basic data reflecting the topographic characteristics of the forest site and is the core basis for topographic adaptation optimization. It mainly includes slope maps and obstacle data. The slope map can be in GeoTIFF format with a pixel resolution of 1 meter × 1 meter, recording the slope value at each location of the site, used to classify slope levels and match the corresponding spacing magnification factor; the obstacle data can be labeled in JSON format, recording the boundary coordinates and types of obstacles such as rocks and gullies, used to avoid unsuitable planting areas and ensure the topographic adaptability of the planting plan.

[0025] Optionally, the above-mentioned planting design data can be directly imported from external devices, obtained from blockchain or databases, or obtained in real time after sending data acquisition requests to different data sources. This embodiment does not limit the method of obtaining planting design data.

[0026] After obtaining the planting design data, it is standardized through three dimensions: coordinate unification, numerical unification, and semantic unification. This transforms planting design data from different sources into structured, unambiguous, and computable standardized data. Coordinate unification includes coordinate system standardization, coordinate accuracy standardization, and projection transformation standardization. For coordinate system standardization, a relative coordinate system is uniformly adopted, with the lower left corner of the forest plot determined by the tree generation measurement as the origin (0,0), the horizontal axis as the X-axis, and the vertical axis as the Y-axis. The positions of all planting units, obstacles, and pits are calibrated based on this relative coordinate system, avoiding positional offset errors caused by different coordinate system conversions. For coordinate accuracy standardization, the coordinate numerical accuracy is specified as 0.1 meters. Whether it's the coordinates of obstacle boundaries, the center coordinates of planting units, or the relative coordinates of pits, all are entered and stored according to this accuracy specification, ensuring the accuracy of spatial location calculations (such as Euclidean distance and adjacency determination). Standardizing projection transformation can be achieved by performing a unified projection transformation on terrain data from different sources, ensuring that all spatial data are aligned under the same projection rules, eliminating positional deviations caused by projection differences, and guaranteeing the collaborative accuracy of "row pointers traversing along contour lines and column pointers dynamically allocating" in the subsequent dual-pointer matching algorithm.

[0027] The process of standardizing numerical values ​​includes standardizing units, precision, and numerical range. For unit standardization, all quantification parameters adopt industry-standard units. For precision standardization, the precision requirements for various numerical values ​​are clearly defined. For example, the precision of slope values ​​is 0.1°, the pixel resolution of slope maps is standardized to 1 meter × 1 meter, and the precision of parameters such as planting unit side length and spacing magnification factor is 0.1. By fixing the number of decimal places and standardizing quantification, the repeatability and accuracy of objective function calculations and slope grade classification are ensured. For numerical range standardization, reasonable ranges are defined for key numerical parameters and verified. For example, the slope value range must match the actual terrain, and the planting unit matrix size must be a positive integer to avoid abnormal numerical input affecting algorithm convergence or constraint determination.

[0028] The process of semantic unification includes data format unification, semantic definition unification, and structured organization unification. For data format unification, various data storage formats can be standardized according to preset standard formats to ensure direct machine parsing. For example, parental information, block parameters, and obstacle data can all use JSON format with fixed field names and hierarchical structures to avoid field redundancy or missing data; slope maps can all use GeoTIFF format with a fixed mapping relationship between pixel values ​​and slope values ​​to ensure the model can directly extract terrain features; output layout coordinates can all be in CSV format, and planting layout maps can all be in PNG format to ensure consistency in data transmission and application. For semantic definition unification, the meanings of key terms and parameters can be clearly defined to build an unambiguous semantic system and avoid misunderstandings of unit specifications among different users. For structured organization unification, all data can be organized according to a "classification hierarchy + core fields" structure to form a standardized dataset.

[0029] Step S202: Construct quantitative constraints for parental distribution optimization based on standardized data, and based on the quantitative constraints, perform iterative optimization through the optimization model according to the criterion of maximizing parental distance to obtain preliminary parental spatial configuration results; the optimization model is built using simulated annealing algorithm and Delaunay triangulation algorithm.

[0030] The aforementioned distance constraints include coordinate system standardization, inter-block spacing constraints, and intra-block spacing constraints. Coordinate system standardization includes: using a relative coordinate system to calibrate positions, with the lower left corner of the planting plot as the origin, the horizontal axis as the X-axis, and the vertical axis as the Y-axis; each planting unit is a square with sides of two meters, and the center coordinates of the planting unit are (x×2, y×2), where x and y are the row and column numbers of the planting unit; inter-block spacing constraints include: the minimum distance between the same parent in adjacent blocks is no less than three planting units, with a corresponding Euclidean distance ≥6; intra-block spacing constraints include: the horizontal and vertical intervals of the same parent within the same block are both no less than two planting units, and diagonal adjacency of the same parent is allowed.

[0031] The terrain adaptation rules include: slope grading standards and dynamic coordinate adjustment rules; the slope grading standards are divided into three slope grades according to the slope range, with each slope grade corresponding to a spacing amplification factor; when the slope range is the first slope grade, the corresponding spacing amplification factor is 1.0; when the slope range is the second slope grade, the corresponding spacing amplification factor is 1.2; when the slope range is the third slope grade, the corresponding spacing amplification factor is 1.5; the first slope grade is a slope less than or equal to 25°, the second slope grade is a slope greater than 25° and less than 35°, and the third slope grade is a slope greater than or equal to 35°.

[0032] The rules for dynamic coordinate adjustment include: increasing the spacing by adjusting the center coordinates of the planting units, keeping the row and column numbers of the planting units unchanged, and multiplying the value of the center coordinates of the planting units in the vertical direction along the slope by the corresponding magnification factor.

[0033] It should be noted that a unified relative coordinate system is used to define all spatial locations within the forest. The lower left corner of the measured forest plot is taken as the origin (0, 0), with the horizontal axis set as the X-axis and the vertical axis as the Y-axis. Each planting unit is defined as a square with a side length of 2 meters, and its center coordinates are (x×2, y×2), where x and y are the row and column numbers of the planting unit, respectively. This standardized calibration provides a unified and accurate spatial reference for subsequent spacing calculations. The aforementioned cross-block spacing constraints are for the distribution of the same parent in adjacent blocks, specifying a minimum distance limit. The minimum distance between the same parent in adjacent blocks is no less than three planting units, corresponding to an Euclidean distance of ≥6 meters, to avoid genetic interference caused by the same parent being too close in different blocks. For the distribution of the same parent within the same block, the horizontal and vertical intervals are limited to no less than 2 planting units, i.e., the difference in row and column numbers is ≥2; at the same time, diagonal adjacency of the same parent is allowed, ensuring that genetic interference is controllable while taking into account the flexibility of the planting layout.

[0034] The aforementioned terrain adaptation rules employ a quantitative strategy of "slope grading + dynamic coordinate adjustment" to achieve precise adaptation to complex mountainous terrain. Measured slope information is obtained, and the terrain is divided into three levels based on the measured slope range, with corresponding differential spacing amplification coefficients. When the slope is ≤25°, it is characterized as a gentle slope, with a spacing amplification coefficient of 1.0; when the slope is between 25° and 35°, it is characterized as a steep slope, with a spacing amplification coefficient of 1.2; and when the slope is >35°, it is characterized as an extremely steep slope, with a spacing amplification coefficient of 1.5. This grading coefficient enables quantitative control of terrain adaptation. During dynamic coordinate adjustment, spacing amplification is achieved by adjusting the center coordinates of the planting unit. The core principle is to keep the row and column numbers of the planting unit unchanged, only multiplying the value of the center coordinates of the planting unit along the vertical direction of the slope by the corresponding spacing amplification coefficient.

[0035] Delaunay triangulation provides the foundation for "precise verification of local adjacency": the empty circle characteristic of Delaunay triangulation determines that there are no other discrete points (i.e., parental planting units) within the circumcircle of any triangle. Based on this characteristic, the algorithm can quickly construct the spatial topological adjacency relationship of all planting units, directly locking the adjacent units of each parent, avoiding the problems of "missing adjacent parents" and "misjudging spacing" in traditional grid search, and eliminating close-range clustering of parents at the local level. Simulated annealing algorithm provides the capability of "global optimal search": the simulated annealing algorithm breaks through the local optimum trap through the mechanism of "allowing poor solutions at initial high temperature → gradually cooling to converge to the optimal solution". In the parent configuration iteration, even if a certain round of adjustment leads to a shortening of the spacing between individual parents (poor solution), the algorithm may still accept the adjustment based on the current temperature, avoiding the dilemma of "optimal local spacing but uneven global distribution", and ensuring the uniform dispersion of parents throughout the entire measured forest area.

[0036] The two work together to achieve a dual guarantee of "local compliance + global optimization": Delaunay triangulation is responsible for "verifying whether the distance between adjacent parents meets the constraint of ≥6 meters" after each iteration, ensuring local configuration compliance; simulated annealing algorithm is responsible for driving the overall layout to optimize towards "maximizing the average Euclidean distance of the same parent". The two form a closed loop of "verification-optimization-re-verification", which ultimately achieves global optimization of the spatial distribution of parents.

[0037] In existing technologies, Delaunay triangulation is mostly used alone for terrain modeling and spatial adjacency analysis, while simulated annealing algorithms are mostly used for single objective function optimization (such as path planning). In this application, the two are combined and integrated with the dual rules of "distance constraint + terrain adaptation" in forest breeding. The local adjacency determination capability of triangulation is deeply coupled with the global optimization capability of simulated annealing. At the same time, the coordinate dynamic correction rules corresponding to slope grading are embedded, which solves the technical pain point that traditional algorithms can only optimize the spacing and cannot adapt to complex mountainous terrain.

[0038] In this embodiment, by setting distance constraints, the spatial spacing requirements of planting units of the same parent line can be clearly defined. This enforces a scientifically dispersed layout of parents at the algorithmic level, completely solving the problem of clustered distribution of the same parents in traditional artificial planting. It fundamentally avoids the bias in the evaluation of genetic traits caused by excessively high parental adjacency rates, significantly improving the overall accuracy of forest progeny testing forest breeding experiments. At the same time, this constraint conforms to the industry norms and technical standards of forest genetic breeding experiments, ensuring that the spatial configuration of parents meets the professional requirements of the experimental design, effectively reducing experimental system errors caused by non-standard layout. In addition, distance constraints promote a uniform spatial distribution of parents within the testing forest area, avoiding excessive concentration or vacancy of parents in local areas, optimizing the uniformity of parental spatial distribution, and ensuring a reasonable allocation of planting units and experimental resources, effectively guaranteeing the resource utilization efficiency of breeding experiments. Furthermore, the terrain adaptation rules standardize and correct the coordinates of planting units based on the actual terrain characteristics of complex mountainous areas. This ensures that the parent plant space configuration scheme closely matches the actual terrain features, slopes, and pit distribution, avoiding the placement of planting units in unsuitable terrain areas. This significantly improves the adaptability of the scheme to complex mountainous terrain, while reducing mountain disturbance and ecological damage caused by terrain incompatibility, as well as growth losses caused by unsuitable terrain environments after seedling planting. The planting coordinates, corrected based on actual terrain, ensure that the scheme perfectly matches the on-site construction conditions, eliminating the need for extensive on-site coordinate adjustments during the construction phase. This guarantees the actual constructability of the planting scheme and significantly reduces the difficulty and cost of on-site adjustments. In addition, the terrain adaptation rules can flexibly adjust the configuration logic of planting units according to different terrain features such as gentle slopes, steep slopes, and extremely steep slopes, adapting to the ecological needs of tree planting in various terrains. This breaks the limitations of traditional methods on terrain and is applicable to different mountain afforestation scenarios.

