A plant recognition method based on laser point cloud considering stem shape and leaf inclination angle

Abstract

The application provides a plant recognition method based on a laser point cloud, considering a stem shape and a leaf inclination angle, comprising: identifying a kind of plant according to a plant structure phenotype parameter; analyzing a state of the plant according to a recognition result of the plant; monitoring a health degree of the plant based on the state of the plant, specifically comprising: establishing a plant health condition recognition model; predicting a growth trend of the plant based on the health degree of the plant; designing a preliminary growth intervention strategy of the plant based on the growth trend of the plant, specifically comprising: establishing a plant classic modeling library, carrying out reverse engineering on a plant classic modeling, determining a modeling scheme of the plant, and generating a preliminary growth intervention strategy of the plant; judging ornamental of the plant according to the health degree of the plant and the preliminary growth intervention strategy of the plant; and determining a regrowth intervention strategy of the plant based on the ornamental of the plant. The application makes appropriate artificial intervention on the plant, is beneficial to promoting healthy growth of the plant, and enhances ornamental value through design of the plant modeling.

Description

[Technical Field]

[0001] This invention relates to the field of information technology, and in particular to a plant identification method based on laser point clouds that takes into account stem shape and leaf tilt angle. [Background Technology]

[0002] Large-scale flowering plants and botanical gardens cultivate many valuable and precious plants. These plants have their own growth patterns, but their ornamental value is also a characteristic that needs to be protected and assessed. Three-dimensional laser scanning can instantly and efficiently acquire point cloud information, offering unique advantages in obtaining plant structural phenotypic features. Point cloud information of plants is collected using a laser scanner and digitized into a three-dimensional model to accurately calculate structural phenotypic parameters. During the plant's growth period, three-dimensional laser modeling automatically identifies plant morphology, monitors its health, and predicts its growth trend. Based on this growth trend, preliminary growth interventions are implemented to help the plant achieve better styling design requirements. Simultaneously, the plant's ornamental value is assessed, and further interventions are conducted based on the assessment results. By implementing specific further intervention measures, the most valuable and economically beneficial plants can be obtained from large-scale plant cultivation in botanical gardens. [Summary of the Invention]

[0003] This invention provides a plant identification method based on laser point clouds that takes into account stem shape and leaf tilt angle, mainly including:

[0004] Based on laser point cloud computing, plant structural phenotypic parameters are calculated; plant species are identified based on these parameters; plant status is analyzed based on the identification results; plant health is monitored based on the plant status; plant growth trends are predicted based on the plant health; preliminary growth intervention strategies are designed based on the plant growth trends; the ornamental value of the plant is assessed based on its health and the preliminary growth intervention strategies; and plant regeneration intervention strategies are determined based on the plant's ornamental value.

[0005] Further optionally, the plant structural phenotypic parameters based on laser point cloud computing include:

[0006] The plant structural phenotypic parameters refer to the leaf tilt angle distribution of the plant. High-precision point cloud information of each plant is acquired through 3D laser scanning and a high-precision camera. The point cloud information is then preprocessed, and a 3D model of the plant's stems, twigs, and leaves is obtained using a spatially irregular triangular mesh. Finally, the format of each triangular facet in the 3D model is parsed to obtain the normal vector data and vertex coordinate data of the triangular facet. Based on the normal vector data and vertex coordinate data of the triangular facets, the leaf tilt angle distribution of the plant is calculated, thus obtaining the plant structural phenotypic parameters. This process includes: acquiring high-precision point cloud information of the plant; establishing a 3D model of the plant's stems, twigs, and leaves; and calculating the leaf tilt angle distribution of the plant.

[0007] The high-precision point cloud information of the collected plants specifically includes:

[0008] Starting from the plant's growth and development stage, high-precision point cloud information of the plant is collected. This collection includes: periodically scanning the plant surface with a 3D laser scanner to receive the reflected signal from each point on the plant; then converting this signal into directly identifiable data using a converter; obtaining the 3D coordinates of each point within a preset range through software processing; then acquiring a color image of the plant using a high-precision camera; and assigning RGB color information to each point corresponding to the 3D coordinates in the color image. All recorded points of the plant constitute the plant's point cloud, and the RGB color information of all recorded points constitutes the plant's high-precision point cloud information.

[0009] The establishment of a three-dimensional model of the plant's stems, twigs, and leaves specifically includes:

[0010] A 3D model of the plant's stems, twigs, and leaves is established based on the plant's point cloud information. First, the original point cloud information is optimized through preprocessing; this preprocessing includes point cloud stitching and point cloud denoising. Point cloud stitching involves aligning points exceeding a preset range of 3D coordinates according to a preset point value before stitching the point cloud information together. Point cloud denoising refers to filtering out noise information using a smoothing low-pass filter. Noise points are points in the neighborhood where the variance of the point cloud information exceeds a first preset threshold of the average variance of the neighboring point cloud information. Then, a preset number of points for modeling are obtained from the point cloud through random point sampling. Finally, a TIN triangle with the points used for modeling as vertices is constructed using a spatially irregular triangular mesh to obtain the 3D model of the plant's stems, twigs, and leaves.

[0011] The calculation of the leaf tilt angle distribution of the plant specifically includes:

[0012] By parsing the format of each triangular facet in the irregular triangular network, the normal vector data and vertex coordinate data of the triangular facet are obtained. Based on the normal vector data and vertex coordinate data of the triangular facet, the leaf tilt angle distribution of the plant's three-dimensional model is calculated. The leaf tilt angle refers to the angle between the normal of the triangular facet and the vertical direction; the larger the value, the more upright the leaf. The leaf tilt angle distribution refers to the probability density of the leaf tilt angle. The calculation method includes: calculating the angle between the normal of each triangular facet and the vertical direction, and using the reciprocal of the area of ​​the triangular facet as the weight, taking the angle between the normal of all triangular facets and the vertical direction for weighting, and accumulating the probability density of the tilt angles of all triangular facets within the corresponding leaf tilt angle distribution interval, which is the leaf tilt angle distribution.

[0013] Further optionally, the identification of plant species based on plant structural phenotypic parameters includes:

[0014] Based on plant structural phenotypic parameters, the plant species are identified using a K-means clustering model. This identification process includes: establishing a plant structural phenotypic parameter template library; randomly sampling to obtain plant structural phenotypic parameters for different plant species, which are then entered into the template library as typical template parameters for each species; collecting point cloud information of the plant to be tested, establishing a three-dimensional model of the stems, branches, and leaves, and calculating the plant structural phenotypic parameters; matching the plant structural phenotypic parameters with the template library, calculating the similarity between the parameters and typical template parameters, and clustering plant structural phenotypic parameters with similarity exceeding a second preset threshold; finally, identifying the plant species as the model's output.

