Construction method of landscaping street tree growth prediction model based on artificial intelligence

Abstract

This invention discloses a method for constructing an artificial intelligence-based growth prediction model for roadside trees in landscaping. The method includes: acquiring time-series image data of trees, extracting key point descriptors and identifying matching terms, calculating their difference values, analyzing the variation patterns of the difference values ​​of matching terms at each growth stage, constructing a growth stage transition model to predict the duration of each growth stage, obtaining the predicted duration of the remaining growth stage, determining linear and nonlinear correlation functions based on historical tree growth data, calculating the environmental fitness within the predicted duration of the remaining growth stage, judging the growth potential based on tree parameters, and combining the growth potential with the environmental fitness to construct a growth prediction model using a neural network. The neural network model constructed by this method comprehensively considers multiple factors, effectively improving the accuracy and reliability of roadside tree growth prediction, assisting in the scientific planning and management of landscaping, and also exhibiting good interpretability.

Description

Technical Field

[0001] This invention relates to the field of digital landscaping, and more particularly to a method for constructing a growth prediction model for roadside trees in landscaping based on artificial intelligence. Background Technology

[0002] With the acceleration of urbanization, landscaping plays an increasingly crucial role in improving the urban ecological environment and enhancing the quality of life for residents. As an important component of urban greening, street trees, through their healthy growth and rational layout, are irreplaceable in shaping the urban landscape, purifying the air, regulating the climate, and providing ecological habitats.

[0003] Traditional methods of street tree management rely heavily on manual experience and regular on-site observations, which have several limitations. Firstly, manual observation struggles to achieve real-time, comprehensive monitoring of large-scale street tree populations, resulting in poor timeliness and completeness of data acquisition. Secondly, manual assessments of tree growth often lack precision and objectivity; differences in the experience and judgment standards of different observers lead to inconsistent and inaccurate evaluations of tree growth, hindering the development of scientifically sound maintenance and management strategies.

[0004] In recent years, computer vision technology, sensor technology, and artificial intelligence algorithms have made significant progress. Integrating these advanced technologies into the field of predicting the growth of street trees in landscaping can help improve the efficiency and scientific nature of street tree management, reduce maintenance costs, and provide strong data support and decision-making basis for urban greening planning, thus promoting the development of urban greening towards intelligence and sustainability. Summary of the Invention

[0005] The purpose of this invention is to provide a method for constructing an artificial intelligence-based prediction model for the growth of roadside trees in landscaping.

[0006] To achieve the above objectives, the present invention is implemented according to the following technical solution:

[0007] The first aspect of this invention provides a method for constructing an artificial intelligence-based growth prediction model for roadside trees in landscaping, comprising:

[0008] Acquire image data of trees over time series, extract descriptors of key points, extract matching items from the descriptors, and calculate the difference value of the matching items;

[0009] Analyze the variation pattern of the difference values ​​of the matching items at each growth stage of the tree, establish a growth stage transition model, predict the time of the growth stage, and obtain the predicted time of all remaining growth stages of the tree.

[0010] Based on historical tree growth data, linear and nonlinear correlation functions of tree growth are obtained, and the environmental fitness of trees during the predicted time of all remaining growth stages is calculated based on the linear and nonlinear correlation functions.

[0011] The growth potential of trees is determined based on tree parameters, and a growth prediction model is constructed based on the growth potential and the environmental adaptability. The tree parameters include tree height, diameter at breast height (DBH), crown width, root system characteristics, and tree age.

[0012] As a further method, the method for extracting key point descriptors includes:

[0013] Obtain images of the same location on a tree before and after its growth stage, and calculate the dense image features using the following expression:

[0014]

[0015] Where x and y are the horizontal and vertical coordinates of a pixel in the image, respectively; R(x,y), G(x,y) and B(x,y) are the pixel values ​​of red, green and blue in the RGB image, respectively; H(x,y), S(x,y) and V(x,y) are the values ​​of hue, saturation and brightness in the HSV image, respectively; LBP(x,y) is the local binary mode feature; HOG(x,y) is the histogram of oriented gradients feature; and λ1, λ2 and λ3 are the color space transformation weight, local binary model weight and histogram of oriented gradients weight, respectively.

[0016] Descriptors are extracted using equidistant grids and perspective projection. Differentiable sampling is performed on two dense image features to obtain descriptors for key points.

[0017] Calculate the similarity function value between two dense image features, and use the gradient descent optimization algorithm to adjust the positions of keypoints to maximize the similarity function value. The expression for the similarity function is:

[0018]

[0019] Where, ω i,j d represents the degree of variation in tree features within the region where the descriptor is located. 1,i and d 2,j Let i and j be the descriptors extracted from the two images, respectively, and let i and j be the indices of the descriptors in the two images. P1(x) 1,i ,y 1,i ) and P1(x 2,j ,y 2,j ) are descriptors d 1,i and d 2,jThe coordinates of the position in the image after perspective projection, σ1 and σ2 are scale parameters used to control the degree of influence of feature point descriptor differences and positional differences, respectively.