[0039] In one embodiment, a specific implementation method is also provided, which uses an optimization model to iteratively optimize based on quantified constraints according to the criterion of maximizing parental distance to obtain preliminary parental spatial configuration results. The method includes: Based on terrain adaptation rules, the coordinates of the centers of all planting units are dynamically corrected to obtain a modified standardized coordinate system. Following this standardized coordinate system, the centers of all planting units in each block are used as discrete point sets to construct a Delaunay triangulation for all planting units and determine the adjacency relationships within the triangulation. An optimization model integrating simulated annealing and Delaunay triangulation is constructed, and the model parameters are initialized. These parameters include initial temperature, cooling rate, and iteration step size. Based on standardized data, the average Euclidean distance between all planting units of the same parent is calculated to construct the objective function. Parent configuration candidate schemes are generated within the standardized coordinate system. Based on the model parameters and the adjacency relationships of the Delaunay triangulation, the parent configuration candidate schemes are iteratively verified using distance constraints as the validation standard, terrain adaptation rules as the spatial boundary, and maximizing the objective function as the optimization direction, until the iteration termination condition is met. The iteration termination condition includes the change in the objective function being less than or equal to a preset threshold or the number of iterations reaching a preset number. When the iteration termination condition is met, a preliminary parent spatial configuration result conforming to the terrain adaptation rules and distance constraints is obtained.

[0040] The preset threshold and preset number of times can be customized according to the actual scenario. For example, the preset threshold can be 0.01 meters and the preset number of times can be three.

[0041] Specifically, before starting the simulated annealing algorithm, distance constraints can be entered into the algorithm parameter library. These constraints include a Euclidean distance of ≥6 meters between adjacent blocks of the same parent and a difference of ≥2 between row and column numbers within the same block. These serve as screening conditions for the initial parent configuration. During the initialization phase, only parent layout schemes that satisfy the distance constraints are generated, eliminating invalid solutions that do not meet the constraints, reducing the algorithm's search range, and improving iteration efficiency. Based on the topographic data in the standardized data, the center coordinates of all planting units are dynamically corrected according to the "slope grading - spacing coefficient" quantization rule. In this process, for conventional areas with a slope of ≤25°, the coordinates of the planting units remain unchanged with a spacing magnification factor of 1.0; for medium-steep slope areas of 25°-35°, the coordinates are multiplied by a spacing magnification factor of 1.2 along the vertical slope direction; and for extremely steep slope areas of >35°, the coordinates are multiplied by a spacing magnification factor of 1.5. This dynamic correction ensures that the coordinates of the planting units are highly adapted to the actual terrain and slope characteristics of the mountain area. The standardized coordinate system formed after the correction becomes the unified spatial carrier for all subsequent algorithm operations, ensuring that the terrain adaptation rules are integrated into the parent configuration design from the source, and avoiding the problem of poor terrain adaptability of traditional solutions from the algorithm level.

[0042] In the revised standardized coordinate system, the centers of all planting units in all blocks are used as discrete point sets. A Delaunay triangulation algorithm is used to construct a Delaunay triangulation network covering the entire measured forest area, thereby accurately identifying the spatial topological adjacency relationships between all planting units. By transforming originally isolated planting units into triangulation network nodes with clear adjacency relationships, the subsequent distance constraint verification is upgraded from the traditional coarse-grained grid judgment to the precise judgment of triangulation network adjacency nodes. This completely avoids the problem of missed adjacent planting units that may occur in grid search, providing a precise spatial relationship basis for the strict implementation of distance constraints. When updating parental positions, based on the empty circle characteristic of the triangulation network, acute-angle adjacency between newly adjusted parental planting units and the same parental unit in adjacent blocks is avoided.

[0043] Furthermore, based on a standardized coordinate system adapted to terrain and the adjacency relationship of planting units, an optimization model integrating simulated annealing and Delaunay triangulation techniques was constructed. Simultaneously, the core parameters of the model were initialized, including the initial temperature and cooling rate of the simulated annealing algorithm, as well as the iteration step size. For example, the initial temperature was set to 1.0℃, the cooling rate to 0.99, and the iteration step size to 50. This optimization model is the core algorithmic carrier for implementing the quantified constraint system. The global optimization capability of the simulated annealing algorithm can prevent parent configurations from getting trapped in local optima, while the precise adjacency determination capability of Delaunay triangulation ensures the accuracy of constraint verification. The reasonable setting of initialization parameters provides a foundation for efficient and stable iteration of the algorithm, avoiding problems such as slow convergence and distorted optimization results caused by improper parameters.

[0044] Based on standardized data, the total number of planting units of the same parent and the center coordinates of all planting units of the same parent are determined. The Euclidean distances between these units are calculated based on the total number of planting units and their center coordinates, and the average value is obtained. An objective function maximizing parental distance is then constructed, which can be expressed by the following formula: ; Where n is the total number of planting units of the same parent (number of units), , ) represents the center coordinates of the i-th planting unit. , Let be the center coordinates of the j-th planting unit, and f be the average Euclidean distance (in meters) between all planting units of the same parent. It can be understood that this objective function, by maximizing the average Euclidean distance between all planting units of the same parent, mathematically ensures the uniform distribution of the parent throughout the entire forest area, directly addressing the core problem of parent aggregation in traditional artificial planting.

[0045] The termination condition of the above simulated annealing algorithm can be customized according to the actual scenario, for example: the change in the objective function is ≤0.01m for 30 consecutive rounds.

[0046] Within the standardized coordinate system corrected by terrain adaptation rules, multiple sets of parent plant configuration candidate schemes are randomly generated. These parent plant configuration candidate schemes refer to the initial correspondence between the parent plant and the planting unit. All parent plant configuration candidate schemes use standardized coordinates as spatial boundaries and do not exceed the planting range after terrain adaptation, ensuring that the initially generated candidate schemes all meet the basic requirements of terrain adaptation rules. This avoids the algorithm blindly searching in invalid terrain space, effectively reducing the search range of the algorithm and improving the efficiency of subsequent iterative optimization. Then, according to the initialized model parameters and based on the adjacency relationship of the planting unit defined by the Delaunay triangulation, multiple rounds of iterative verification and optimization are carried out on the generated parent plant configuration candidate schemes. The entire iterative process always takes the quantitative constraints as the core criterion and the maximization of the objective function as the optimization direction. Specifically, the distance constraint is used as a rigid verification standard. After each iteration generates a new candidate scheme, the spatial distance of the same parent plant is immediately verified through the adjacent nodes of the Delaunay triangulation. If the distance of the same parent plant in adjacent blocks is <6 meters or the difference in row and column numbers within the same block is <2, it is directly judged as an invalid solution and a new candidate scheme is regenerated. The terrain adaptation rules are used as the basis for non-aggression. The spatial boundary is broken, and all iterative operations are performed within the corrected standardized coordinate system to avoid configuring parent plants outside the coordinate system. Simultaneously, it verifies whether planting units in areas with slopes >25° have completed the spacing enlargement according to the rules, ensuring that terrain adaptation requirements are met throughout the process. The core optimization direction is maximizing the objective function. If a new scheme satisfies distance constraints and terrain adaptation rules, and the objective function value (average Euclidean distance of the same parent plant) is better than the current optimal solution, then the scheme is directly accepted and the optimal solution is updated. If the objective function value differs from the current optimal solution but does not exceed the acceptance probability of the simulated annealing algorithm, then the scheme is temporarily retained to avoid the algorithm getting trapped in local optima and to ensure the global optimality of parent plant configuration. The entire iterative process continues until a preset iteration termination condition is met. This termination condition can be either the change in the objective function ≤ a preset threshold, indicating that the improvement in parent plant distance tends to stabilize, or the number of algorithm iterations reaching a preset number. For example, if the change in the objective function is ≤0.01 meters for 30 consecutive iterations or the number of iterations reaches 500, the iteration process can be stopped when either condition is met. The output optimization result at this point is the preliminary parent plant spatial configuration result. The preliminary parent space configuration result is the optimal solution after multiple rounds of algorithm optimization and multi-dimensional verification, and it meets the terrain adaptation rules and distance constraints.

[0047] In this embodiment, by constructing an objective function, the scientific and dispersed layout of parent plants can be achieved at the algorithmic level, completely solving the problems of clustered distribution of the same parent plants and excessively high adjacency rates in traditional artificial planting, thus providing algorithmic support for improving the accuracy of genetic trait evaluation. Furthermore, by iteratively maximizing the objective function, not only is the spatial distance of the same parent plant maximized, effectively reducing the adjacency rate of the same parent plant and meeting the scientific standards of forest tree genetic breeding experiments, but the parent plant configuration is also highly adapted to the actual slope and terrain of complex mountainous areas, ensuring the terrain adaptability and construction feasibility of the planting scheme.

[0048] Step S203: Fine-tune the preset basic model with knowledge of forestry and grassland to obtain the fine-tuned model, and use the fine-tuned model to process the standardized data to generate structured feature text.

[0049] It is understandable that the aforementioned pre-defined base model possesses natural language understanding and text generation capabilities, and is adapted to the feature parsing requirements of forestry scenarios, making it a suitable candidate for GPT-3.5 Turbo. During the fine-tuning of the pre-defined base model with forestry and grassland domain knowledge, a low-rank adaptation (LoRA) strategy can be employed. This involves training only the adapter matrix of the base model, freezing the model's backbone parameters, and avoiding the resource consumption and overfitting issues caused by full training while achieving efficient transfer of domain knowledge, thus obtaining the fine-tuned model. The aforementioned structured feature text includes basic information on forest progeny determination, a list of pit features, and obstacle information. In one embodiment, a specific implementation method is also provided for fine-tuning the preset basic model using forestry and grassland domain knowledge to obtain the fine-tuned model. Please refer to [link to relevant documentation]. Figure 3 As shown, the method includes: Step S301: Construct a fine-tuning dataset and divide it into a training set and a test set. The fine-tuning dataset includes a visual text pairing dataset and a rule knowledge base dataset. The visual text pairing dataset includes historical mountain cave maps and corresponding structured annotation results. The rule knowledge base dataset includes the mapping results between rule keywords and parameter values.