[0015] Further, optionally, the analysis of the plant's state based on the plant identification results includes:

[0016] Based on the plant species identification results, the plant status is analyzed. The plant status includes plant height, leaf morphology, stem shape, and petal characteristics. Leaf morphology includes leaf length, leaf width, and leaf shape; petal characteristics include flower diameter, flowering period, and lifespan. The Apriori association rule algorithm is used to mine the association relationships between plant species corresponding to their status. The association rule algorithm process is as follows: Through iterative input, all frequent itemsets of morphological factors in the plant database are retrieved, i.e., itemsets with support not lower than a preset third threshold. Support is the percentage of each factor itemset appearing in the database relative to the total number of its respective itemsets. Then, rules satisfying the confidence level are constructed using the frequent itemsets. Confidence is the percentage of each factor's total itemsets relative to all itemsets in the database. By comparing support and confidence, the strength of the association with the plant species is determined, and a plant species association network diagram is drawn based on the association rule algorithm. The plant species association network diagram indicates the plant status.

[0017] Further optionally, the monitoring of plant health based on plant condition includes:

[0018] Based on the plant's condition, a plant health status identification model is used to monitor the plant's health level. During the plant's growth period, a plant health status evaluation function is constructed: NH = W*W1 + N*W2 + I*W3. Here, NH represents the plant's unhealthy index during the growth period; a higher NH value indicates a poorer plant health. W, N, and I represent unhealthy conditions, including: W represents the plant's water deficiency during the growth period (W = 1 if the plant health status identification model determines water deficiency, otherwise W = 0); N represents the plant's nutrient deficiency during the growth period. The plant's health status is assessed using a plant health status identification model. If the model determines that the plant is malnourished, then N = 1; otherwise, N = 0. I represents the plant's status during its growth period, indicating the presence of pests or diseases. If the model determines that the plant is malnourished, then I = 1; otherwise, I = 0. W1, W2, and W3 represent the percentage of time the corresponding malnourished status lasts within the total growth period. Based on the plant's health evaluation function, an unhealthy index is calculated for the plant during its growth period to monitor its health status. This includes establishing a plant health status identification model.

[0019] The establishment of the plant health status identification model specifically includes:

[0020] A plant health status identification model is established using deep learning by randomly collecting more than a fourth preset threshold number of plants. This model includes: using plant morphology as both a training and testing set; extracting features from the training set (including plant height, leaf morphology, stem shape, and flower morphology); preprocessing the feature data (labeling the plant's health status based on point cloud information); and using these labels to train the model to determine the relationship between plant health status and plant morphology, recording the results in a preset format. Finally, the model parameters are continuously adjusted using test set data to improve the accuracy of the plant health status identification model.

[0021] Further optionally, the prediction of plant growth trends based on plant health includes:

[0022] By recording the health status of plants, the growth trend of plants is predicted. The plant growth trend includes the number of leaves, the direction of branch extension, and the number of flowers. Historical plant growth trends are obtained by crawling large amounts of plant data. A plant growth trend prediction model is then used to predict the plant growth trend. The plant growth trend prediction model is as follows: Historical plant growth trends are used as sample data. K-means clustering algorithm is used to analyze the data and select multiple centroids based on the historical plant growth trend types. The Euclidean distance from each sample to the centroid is calculated, and each sample is assigned to the nearest centroid. The average value of the sample data for each category is then used to calculate the new centroid for each category. Finally, the sample classification and centroid calculation are iteratively performed until the clustering converges and the old and new centroids no longer change. The plant health status and corresponding health status are input as parameters into the plant growth trend prediction model. The plant health status is classified into the type most similar to it, and the current plant growth trend is determined by the basic plant growth trend of that type of plant health status.

[0023] Further, optionally, the design of preliminary plant growth intervention strategies based on plant growth trends includes:

[0024] Based on the plant's growth trend, a preliminary plant growth intervention strategy is designed. This design includes: establishing a classic plant styling library to collect classic plant shapes; then performing reverse engineering on these classic shapes to derive key methods for cultivating them; calculating the similarity between the plant's 3D model and its growth trend and the classic plant styling library to determine the plant's styling scheme; and then, combining the plant's growth trend prediction results with the styling scheme, generating a preliminary plant growth intervention strategy. This includes: establishing a classic plant styling library; performing reverse engineering on classic plant shapes; determining the plant's styling scheme; and generating the preliminary plant growth intervention strategy.

[0025] The establishment of the classic plant styling library specifically includes:

[0026] The classic plant styling library refers to a collection of classic plant styling examples. Establishing this library includes: crawling the internet to retrieve the names and plant compositions of classic plant styling examples; creating entries for each classic plant styling example based on its name; the plant composition of each classic plant styling example includes: plant type, form, and quantity; entering the plant composition of each classic plant styling example into the corresponding entry; and arranging the entries in a preset order to complete the establishment of the classic plant styling library.

[0027] The reverse engineering of classic plant shapes specifically includes:

[0028] Reverse engineering is performed on classic plant shapes to analyze key methods for cultivating classic plant shapes. The reverse engineering includes: obtaining the cultivation process for the classic plant shape; comparing the cultivation process for the classic plant shape with the cultivation process for unshaped plants of the same species to identify methods that differ from the cultivation process for unshaped plants of the same species; and then recording these differing methods as key methods for cultivating classic plant shapes under the corresponding entry in the classic plant shape database.

[0029] The determination of the plant's styling scheme specifically includes:

[0030] Determining the plant's styling scheme includes: calculating the similarity between the plant's 3D model and the 3D models of classic plant styles of the same type in a classic plant style library; using at least one classic plant style with a similarity exceeding a fifth preset threshold as the plant's styling template; when there is only one styling template, the styling template is determined as the plant's styling scheme; or, when there are multiple styling templates, the styling template with the highest similarity between the plant's growth trend and the plant's morphology in the classic plant styles is selected and determined as the plant's styling scheme.

[0031] The initial growth intervention strategy for the generated plants specifically includes:

[0032] Based on the plant's styling plan, a preliminary plant growth intervention strategy is generated. This strategy includes a data processing module, a neural network model construction module, and a preliminary plant growth intervention strategy generation module. Based on the plant's growth trend, preliminary growth intervention is implemented for plants whose morphological differences from the styling plan exceed a preset level. This includes: determining whether preliminary growth intervention is needed based on the predicted growth trend; if the predicted growth trend deviates from the plant's morphological shape by more than a preset percentage, preliminary growth intervention is determined, and corresponding aspects of the deviation are addressed for those deviations. Preliminary intervention; the preliminary intervention includes: water intervention, fertilizer intervention, temperature intervention, light intervention, and pest and disease control intervention; the data processing module obtains the plant growth trend prediction results data, and uses a deep learning RNN algorithm to analyze the plant growth trend prediction results and the numerical values ​​of the plant morphology that deviate from the plant shaping scheme, determine the corresponding aspects, and construct morphology corresponding aspect vectors; the neural network model construction module uses the morphology corresponding aspect vectors to construct and train a convolutional neural network model; the plant preliminary growth intervention strategy generation module uses the trained convolutional neural model to perform preliminary intervention on specific aspects, thereby realizing preliminary plant growth intervention.