[0020] As a further method, the method for extracting matches from descriptors and calculating the difference values ​​of the matches includes:

[0021] For the descriptor sets of images before and after the growth stage, calculate the pairwise dot product between all descriptors, and obtain the similarity matrix by summing the products of the descriptors in each dimension;

[0022] The maximum value in each row and column is found using the similarity matrix. If a descriptor satisfies the maximum value in both rows and columns, it is identified as a match.

[0023] The features of the matching items are normalized, and the difference value is calculated using Euclidean distance.

[0024] As a further method, the method for analyzing the changing patterns of the difference values ​​of the matching items at each growth stage of the tree and establishing a growth stage transition model includes:

[0025] The matching item difference data of historical tree samples before and after each growth stage were obtained. The K-Means clustering algorithm was used to divide the matching item difference data into different clusters, each cluster representing a growth stage, and the range of difference values ​​corresponding to each growth stage was determined.

[0026] A growth stage transition model is constructed using the decision tree algorithm, and the training set is used to train the model with the range of differences corresponding to each growth stage and the historical growth stage transition.

[0027] The timing of the next growth stage is determined by a growth stage transition model, and this process is repeated multiple times until the predicted timing of all remaining growth stages of the tree is obtained.

[0028] As a further method, the method for obtaining linear and nonlinear correlation functions of tree growth based on historical tree growth data includes:

[0029] Collect historical tree growth data and historical environmental data, using factors from the historical environmental data as independent variables and factors from the historical tree growth data as dependent variables;

[0030] Calculate the Pearson coefficient r between each independent variable and the dependent variable. If |r|≥0.3, there is a linear correlation; if |r|<0.3, there is a non-linear correlation.

[0031] For linear correlation, if it is a single-variable linear regression, use a single-variable linear model; if it is a multiple linear regression, use a stepwise regression method to select the combination of independent variables and use a multivariate linear model. For nonlinear correlation, fit multiple nonlinear models. For each nonlinear model, calculate the goodness-of-fit index and select the model with the best goodness-of-fit index. The calculation formula is as follows:

[0032]

[0033] Where P is the number of parameters in the model, l(θ) is the log-likelihood function of the model, n is the number of samples, σ is the error term, and y i For the i-th observation, f(x) i x, θ) are the predicted values ​​of the model, and x i Let θ be the value of the i-th independent variable, and θ be the set of parameters of the model.

[0034] As a further method, the method for calculating the environmental fitness of trees over the predicted time span of all remaining growth stages based on linear and nonlinear correlation functions includes:

[0035] The predicted times of all remaining growth stages of the trees are input into a pre-trained random forest model to obtain predicted environmental data, wherein the random forest model is trained based on historical time series and historical environmental data.

[0036] Environmental data consistent with the historical environmental data variables used when establishing linear and nonlinear correlation functions are obtained from the predicted environmental data. The growth value sequence is then calculated by averaging the linear and nonlinear correlation functions.

[0037] Environmental fitness is calculated based on the predicted growth value sequence, and the expression is as follows:

[0038]

[0039] Where ξ is the total number of observed data points, y i For the i-th predicted growth value, σ is the average of all predicted growth values. y x is the standard deviation of all predicted growth values. i Let μ be the environmental variable vector of the i-th observation, and σ be the average value vector of the environmental variable vector. x represents the standard deviation of the environmental variable vector.

[0040] As a further method, the method for determining the growth potential of trees based on tree parameters includes:

[0041] To obtain the tree's current parameters and quantify its growth potential, the expression is:

[0042]

[0043] Among them, V H To the tree's height growth rate, L represents the average tree height growth rate of this tree species. a W represents the effective light ratio available to trees. f D is the limiting factor for wind on tree height growth. c For the current chest diameter, D s For the same tree species and age, the standard diameter at breast height (DBH) is S r V is the root growth space coefficient. C For the crown expansion speed, L represents the average crown spread rate of this tree species. i N is the ratio of the average light intensity within the canopy to the optimum light intensity for this tree species. a S represents the proportion of available nutrients in the soil. d S represents the depth of the root system in the soil. q For soil quality, B r B is the ratio of root biomass to the aboveground biomass of a tree. s A represents the ratio of the standard root system to the aboveground biomass of this tree species. c Given the current age of the tree, G represents the average age of this tree species. s This represents the growth stage coefficient.

[0044] As a further method, the method for constructing a growth prediction model based on the growth potential and the environmental adaptability includes:

[0045] A growth prediction model is constructed using a neural network to predict the relationship between growth potential, environmental adaptability, and tree growth rate. The model has two output layers, which output the predicted tree height and diameter at breast height over time series.

[0046] A decay function is introduced into the growth prediction model based on the growth potential. The smaller the growth potential, the larger the value of the decay function, which is applied to the hidden layer of the model.