[0050] Step S302: Initialize the training parameters of the base model; the training parameters include the learning rate, training epochs, and batch size; the model parameters include backbone parameters and other parameters.

[0051] Step S303: Fix the backbone parameters of the base model, input the training set into the base model for feature parsing, and obtain the output result.

[0052] Step S304: Construct the cross-entropy loss function based on the output results and structured annotation results.

[0053] Step S305: Minimize the cross-entropy loss function, and iteratively train the other parameters of the base model based on the training parameters to obtain the trained model.

[0054] Step S306: Input the test set into the trained model for testing to obtain the fine-tuned model; the fine-tuned model has the ability to analyze features in the forestry and grassland domain.

[0055] In this embodiment, a fine-tuning dataset is first constructed. This dataset includes a visual-text pairing dataset and a rule knowledge base dataset. The visual-text pairing dataset includes historical mountain pit maps and corresponding structured annotation results. Specifically, the visual-text pairing dataset can use historical mountain pit maps of forestry planting scenarios as core visual materials, and perform pixel-level structured annotation on each historical mountain pit map to obtain the corresponding structured annotation results. These annotation results can include 12 core features of the forestry and grassland domain, such as pit number, coordinates, slope, and obstacles. The rule knowledge base dataset can be constructed by first obtaining forestry topography specifications and then constructing a mapping result of "rule keywords - parameter values" based on these specifications. This fine-tuning dataset, for example, includes: ① 5000 historical mountain pit maps, each with pixel-level structured annotation (annotating key information such as pit number, location, and slope); ② 200 forestry topography specifications, from which core rules are extracted and a mapping result of "rule keywords - parameter values" is constructed and stored in a mapping library, providing standardized domain knowledge for the model. The mapping results between the above keywords and parameter values ​​can be illustrated in Table 1 below: Table 1. Mapping results between keywords and parameter values

[0056] After constructing the fine-tuning dataset, it is divided into a training set and a test set according to a preset ratio. The training set is used for iterative learning of domain knowledge and parameter optimization of the model, while the test set is used for effect verification and performance evaluation after model fine-tuning. This ensures an objective judgment of the model's actual analytical capabilities and provides scientific, complete, and professional data support for subsequent targeted fine-tuning.

[0057] For the preset base model, initialization and configuration of training parameters are performed. These model parameters include backbone parameters and other parameters. The backbone parameters are the core, underlying parameters of the base model, determining its basic capabilities such as general feature extraction and text generation. The other parameters are the model's adaptability parameters, which can achieve targeted adaptation to domain knowledge through iterative training. The reasonable initialization of quantitative parameters such as learning rate, training epochs, and batch size is crucial to ensuring model training efficiency and avoiding overfitting or underfitting. Specifically, the learning rate can be 2e-4, the training epochs 10, and the batch size 8.

[0058] The core parameters of the base model are completely frozen. The training set is then input into the base model to initiate the feature parsing process. This allows the model to perform preliminary feature extraction and parsing of the forestry and grassland data in the training set based on its existing general capabilities, generating corresponding output results. During this process, the model only adaptively learns from the forestry and grassland data using other parameters that are not yet fixed. The core objective is for the model to initially learn the association logic between visual features and textual descriptions specific to the forestry and grassland domain, as well as the rules governing forestry planting, achieving targeted transfer of domain knowledge rather than a reconstruction of the model's basic capabilities. The output results are used as the model's predicted values, and the structured annotations inherent in the visual-text pairing data in the training set are used as the ground truth labels. A cross-entropy loss function is constructed based on the deviation between the two. This cross-entropy loss function accurately quantifies the degree of deviation between the model's parsing results and the standard structured annotation results. A larger deviation value results in a higher loss function value, indicating a weaker domain feature parsing ability; a smaller deviation value results in a lower loss function value, indicating that the model's parsing results are closer to the standard annotation requirements.

[0059] By minimizing the cross-entropy loss function and using a LoRA fine-tuning strategy, the core parameters of the model are frozen, and the adapter matrix is ​​trained (rank set to 8). Following the initial learning rate, training epochs, and batch size, multiple epochs of iterative training and optimization are performed on the other parameters of the base model. This allows the model's parsing results for forestry and grassland data to continuously approach the standard structured annotation results, while gradually learning forestry planting rules from the rule knowledge base dataset. When the training reaches a preset number of epochs and the loss function value stabilizes and no longer decreases significantly, the iterative training terminates, resulting in the trained model. The test set, consisting of entirely new forestry and grassland data not encountered by the model, is input into the trained model to verify the overall effectiveness of feature parsing in the forestry and grassland domain. This test set objectively assesses the model's actual domain parsing and generalization capabilities. If the model's output on the test set meets the preset performance indicators, such as feature recognition accuracy reaching the threshold, the model is considered to have passed the fine-tuning test, and the model trained at this point is the final fine-tuned model. If the parsing results do not reach the threshold, the training parameters (such as learning rate and training epochs) need to be readjusted, and the model training and testing should be carried out again until the model performance meets the standards. This fine-tuned model, while retaining the basic capabilities of the general basic model such as text generation and feature parsing, adds a feature parsing capability specifically for the forestry and grassland field. It can accurately parse 12 core features in the mountain pit map, and at the same time, it can strictly follow the professional rules of forestry topography specifications to generate structured feature text that conforms to industry standards, providing accurate, standardized, and directly identifiable feature parsing results for the subsequent operation of the two-pointer matching algorithm.

[0060] This step, by freezing the core parameters of the basic model, avoids the loss of general basic capabilities caused by training with all parameters. It also significantly reduces computational costs and improves training efficiency. Furthermore, by constructing a cross-entropy loss function, it sets the core optimization objective for subsequent iterative optimization of other model parameters: continuously reducing the loss function value through iterative training until the model output highly matches the standard annotation results. This achieves the targeted transformation of the general basic model into a forestry and grassland-specific model, solving the problem of the general model lacking forestry and grassland expertise and being unable to accurately analyze the features of mountain pit maps. It provides crucial model support for the automated and intelligent advancement of intelligent planting methods for forest progeny determination. Simultaneously, this lightweight fine-tuning strategy significantly reduces the computational and time costs of model training and has good engineering feasibility.

[0061] Optionally, the process of generating structured feature text includes: obtaining a mountain pit map from standardized data, processing the mountain pit map through a fusion model to extract visual features; the visual features include spatial information, terrain features, and pit distribution information; the fusion model has cross-modal feature learning capabilities; inputting the visual features into the fine-tuned model to generate structured feature text; the structured feature text includes basic information of the forest progeny survey, a list of pit features, and obstacle information.

[0062] Among them, the aforementioned fusion model can be the Contrastive Language-Image Pre-training (CLIP) model, which is a cross-modal foundational model. Its core objective is to learn transferable visual models from natural language to achieve cross-modal association and feature alignment between images and text.

[0063] The process begins by obtaining cave images from standardized data. These images are then processed using a fusion model. First, the model's internal visual encoder performs multi-layer convolution and feature extraction on the pixel information of the cave images, generating low-level visual features containing core visual information from the forestry and grassland domain, such as cave location, slope, obstacle distribution, and block boundaries. Next, the model's cross-modal feature fusion module maps these low-level visual features into high-dimensional feature vectors aligned with the semantic dimension of the text. Simultaneously, forestry and grassland domain rules are used to optimize and filter these feature vectors, removing invalid visual interference. The final output is a finely tuned visual feature that can be recognized by the model and adapted to the requirements of structured text parsing. This visual feature can be represented by a feature vector. The visual feature is then input into the finely tuned model. Based on learned domain rules and feature recognition capabilities, the model parses 12 key features from the cave images, ultimately generating structured feature text in JSON format. The structured feature text contains three core modules: ① basic information (such as forest number, number of blocks, and total number of planting units); ② pit feature list (such as pit number, relative coordinates, slope, spacing magnification factor, availability status, etc.); ③ obstacle details (such as obstacle type, boundary coordinates, covered pit number, etc.), ensuring that the output data is structured, unambiguous, and can be directly used for subsequent algorithm matching.

[0064] The basic information comprises the core parameters of the test forest, reflecting the overall situation of the test forest in its progeny. It clarifies the boundary conditions and core objectives of the planting design and can include the following six categories of key information: basic identification information of the test forest, core parameters of the experimental design, global topographic features, basic standards for planting construction, data accuracy and source description, and experimental objectives and cycle. The basic identification information of the test forest can include: test forest number, name of the forest farm, geographical location (latitude and longitude range), and experimental tree species (e.g., "Masson pine"), ensuring the unique and traceable identity of the test forest. Core parameters of the experimental design include: number of blocks, matrix size of planting units in each block, total number of planting units, number and number of parent trees, directly related to subsequent parent tree configuration and pit allocation logic. Global topographic features include: overall slope range, proportion of main terrain types (e.g., "gentle slopes account for 70%, steep slopes account for 30%), and soil type (e.g., "red soil"), referencing forest area topographic classification standards to provide a macro-level basis for the application of terrain adaptation rules (e.g., spacing amplification factor). The basic standards for planting construction include: the side length of the planting unit (e.g., "2 meters", corresponding to a square unit), the size of the planting hole (e.g., "60cm×60cm×40cm"), and the standard for basal fertilizer application (e.g., "5-10kg of basal fertilizer per hole"), ensuring that the design scheme conforms to the construction technical specifications. Data accuracy and source specifications include: slope map accuracy (e.g., "GeoTIFF format, pixel resolution 1 meter × 1 meter, slope error ≤ 0.5°"), coordinate system (e.g., "relative coordinates, origin at the lower left corner of the test forest (0,0)"), and data acquisition method (e.g., "drone aerial photography + GPS field mapping"), ensuring the accuracy of subsequent feature matching and distance calculation. Experimental objectives and cycles include: the type of breeding experiment (e.g., "progeny genetic gain determination") and the observation period (e.g., "5 years for fast-growing tree species / 10 years for slow-growing tree species"), providing guidance for subsequent parent selection and growth monitoring.