[0033] Further, optionally, the determination of the plant's ornamental value based on the plant's health status and initial growth intervention strategies includes:

[0034] The ornamental value of the plant is judged based on its health status and the initial growth intervention strategy. This judgment includes: establishing an ornamental value evaluation system, PA = (HD + SP) / 2; where PA represents the ornamental value evaluation index; HD represents the plant's health status evaluation index (HD = 1 if the unhealthy index in the health status evaluation function is below a preset threshold, otherwise HD = 0); SP represents the plant's shape evaluation index (SP = 1 if the probability of the plant's shape appearing in historical garden designs is greater than a sixth preset threshold, otherwise SP = 0); analyzing the ornamental value of the plant using a BP neural network; and determining training and testing samples based on the plant's health status evaluation function and shape scheme. The training samples are used for neural network training, and the testing samples are used to detect the relative error between the actual and predicted values.

[0035] Further optionally, the determination of plant regrowth intervention strategies based on plant ornamental value includes:

[0036] Based on the plant's ornamental value, a plant regrowth intervention strategy is determined. This strategy includes increasing or decreasing initial intervention, changing the styling scheme, and replacing the plant species. The determination of the plant regrowth intervention strategy involves: using the initial plant regrowth intervention strategy as a sample, constructing a single-class SVM model based on influencing factors. These influencing factors include the plant's health evaluation index and styling evaluation index within the plant ornamental value evaluation system. Then, using an activation function, the distance range from the intermediate output sample of the single-class SVM model to the center of the hypersphere is mapped to [0,1]. The mapping result represents the probability of performing plant regrowth intervention. If the mapping result is greater than a seventh preset threshold, plant regrowth intervention is performed; otherwise, the initial plant regrowth intervention is maintained. Finally, based on the change in the probability of performing plant regrowth intervention, the plant regrowth intervention strategy is determined.

[0037] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0038] Appropriate artificial intervention in plant cultivation can promote healthy growth and maintain a good shape. Designing plant shapes can create unique scenic lines, thereby enhancing the ornamental value of the plants. At the same time, strengthening the construction of garden landscapes can also make gardens meet people's aesthetic needs, making people's lives more comfortable, improving their visual environment, and allowing them to cultivate their minds in a healthy, comfortable, and tranquil setting, bringing better economic and ecological benefits to society. [Attached Image Description]

[0039] Figure 1 This is a flowchart of a plant identification method based on laser point cloud that takes into account stem shape and leaf tilt angle according to the present invention.

[0040] Figure 2 This is another flowchart illustrating the initial plant growth intervention strategy designed based on the plant's growth trend, as per the present invention.

Detailed Implementation Methods

[0041] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0042] Figure 1 This is a flowchart illustrating a plant identification method based on laser point clouds that takes into account stem shape and leaf tilt angle, according to the present invention. Figure 1 As shown, this embodiment of a plant identification method based on laser point clouds that takes into account stem shape and leaf tilt angle may specifically include:

[0043] Step 101: Compute plant structural phenotypic parameters based on laser point cloud computing.

[0044] The plant structural phenotypic parameters refer to the leaf tilt angle distribution of the plant. High-precision point cloud information of each plant is acquired through 3D laser scanning and a high-precision camera. The point cloud information of the plant is then preprocessed, and a 3D model of the plant stems, branches, and leaves is obtained using a spatially irregular triangular mesh. Finally, the format of each triangular facet in the 3D model is parsed to obtain the normal vector data and vertex coordinate data of the triangular facet. Based on the normal vector data and vertex coordinate data of the triangular facet, the leaf tilt angle distribution of the plant is calculated, thus obtaining the plant structural phenotypic parameters.

[0045] Collect high-precision point cloud information of plants.

[0046] Starting from the plant's growth and development stage, high-precision point cloud information of the plant is collected. This collection includes: periodically scanning the plant surface with a 3D laser scanner to receive the reflected signal from each point on the plant; then converting this signal into directly identifiable data using a converter; obtaining the 3D coordinates of each point within a preset range through software processing; then acquiring a color image of the plant using a high-precision camera; and assigning RGB color information to each point corresponding to the 3D coordinates in the color image. All recorded points of the plant constitute the plant's point cloud, and the RGB color information of all recorded points constitutes the plant's high-precision point cloud information.

[0047] Create a three-dimensional model of the plant's stems, twigs, and leaves.

[0048] A 3D model of the plant's stems, twigs, and leaves is established based on the plant's point cloud information. First, the original point cloud information is optimized through preprocessing; this preprocessing includes point cloud stitching and point cloud denoising. Point cloud stitching involves aligning points exceeding a preset range of 3D coordinates according to a preset point value before stitching the point cloud information together. Point cloud denoising refers to filtering out noise information using a smoothing low-pass filter. Noise points are points in the neighborhood where the variance of the point cloud information exceeds a first preset threshold of the average variance of the neighboring point cloud information. Then, a preset number of points for modeling are obtained from the point cloud through random point sampling. Finally, a TIN triangle with the points used for modeling as vertices is constructed using a spatially irregular triangular mesh to obtain the 3D model of the plant's stems, twigs, and leaves. For example, during the cultivation of flowers at the flower cultivation base, four plants with different morphological characteristics were selected: peace lily (uniform type), dragon's beard orchid (slanting type), pothos (spherical type), and bird's nest fern (slanting type). In the experimental field of Huazhong Agricultural University in Wuhan, Hubei Province, the laser point clouds of the four experimental objects were measured using a Mantis-Vision F5 handheld 3D laser scanner and an ASDFieldSpec4 portable ground object spectrometer to obtain the corresponding high-precision point cloud information. Preprocessing was performed in Geomagic Studio software to optimize the original scan data. A spatially irregular triangular mesh was used to approximate the data model of the 3D solid surface. The established 3D model was stored in STL file format. A complete STL file records the normal vector data and vertex coordinate data of all triangular faces that make up the solid model. By parsing the format of each triangular face, the leaf tilt angle distribution of each 3D model was calculated.

[0049] Calculate the leaf tilt angle distribution of the plant.