[0047] The randomness of tree growth and the environment is represented by a probability density distribution. This probability density distribution is added to the output layer of the growth prediction model, and then updated using Bayes' theorem. The expression for Bayes' theorem is:

[0048]

[0049] Where Θ represents the observed data, G represents the tree growth state, E represents the environmental factors, P(Θ|G,E) represents the probability of the observed data Θ occurring given the tree growth state G and the environmental factors E, P(G|Θ,E) represents the probability of the tree being in state G given the observed data Θ and the environmental factors E, and P(Θ|E) represents the probability distribution of the observed data Θ given the environmental factors E.

[0050] Where P(G|Θ,E) follows a multi-source normal distribution, specifically:

[0051]

[0052] Where ∑(Θ,E) represents the covariance matrix of tree growth state G given the observed data Θ and environmental factors E, μ(Θ,E) represents the mean of tree growth state G given the observed data Θ and environmental factors E, and T represents the transpose operation.

[0053] Where P(Θ|E) follows a mixed distribution, specifically:

[0054]

[0055] Where K is the number of components in the mixed distribution, ω k Let the weight of the k-th component be . This indicates that under environmental factor E, with Let μ be the conditional probability density function of a normal distribution, where μ k and ∑ k It is the mean vector and covariance matrix of the k-th component;

[0056] The growth prediction model was trained using historical environmental fitness and combinations of historical tree height and diameter at breast height as the training set.

[0057] Secondly, embodiments of this application also provide an electronic device, including: a processor; and a memory arranged to store computer-executable instructions, which, when executed, cause the processor to perform the steps of the method described in the first aspect.

[0058] Thirdly, embodiments of this application also provide a computer-readable storage medium storing one or more programs that, when executed by an electronic device including multiple applications, cause the electronic device to perform the steps of the method described in the first aspect.

[0059] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects:

[0060] (1) By acquiring image data of trees over time and extracting descriptors of key points, this invention can more accurately analyze the growth changes of trees. Based on actual image data rather than relying solely on historical statistical data, this method can improve the accuracy of prediction.

[0061] (2) The growth stage transition model established by the present invention based on the analysis of the difference value change pattern predicts the time when the tree enters the next growth stage, predicts and divides the growth stages of the tree, helps to identify the different growth situations of the tree in different growth stages, and makes the model construction more comprehensive by taking time factors into account.

[0062] (3) The present invention judges the growth potential of trees based on tree parameters and constructs a growth prediction model in combination with environmental adaptability. It introduces a decay function into the model based on growth potential, takes into account the natural decay of tree growth potential over time, and the long-term impact of environmental factors on tree growth, which helps to more accurately predict the performance of trees at different growth stages. Attached Figure Description

[0063] Figure 1 This is a flowchart illustrating the steps of constructing an artificial intelligence-based prediction model for the growth of roadside trees in landscaping, as described in this embodiment of the invention.

[0064] Figure 2 This is a schematic diagram of the structure of an electronic device in an embodiment of this specification. Detailed Implementation

[0065] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0066] Reference Figure 1 As shown, this invention provides a method for constructing an artificial intelligence-based growth prediction model for roadside trees in landscaping, including:

[0067] S100 acquires image data from the time series of trees, extracts descriptors of key points, extracts matching items from the descriptors, and calculates the difference value of the matching items;

[0068] It should be explained that this step involves acquiring time-series images of trees and extracting key point descriptors. This process utilizes multiple attributes and features of pixels, and through specific calculations and processing, accurately characterizes the key features of trees, laying the foundation for subsequent comparisons. Matching items are found from the descriptors, and the corresponding features in images at different times are determined by using a similarity matrix, reflecting the continuity and changes in tree growth. These data combined can clearly present the growth trend of trees, providing strong support for growth stage judgment and model construction.

[0069] In one specific embodiment, a batch of Chinese scholar trees (Sophora japonica) were newly planted as street trees on a newly built urban street. The plan is to use the artificial intelligence-based landscape greening street tree growth prediction model of this invention to predict the future growth status of these Chinese scholar trees, in order to rationally plan street landscape maintenance, tree care resource allocation, and other related work.

[0070] In the actual evaluation, starting from the first month after the completion of the planting of Chinese locust trees, images of the trees were taken at 9:00 AM at the beginning of each month using a high-definition camera at a distance of 5 meters and a horizontal angle of 45 degrees. Images from the first and second months were selected for processing. For the pixel coordinates (300, 400) in the image, the RGB values ​​were (100, 80, 60), the HSV values ​​were (35, 0.4, 0.6), the local binary mode feature value was 0.25, the histogram of oriented gradients feature value was 0.4, and the color space conversion weight was set to 0.35. The dense image feature value of the pixel was obtained through the dense image feature calculation formula.