[0065] The planting hole feature list contains refined parameters for each individual planting unit and is the core basis for dual-pointer matching. It includes nine categories of information strongly related to parent plant matching, terrain adaptation, and construction implementation, ensuring that each planting hole accurately corresponds to the optimal parent plant. This information includes: planting hole identification, spatial location parameters, terrain adaptation parameters, availability status, adjacency information, obstacle association information, soil and fertility information, parent plant adaptation suggestions, and construction notes. The planting hole identification includes: a planting hole number, used to uniquely identify each planting hole and avoid numbering errors during subsequent matching. Spatial location parameters include: relative coordinates and altitude, used to calculate the Euclidean distance and slope adaptation coefficient between parents. Terrain adaptation parameters include: the slope of the area where the planting hole is located (e.g., "22.5°", accuracy 0.1°) and the spacing amplification coefficient (assigned according to slope grading rules, e.g., "1.0" corresponds to a gentle slope ≤25°, "1.5" corresponds to an extremely steep slope >35°), directly related to the planting unit spacing adjustment logic. Availability status includes "Yes / No," with criteria including whether the planting hole is covered by obstacles (such as rocks or gullies) and whether it meets planting hole specifications (e.g., depth ≥ 40cm), avoiding assigning parent plants to invalid planting holes. Adjacency information includes adjacent planting hole numbers, identified based on Delaunay triangulation adjacency relationships, used to verify parent plant distance constraints. Obstacle association information includes the distance to the nearest obstacle and whether the plant is within the obstacle's influence range, referencing forest area obstacle identification standards to avoid planting holes being affected by obstacles. Soil and fertility information includes soil moisture and pre-allocated basal fertilizer application amount, providing a basis for subsequent planting decisions. Parent plant adaptation suggestions include recommended parent plant priorities (e.g., "High, adapt to P1 (high genetic priority)"), generated based on planting hole terrain quality and parent plant requirements, assisting in dual-pointer matching decisions. Construction notes include special construction requirements, such as "surface gravel needs to be cleared to ensure planting hole depth meets standards" and "small earthen embankments need to be built near gullies to prevent water accumulation," directly guiding on-site construction operations and reducing rectification rates.

[0066] Obstacle information refers to precise descriptions of disturbance factors in forest areas, providing support for pit avoidance and safe construction. Referring to forest obstacle classification standards (natural / artificial obstacles) and ultrasonic detection technical specifications, obstacle information must clearly define "location, type, and impact range" to avoid conflicts between parent plantations and obstacles. Specifically, it includes five categories of information: basic obstacle attributes, spatial range parameters, impact range and risk level, detection and verification information, and construction response measures. Among these, basic obstacle attributes include obstacle type (classified by source: natural obstacles such as "rocks," "gullies," and "fallen trees," artificial obstacles such as "abandoned building foundations" and "leftover equipment"), and core feature descriptions (e.g., "rocks: hard surface, no vegetation cover; gullies: 2-3 meters wide, 1-1.5 meters deep, with water accumulation during the rainy season"). Refer to forest obstacle identification technical documents. Spatial range parameters include boundary coordinates (e.g., rock boundary "[(10.2,20.5),(10.2,25.3),(15.6,25.3),(15.6,20.5)]", unit: meters, based on a relative coordinate system) and area (e.g., "16 square meters"), precisely delineating the obstacle's influence area to exclude pits within that range. The influence range and risk level include: directly affected pit numbers (e.g., "["C04","C05"]", indicating pits completely covered by the obstacle), indirectly affected areas, and risk level (e.g., "High risk: Planting is prohibited within 2 meters of the gully to prevent soil erosion and seedling lodging"), referencing ultrasonic obstacle risk assessment standards. Detection and verification information includes: detection method (e.g., "initial assessment by drone aerial photography + on-site ultrasonic detection", referencing methods for using ultrasonic technology for obstacle detection with tree planting machines), detection accuracy, and verification results, ensuring the obstacle information is accurate and reliable. Construction response measures include: avoidance suggestions and safety tips, which are directly related to subsequent assessment of pit availability and construction safety management.

[0067] In this embodiment, CLIP overcomes the technical bottleneck that GPT cannot directly process images: the GPT model natively only supports text input and cannot parse the visual features of mountain cave maps (such as cave location, slope texture, and obstacle shape). As a cross-modal model, CLIP's visual encoder (such as ViT-L / 14) can extract high-dimensional visual feature vectors from images, transforming the visual information of "cave-slope-obstacle" into feature data that GPT can recognize, thus building a bridge for the conversion between "image → feature → text". GPT Forestry and Grassland Fine-tuning Injects Professional Rule Analysis Capabilities: Through fine-tuning with two core datasets, the GPT model achieves a dual improvement in "feature recognition + rule verification": First, a visual-text pairing dataset (5000 mountain pit maps + structured annotations) allows the model to learn the mapping relationship between "visual features → forestry features" (e.g., "steep slope texture → slope 25°-35° → spacing coefficient 1.2"); second, a rule knowledge base dataset (200 "rule keyword-parameter value" mappings extracted from forestry standards) allows the model to understand professional constraints (e.g., "high-priority parents cannot be assigned to pits on extremely steep slopes").

[0068] Accurate analysis and compliant output are achieved through the synergistic effect of CLIP and GPT: CLIP ensures the integrity of visual feature extraction (without missing key information such as burrows and obstacles), while the fine-tuned GPT ensures the professionalism of feature analysis (without confusing slope grades or violating forestry regulations). At the same time, feature consistency is automatically checked when generating structured text (such as matching slope and spacing coefficients, and checking for conflicts between obstacle boundaries and burrow locations). Ultimately, the accuracy of terrain analysis is improved from 82.3% of existing AI methods (such as the automated tree placement scheme based on Arcpy) to 92.7%.

[0069] In existing technologies, CLIP is mostly used for general image retrieval and visual question answering, while GPT-type models are mostly used for general text generation. Neither of these technologies is optimized for the "complex mountainous terrain analysis" in the forestry and grassland field. The innovation of this solution lies in combining cross-modal feature transformation (CLIP) with domain knowledge fine-tuning (GPT). This not only solves the problem that "image features cannot be directly converted into forestry structured data," but also enables the model to have "forestry rule self-verification" capabilities, overcoming the limitation of existing technologies that "can only recognize basic features and cannot adapt to forestry professional scenarios." This combination of "cross-modal transformation + domain rule injection" is being applied for the first time in the forest progeny determination and planting design scenario, overcoming the limitations of existing technologies that cannot simultaneously achieve image analysis, professional rule adaptation, and accurate matching.

[0070] Step S204: Based on the structured feature text, a two-pointer matching algorithm is used to match the parent configuration map in the preliminary parent spatial configuration result with the mountain and nest map in the standardized data to obtain the matching result.

[0071] The aforementioned dual-pointer system includes row and column pointers. The dual-pointer matching algorithm comprises a contour line tracing algorithm and a hybrid allocation strategy. The contour interval in the contour line tracing algorithm can be set to 1 meter. The contour line tracing algorithm uses Digital Elevation Model (DEM) terrain data as its core basis, employing a spatial path planning algorithm that follows the natural terrain trend of contour lines in the mountain pit map to complete the traversal of planting units, thus defining traversal boundaries that conform to the actual mountain terrain for parent allocation. The hybrid allocation strategy combines parent genetic priority with the ratio of planting unit demand, using logical rules to pre-allocate parents to pits along the traversal path, achieving accurate matching of high-quality terrain with high-priority parents and a scientific ratio of the planting quantity of each parent.

[0072] In the process of obtaining the matching results, based on structured feature text, a two-pointer matching algorithm is used to match the parental configuration map in the preliminary parental spatial configuration results with the mountain and nest map in the standardized data, and the matching results include: Based on the digital elevation model in the structured feature text, a contour line traversal algorithm is used to iterate row by row along the contour lines of the mountain pit map, outputting a list of available pits along the traversal path. Using column pointers, a hybrid allocation strategy is employed to pre-allocate parent plants according to the traversal path, forming a pre-matching list between parent planting units and available pits in the list of available pits. The hybrid allocation strategy includes: obtaining the required planting unit ratio for each parent and the parent genetic priority information from the structured feature text; prioritizing the parent genetic priority information to obtain a ranking result; pre-allocating available pits along the traversal path with different parent planting units based on the ranking result and the required planting unit ratio for each parent; and matching according to the pre-matching list and preset matching rules to obtain the matching result.

[0073] It is understandable that the aforementioned Digital Elevation Model (DEM) is a terrain model that digitally expresses the elevation information of mountain terrain in the form of a numerical array. It can be understood as the "terrain data source" of structured feature text. It discretizes the three-dimensional terrain space of the target mountain into a set of elevation points in a regular grid, accurately recording the core terrain parameters such as altitude, slope, and aspect of each grid point. It is the core spatial data foundation for contour line tracking algorithms to carry out terrain traversal and terrain adaptation rules to correct the coordinates of planting units. It is also the core data source of terrain information in structured feature text.

[0074] Specifically, the row pointer uses the DEM data in the structured feature text as the core basis and executes the contour line tracking algorithm to systematically traverse the contour lines of the mountain pit map row by row. This traversal method conforms to the natural trend of complex mountains and avoids the problem of terrain mismatch of planting units caused by traversing across contour lines. During the traversal process, effective pits that meet the quantitative constraints such as slope and obstacles are simultaneously screened out. Finally, a list of available pits is output, which includes key information such as pit number, actual coordinates, and terrain quality level.

[0075] The column pointer, based on the traversal path and available nest list output by the row pointer, executes a "priority + proportional" hybrid allocation strategy to complete the parent pre-allocation: First, it extracts the demand ratio of each parent planting unit and the parent genetic priority information (high / medium / low) from the structured feature text, sorts the parents from high to low priority, and prioritizes the allocation of high-priority nests (such as nests with a slope ≤25° and no obstacles) to high-priority parents; then, combined with the demand ratio, it matches the available nests on the traversal path with different parent planting units one by one, forming a pre-matching list of "parent planting unit - nest".

[0076] In this embodiment, the row pointer traverses the contour lines line by line to output a list of available nests, which can define accurate and compliant terrain boundaries and optional ranges for subsequent pre-allocation of parent stock; and the column pointer achieves fully automatic matching and conversion between the parent stock configuration map and the mountain nest map through a dynamic allocation collaborative mode.

[0077] In one embodiment, this application also provides a specific implementation method for obtaining matching results by matching according to a pre-matching list and preset matching rules. Please refer to [link to relevant documentation]. Figure 4 As shown, the method includes: Step S401: Based on the pre-matching list, obtain the coordinates of available pits in the mountain pit map marked in the structured feature text, and the center coordinates of the planting units of the parent planting units in the parent configuration map.

[0078] Step S402: Calculate the Euclidean distance between the coordinates of the currently available planting holes and the coordinates of the center of the planting unit.

[0079] Step S403: When the Euclidean distance meets the preset error condition range, bind the currently available nests with the parent planting units.

[0080] Step S404: When the Euclidean distance does not meet the preset error condition range, the column pointer backtracks to the previous unassigned available hole and matches the previous unassigned available hole with the current parent planting unit.

[0081] Step S405: When a preset number of consecutive matching failures occur, the column pointer stops backtracking, triggers the row pointer to perform a jump operation, and starts restarting the column pointer to perform the matching operation to obtain the matching result; triggering the row pointer to perform a jump operation and starting restarting the column pointer to perform the matching operation includes: triggering the row pointer to jump to the adjacent contour path and taking the adjacent contour path as the new path, and restarting the column pointer to perform the matching operation on the new path; if the matching still fails, the area is determined as the area to be manually adjusted.