[0050] By parsing the format of each triangular facet in the irregular triangular network, the normal vector data and vertex coordinate data of the triangular facet are obtained. Based on the normal vector data and vertex coordinate data of the triangular facet, the leaf tilt angle distribution of the plant's three-dimensional model is calculated. The leaf tilt angle refers to the angle between the normal of the triangular facet and the vertical direction; the larger the value, the more upright the leaf. The leaf tilt angle distribution refers to the probability density of the leaf tilt angle. The calculation method includes: calculating the angle between the normal of each triangular facet and the vertical direction, and using the reciprocal of the area of ​​the triangular facet as the weight, taking the angle between the normal of all triangular facets and the vertical direction for weighting, and accumulating the probability density of the tilt angles of all triangular facets within the corresponding leaf tilt angle distribution interval, which is the leaf tilt angle distribution. For example, a 3D laser scanner scans the entire surface of a *Gymnocalycium mihanovichii* plant every other day, receiving the reflected waveform of the plant's outline. A converter decodes the waveform, converting it into *Gymnocalycium mihanovichii* outline data. Next, a data processing system organizes the outline data, outputting the 3D coordinates (x, y, z) of each point in the outline, ranging from (0, 0, 0) to (1024, 1024, 1024). Then, a high-precision camera captures images of the *Gymnocalycium mihanovichii*, creating a color image and assigning each point in the outline its corresponding RGB color information (R, G, B) at its 3D coordinate pixel position. The RGB color information ranges from (0, 0, 0) to (255, 255, 255), with larger values ​​indicating stronger color characteristics. By acquiring the RGB color information of all points, high-precision point cloud information of the *Gymnocalycium mihanovichii* is obtained.

[0051] Step 102: Identify the plant species based on plant structural phenotypic parameters.

[0052] Based on plant structural phenotypic parameters, the plant species are identified using a K-means clustering model. This identification process includes: establishing a plant structural phenotypic parameter template library; randomly sampling to obtain plant structural phenotypic parameters for different plant species, which are then entered into the template library as typical template parameters for each species; collecting point cloud information of the plant to be tested, establishing a three-dimensional model of the stems, branches, and leaves, and calculating the plant structural phenotypic parameters; matching the plant structural phenotypic parameters with the template library, calculating the similarity between the parameters and typical template parameters, and clustering plant structural phenotypic parameters with similarity exceeding a second preset threshold; finally, identifying the plant species as the model's output. For example, the 3D coordinates (x, y, z) of the *Gynostemma pentaphyllum* outline range from (0, 0, 0) to (1024, 1024, 1024). A 3D laser scanner scans the area behind and to the right of the *Gynostemma pentaphyllum* at 45° angles, and a high-precision camera acquires high-precision point cloud information of the *Gynostemma pentaphyllum*. In the two point cloud areas, some point cloud information behind the *Gynostemma pentaphyllum* overlaps. After aligning the two areas in 3D coordinates according to the midpoint (512, 512, 512) of their coordinate systems, the two areas are... The point cloud information of the corresponding points is connected and spliced ​​to obtain a merged set of point cloud information of the non-overlapping parts; the point cloud information processing system monitors that in the neighborhood of (255,243,265), the average variance of point cloud information (R,G,B) is (2,3,3), while the variance of point cloud information of (255,243,266) is (3,4,4), so this point is a noisy point; then, 100 points are randomly sampled for modeling and TIN triangles are constructed to obtain a three-dimensional model of the stem and leaves of the dragon's beard orchid.

[0053] Step 103: Analyze the plant's condition based on the plant identification results.

[0054] Based on the plant species identification results, the plant status is analyzed. The plant status includes plant height, leaf morphology, stem shape, and petal characteristics. Leaf morphology includes leaf length, leaf width, and leaf shape; petal characteristics include flower diameter, flowering period, and lifespan. The Apriori association rule algorithm is used to mine the association relationships between plant species corresponding to their status. The algorithm process is as follows: Through iterative input, all frequent itemsets of morphological factors in the plant database are retrieved, i.e., itemsets with support not lower than a preset third threshold. Support is the percentage of each factor itemset appearing in the database relative to the total number of its respective itemsets. Then, rules satisfying a certain confidence level are constructed using these frequent itemsets. Confidence is the percentage of each factor's total itemsets relative to all itemsets in the database. By comparing support and confidence, the strength of the association with the plant species is determined, and a plant species association network diagram is drawn based on the association rule algorithm. This plant species association network diagram indicates the plant's status. For example, the analytical format of the triangular facets of the irregular triangular network of *Gymnocalycium mihanovichii* is {the angle between the leaf normal and the vertical direction, the 3D coordinates of vertex 1, the 3D coordinates of vertex 2, and the 3D coordinates of vertex 3}; the analytical result of a triangular facet is {35°, (20,20,20), (22.20,20), (21,22,20)}, and its area is 2; the product of the reciprocal of its area and its included angle is 35 / 2, which is accumulated in the interval [30°, 40°] of the leaf tilt angle distribution; the next triangular facet is calculated until all triangular facets are calculated, and the leaf tilt angle distribution of *Gymnocalycium mihanovichii* is obtained.

[0055] Step 104: Monitor the health of the plant based on its condition.

[0056] Based on the plant's condition, a plant health status identification model is used to monitor the plant's health level. During the plant's growth period, a plant health status evaluation function is constructed: NH = W*W1 + N*W2 + I*W3. Here, NH represents the plant's unhealthy index during the growth period; a higher NH value indicates a worse health status. W, N, and I represent unhealthy conditions, including: W represents water deficiency during the growth period (W = 1 if the plant health status identification model determines water deficiency, otherwise W = 0); N represents nutrient deficiency during the growth period (N = 1 if nutrient deficiency is determined, otherwise N = 0); and I represents pest and disease damage during the growth period (I = 1 if pest and disease are determined, otherwise I = 0). W1, W2, and W3 represent the percentage of time the corresponding unhealthy condition persists within the total plant growth period. Based on the plant health evaluation function, the unhealthy index of the plant during its growth period is calculated to monitor the plant's health. For example, by collecting point cloud information of a plant to be tested in a botanical garden, and establishing a three-dimensional model of the stem and leaves, the probability densities of the structural phenotypic parameters in the intervals [30°, 40°], [40°, 50°], and [50°, 60°] are calculated to be 0.08, 0.07, and 0.09, respectively. This data is then input into a K-means clustering model, and plant species with a similarity exceeding 80% to typical template parameters in the plant structural phenotypic parameter template library are selected as outputs. In the plant structural phenotypic parameter template library, the probability densities of the leaf tilt angles of *Gymnocalycium mihanovichii* in the intervals [30°, 40°], [40°, 50°], and [50°, 60°] are 0.07, 0.07, and 0.09, respectively. Since the similarity exceeds 80%, the result analysis indicates that the plant species is *Gymnocalycium mihanovichii*.

[0057] Establish a plant health status identification model.