[0071] In the actual evaluation, for the descriptor sets of images from these two months, the pairwise dot product between all descriptors was calculated to obtain a similarity matrix. The maximum value of each row and column in the similarity matrix was determined to identify the matching items, resulting in a total of 6 matching items. After normalizing the features of the matching items, the Euclidean distance between the feature vectors of two matching items [0.1, 0.18, 0.25] and [0.13, 0.2, 0.28] was calculated, and the difference value was 0.035.

[0072] S200 analyzes the variation pattern of the difference values ​​of the matching items in each growth stage of the tree, establishes a growth stage transition model, predicts the time of the growth stage, and obtains the predicted time of all remaining growth stages of the tree.

[0073] It should be explained that by analyzing the variation patterns of the difference values ​​of the matching items, we can gain insight into the evolution rhythm of the characteristics of each growth stage. For example, the difference values ​​fluctuate little in the juvenile stage and fluctuate greatly and have obvious trends in the mature stage. By establishing a growth stage transition model, we can quantify the relationship between this evolution and stage transition in a data-driven manner. By predicting the growth stage time through the model, garden managers can plan maintenance strategies in advance and allocate resources rationally.

[0074] In the actual assessment, matching difference values ​​were calculated from images of Chinese scholar trees (Sophora japonica) before and after their growth stages in other areas of the city over the past 15 years. Using the K-Means clustering algorithm, these data were divided into 5 clusters, corresponding to the budding, seedling, rapid growth, stable, and senescent stages of the Chinese scholar trees. The range of difference values ​​for each stage was determined, with the budding stage range being [0-0.05]. A growth stage transition model was constructed using a decision tree algorithm, training the model with the difference value range and historical growth stage transition data as the training set. The current batch of Chinese scholar trees has a matching difference value of 0.03, indicating it is in the budding stage according to the model. It is expected to enter the seedling stage in 6 months. The predicted time for the remaining growth stages is shown in Table 1.

[0075] Table 1. Predicted time of remaining growth stages

[0076] growth stage Expected entry time Expected duration Seedling stage 6 months after planting 12 months Rapid growth period 19 months after planting 24 months Stable period 43 months after planting 36 months Aging period 10 years after planting 20 years

[0077] S300 obtains the linear and nonlinear correlation functions of tree growth based on historical tree growth data, and calculates the environmental fitness of the tree within the predicted time of all remaining growth stages based on the linear and nonlinear correlation functions.

[0078] It's important to explain that by deeply exploring the intrinsic relationship between tree growth and environmental data, constructing precise functional relationships to quantify the impact of the environment on tree growth, and then calculating environmental fitness, the significance lies in providing key parameters and basis for growth prediction models. This allows the models to more realistically simulate the growth dynamics of trees in different environments. Through accurate environmental fitness calculations, the models can more accurately predict the growth rate and morphological changes of trees at various future growth stages, thereby helping landscape managers to develop targeted maintenance plans in advance and allocate resources rationally.

[0079] In the actual assessment, the predicted time of the remaining growth stage of Sophora japonica (based on the growth stage transition model) is input into a pre-trained random forest model to obtain predicted environmental data. This data predicts the annual light duration, temperature, and soil moisture for the next three years. Environmental data consistent with those used when establishing the correlation function are selected from the predicted environmental data, and the predicted growth value sequence is calculated using weighted linear and nonlinear correlation functions. Assuming the predicted tree height growth value sequence for the next three years is calculated to be [0.5, 0.8, 1.2], the environmental fitness is calculated using a total of 3 observation data points, yielding a fitness of 0.48 according to the formula.

[0080] S400 determines the growth potential of trees based on tree parameters and constructs a growth prediction model based on the growth potential and the environmental adaptability; the tree parameters include tree height, diameter at breast height, crown width, root system characteristics and tree age.

[0081] It's important to explain that tree parameters allow us to understand a tree's current growth status from different dimensions. For example, tree height and diameter at breast height (DBH) reflect the tree's longitudinal and lateral growth, crown width relates to its photosynthetic range, root characteristics affect nutrient absorption, and tree age reflects its growth stage. By combining these parameters, we can accurately determine how much space and capacity a tree still has for growth. By combining the determined growth potential with the previously calculated environmental adaptability, we can build a growth prediction model. This model allows us to fully consider the tree's own characteristics and its adaptation to its environment. The resulting model is no longer a simple prediction based on a single factor, but a scientific prediction that comprehensively considers multiple factors, and can more accurately predict the tree's growth rate and morphological changes at different stages in the future.

[0082] In the actual assessment, the current height of the Chinese scholar tree was measured to be 0.3 meters. The average height growth rate of this tree species is 0.45 meters / year. The effective light penetration ratio is 0.55. The wind-induced height growth limitation coefficient is 0.12. The current diameter at breast height (DBH) is 2 centimeters, while the standard DBH for the same species and age is 1.8 centimeters. The root growth space coefficient is 0.65. The crown expansion rate is 0.15 meters / year, while the average crown expansion rate for this tree species is 0.2 meters / year. The average light intensity within the crown is... The optimal light intensity ratio for this tree species is 0.45, the proportion of available nutrients in the soil is 0.35, the root depth in the soil is 0.5 meters, the soil mass is 0.55, the ratio of root biomass to aboveground biomass is 0.15, the ratio of standard root biomass to aboveground biomass is 0.13, the current age of the tree is 0.5 years, the average age of this tree species is 1 year, the growth stage coefficient is 0.7, and the calculated growth potential value is 0.52.