[0082] In this embodiment, during the dual-pointer collaborative matching process, the Euclidean distance between the available pit coordinates in the hillside pit map marked in the structured feature text and the center coordinates of the parent planting unit in the parent plant configuration map is calculated. The system then determines whether the Euclidean distance meets a preset error condition range, which can be a tolerance range of ±0.3 meters. If the Euclidean distance meets the preset error condition range (i.e., within ±0.3 meters), the match is considered successful, and the available pit is bound to the corresponding parent planting unit. If the Euclidean distance does not meet the preset error condition range (i.e., exceeds ±0.3 meters), the match is considered a failure. By setting a tolerance range, both terrain errors and construction accuracy can be considered, ensuring matching precision and avoiding matching failures caused by minor coordinate deviations.

[0083] Optionally, a match may be deemed unsuccessful if any of the following conditions are met: inadequate terrain suitability, failure to meet distance constraints, inadequate pit availability, or excessive demand ratio. Triggering conditions for inadequate terrain suitability include: the slope of the pit to be assigned to a high-priority parent (e.g., P1) is >25° (not a high-quality gentle slope), which does not conform to the allocation logic of "high-priority parents are preferentially matched with high-quality terrain"; the pit spacing amplification factor does not match the growth requirements of the parent, such as the pit coefficient for the tolerant parent P4 being 1.0, which does not meet the suitability requirement of 1.5 for extremely steep slopes. Triggering conditions for failure to meet distance constraints include: the Euclidean distance between the pit to be assigned and the already assigned pits of the same parent is <6 meters (minimum distance requirement for adjacent blocks); the difference in row and column numbers between the pit to be assigned and the already assigned pits of the same parent within the same block is <2, indicating a risk of direct adjacency.

[0084] The triggering conditions for unsatisfactory seedling availability include: the seedling marked as available in the text by the row pointer (e.g., covered by rocks or gullies); and the seedling being marked as available but located in the "indirectly affected area" of an obstacle (e.g., less than 1 meter from the edge of a rock), posing a risk to seedling growth. The triggering conditions for exceeding the demand limit include: the number of seedlings allocated to a parent line has reached or exceeded its demand limit, but the column pointer still attempts to allocate new seedlings to it.

[0085] As an feasible approach, a tiered backtracking / jump mechanism is implemented to address matching failure scenarios. This triggers the row pointer to perform a jump operation and restarts the column pointer to perform the matching operation, obtaining the matching result and ensuring matching efficiency and integrity. When the Euclidean distance does not meet the preset error condition range, it indicates a matching failure. The column pointer can backtrack to the previous unassigned available nest and retry matching the previous unassigned available nest with the current parent planting unit. When matching fails for a preset number of consecutive times, the column pointer stops backtracking, triggers the row pointer to jump to the adjacent contour path and use it as the new path, restarting the column pointer allocation process on the new path. The preset number of times can be customized according to actual needs, such as three times.

[0086] When the row pointer jumps to the adjacent contour path, a matching process similar to the above steps is executed. If the matching still fails, the area is marked as an area to be manually adjusted, and the marking information is recorded in the structured feature text for subsequent manual verification and fine-tuning to ensure that there are no blind spots in the matching.

[0087] It should be noted that this embodiment provides a specific scenario example of three consecutive matching failures. For example, when the column pointer is assigned to the cavity C06, the Euclidean distance exceeds the tolerance of ±0.3 meters. The backtracking to the cavity C05 still fails. The row pointer jumps to the adjacent contour line (contour distance 1 meter) and the matching is successful.

[0088] The dual-pointer matching algorithm in this embodiment achieves a precise conversion of the parent stock configuration map from "theoretical coordinates" to "actual planting sites" by combining the collaborative logic of row pointers fitting the terrain and column pointers adapting to priority, along with refined Euclidean distance matching rules and hierarchical anomaly handling. This not only ensures the terrain adaptability of the planting scheme and the rationality of parent stock allocation, but also balances automation efficiency with the reliability of manual backup through the anomaly marking mechanism.

[0089] Step S205: Mark abnormal regions based on matching results, perform visual fine-tuning and constraint verification on abnormal regions to obtain fine-tuning results, and integrate the fine-tuning results with the matching results to generate target planting results.

[0090] Understandably, the above matching results represent a one-to-one correspondence between parent plantation units and actual planting holes, obtained after the dual-pointer matching algorithm performs contour line traversal, mixed allocation pre-allocation, and Euclidean distance tolerance rules to match the initial parent plantation spatial configuration results with the mountain planting hole map. Abnormal areas are local planting areas verified based on the matching results within the local judgment units. Fine-tuning the distance involves adjusting the position of parent plant icons in abnormal areas within the dual-layer overlay interface, and after real-time system verification to ensure compliance with distance constraints and terrain adaptation rules, resulting in a corrected parent-plantation hole matching result—manually optimized and compliant matching data. The target planting result integrates the automatically output matching results from the algorithm with the manually adjusted fine-tuning results for abnormal areas, forming a complete parent plantation plan that simultaneously meets breeding experiment specifications, terrain adaptation requirements, and can be directly used for on-site planting construction.

[0091] In one embodiment, abnormal regions are marked based on the matching results, and the abnormal regions are then visually fine-tuned and constraint-checked to obtain the fine-tuning results, including: Based on the matching results, local judgment units are defined; the unit parameters of the local judgment units are obtained, and based on the unit parameters, it is determined whether the local judgment unit is an abnormal area; the unit parameters include: pit missing rate and average slope; when it is an abnormal area, the abnormal area is marked; the parent plant configuration map and the mountain pit map are overlaid to build a double-layer overlay interface; the user operation command is received and responded to, and the planting position of the parent plant icon in the abnormal area is adjusted to obtain the adjusted planting result; the user operation command is triggered after the user performs a drag operation on the double-layer overlay interface; other constraints are obtained; other constraints include: pit availability constraints, demand ratio constraints, and construction compliance constraints; the adjusted planting result is constrained and verified based on other constraints and quantitative constraints; when all other constraints and quantitative constraints pass the verification, the fine-tuning result is generated.

[0092] Specifically, after obtaining the matching results, local judgment units are defined according to a preset specification, which can be a fixed specification of 10×10 meters. For each local judgment unit, each local judgment unit is checked one by one to see if it meets the abnormal conditions. Two core unit parameters are extracted from the local judgment unit: the missing rate of pits and the average slope. The missing rate of pits is compared with the corresponding missing rate threshold, and the average slope is compared with the corresponding slope judgment threshold. When the missing rate of pits is greater than the missing rate threshold, or the average slope is greater than the slope judgment threshold, the local judgment unit is determined to meet the abnormal conditions. Local judgment units that meet the abnormal conditions are marked as abnormal areas. The missing rate threshold can be 35°, and the slope judgment threshold can be 20%. That is, when the average slope is >35° or the missing rate of pits is >20%, the local judgment unit is determined to meet the abnormal conditions.

[0093] Furthermore, the parent plant configuration map and the mountain planting site map are precisely overlaid to create a dual-layer overlay operation interface that allows users to intuitively view the theoretical positions of the parents and the actual positions of the planting sites. In this interface, users can adjust the planting position of the parents by dragging the parent plant icon for marked abnormal areas. When the user performs the drag operation, the system receives and responds to this operation command in real time, converting the dragged parent plant position into the adjusted planting result.

[0094] Next, other constraints such as planting hole availability, demand ratio, and construction compliance are retrieved. These, combined with the aforementioned quantitative constraints including distance constraints and terrain adaptation rules, form a comprehensive verification standard. Based on these two constraints, the manually adjusted planting results are automatically verified. When all other constraints and quantitative constraints pass verification, the adjusted results are confirmed as compliant and valid, and fine-tuning results for abnormal areas are generated. When at least one of the other constraints or quantitative constraints fails verification, the user is prompted to readjust until all constraints are met, ensuring that the fine-tuning results both conform to the actual terrain and comply with the comprehensive specifications of breeding trials and construction. It can be understood that the aforementioned planting hole availability constraint can be interpreted as a feasibility constraint, specifying the availability status based on structured feature text to ensure that parent plants are only assigned to usable planting holes. Basic availability constraints include obstacle avoidance constraints: the planting hole must be at least 1 meter away from the edge of an obstacle, and at least 1.5 meters away in extremely steep slope areas, to prevent seedlings from being eroded / squeezed. The demand ratio constraint can be understood as an experimental design specification constraint, which is formulated based on the preliminary parental spatial configuration results to ensure that the allocation meets the experimental design requirements. For example, it includes: the number of allocated holes for each parent ≤ its upper limit of demand (e.g., P1 ≤ 9 holes, P4 ≤ 81 holes); parents of the same priority (e.g., P2, P3) have the same demand, and the allocated number must be equal (9 holes each), with a deviation ≤ 1. Construction compliance constraints are industry standard constraints and can include: planting hole specification constraints; protection row constraints. Among these, construction compliance constraints include planting hole specifications ≥ 60cm × 60cm × 40cm, and the allocation of only ordinary parents to the two rows of holes around the perimeter of the test forest, specifically the two planting units outside the test forest boundary.

[0095] After obtaining the fine-tuning results, they are fully integrated with the original matching results output by the dual-pointer matching algorithm in non-abnormal areas. This can be done by standardizing and verifying the parent-nest correspondence and planting coordinate information of the two types of results, eliminating duplicate data, and supplementing information dimensions. Finally, a target planting result is generated that meets all-dimensional constraints such as distance constraints, terrain adaptation rules, nest availability, demand ratio, and construction compliance throughout the entire domain. It can also be directly connected to on-site operations and adapted to complex mountainous terrain.

[0096] It should be noted that the above-mentioned target planting results may include optimized parent plant layout coordinates in CSV format and parent plant planting layout map in PNG format. The parent plant layout coordinates include block number, parent number, planting unit row and column number, relative coordinates, slope range, and spacing magnification factor. The parent plant planting layout map can be 1:500 in scale and includes planting unit grid, parent plant markings, slope zoning, obstacle markings, and coordinate scale.

[0097] In this embodiment, by determining whether an area is abnormal based on abnormal parameters, the core scope for subsequent manual fine-tuning can be clearly identified, ensuring that no problem area is missed or expanded. Furthermore, by building a dual-layer overlay interface, manual adjustment operations become intuitive and convenient, requiring no specialized coordinate input and lowering the operational threshold.

[0098] For example, the experimental site is Xishan Forest Farm in a certain city (29°23′-29°52′N, 105°28′-106°02′E), covering an area of ​​1 hectare with a slope of 15°-40° and a soil type of red-yellow soil, which conforms to the topographic characteristics of major forest-producing areas in southern China. There are two rocky areas and one gully area within the experimental area, which need to be avoided in the planting design. The core experimental parameters include parental information, blocks and planting units, and topographic parameters. Parental information can be selected from four superior *Scleroderma purpurea* parents, numbered P1-P4, with genetic priority in the order of high, medium, medium, and low. The blocks and planting units include: 3 blocks, each with a 6×6 matrix layout (36 planting units), and each planting unit is a square with a side length of 2 meters. A single block occupies an area of ​​144 square meters (12 meters × 12 meters). Terrain parameters include slope map, obstacle information, etc. The slope map is in GeoTIFF format with a pixel resolution of 1 meter × 1 meter and a pixel value accuracy of 0.1°; obstacles are labeled in JSON format with a boundary coordinate accuracy of 0.1 meters.