[0058] A plant health status identification model is established using deep learning by randomly collecting more than a fourth preset threshold number of plants. This model includes: using plant morphology as both a training and testing set; extracting features from the training set (including plant height, leaf morphology, stem shape, and flower morphology); preprocessing the feature data (labeling the plant's health status based on point cloud information); and using these labels to train the model to determine the relationship between plant health status and plant morphology, recording the results in a preset format. Finally, the model parameters are continuously adjusted using test set data to improve the accuracy of the plant health status identification model. For example, if a plant in a garden cultivation base is identified as a daffodil, the Apriori algorithm is used to mine the associations between the plant species corresponding to the states of all plants. The first association rule is "Daffodil => plant height 20-30cm, support = 0.789, confidence = 0.933"; the second association rule is "Daffodil => leaf length 30-50cm, leaf width 1-5cm, relatively wide and flat leaves, support = 0.746, confidence = 0.921"; the third association rule is "Daffodil => nearly spherical stem, bulb, support = 0.824, confidence = 0.896"; and the fourth association rule is "Daffodil => flower diameter 5-5.5cm". Flowering period in spring, lifespan of flowers 15-20 days, support = 0.846, confidence = 0.916; This indicates that the morphology of a daffodil plant is likely to be 20-30cm tall, with leaves 30-50cm long, 1-5cm wide, relatively wide and flat leaves, a nearly spherical stem, a bulb, and flowers 5-5.5cm in diameter, flowering period in spring, and a lifespan of 15-20 days. In other words, there is a correlation between the plant species "daffodil" and the plant state characterized by "20-30cm" height, "30-50cm long, 1-5cm wide, relatively wide and flat leaves," "nearly spherical stem, bulb," and "5-5.5cm flower diameter, spring flowering period, and 15-20 days." Summarizing multiple association rules and drawing a species association network diagram for the plant shows that the corresponding association results on the species association network diagram represent the state of the plant.

[0059] Step 105: Based on the health of the plant, predict the plant's growth trend.

[0060] By recording the health status of plants, the growth trend of plants is predicted. The plant growth trend includes the number of leaves, the direction of branch extension, and the number of flowers. Historical plant growth trends are obtained by crawling large amounts of plant data. A plant growth trend prediction model is then used to predict the plant's growth trend. The plant growth trend prediction model works as follows: Historical plant growth trends are used as sample data. K-means clustering is used to analyze the data and select multiple centroids based on the historical plant growth trend types. The Euclidean distance from each sample to the centroid is calculated, and each sample is assigned to the nearest centroid. The average value of the sample data for each category is then used to calculate a new centroid for each category. Finally, the sample classification and centroid calculation are iteratively performed until the clustering converges and the old and new centroids no longer change. The plant's health status and corresponding health condition are input as parameters into the plant growth trend prediction model. The plant's health status is classified into the most similar type, and the current plant's growth trend is determined by the basic plant growth trend of that type of plant. For example, a plant health status identification model determines that a *Dracaena sanderiana* plant is suffering from both water deficiency and pest / disease damage; specifically, it has been water deficient for 20 days and affected by pests / diseases for 30 days, with a total growing season of 40 days. Based on the plant health evaluation function, the *Dracaena sanderiana*'s unhealthy index during its growing season is:

[0061] NH=1×0.5+0+1×0.75=1.25.

[0062] Step 106: Based on the plant's growth trend, design a preliminary plant growth intervention strategy.

[0063] Based on the plant's growth trend, a preliminary plant growth intervention strategy is designed. This strategy includes: establishing a classic plant styling library to collect classic plant shapes; then performing reverse engineering on these classic shapes to derive key methods for cultivating them; calculating the similarity between the plant's 3D model and its growth trend and the classic plant styling library to determine the desired styling scheme; and finally, combining the plant's growth trend prediction results with the chosen styling scheme to generate a preliminary plant growth intervention strategy. For example, a plant health status identification model was established by randomly collecting over 100 daffodil plants. Through model analysis, plants with a height "below 20cm" were identified as lacking nutrients; leaves with a length less than 30cm and a width less than 1cm were identified as suffering from pests or diseases; and flowers with a diameter less than 5cm were identified as lacking water. Based on the collected point cloud information, the plant heights of daffodils A, B, C, and D were analyzed to be 19cm, 21cm, 21cm, and 25cm, respectively; leaf length × width were 31×1.1, 28×0.8, 36×3, and 43×2.6, respectively; and flower diameters were 5.1, 5.2, 4.6, and 5.5, respectively. The training results of the plant health status identification model were recorded in the format {plant number, health status}. The final results were: {daffodil A, lacking nutrients}, {daffodil B, suffering from pests or diseases}, {daffodil C, lacking water}, and {daffodil D, normal status}.

[0064] Establish a library of classic plant shapes.

[0065] The classic plant styling library refers to a collection of classic plant styling examples. Establishing this library includes: crawling the internet to retrieve the names and plant compositions of classic plant styling examples; creating entries for each classic plant styling example based on its name; the plant composition of each classic plant styling example includes: plant type, form, and quantity; entering the plant composition of each classic plant styling example into the corresponding entry; and arranging the entries in a preset order to complete the establishment of the classic plant styling library. For example, a certain *Dracaena sanderiana* plant is in poor health, with an "unhealthy index of 1.25," corresponding to health conditions including "50% water deficiency during the total growth period" and "75% susceptibility to pests and diseases during the total growth period." These three parameters are used as inputs into the plant's growth trend prediction model. K-means clustering analysis categorizes the plant as "insufficient human care." Finally, considering that the typical growth trend for this type is "approximately 70% of the normal number of leaves, branches extending at approximately 30° off the normal direction, and approximately 60% of the normal number of flowers," the predicted growth trend for this *Dracaena sanderiana* plant is "approximately 70% of the normal number of leaves, branches extending at approximately 30° off the normal direction, and approximately 60% of the normal number of flowers."

[0066] Reverse engineering classic plant shapes.

[0067] Reverse engineering is performed on classic plant shapes to analyze key methods for cultivating classic plant shapes. The reverse engineering includes: obtaining the cultivation process for the classic plant shape; comparing the cultivation process for the classic plant shape with the cultivation process for unshaped plants of the same species to identify methods that differ from the cultivation process for unshaped plants of the same species; and then recording these differing methods as key methods for cultivating classic plant shapes under the corresponding entry in the classic plant shape database. For example, three classic plant shapes were collected: a hexagonal pavilion shape, an elephant shape, and a "Happy New Year" shape. The hexagonal pavilion shape requires six small-leaved privet trees approximately 4 meters tall with straight branches and good leaf density. Reverse engineering of the hexagonal pavilion shape revealed that the key method is the use of a low-concentration paclobutrazol during the vigorous growth period. Similarity calculations were performed between the 3D model and production trend of the small-leaved privet in the cultivation base and the morphology of the small-leaved privet required for the classic shapes. The highest similarity was found with the hexagonal pavilion shape, so this batch of small-leaved privet was selected for the hexagonal pavilion shape. Under the premise of maintaining normal growth and good health, preliminary interventions were implemented during the subsequent growth period, including watering and fertilization, to help the small-leaved privet grow into the shape required for the hexagonal pavilion shape.

[0068] Determine the plant's design scheme.