[0083] In the actual assessment, a neural network growth prediction model was constructed, with two output layers predicting tree height and diameter at breast height (DBH) respectively. A decay function based on growth potential was introduced and applied to the hidden layers. Randomness was represented by a probability density distribution and added to the output layer, updated using Bayes' theorem. Specifically, the model was trained using historical environmental adaptability, tree height, and DBH data of previous Chinese scholar trees in the same city as the training set. After training, the model predicted that tree height would reach 0.6 meters and 0.9 meters in the next two years, and DBH would reach 3 centimeters and 4 centimeters respectively.

[0084] Specifically, through testing the model, it was found that the model's accuracy in predicting tree height was over 94%, and its accuracy in predicting diameter at breast height (DBH) was over 87%, with deviations of no more than 2% and 3% respectively, providing an effective basis for maintenance management decisions.

[0085] In this embodiment, the method for extracting key point descriptors includes:

[0086] Obtain images of the same location on a tree before and after its growth stage, and calculate the dense image features using the following expression:

[0087]

[0088] Where x and y are the horizontal and vertical coordinates of a pixel in the image, respectively; R(x,y), G(x,y) and B(x,y) are the pixel values ​​of red, green and blue in the RGB image, respectively; H(x,y), S(x,y) and V(x,y) are the values ​​of hue, saturation and brightness in the HSV image, respectively; LBP(x,y) is the local binary mode feature; HOG(x,y) is the histogram of oriented gradients feature; and λ1, λ2 and λ3 are the color space transformation weight, local binary model weight and histogram of oriented gradients weight, respectively.

[0089] Descriptors are extracted using equidistant grids and perspective projection. Differentiable sampling is performed on two dense image features to obtain descriptors for key points.

[0090] Calculate the similarity function value between two dense image features, and use the gradient descent optimization algorithm to adjust the positions of keypoints to maximize the similarity function value. The expression for the similarity function is:

[0091]

[0092] Where, ω i,j d represents the degree of variation in tree features within the region where the descriptor is located. 1,i and d 2,j Let i and j be the descriptors extracted from the two images, respectively, and let i and j be the indices of the descriptors in the two images. P1(x) 1,i ,y 1,i ) and P1(x 2,j ,y 2,j ) are descriptors d 1,i and d 2,j The coordinates of the position in the image after perspective projection, σ1 and σ2 are scale parameters used to control the degree of influence of feature point descriptor differences and positional differences, respectively.

[0093] In this embodiment, the method for extracting matches from descriptors and calculating the difference values ​​of the matches includes:

[0094] For the descriptor sets of images before and after the growth stage, calculate the pairwise dot product between all descriptors, and obtain the similarity matrix by summing the products of the descriptors in each dimension;

[0095] The maximum value in each row and column is found using the similarity matrix. If a descriptor satisfies the maximum value in both rows and columns, it is identified as a match.

[0096] The features of the matching items are normalized, and the difference value is calculated using Euclidean distance.

[0097] In this embodiment, the method for analyzing the variation pattern of the difference values ​​of the matching items at each growth stage of the tree and establishing a growth stage transformation model includes:

[0098] The matching item difference data of historical tree samples before and after each growth stage were obtained. The K-Means clustering algorithm was used to divide the matching item difference data into different clusters, each cluster representing a growth stage, and the range of difference values ​​corresponding to each growth stage was determined.

[0099] A growth stage transition model is constructed using the decision tree algorithm, and the training set is used to train the model with the range of differences corresponding to each growth stage and the historical growth stage transition.

[0100] The timing of the next growth stage is determined by a growth stage transition model, and this process is repeated multiple times until the predicted timing of all remaining growth stages of the tree is obtained.

[0101] In this embodiment, the method for obtaining linear and nonlinear correlation functions of tree growth based on historical tree growth data includes:

[0102] Collect historical tree growth data and historical environmental data, using factors from the historical environmental data as independent variables and factors from the historical tree growth data as dependent variables;

[0103] Calculate the Pearson coefficient r between each independent variable and the dependent variable. If |r|≥0.3, there is a linear correlation; if |r|<0.3, there is a non-linear correlation.