[0099] In the intelligent planting process, planting design data for progeny testing of forest trees can be obtained first. This planting design data includes: parental information, block parameters, and topographic data. Parental information can be represented in JSON format as a parental list, for example: [{"Parental Number":"P1","Genetic Priority":"High"},{"Parental Number":"P2","Genetic Priority":"Medium"},{"Parental Number":"P3","Genetic Priority":"Medium"},{"Parental Number":"P4","Genetic Priority":"Low"}]; Block parameters can be represented in JSON format, for example: {"Number of Blocks":3,"Planting Unit Matrix for Each Block":"6×6 (rows)"} × column), "Planting unit side length":2,"Unit":"meter"}; Topographic data can include slope map and obstacle information. The slope map can be a 36m×36m area covering the test area, with pixel values ​​ranging from 15.2° to 39.7°. Obstacle information is represented in JSON format, for example: [{"Type":"Rock","Boundary coordinates":[(10.2,20.5),(10.2,25.3),(15.6,25.3),(15.6,20.5)]},{"Type":"Grat","Boundary coordinates":[(30.1,18.4),(30.1,22.7),(35.8,22.7),(35.8,18.4)]}). The planting design data was standardized, for example, the slope map was pixel-calibrated so that the slope value error was ≤0.5°; the obstacle coordinates were projected and transformed, and a relative coordinate system was uniformly adopted (the lower left corner of the test site is the origin (0,0)).

[0100] Next, quantitative constraints for optimizing parental distribution were constructed, including distance constraints and terrain adaptation rules. Distance constraints included: the minimum distance between the same parent in adjacent blocks ≥ 6 meters (i.e., 3 planting units), verified by Euclidean distance calculation of the center coordinates of the planting units; and the difference in the horizontal / vertical row / column number of the same parent within the same block ≥ 2. For example, planting units (1,1) meet the requirements with (1,3) and (3,1), but not with (1,2) and (2,1). Terrain adaptation rules included: slope grading standards and dynamic coordinate adjustment rules. The slope grading standards included the correspondence between slope grading and spacing coefficients, as shown in Table 2 below. Table 2 Correspondence between Slope Grade and Spacing Coefficient

[0101] For example, the original coordinates of a planting unit with a slope of 28° are (6.0, 8.0). After adjustment using the above method, the adjusted coordinates are (6.0×1.2, 8.0×1.2)=(7.2, 9.6).

[0102] An optimization model combining simulated annealing and Delaunay triangulation was adopted. The model parameters were initialized, for example, the initial temperature was set to 1.0, the cooling rate to 0.99, the iteration step size to 50, and the maximum number of iterations to 500. Then, the objective function was calculated: each parent was assigned 9 planting units (n=9). The average Euclidean distance between all planting units of the same parent was calculated using the objective function. In this embodiment, the convergence condition was met after 320 iterations, and the objective function value stabilized at 8.62 meters. Delaunay triangulation was then applied: the centers of 108 planting units in 3 blocks were used as discrete point sets to construct a triangulation network, identifying adjacent units and avoiding acute-angle adjacencies. Finally, convergence verification was performed: if the change in the objective function was ≤0.01 meters after 30 consecutive iterations, iteration was stopped, and the preliminary parental spatial configuration results were output. This could be a standardized, structured dataset, with core information including basic planting unit information, coordinate information, and constraint satisfaction verification results. The total number of planting units for parent P1 is n=9, with 3 blocks × 3 planting units per block, allocated proportionally. The side length of each planting unit is 2 meters, and the coordinate system is relative (the lower left corner of the experimental plot is the origin (0,0)). The terrain adaptation rule has been applied: for some planting units with a slope >25°, the coordinates are corrected by a spacing coefficient of 1.2. The preliminary spatial configuration scheme of parent P1 is shown in Table 3 below. Table 3 Preliminary Space Configuration Scheme for Parent P1

[0103] A fine-tuning dataset was constructed, comprising a visual text pairing dataset and a rule knowledge base dataset. 5000 mountain cave maps were obtained and labeled with cave numbers, terrain types, and other information, along with 12 core rule categories extracted from 200 forestry topography specifications. The fine-tuning process was then performed: the adapter matrix was trained using LoRA technology. After 10 rounds of training, the model's feature recognition accuracy reached 95.3%, resulting in the fine-tuned model. The cave maps from the experimental site were input into the fine-tuned model, generating structured feature text in JSON format. This text includes information such as the number, coordinates, and slope of 108 caves. This structured JSON text can be processed as follows: json { "Basic Information":{"Forest Number":"SD-2025-001","Number of Blocks":3,"Total Number of Planting Units":108}, "List of acupoint characteristics":[ {"Cave ID":"A01","Relative Coordinates (X,Y):(2.0,2.0),"Slope (°):22.5,"Spacing Amplification Factor":1.0,"Availability":"Yes"}, {"Cave ID":"A02","Relative Coordinates (X,Y):(4.0,2.0),"Slope (°):28.3,"Spacing Amplification Factor":1.2,"Availability":"Yes"} ], "Obstacle Details":[ {"Type":"Rock","Cave Number":["C04","C05"],"Boundary Coordinates":[(8.0,6.0),(8.0,10.0),(12.0,10.0),(12.0,6.0)]} ] } Furthermore, contour lines (1-meter interval) are extracted from the DEM data in the structured feature text. Holes are traversed along these contour lines, and column pointers are used to allocate high-quality terrain according to priority (P1 > P2 = P3 > P4). The remaining holes are then allocated proportionally, forming a pre-matching list between parent planting units and holes in the available hole list. The Euclidean distance between the coordinates of the currently available hole and the center coordinates of the planting unit is calculated. The Euclidean distance tolerance is checked to see if it is ±0.3 meters. If it is within ±0.3 meters, the currently available hole is bound to the parent planting unit, thus obtaining the matching result. This embodiment achieves a matching success rate of 92.7%. Furthermore, the intelligent matching of 108 holes takes 180 minutes per hectare, which is 8 times faster than traditional manual matching (4320 minutes per hectare) and 1 time faster than existing AI methods (360 minutes per hectare).

[0104] Based on the matching results, anomaly detection and manual fine-tuning were performed. First, using 10×10 meters as the anomaly judgment unit, the average slope of one unit in the southeast of the experimental site was obtained: 37.2°, and a pit loss rate of 24%. Based on the pit loss rate and average slope, it was determined whether it was an anomaly area and marked accordingly. Then, the parent plant configuration map and the mountain pit map were overlaid to create a double-layer overlay interface. Users performed manual fine-tuning operations on this interface, for example, dragging the P4 parent plant icon to coordinates (28.6, 32.4). The system verified and displayed a slope of 36.8°, a spacing magnification factor of 1.5, and a distance of 7.8 meters from adjacent parent plants. The adjusted planting results were then constrained and verified to meet the constraints. The fine-tuning took 6 minutes. This embodiment provides the conversion accuracy and fine-tuning effect for different slope areas, including a comparison table of planting effects for different terrain types, as shown in Table 4 below. Table 4 Comparison of Planting Effects in Different Terrain Types

[0105] This embodiment also includes four control experiments to verify the effectiveness of the intelligent planting method for progeny testing of forest trees provided in this application. The four control experiments include: control group (manual), experimental group 1 (algorithm optimization only), existing AI method, and experimental group 2 (method of this application). Corresponding operations were performed on the four control experiments to obtain the experimental results for each group, including planting accuracy, design time, manual error rate, and adjacent rate of the same parent line. The comparison table of planting effects of different experiments can be seen in Table 5 below: Table 5 Comparison of Planting Effects in Different Trials

[0106] It should be noted that the data in Table 5 above are the result of the combination of the mean and standard deviation obtained after three repeated experiments. The mean reflects the central tendency of the data, and the standard deviation reflects the dispersion of the data, thus ensuring the reliability and statistical significance of the above experimental data.

[0107] The above experimental results show that the intelligent planting method for progeny testing forests provided in this application is significantly superior to traditional manual methods and existing AI methods in terms of planting accuracy, design efficiency, error control, and parent distribution optimization. It also has strong terrain adaptability and fully meets the planting design requirements for progeny testing forests in complex mountainous areas.

[0108] The intelligent planting method for progeny testing of native trees provided in this application has significant advantages over traditional manual planting methods in multiple dimensions: It solves the problem of clustering of identical parents by using quantitative distance constraints and optimization algorithms, reducing the adjacent rate of identical parents to 0%, significantly reducing experimental error and improving the accuracy of genetic trait evaluation compared to the 28.7%±3.5% of traditional manual methods; the automatic conversion accuracy for different terrains such as gentle slopes, steep slopes, and extremely steep slopes reaches 96.5%, 89.2%, and 78.3% respectively, and the terrain conformity after fine-tuning all exceed 99%, effectively breaking through the bottleneck of adapting to complex mountainous terrain and making it highly adaptable to terrain; the planting design accuracy reaches 92.7%, an improvement of 32% compared to the 60.3%±4.2% of traditional manual methods. The efficiency is improved by 4 percentage points, with a design time of 180±30 minutes / hectare, which is 8 times more efficient than the traditional manual method of 4320±360 minutes / hectare. At the same time, the manual error rate is reduced from 15.3% to 1.2%, achieving a dual improvement in efficiency and accuracy. Furthermore, the use of a dual-pointer algorithm reduces the time complexity to O(n). When the number of nests increases from 50 to 200, the conversion time only increases from 2.1 minutes to 6.8 minutes, demonstrating excellent scalability and meeting the needs of large-scale breeding experiments. Moreover, the method is fully software-based for automated operation, and the manual fine-tuning function is easy to operate. Adjustments in extreme terrain areas only take 5-8 minutes, significantly reducing the professional operation threshold and balancing the scientific nature of the technology with the practicality of engineering.