[0069] Determining the plant's styling scheme includes: calculating the similarity between the plant's 3D model and the 3D models of classic plant styles of the same type in a classic plant style library; using at least one classic plant style with a similarity exceeding a fifth preset threshold as the plant's styling template; when there is only one styling template, the styling template is determined as the plant's styling scheme; or, when there are multiple styling templates, the styling template with the highest similarity between the plant's growth trend and the plant's morphology in the classic plant styles is selected and determined as the plant's styling scheme. For example, information was collected from the internet about hexagonal pavilions, elephants, and the words "Happy New Year." First, major categories were created for "Architectural Forms," ​​"Animal Forms," ​​and "Text Forms." Then, the three forms were placed into their respective major categories and named "Hexagonal Pavilion," "Elephant," and "Happy New Year" subcategories. Further research revealed that the hexagonal pavilion design required six small-leaved privet trees, each approximately 4 meters tall, with straight branches and dense foliage. These components were then entered into the "Hexagonal Pavilion" subcategory. After completing the search and entry of all other forms, a classic plant form library was established.

[0070] Develop preliminary plant growth intervention strategies.

[0071] Based on the plant's styling plan, a preliminary plant growth intervention strategy is generated. This strategy includes a data processing module, a neural network model construction module, and a preliminary plant growth intervention strategy generation module. Based on the plant's growth trend, preliminary growth intervention is implemented for plants whose morphological differences from the styling plan exceed a preset level. This includes: determining whether preliminary growth intervention is needed based on the predicted growth trend; if the predicted growth trend deviates from the plant's morphological shape by more than a preset percentage, preliminary growth intervention is determined, and corresponding aspects of the deviation are addressed for those deviations. Preliminary intervention; the preliminary intervention includes: water intervention, fertilizer intervention, temperature intervention, light intervention, and pest and disease control intervention; the data processing module obtains the plant growth trend prediction results data, and uses a deep learning RNN algorithm to analyze the plant growth trend prediction results and the numerical values ​​of the plant morphology that deviate from the plant shaping scheme, determine the corresponding aspects, and construct morphology corresponding aspect vectors; the neural network model construction module uses the morphology corresponding aspect vectors to construct and train a convolutional neural network model; the plant preliminary growth intervention strategy generation module uses the trained convolutional neural model to perform preliminary intervention on specific aspects, thereby realizing preliminary plant growth intervention. For example, data crawled from the internet showed that when cultivating Ligustrum lucidum for hexagonal pavilion shapes, nitrogen fertilizer was used during the vigorous growth period, along with a diluted concentration of paclobutrazol, and phosphorus and potassium fertilizer was used during the flowering and fruiting period. However, when cultivating Ligustrum lucidum for non-shaped purposes, nitrogen fertilizer was used during the vigorous growth period, and phosphorus and potassium fertilizer was sufficient during the flowering and fruiting period. By comparing the two, the analysis revealed that the key difference was the use of a diluted concentration of paclobutrazol during the vigorous growth period. Finally, this key method was recorded under the "Hexagonal Pavilion" subcategory.

[0072] Step 107: Determine the ornamental value of the plant based on its health status and the initial plant growth intervention strategy.

[0073] The ornamental value of the plant is judged based on its health status and the initial growth intervention strategy. This judgment includes: establishing an ornamental value evaluation system, PA = (HD + SP) / 2; where PA represents the ornamental value evaluation index; HD represents the plant's health status evaluation index (HD = 1 if the unhealthy index in the health status evaluation function is below a preset threshold, otherwise HD = 0); SP represents the plant's shape evaluation index (SP = 1 if the probability of the plant's shape appearing in historical garden designs is greater than a sixth preset threshold, otherwise SP = 0); analyzing the ornamental value of the plant using a BP neural network; and determining training and testing samples based on the plant's health status evaluation function and shape scheme. The training samples are used for neural network training, and the testing samples are used to detect the relative error between the actual and predicted values. For example, when designing a styling scheme for Ligustrum lucidum, the plant species in the "Hexagonal Pavilion," "Elephant," and "Happy New Year" subcategories of the classic plant styling library include Ligustrum lucidum. The similarity between the current 3D model of Ligustrum lucidum and the 3D models of Ligustrum lucidum in the three subcategories is calculated. The calculated similarity is 80%, 80%, and 70%, respectively. Further selection is made based on the current growth trend of Ligustrum lucidum. The current growth trend of Ligustrum lucidum is "the number of leaves is about 70% of the normal state," the morphology of the "Hexagonal Pavilion" styling Ligustrum lucidum is "the number of leaves is about 68% of the normal state," and the morphology of the "Elephant" styling Ligustrum lucidum is "the number of leaves is about 90% of the normal state." Therefore, the "Hexagonal Pavilion" styling is determined to be the current styling scheme for Ligustrum lucidum.

[0074] Step 108: Determine the plant regrowth intervention strategy based on the plant's ornamental value.

[0075] Based on the plant's ornamental value, a plant regrowth intervention strategy is determined. This strategy includes increasing or decreasing initial intervention, changing the styling scheme, and replacing the plant species. The determination of the plant regrowth intervention strategy involves: using the initial plant regrowth intervention strategy as a sample, constructing a single-class SVM model based on influencing factors. These influencing factors include the plant's health evaluation index and styling evaluation index within the plant ornamental value evaluation system. Then, using an activation function, the distance range from the intermediate output sample of the single-class SVM model to the center of the hypersphere is mapped to [0,1]. The mapping result represents the probability of performing plant regrowth intervention. If the mapping result is greater than a seventh preset threshold, plant regrowth intervention is performed; otherwise, the initial plant regrowth intervention is maintained. Finally, based on the change in the probability of performing plant regrowth intervention, the plant regrowth intervention strategy is determined. For example, the predicted growth trend of *Ligustrum lucidum* is "the number of leaves is about 60% of the normal state, the branches extend about 32° away from the normal direction, and the number of flowers is about 80% of the normal state"; the morphology of the "Hexagonal Pavilion" styling scheme is "the number of leaves is about 70% of the normal state, the branches extend about 30° away from the normal direction, and the number of flowers is about 70% of the normal state"; if the difference between the two deviates from the styling scheme values ​​by 10%, preliminary intervention is required, and intervention is applied to both "number of leaves" and "number of flowers"; using the RNN algorithm to analyze the corresponding aspects of *Ligustrum lucidum* morphology, a morphology-corresponding aspect vector SC = {c1,c2,c3,…,cn} is constructed, and the result is SC = {lack of leaf quantity, lack of flower quantity}; then, a convolutional neural network is constructed using the growth trend and morphology-corresponding aspect vector of *Ligustrum lucidum* and the neural network is trained; finally, the trained model is used for preliminary intervention, with water intervention for lack of leaf quantity and fertilizer intervention for lack of flower quantity.