[0104] For linear correlation, if it is a single-variable linear regression, use a single-variable linear model; if it is a multiple linear regression, use a stepwise regression method to select the combination of independent variables and use a multivariate linear model. For nonlinear correlation, fit multiple nonlinear models. For each nonlinear model, calculate the goodness-of-fit index and select the model with the best goodness-of-fit index. The calculation formula is as follows:

[0105]

[0106] Where P is the number of parameters in the model, l(θ) is the log-likelihood function of the model, n is the number of samples, σ is the error term, and y i For the i-th observation, f(x) i x, θ) are the predicted values ​​of the model, and x i Let θ be the value of the i-th independent variable, and θ be the set of parameters of the model.

[0107] In this embodiment, the method for calculating the environmental fitness of trees over the predicted timeframes of all remaining growth stages based on linear and nonlinear correlation functions includes:

[0108] The predicted times of all remaining growth stages of the trees are input into a pre-trained random forest model to obtain predicted environmental data, wherein the random forest model is trained based on historical time series and historical environmental data.

[0109] Environmental data consistent with the historical environmental data variables used when establishing linear and nonlinear correlation functions are obtained from the predicted environmental data. The growth value sequence is then calculated by averaging the linear and nonlinear correlation functions.

[0110] Environmental fitness is calculated based on the predicted growth value sequence, and the expression is as follows:

[0111]

[0112] Where ξ is the total number of observed data points, y i For the i-th predicted growth value, σ is the average of all predicted growth values. y x is the standard deviation of all predicted growth values. i Let μ be the environmental variable vector of the i-th observation, and σ be the average value vector of the environmental variable vector. x represents the standard deviation of the environmental variable vector.

[0113] In this embodiment, the method for determining the growth potential of trees based on tree parameters includes:

[0114] To obtain the tree's current parameters and quantify its growth potential, the expression is:

[0115]

[0116] Among them, V H To the tree's height growth rate, L represents the average tree height growth rate of this tree species. a W represents the effective light ratio available to trees. f D is the limiting factor for wind on tree height growth. c For the current chest diameter, D s For the same tree species and age, the standard diameter at breast height (DBH) is S r V is the root growth space coefficient. C For the crown expansion speed, L represents the average crown spread rate of this tree species. i N is the ratio of the average light intensity within the canopy to the optimum light intensity for this tree species. a S represents the proportion of available nutrients in the soil. d S represents the depth of the root system in the soil. q For soil quality, B rB is the ratio of root biomass to the aboveground biomass of a tree. s A represents the ratio of the standard root system to the aboveground biomass of this tree species. c Given the current age of the tree, G represents the average age of this tree species. s This represents the growth stage coefficient.

[0117] In this embodiment, the method for constructing a growth prediction model based on the growth potential and the environmental adaptability includes:

[0118] A growth prediction model is constructed using a neural network to predict the relationship between growth potential, environmental adaptability, and tree growth rate. The model has two output layers, which output the predicted tree height and diameter at breast height over time series.

[0119] A decay function is introduced into the growth prediction model based on the growth potential. The smaller the growth potential, the larger the value of the decay function, which is applied to the hidden layer of the model.

[0120] The randomness of tree growth and the environment is represented by a probability density distribution. This probability density distribution is added to the output layer of the growth prediction model, and then updated using Bayes' theorem. The expression for Bayes' theorem is:

[0121]

[0122] Where Θ represents the observed data, G represents the tree growth state, E represents the environmental factors, P(Θ|G,E) represents the probability of the observed data Θ occurring given the tree growth state G and the environmental factors E, P(G|Θ,E) represents the probability of the tree being in state G given the observed data Θ and the environmental factors E, and P(Θ|E) represents the probability distribution of the observed data Θ given the environmental factors E.

[0123] Where P(G|Θ,E) follows a multi-source normal distribution, specifically:

[0124]

[0125] Where ∑(Θ,E) represents the covariance matrix of tree growth state G given the observed data Θ and environmental factors E, μ(Θ,E) represents the mean of tree growth state G given the observed data Θ and environmental factors E, and T represents the transpose operation.

[0126] Where P(Θ|E) follows a mixed distribution, specifically:

[0127]

[0128] Where K is the number of components in the mixed distribution, ωk Let the weight of the k-th component be . This indicates that under environmental factor E, with Let μ be the conditional probability density function of a normal distribution, where μ k and ∑ k It is the mean vector and covariance matrix of the k-th component;

[0129] The growth prediction model was trained using historical environmental fitness and combinations of historical tree height and diameter at breast height as the training set.

[0130] Figure 2 This is a schematic diagram of the structure of an electronic device according to an embodiment of this application. Please refer to it. Figure 2 At the hardware level, the electronic device includes a processor, and optionally also includes an internal bus, a network interface, and memory. The memory may include main memory, such as high-speed random-access memory (RAM), or non-volatile memory, such as at least one disk drive. Of course, the electronic device may also include other hardware required for other business operations.

[0131] The processor, network interface, and memory can be interconnected via an internal bus, which can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. This bus can be categorized into address bus, data bus, control bus, etc.

[0132] The memory is used to store programs. Specifically, the program may include program code, which includes computer operating instructions. The memory may include internal memory and non-volatile memory, and provides instructions and data to the processor.