[0109] This application provides a method for intelligent planting of trees based on progeny generation. Compared with existing technologies, this method acquires and standardizes planting design data, providing more comprehensive data guidance for subsequent planting processes. It eliminates the reliance on experience in traditional manual operations, avoiding issues like incorrect numbering and positional deviations caused by manual lot drawing and hand-drawn maps. Based on standardized data, quantitative constraints are constructed, and an optimization model integrating simulated annealing and triangulation algorithms is iteratively optimized according to the parental distance maximization criterion to obtain preliminary configuration results. This algorithmically achieves a scientifically dispersed layout of parents, effectively reducing the clustering of identical parents and improving the accuracy of genetic trait evaluation. Furthermore, the basic model is fine-tuned using forestry and grassland knowledge, allowing the fine-tuned model to accurately analyze terrain and pit features, laying a precise data foundation for subsequent matching. The process involves generating structured feature text based on the fine-tuned model, addressing the poor adaptability of traditional methods to complex mountainous terrain. A dual-pointer matching algorithm then ensures precise matching between the parent plant configuration map and the mountain pit map, allowing the initial configuration results to align with the actual terrain and improving the terrain adaptability and efficiency of the design. Finally, abnormal areas in the matching results are marked, and visual fine-tuning and constraint verification are performed. The fine-tuning and matching results are integrated to generate the target planting results. This approach significantly improves planting design efficiency through automation while ensuring compliance and operability through manual fine-tuning. Overall, it solves many problems associated with traditional manual planting methods, such as unreasonable parent plant distribution, poor terrain adaptability, low design efficiency, and high error rates. This enables intelligent, standardized, and precise planting of parent plants for forest progeny testing, significantly improving the scientific rigor of planting schemes.

[0110] Based on the same inventive concept, this application also provides an intelligent planting device for implementing the above-mentioned intelligent planting method for progeny testing forests. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the intelligent planting device for progeny testing forests provided below can be found in the limitations of the intelligent planting method for progeny testing forests described above, and will not be repeated here.

[0111] In one exemplary embodiment, such as Figure 5 As shown, a smart planting device for determining the progeny of trees is provided. The device includes: The acquisition module 510 is used to acquire the planting design data of the progeny test forest and perform standardization processing to obtain standardized data; the planting design data includes: parent information, block parameters and topographic data; The iterative optimization module 520 is used to construct quantitative constraints for parental distribution optimization based on standardized data, and based on the quantitative constraints, iteratively optimizes the optimization model according to the criterion of maximizing parental distance to obtain preliminary parental spatial configuration results; the optimization model is built using simulated annealing algorithm and Delaunay triangulation algorithm. The processing module 530 is used to fine-tune the preset basic model with knowledge of forestry and grassland, obtain the fine-tuned model, and use the fine-tuned model to process standardized data to generate structured feature text. The matching module 540 is used to match the parent configuration map in the preliminary parent spatial configuration result with the mountain and nest map in the standardized data based on the structured feature text and a two-pointer matching algorithm to obtain the matching result; The generation module 550 is used to mark abnormal regions based on the matching results, perform visual fine-tuning and constraint verification on the abnormal regions, obtain fine-tuning results, and integrate the fine-tuning results with the matching results to generate target planting results.

[0112] As an optional implementation, the quantification constraints include: distance constraints and terrain adaptation rules; Distance constraints include coordinate system standardization, inter-block spacing constraints, and intra-block spacing constraints. Coordinate system standardization includes: using a relative coordinate system to calibrate the position, with the lower left corner of the forest plot for progeny testing as the origin, the horizontal axis as the X-axis, and the vertical axis as the Y-axis; each planting unit is a square with a side length of two meters, and the center coordinates of the planting unit are (x×2, y×2), where x and y are the row and column numbers of the planting unit; inter-block spacing constraints include: the minimum distance between the same parent in adjacent blocks is not less than three planting units, and the corresponding Euclidean distance is ≥6; intra-block spacing constraints include: the horizontal and vertical spacing of the same parent within the same block is not less than two planting units, and the same parent is allowed to be diagonally adjacent. The terrain adaptation rules include: slope grading standards and dynamic coordinate adjustment rules; the slope grading standards are divided into three slope levels according to the slope range, with each slope level corresponding to a spacing amplification factor; when the slope range is the first slope level, the corresponding spacing amplification factor is 1.0; when the slope range is the second slope level, the corresponding spacing amplification factor is 1.2; when the slope range is the third slope level, the corresponding spacing amplification factor is 1.5; the first slope level is a slope less than or equal to 25°, the second slope level is a slope greater than 25° and less than 35°, and the third slope level is a slope greater than or equal to 35°; The dynamic coordinate adjustment rules include: increasing the spacing by adjusting the center coordinates of the planting units, keeping the row and column numbers of the planting units unchanged, and multiplying the value of the center coordinates of the planting units in the vertical direction along the slope by the corresponding spacing amplification factor.

[0113] As an optional implementation, the iterative optimization module 520 is specifically used for: Based on terrain adaptation rules, the center coordinates of all planting units are dynamically corrected to obtain a modified standardized coordinate system. According to the standardized coordinate system, the centers of all planting units in all blocks are taken as discrete point sets to construct the Delaunay triangulation of all planting units and determine the adjacency relationship of the Delaunay triangulation. An optimization model integrating simulated annealing and Delaunay triangulation is constructed, and the model parameters are initialized. The model parameters include initial temperature, cooling rate, and iteration step size. Based on standardized data, the average Euclidean distance between all planting units of the same parent was calculated to construct the objective function; Generate parental configuration candidate schemes within a standardized coordinate system; Based on the model parameters and the adjacency relationship of the Delaunay triangulation, distance constraints are used as the verification standard, terrain adaptation rules are used as the spatial boundary, and the maximization of the objective function is used as the optimization direction. The parent configuration candidate schemes are iteratively verified until the iteration termination condition is met. The iteration termination condition includes the change in the objective function being less than or equal to a preset threshold or the number of iterations reaching a preset number. When the iteration termination condition is met, the preliminary parental spatial configuration results that conform to the terrain adaptation rules and distance constraints are obtained.

[0114] As an optional implementation, the processing module 530 is specifically used for: A fine-tuning dataset is constructed and divided into a training set and a test set. The fine-tuning dataset includes a visual text pairing dataset and a rule knowledge base dataset. The visual text pairing dataset includes historical mountain cave maps and corresponding structured annotation results. The rule knowledge base dataset includes the mapping results between rule keywords and parameter values. Initialize the training parameters of the base model; training parameters include learning rate, number of training epochs, and batch size; model parameters include backbone parameters and other parameters; With the core parameters of the base model fixed, the training set is input into the base model for feature parsing to obtain the output results. Based on the output results and structured annotation results, construct the cross-entropy loss function; By minimizing the cross-entropy loss function, the other parameters of the base model are iteratively trained based on the training parameters to obtain the trained model. The test set is input into the trained model for testing, resulting in a fine-tuned model; the fine-tuned model has the ability to analyze features in the forestry and grassland domain.

[0115] As an optional implementation, the processing module 530 is also used for: Mountain cave maps are obtained from standardized data. These maps are then processed using a fusion model to extract visual features. The visual features include spatial information, terrain features, and cave distribution information. The fusion model has cross-modal feature learning capabilities. Visual features are input into the fine-tuned model to generate structured feature text; the structured feature text includes basic information about the forest progeny, a list of pit features, and obstacle information.

[0116] As an optional implementation, the matching module 540 is specifically used for: Based on the digital elevation model in the structured feature text, the contour line tracing algorithm is used to traverse the contour line direction of the mountain cave map line by line using row pointers, and outputs a list of available caves on the traversal path; By using column pointers to follow the traversal path, a hybrid allocation strategy is employed to perform parent pre-allocation, forming a pre-matching list between parent planting units and available planting holes in the available hole list. The hybrid allocation strategy includes: obtaining the required proportion of planting units for each parent and the parent genetic priority information in the structured feature text; sorting the parent genetic priority information according to priority to obtain the sorting result; and pre-allocating the available holes on the traversal path with different parent planting units based on the sorting result and the required proportion of planting units for each parent. Based on the pre-match list, the matching is performed according to the preset matching rules to obtain the matching results.

[0117] As an optional implementation, the matching module 540 is also used for: Based on the pre-matching list, obtain the coordinates of available pits in the mountain pit map and the center coordinates of the planting units of the parent planting units in the parent configuration map, which are marked in the structured feature text. Calculate the Euclidean distance between the coordinates of the currently available planting holes and the coordinates of the center of the planting unit; When the Euclidean distance meets the preset error condition range, the currently available nests are bound to the parent planting units; When the Euclidean distance does not meet the preset error condition range, the column pointer backtracks to the previous unassigned available hole and matches the previous unassigned available hole with the current parent planting unit; When a preset number of consecutive matching attempts fail, the column pointer stops backtracking, triggers the row pointer to perform a jump operation, and restarts the column pointer to perform the matching operation, thus obtaining the matching result. Triggering the row pointer to perform a jump operation and restarting the column pointer to perform the matching operation includes: triggering the row pointer to jump to the adjacent contour path and using the adjacent contour path as the new path, and restarting the column pointer to perform the matching operation on the new path; if the matching still fails, the area is determined as an area to be manually adjusted.

[0118] As an optional implementation, the generation module 550 is specifically used for: Based on the matching results, local decision units are defined; Obtain the unit parameters of the local judgment unit and determine whether the local judgment unit is an abnormal area based on the unit parameters; the unit parameters include: pit loss rate and average slope; When a region is considered abnormal, mark it as abnormal. The parent plant configuration map and the mountain area and nest map are overlaid to create a dual-layer overlay interface; It receives and responds to user operation commands, adjusts the planting position of the parent plant icon in the abnormal area, and obtains the adjusted planting result; the user operation command is triggered after the user performs a drag operation on the dual-layer overlay interface; Obtain other constraints; other constraints include: pit availability constraints, demand ratio constraints, and construction compliance constraints; The adjusted planting results are constrained and verified based on other constraints and quantitative constraints. When all other constraints and quantization constraints pass the verification, the fine-tuning result is generated.

[0119] The intelligent planting device for progeny testing of forest trees provided in this application embodiment acquires and standardizes planting design data, providing more comprehensive data guidance for subsequent planting processes. This eliminates the reliance on experience in traditional manual operations, avoiding the numbering errors and positional deviations caused by manual lot drawing and hand-drawn maps. Based on standardized data, quantitative constraints are constructed, and an optimization model integrating simulated annealing and triangulation algorithms is iteratively optimized according to the parental distance maximization criterion to obtain preliminary configuration results. This achieves a scientifically dispersed layout of parents at the algorithmic level, effectively reducing the clustering of identical parents and improving the accuracy of genetic trait evaluation. Furthermore, the basic model is fine-tuned using forestry and grassland knowledge, allowing the fine-tuned model to accurately analyze terrain and pit features, laying a precise data foundation for subsequent matching. The structured feature text generated based on the fine-tuned model solves the problem of poor adaptability of traditional methods to complex mountainous terrain. Then, the dual-pointer matching algorithm achieves accurate matching between the parent plant configuration map and the mountain pit map, so that the initial configuration results fit the actual terrain and improve the terrain adaptability and efficiency of the design. Finally, abnormal areas of the matching results are marked and visual fine-tuning and constraint verification are carried out. The fine-tuning and matching results are integrated to generate the target planting results. The automated process greatly improves the efficiency of planting design, while the manual fine-tuning ensures the compliance and operability of the plan. Overall, it solves many problems of unreasonable parent plant distribution, poor terrain adaptability, low design efficiency and high error rate of traditional manual planting methods. It realizes the intelligent, standardized and precise planting of parent plants for forest progeny determination, and greatly improves the scientific nature of the planting plan.