Claims

1. A plant identification method based on laser point clouds that takes into account stem shape and leaf tilt angle, characterized in that, The method includes: The plant structural phenotypic parameters based on laser point cloud computing specifically include: collecting high-precision point cloud information of the plant, establishing a three-dimensional model of the plant's stems, branches, and leaves, and calculating the leaf tilt angle distribution; identifying the plant species based on the plant structural phenotypic parameters; analyzing the plant's state based on the identification results; monitoring the plant's health based on its state, specifically including: establishing a plant health status identification model; predicting the plant's growth trend based on its health status; and designing a preliminary plant growth intervention strategy based on the plant's growth trend, specifically including: establishing a classic plant shape library, reverse engineering classic plant shapes to determine the plant's shape scheme, and generating a preliminary plant growth intervention strategy; judging the plant's ornamental value based on its health status and the preliminary plant growth intervention strategy; and determining a plant regeneration intervention strategy based on the plant's ornamental value. The calculation of the leaf tilt angle distribution of the plant includes: By parsing the format of each triangular facet in the spatial irregular triangular network, the normal vector data and vertex coordinate data of the triangular facet are obtained. Based on the normal vector data and vertex coordinate data of the triangular facet, the leaf tilt angle distribution of the plant's three-dimensional model is calculated. The leaf tilt angle refers to the angle between the normal of the triangular facet and the vertical direction; the larger the value, the more upright the leaf. The leaf tilt angle distribution refers to the probability density of the leaf tilt angle. The calculation method includes: calculating the angle between the normal of each triangular facet and the vertical direction, and using the reciprocal of the area of ​​the triangular facet as the weight, taking the angle between the normal of all triangular facets and the vertical direction for weighting, and accumulating the probability density of the tilt angles of all triangular facets within the corresponding leaf tilt angle distribution interval, which is the leaf tilt angle distribution. The initial growth intervention strategy for plants, based on their growth trends, includes: Based on the plant's growth trend, a preliminary plant growth intervention strategy is designed. This design includes: establishing a classic plant styling library to collect classic plant shapes; then performing reverse engineering on these classic shapes to derive key methods for cultivating them; calculating the similarity between the plant's 3D model and its growth trend and the classic plant styling library to determine the plant's styling scheme; and then, combining the plant's growth trend prediction results with the styling scheme, generating a preliminary plant growth intervention strategy. This includes: establishing a classic plant styling library; performing reverse engineering on classic plant shapes; determining the plant's styling scheme; and generating the preliminary plant growth intervention strategy. The establishment of the classic plant styling library specifically includes: The plant classic shape library refers to a collection of classic plant shapes. Establishing the plant classic shape library includes: crawling the internet to retrieve the names of classic plant shapes and their plant composition; creating entries for each classic plant shape based on its name; the plant composition of each classic plant shape includes the plant type, form, and quantity; entering the plant composition of each classic plant shape into the content of the corresponding entry; and arranging the entries in a preset order to complete the establishment of the plant classic shape library. The reverse engineering of classic plant shapes specifically includes: Reverse engineering is performed on classic plant shapes to analyze key methods for cultivating classic plant shapes. The reverse engineering includes: obtaining the cultivation process for the classic plant shape; comparing the cultivation process for the classic plant shape with the cultivation process for unshaped plants of the same species to identify methods that differ from the cultivation process for unshaped plants of the same species; and then recording these differing methods as key methods for cultivating classic plant shapes under the corresponding entry in the classic plant shape database. The determination of the plant's styling scheme specifically includes: Determining the plant's styling scheme includes: calculating the similarity between the plant's 3D model and the 3D models of classic plant styles of the same type in a classic plant style library; using at least one classic plant style with a similarity exceeding a fifth preset threshold as the plant's styling template; when there is only one styling template, the styling template is determined as the plant's styling scheme; or, when there are multiple styling templates, the styling template with the highest similarity between the plant's growth trend and the plant's morphology in the classic plant styles is selected and determined as the plant's styling scheme. The initial growth intervention strategy for the generated plants specifically includes: Based on the plant's styling plan, a preliminary plant growth intervention strategy is generated. This strategy includes a data processing module, a neural network model construction module, and a preliminary plant growth intervention strategy generation module. Based on the plant's growth trend, preliminary growth intervention is implemented for plants whose morphological differences from the styling plan exceed a preset level. This includes: determining whether preliminary growth intervention is needed based on the predicted growth trend; if the predicted growth trend deviates from the plant's morphological shape by more than a preset percentage, preliminary growth intervention is determined, and corresponding aspects of the deviation are addressed for those where the predicted growth trend deviates from the plant's morphological shape by more than a preset percentage. Preliminary intervention; the preliminary intervention includes: water intervention, fertilizer intervention, temperature intervention, light intervention, and pest and disease control intervention; the data processing module obtains the plant growth trend prediction results, and uses a deep learning RNN algorithm to analyze the plant growth trend prediction results and the numerical values ​​of the plant morphology that deviate from the plant shaping scheme, determining the corresponding aspects and constructing morphology-corresponding aspect vectors; the neural network model construction module uses the morphology-corresponding aspect vectors to construct and train a convolutional neural network model; the plant preliminary growth intervention strategy generation module uses the trained convolutional neural model to perform preliminary intervention on specific aspects, realizing preliminary plant growth intervention. The identification of plant species based on plant structural phenotypic parameters includes: Based on plant structural phenotypic parameters, plant species are identified using a K-means clustering model. This identification process includes: establishing a plant structural phenotypic parameter template library; randomly sampling and acquiring plant structural phenotypic parameters for different plant species, which are then entered into the template library as typical template parameters for each species; collecting point cloud information of the plant to be tested, establishing a three-dimensional model of the stems, branches, and leaves, and calculating the plant structural phenotypic parameters; matching the plant structural phenotypic parameters with the template library, calculating the similarity between the parameters and typical template parameters, and clustering plant structural phenotypic parameters with similarity exceeding a second preset threshold; finally, identifying the plant species as the model's output. The analysis of the plant's condition based on the plant identification results includes: Based on the plant species identification results, the plant status is analyzed. The plant status includes plant height, leaf morphology, stem shape, and petal characteristics. Leaf morphology includes leaf length, leaf width, and leaf shape; petal characteristics include flower diameter, flowering period, and lifespan. The Apriori association rule algorithm is used to mine the association relationships between plant species corresponding to their status. The association rule algorithm process is as follows: Through iterative input, all frequent itemsets of morphological factors in the plant database are retrieved, i.e., itemsets with support not lower than a preset third threshold. Support is the percentage of each factor itemset appearing in the database relative to the total number of its respective itemsets. Then, rules satisfying the confidence level are constructed using the frequent itemsets. Confidence is the percentage of each factor's total itemsets relative to all itemsets in the database. By comparing support and confidence, the strength of the association with the plant species is determined, and a plant species association network diagram is drawn based on the association rule algorithm. The plant species association network diagram indicates the plant status.