[0133] The processor reads the corresponding computer program from non-volatile memory into memory and then runs it, forming a method for constructing an AI-based prediction model for the growth of roadside trees in landscaping at the logical level. The processor executes the program stored in memory, specifically executing any of the aforementioned methods for constructing an AI-based prediction model for the growth of roadside trees in landscaping.

[0134] This invention can be applied to or implemented by a processor. The processor may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method can be completed by integrated logic circuits in the processor's hardware or by instructions in software form. The processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The storage medium is located in the memory, and the processor reads the information in the memory and, in conjunction with its hardware, completes the steps of the above method.

[0135] This application also proposes a computer-readable storage medium that stores one or more programs, the programs including instructions that, when executed by an electronic device including multiple applications, perform the aforementioned method for constructing any AI-based prediction model for the growth of roadside trees in landscaping.

[0136] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined in the claims, all of which should fall within the protection scope of the present invention.

Claims

1. A method for constructing an artificial intelligence-based growth prediction model for street trees in landscaping, characterized in that, Includes the following steps: Acquire image data of trees over time series, extract descriptors of key points, extract matching items from the descriptors, and calculate the difference value of the matching items; Analyze the variation pattern of the difference values ​​of the matching items at each growth stage of the tree, establish a growth stage transition model, predict the time of the growth stage, and obtain the predicted time of all remaining growth stages of the tree. Based on historical tree growth data, linear and nonlinear correlation functions of tree growth are obtained, and the environmental fitness of trees during the predicted time of all remaining growth stages is calculated based on the linear and nonlinear correlation functions. The growth potential of trees is determined based on tree parameters, and a growth prediction model is constructed based on the growth potential and the environmental adaptability; the tree parameters include tree height, diameter at breast height, crown width, root system characteristics, and tree age; The method for extracting matches from descriptors and calculating the difference values ​​of the matches includes: For the descriptor sets of images before and after the growth stage, calculate the pairwise dot product between all descriptors, and obtain the similarity matrix by summing the products of the descriptors in each dimension; The maximum value in each row and column is found using the similarity matrix. If a descriptor satisfies the maximum value in both rows and columns, it is identified as a match. The features of the matching items are normalized, and the difference value is calculated using Euclidean distance; The method for analyzing the changing patterns of the difference values ​​of the matching items at each growth stage of a tree and establishing a growth stage transition model includes: The matching item difference data of historical tree samples before and after each growth stage were obtained. The K-Means clustering algorithm was used to divide the matching item difference data into different clusters, each cluster representing a growth stage, and the range of difference values ​​corresponding to each growth stage was determined. A growth stage transition model is constructed using the decision tree algorithm, and the training set is used to train the model with the range of differences corresponding to each growth stage and the historical growth stage transition. The timing of the next growth stage is determined by a growth stage transition model, and this process is repeated multiple times until the predicted timing of all remaining growth stages of the tree is obtained. The method for calculating the environmental fitness of trees over the predicted time span of all remaining growth stages based on linear and nonlinear correlation functions includes: The predicted times of all remaining growth stages of the trees are input into a pre-trained random forest model to obtain predicted environmental data, wherein the random forest model is trained based on historical time series and historical environmental data. Environmental data consistent with the historical environmental data variables used when establishing linear and nonlinear correlation functions are obtained from the predicted environmental data. The growth value sequence is then calculated by averaging the linear and nonlinear correlation functions. Environmental fitness is calculated based on the predicted growth value sequence, and the expression is as follows: Where ξ is the number of data points in the predicted growth value sequence, and y i For the i-th predicted growth value, σ is the average of all predicted growth values ​​during the growth phase. y x represents the standard deviation of all predicted growth values ​​during the growth phase. i Let μ be the environmental data vector corresponding to the i-th growth value, and σ be the average vector of all environmental data vectors. x The standard deviation vector of all environmental data vectors; The method for constructing a growth prediction model based on the growth potential and the environmental adaptability includes: A growth prediction model is constructed using a neural network to predict the relationship between growth potential, environmental adaptability, and tree growth rate. The model has two output layers, which output the predicted tree height and diameter at breast height over time series. A decay function is introduced into the growth prediction model based on the growth potential. The smaller the growth potential, the larger the value of the decay function, which is applied to the hidden layer of the model. The randomness of tree growth and the environment is represented by a probability density distribution. This probability density distribution is added to the output layer of the growth prediction model, and then updated using Bayes' theorem. The expression for Bayes' theorem is: Where Θ represents the observed data, G represents the tree growth state, E represents the environmental factors, P(Θ|G,E) represents the probability of the observed data Θ occurring given the tree growth state G and the environmental factors E, P(G|Θ,E) represents the probability of the tree being in state G given the observed data Θ and the environmental factors E, and P(Θ|E) represents the probability distribution of the observed data Θ given the environmental factors E. Where P(G|Θ,E) follows a multi-source normal distribution, specifically: Where ∑(Θ,E) represents the covariance matrix of tree growth state G given the observed data Θ and environmental factors E, μ(Θ,E) represents the mean of tree growth state G given the observed data Θ and environmental factors E, and T represents the transpose operation. Where P(Θ|E) follows a mixed distribution, specifically: Where K is the number of components in the mixed distribution, ω k Let P(Θ|N(μ) be the weight of the k-th component. k ,∑ k ),E) represents the situation under environmental factor E, with N(μ) k ,∑ k Let μ be the conditional probability density function of the normal distribution, where μ k and ∑ k It is the mean vector and covariance matrix of the k-th component; The growth prediction model was trained using historical environmental fitness and combinations of historical tree height and diameter at breast height as training sets.