[0120] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 6 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores video tag processing data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for intelligent planting of forest trees for progeny determination.

[0121] Those skilled in the art will understand that Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0122] In one exemplary embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0123] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0124] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0125] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0126] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0127] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0128] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0129] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for intelligent planting of forests for determining progeny, characterized in that, The intelligent planting method for forest progeny testing includes: The planting design data of the progeny test forest of the forest tree were obtained and standardized to obtain standardized data; the planting design data included: parent information, block parameters and topographic data; Based on the standardized data, quantitative constraints for optimizing parental distribution are constructed. Based on these quantitative constraints, the optimization model is iteratively optimized according to the criterion of maximizing parental distance to obtain preliminary parental spatial configuration results. The optimization model is constructed using simulated annealing and Delaunay triangulation algorithms. The preset basic model is fine-tuned using knowledge from the forestry and grassland domain to obtain the fine-tuned model. The fine-tuned model is then used to process the standardized data to generate structured feature text. Based on the structured feature text, a two-pointer matching algorithm is used to match the parent configuration map in the preliminary parent spatial configuration result with the mountain and nest map in the standardized data to obtain the matching result; Based on the matching results, abnormal regions are marked, and the abnormal regions are visually fine-tuned and constraint-verified to obtain fine-tuning results. The fine-tuning results are then integrated with the matching results to generate target planting results.

2. The intelligent planting method for determining the progeny of trees according to claim 1, characterized in that, The quantification constraints include: distance constraints and terrain adaptation rules; The distance constraints include coordinate system standardization, inter-block spacing constraints, and intra-block spacing constraints. The coordinate system standardization includes: using a relative coordinate system to calibrate the position, with the lower left corner of the forest plot as the origin, the horizontal axis as the X-axis, and the vertical axis as the Y-axis; each planting unit is a square with sides of two meters, and the center coordinates of the planting unit are (x×2, y×2), where x and y are the row and column numbers of the planting unit. The inter-block spacing constraints include: the minimum distance between the same parent in adjacent blocks is not less than three planting units, with a corresponding Euclidean distance ≥6. The intra-block spacing constraints include: the horizontal and vertical spacing between the same parent within the same block is not less than two planting units, and diagonal adjacency of the same parent is allowed. The terrain adaptation rules include: slope grading standards and coordinate dynamic adjustment rules; the slope grading standards are divided into a first slope level, a second slope level, and a third slope level according to the slope range, with each slope level corresponding to a spacing amplification factor; when the slope range is the first slope level, the corresponding spacing amplification factor is 1.0; when the slope range is the second slope level, the corresponding spacing amplification factor is 1.2; when the slope range is the third slope level, the corresponding spacing amplification factor is 1.5; the first slope level is a slope less than or equal to 25°, the second slope level is a slope greater than 25° and less than 35°, and the third slope level is a slope greater than or equal to 35°; The coordinate dynamic adjustment rules include: increasing the spacing by adjusting the center coordinates of the planting unit, keeping the row and column numbers of the planting unit unchanged, and multiplying the value of the center coordinates of the planting unit in the vertical direction along the slope by the corresponding spacing amplification coefficient.

3. The intelligent planting method for progeny testing of forest trees according to claim 2, characterized in that, Based on the aforementioned quantitative constraints, the optimization model is iteratively optimized according to the criterion of maximizing parental distance to obtain preliminary parental spatial configuration results, including: Based on the terrain adaptation rules, the center coordinates of all planting units are dynamically corrected to obtain the modified standardized coordinate system. According to the standardized coordinate system, the centers of all planting units in all blocks are taken as discrete point sets to construct the Delaunay triangulation of all planting units and determine the adjacency relationship of the Delaunay triangulation. An optimization model integrating simulated annealing and Delaunay triangulation is constructed, and the model parameters of the optimization model are initialized; the model parameters include initial temperature, cooling rate, and iteration step size. Based on the standardized data, the average Euclidean distance between all planting units of the same parent is calculated to construct the objective function; Generate parental configuration candidate schemes within the standardized coordinate system; Based on the model parameters and the adjacency relationship of the Delaunay triangulation, the parent configuration candidate schemes are iteratively verified using distance constraints as the verification standard, terrain adaptation rules as the spatial boundary, and maximizing the objective function as the optimization direction, until the iteration termination condition is met. The iteration termination condition includes the change in the objective function being less than or equal to a preset threshold or the number of iterations reaching a preset number. When the iteration termination condition is met, a preliminary parental spatial configuration result that conforms to the terrain adaptation rules and distance constraints is obtained.

4. The intelligent planting method for determining the progeny of trees according to claim 1, characterized in that, The preset basic model is fine-tuned using knowledge from the forestry and grassland domain to obtain the fine-tuned model, including: A fine-tuning dataset is constructed and divided into a training set and a test set. The fine-tuning dataset includes a visual text pairing dataset and a rule knowledge base dataset. The visual text pairing dataset includes historical mountain cave maps and corresponding structured annotation results. The rule knowledge base dataset includes the mapping results between rule keywords and parameter values. Initialize the training parameters of the base model; the training parameters include the learning rate, training epochs, and batch size; the model parameters include backbone parameters and other parameters. By fixing the backbone parameters of the base model, the training set is input into the base model for feature parsing to obtain the output result; Based on the output results and the structured annotation results, a cross-entropy loss function is constructed; By minimizing the cross-entropy loss function, the other parameters of the base model are iteratively trained based on the training parameters to obtain the trained model. The test set is input into the trained model for testing to obtain a fine-tuned model; the fine-tuned model has the ability to analyze features in the forestry and grassland field.

5. The intelligent planting method for determining the progeny of trees according to claim 4, characterized in that, The standardized data is processed using the fine-tuned model to generate structured feature text, including: A mountain cave map is obtained from the standardized data, and the mountain cave map is processed by a fusion model to extract visual features; the visual features include spatial information, terrain features, and cave distribution information; the fusion model has cross-modal feature learning capabilities; The visual features are input into the fine-tuned model to generate structured feature text; the structured feature text includes basic information of the forest progeny measurement, a list of pit features, and obstacle information.

6. The intelligent planting method for progeny testing of forest trees according to claim 1, characterized in that, The dual-pointer matching algorithm includes: a contour tracing algorithm and a hybrid allocation strategy; the dual pointers include row pointers and column pointers; Based on the structured feature text, a two-pointer matching algorithm is used to match the parental configuration map in the preliminary parental spatial configuration results with the mountain and nest map in the standardized data to obtain the matching results, including: Based on the digital elevation model in the structured feature text, the contour line tracing algorithm is used to traverse the contour line direction of the mountain cave map line by line using row pointers, and outputs a list of available caves on the traversal path; Following the traversal path using column pointers, a hybrid allocation strategy is employed to pre-allocate parent plants, forming a pre-matching list between parent planting units and available planting holes in the available hole list. The hybrid allocation strategy includes: obtaining the required proportion of planting units for each parent and the parent genetic priority information in the structured feature text; prioritizing the parent genetic priority information to obtain a ranking result; and pre-allocating available planting holes along the traversal path with different parent planting units based on the ranking result and the required proportion of planting units for each parent. According to the pre-match list, the matching is performed according to the preset matching rules to obtain the matching result.

7. The intelligent planting method for progeny testing of forest trees according to claim 6, characterized in that, Based on the pre-match list, matching is performed according to preset matching rules to obtain matching results, including: Based on the pre-matching list, obtain the coordinates of available pits in the mountain pit map marked in the structured feature text, and the center coordinates of the planting units of the parent planting units in the parent configuration map; Calculate the Euclidean distance between the coordinates of the currently available planting holes and the center coordinates of the planting unit; When the Euclidean distance meets the preset error condition range, the currently available nest is bound to the parent planting unit; When the Euclidean distance does not meet the preset error condition range, the column pointer backtracks to the previous unassigned available nest and matches the previous unassigned available nest with the current parent planting unit; When a preset number of consecutive matching failures occur, the column pointer stops backtracking, triggers the row pointer to perform a jump operation, and initiates a restart of the column pointer to perform the matching operation, thereby obtaining a matching result. The triggering of the row pointer to perform a jump operation and initiating a restart of the column pointer to perform the matching operation includes: triggering the row pointer to jump to an adjacent contour path and using the adjacent contour path as a new path, and restarting the column pointer to perform the matching operation on the new path. If the matching still fails, the region is determined as a region to be manually adjusted.

8. The intelligent planting method for determining the progeny of trees according to claim 1, characterized in that, Based on the matching results, abnormal regions are marked, and the abnormal regions are then subjected to visual fine-tuning and constraint verification to obtain the fine-tuning results, including: Based on the matching results, local decision units are defined; The unit parameters of the local determination unit are obtained, and the local determination unit is determined as an abnormal region based on the unit parameters; the unit parameters include: pit loss rate and average slope; When a region is considered abnormal, it is marked as abnormal. The parent plant configuration map and the mountain area and nest map are overlaid to create a dual-layer overlay interface; The system receives and responds to user operation commands, adjusts the planting position of the parent plant icon in the abnormal area, and obtains the adjusted planting result; the user operation command is triggered after the user performs a drag operation on the dual-layer overlay interface; Obtain other constraints; these other constraints include: pit availability constraints, demand ratio constraints, and construction compliance constraints. The adjusted planting results are constrained and verified based on the other constraints and the quantitative constraints. The fine-tuning result is generated when all other constraints and quantization constraints pass the verification.

9. A smart planting device for determining the progeny of trees, characterized in that, The intelligent planting device for determining the progeny of trees includes: The acquisition module is used to acquire planting design data of the progeny test forest of trees and perform standardization processing to obtain standardized data; the planting design data includes: parent information, block parameters and topographic data; The iterative optimization module is used to construct quantitative constraints for parental distribution optimization based on the standardized data, and based on the quantitative constraints, iteratively optimizes the optimization model according to the criterion of maximizing parental distance to obtain preliminary parental spatial configuration results; the optimization model is built using simulated annealing algorithm and Delaunay triangulation algorithm. The processing module is used to fine-tune the preset basic model with knowledge of forestry and grassland, to obtain the fine-tuned model, and to process the standardized data using the fine-tuned model to generate structured feature text. The matching module is used to match the parent configuration map in the preliminary parent spatial configuration result with the mountain and nest map in the standardized data based on the structured feature text using a two-pointer matching algorithm to obtain the matching result; The generation module is used to mark abnormal regions based on the matching results, perform visual fine-tuning and constraint verification on the abnormal regions, obtain fine-tuning results, and integrate the fine-tuning results with the matching results to generate target planting results.

10. A computer device, comprising: The memory and processor contain a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the steps of the intelligent planting method for determining the progeny of forest trees according to any one of claims 1-8.

Images