2. The plant identification method based on laser point cloud considering stem shape and leaf tilt angle according to claim 1, characterized in that, The plant structural phenotypic parameters based on laser point cloud computing include: The plant structural phenotypic parameters refer to the leaf tilt angle distribution of the plant. High-precision point cloud information of each plant is acquired through 3D laser scanning and a high-precision camera. The point cloud information is then preprocessed, and a 3D model of the plant's stems, twigs, and leaves is obtained using a spatially irregular triangular mesh. Finally, the format of each triangular facet in the 3D model is parsed to obtain the normal vector data and vertex coordinate data of the triangular facet. Based on the normal vector data and vertex coordinate data of the triangular facets, the leaf tilt angle distribution of the plant is calculated, thus obtaining the plant structural phenotypic parameters. This process includes: acquiring high-precision point cloud information of the plant; establishing a 3D model of the plant's stems, twigs, and leaves; and calculating the leaf tilt angle distribution of the plant. The high-precision point cloud information of the collected plants specifically includes: Starting from the plant's growth and development stage, high-precision point cloud information of the plant is collected. This collection includes: periodically scanning the plant's surface with a 3D laser scanner to receive the reflected signal from each point; then converting this signal into directly identifiable data using a converter; processing this data with software to obtain the 3D coordinates of each point within a preset range; then acquiring a color image of the plant using a high-precision camera and assigning RGB color information to the pixel corresponding to the 3D coordinates in the color image for each point; all recorded points of the plant constitute the plant's point cloud, and the RGB color information of all recorded points constitutes the plant's high-precision point cloud information. The establishment of a three-dimensional model of the plant's stems, twigs, and leaves specifically includes: A 3D model of the plant stems, twigs, and leaves is established based on the point cloud information of the plant. First, the original point cloud information is optimized through preprocessing. The preprocessing includes point cloud stitching and point cloud denoising. Point cloud stitching involves aligning points that exceed the preset range of the 3D coordinates according to preset points and then stitching the point cloud information together. Point cloud denoising refers to filtering out noise information through smoothing low-pass filtering. Noise refers to points in the neighborhood of each point whose variance of the point cloud information exceeds a first preset threshold of the average variance of the neighborhood point cloud information. Then, a preset number of points for modeling are obtained from the point cloud through random point sampling. Finally, a TIN triangle with the points used for modeling as vertices is constructed using a spatial irregular triangular mesh to obtain the 3D model of the plant stems, twigs, and leaves.

3. The plant identification method based on laser point cloud considering stem shape and leaf tilt angle according to claim 1, characterized in that, The monitoring of plant health based on plant condition includes: Based on the plant's condition, a plant health status identification model is used to monitor the plant's health level. During the plant's growth period, a plant health status evaluation function is constructed: NH = W*W1 + N*W2 + I*W3. Here, NH represents the plant's unhealthy index during the growth period; a higher NH value indicates a poorer plant health. W, N, and I represent unhealthy conditions, including: W indicating water deficiency during the growth period (W = 1 if the plant health status identification model determines water deficiency, otherwise W = 0); and N indicating nutrient deficiency during the growth period. The plant's health status is assessed using a plant health status identification model. If the model determines that the plant is malnourished, then N = 1; otherwise, N = 0. I represents the plant's status during its growth period, indicating the presence of pests or diseases. If the model determines that the plant is malnourished, then I = 1; otherwise, I = 0. W1, W2, and W3 represent the percentage of time the corresponding malnourished status lasts within the total growth period. Based on the plant's health evaluation function, an unhealthy index is calculated for the plant during its growth period to monitor its health status. This includes establishing a plant health status identification model. The establishment of the plant health status identification model specifically includes: A plant health status identification model is established using deep learning by randomly collecting more than a fourth preset threshold number of plants. This model includes: using plant morphology as both a training and testing set; extracting features from the training set (including plant height, leaf morphology, stem shape, and flower morphology); preprocessing the feature data (labeling the plant's health status based on point cloud information); and using the health status as labels to train the plant health status identification model, determining the relationship between plant health status and plant morphology, and recording the results according to a preset format. Finally, the model parameters are continuously adjusted using data from the testing set to improve the accuracy of the plant health status identification model.

4. The plant identification method based on laser point cloud considering stem shape and leaf tilt angle according to claim 1, characterized in that, The prediction of plant growth trends based on plant health includes: By recording the health status of plants, the growth trend of plants is predicted. The plant growth trend includes the number of leaves, the direction of branch extension, and the number of flowers. Historical plant growth trends are obtained by crawling large amounts of plant data. A plant growth trend prediction model is then used to predict the plant growth trend. The plant growth trend prediction model is as follows: Historical plant growth trends are used as sample data. K-means clustering algorithm is used to analyze the data and select multiple centroids based on the historical plant growth trend types. The Euclidean distance from each sample to the centroid is calculated, and each sample is assigned to the nearest centroid. The average value of the sample data for each category is then used to calculate the new centroid for each category. Finally, the sample classification and centroid calculation are iteratively performed until the clustering converges and the old and new centroids no longer change. The plant health status and corresponding health status are input as parameters into the plant growth trend prediction model. The plant health status is classified into the type most similar to it, and the current plant growth trend is determined by the basic plant growth trend of that type of plant health status.

5. The plant identification method based on laser point cloud considering stem shape and leaf tilt angle according to claim 1, characterized in that, The assessment of a plant's ornamental value based on its health and initial growth intervention strategies includes: The ornamental value of the plant is judged based on its health status and the initial growth intervention strategy. This judgment includes: establishing an ornamental value evaluation system, PA = (HD + SP) / 2; where PA represents the ornamental value evaluation index; HD represents the plant's health status evaluation index (HD = 1 if the unhealthy index in the health status evaluation function is below a preset threshold, otherwise HD = 0); SP represents the plant's shape evaluation index (SP = 1 if the probability of the plant's shape appearing in historical garden designs is greater than a sixth preset threshold, otherwise SP = 0); analyzing the ornamental value of the plant using a BP neural network; and determining training and testing samples based on the plant's health status evaluation function and shape scheme. The training samples are used for neural network training, and the testing samples are used to detect the relative error between the actual and predicted values.

6. The plant identification method based on laser point cloud considering stem shape and leaf tilt angle according to claim 1, characterized in that, The plant regrowth intervention strategy determined based on the plant's ornamental value includes: Based on the plant's ornamental value, a plant regrowth intervention strategy is determined. This strategy includes increasing or decreasing initial intervention, changing the styling scheme, and replacing the plant species. The determination of the plant regrowth intervention strategy involves: using the initial plant regrowth intervention strategy as a sample, constructing a single-class SVM model based on influencing factors. These influencing factors include the plant's health evaluation index and styling evaluation index within the plant ornamental value evaluation system. Then, using an activation function, the distance range from the intermediate output sample of the single-class SVM model to the center of the hypersphere is mapped to [0,1]. The mapping result represents the probability of performing plant regrowth intervention. If the mapping result is greater than a seventh preset threshold, plant regrowth intervention is performed; otherwise, the initial plant regrowth intervention is maintained. Finally, based on the change in the probability of performing plant regrowth intervention, the plant regrowth intervention strategy is determined.

Images