2. The method for constructing an artificial intelligence-based prediction model for the growth of street trees in landscaping, as described in claim 1, is characterized in that... The method for extracting key point descriptors includes: The expression for calculating dense image features is: Where x and y are the horizontal and vertical coordinates of a pixel in the image, respectively; R(x,y), G(x,y) and B(x,y) are the pixel values ​​of red, green and blue in the RGB image, respectively; H(x,y), S(x,y) and V(x,y) are the values ​​of hue, saturation and brightness in the HSV image, respectively; LBP(x,y) is the local binary mode feature; HOG(x,y) is the histogram of oriented gradients feature; and λ1, λ2 and λ3 are the color space transformation weight, local binary model weight and histogram of oriented gradient weight, respectively. Descriptors are extracted using equidistant grids and perspective projection. Differentiable sampling is performed on two dense image features to obtain descriptors for key points. Calculate the similarity function value between two dense image features, and use the gradient descent optimization algorithm to adjust the positions of keypoints to maximize the similarity function value. The expression for the similarity function is: Where, ω i,j d represents the degree of variation in tree features within the region where the descriptor is located. 1,i and d 2,j Let i and j be the descriptors extracted from the two images, respectively, and let i and j be the indices of the descriptors in the two images. P1(x) 1,i ,y 1,i ) and P1(x 2,j ,y 2,j ) are descriptors d 1,i and d 2,j The coordinates of the position in the image after perspective projection, σ1 and σ2 are scale parameters used to control the degree of influence of feature point descriptor differences and positional differences, respectively.

3. The method for constructing an artificial intelligence-based prediction model for the growth of street trees in landscaping, as described in claim 1, is characterized in that... The method for obtaining linear and nonlinear correlation functions of tree growth based on historical tree growth data includes: Collect historical tree growth data and historical environmental data, using factors from the historical environmental data as independent variables and factors from the historical tree growth data as dependent variables; Calculate the Pearson coefficient r between each independent variable and the dependent variable. If |r|≥0.3, there is a linear correlation; if |r|<0.3, there is a non-linear correlation. For linear correlation, if it is a single-variable linear regression, use a single-variable linear model; if it is a multiple linear regression, use a stepwise regression method to select the combination of independent variables and use a multivariate linear model. For nonlinear correlation, fit multiple nonlinear models. For each nonlinear model, calculate the goodness-of-fit index and select the model with the best goodness-of-fit index. The calculation formula is as follows: Where P is the number of parameters in the model, l(θ) is the log-likelihood function of the model, n is the number of samples, σ is the error term, and y i For the i-th observation, f(x) i x, θ) are the predicted values ​​of the model, and x i Let θ be the value of the i-th independent variable, and let θ be the set of parameters of the model.

4. The method for constructing an artificial intelligence-based prediction model for the growth of roadside trees in landscaping according to claim 1, characterized in that, The method for determining the growth potential of trees based on tree parameters includes: To obtain the tree's current parameters and quantify its growth potential, the expression is: Among them, V H To the tree's height growth rate, L represents the average tree height growth rate of this tree species. a W represents the effective light ratio available to trees. f D is the limiting factor for wind on tree height growth. c For the current chest diameter, D s For the same tree species and age, the standard diameter at breast height (DBH) is S r V is the root growth space coefficient. C For the crown expansion speed, L represents the average crown spread rate of this tree species. i N is the ratio of the average light intensity within the canopy to the optimum light intensity for this tree species. a S represents the proportion of available nutrients in the soil. d S represents the depth of the root system in the soil. q For soil quality, B r B is the ratio of root biomass to the aboveground biomass of a tree. s A represents the ratio of the standard root system to the aboveground biomass of this tree species. c Given the current age of the tree, G represents the average age of this tree species. s This represents the growth stage coefficient.

5. An electronic device, comprising: processor; as well as A memory configured to store computer-executable instructions, which, when executed, cause the processor to perform the method of any one of claims 1 to 4.

6. A computer-readable storage medium storing one or more programs, which, when executed by an electronic device including a plurality of applications, cause the electronic device to perform the method of any one of claims 1 to 4.

Images