A method and system for predicting the strength of controlled-release fertilizer based on phenotypic characteristics

Abstract

The application discloses a kind of based on phenotypic characteristics's controlled-release fertilizer intensity prediction method and system, it is related to fertilizer intensity prediction technical field, including: obtaining the phenotypic characteristics and intensity data of controlled-release fertilizer, constructs sample set;Through principal component analysis method, dimensionality reduction is handled;According to the data after dimensionality reduction, construct controlled-release fertilizer intensity prediction model;Through particle swarm algorithm and k-fold function, controlled-release fertilizer intensity prediction model parameter optimization is carried out;The phenotypic characteristics of controlled-release fertilizer are obtained, input into the optimized controlled-release fertilizer intensity prediction model, and the controlled-release fertilizer intensity prediction result is obtained.The SVM prediction model of the application is optimized by particle swarm algorithm, the optimal parameters of the model are found by K-fold cross validation, the accuracy of model parameter selection is guaranteed when different controlled-release fertilizers are predicted, and the prediction accuracy is improved.The prediction model has good prediction performance, high prediction accuracy, simple and efficient model, and the prediction has objectivity and universality, which provides a theoretical basis for nondestructive testing controlled-release fertilizer intensity method.

Description

Technical Field

[0001] This invention relates to the field of fertilizer intensity prediction technology, and more specifically to a method and system for predicting the intensity of controlled-release fertilizers based on phenotypic characteristics. Background Technology

[0002] Fertilizer application plays a crucial role in agriculture, especially in increasing grain production, thus achieving agricultural output growth. While grain yields have increased year by year, this high output is accompanied by environmental pollution and declining agricultural profits due to excessive fertilizer application and low fertilizer utilization rates. The rapid nutrient release rate of traditional fertilizers, along with their tendency to be lost, migrate, and volatilize, causes environmental pollution, disrupts soil microbial ecosystems, depletes natural resources, and affects crop quality, all of which constrain sustainable agricultural development. Controlling fertilizer application rates, improving fertilizer application efficiency, and using controlled-release fertilizers instead of traditional fertilizers are of profound significance for sustainable agricultural development. The controlled-release fertilizer's ability to control nutrient release stems from its unique nutrient diffusion mechanism. The coating of controlled-release fertilizer isolates the fertilizer's interior from the external environment, creating a concentration difference to control nutrient release. If the coating of controlled-release fertilizer is damaged during production, transportation, or application, it becomes no different from ordinary fertilizer, and may even have the opposite effect. Accurately predicting fertilizer intensity is crucial for reducing controlled-release fertilizer coating damage and ensuring that controlled-release fertilizer releases nutrients at the preset rate.

[0003] Particle phenotypic characteristics influence the mechanical and flow behavior of granular materials and are important parameters for predicting and controlling their properties. Fertilizers play a crucial role in agricultural granules, and their phenotypic characteristics affect fertilizer sphericity, friction, coefficient of restitution, and the effectiveness of mechanized fertilization, making them an indispensable part of agricultural machinery design and research. Studies by Shan Jun and Xu Liming et al. have found that spherical texture characteristics affect the shape accuracy of the sphere. Studies by Mohammad and Ahmad et al. have found that texture characteristics are significant in predicting the surface roughness of objects. Studies by Kan Hongfu et al. have found that higher fertilizer sphericity results in denser fertilizer particles and higher fertilizer strength; higher fertilizer roundness results in smoother, more uniform fertilizer, higher porosity, faster heat dissipation, and better flowability. Hofstee and Huisman's research found that five physical properties affecting fertilizer movement are fertilizer particle size, strength, coefficient of friction, coefficient of recovery, and aerodynamic drag. Among these, fertilizer strength indirectly affects its movement; fertilizers with low strength are more likely to break during the application process, leading to changes in particle size and affecting nutrient distribution. Silverberg et al. found that fertilizer shape and the pore structure formed by fertilizer accumulation affect salt ion diffusion, thus influencing fertilizer performance. Basu, Terry, and others found that fertilizer particle size affects nutrient separation and release time. Hoffmeister et al., through model experiments, found that differences in fertilizer particle size, shape, and density affect the trend of nutrient separation during the transport and propagation of blended fertilizer particles.

[0004] Therefore, how to propose a method and system for predicting the intensity of controlled-release fertilizer based on phenotypic features, predict the intensity of controlled-release fertilizer based on phenotypic features, and reduce the breakage rate of controlled-release fertilizer to improve fertilization efficiency is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] In view of this, the present invention provides a method and system for predicting the intensity of controlled-release fertilizer based on phenotypic features. This method predicts the intensity of controlled-release fertilizer based on phenotypic features, reducing the breakage rate of controlled-release fertilizer and improving fertilization efficiency. To achieve the above objectives, the present invention adopts the following technical solution:

[0006] A method for predicting the intensity of controlled-release fertilizer based on phenotypic features, comprising:

[0007] Obtain phenotypic characteristics and intensity data of controlled-release fertilizers, and construct a sample set;

[0008] Principal component analysis was used to reduce the dimensionality of the sample data.

[0009] A controlled-release fertilizer intensity prediction model based on support vector machine is constructed based on the dimensionality-reduced data.

[0010] Parameter optimization of controlled-release fertilizer intensity prediction model was performed using particle swarm optimization and k-fold function.

[0011] The phenotypic features of controlled-release fertilizer are obtained and input into the optimized controlled-release fertilizer intensity prediction model to obtain the controlled-release fertilizer intensity prediction results.

[0012] Optionally, the phenotypic features include: triaxial features, sphericity, granularity, and texture features.

[0013] Optionally, the phenotypic characteristics and intensity data of the controlled-release fertilizer are obtained in the following way:

[0014] The strength of controlled-release fertilizer was obtained using a universal testing machine;

[0015] Acquire images of controlled-release fertilizer, and obtain the triaxial features, sphericity, and particle size of the controlled-release fertilizer after image preprocessing;

[0016] Texture features in the phenotypic features of controlled-release fertilizer are calculated using the gray-level co-occurrence matrix algorithm.

[0017] Optionally, it also includes data validation of the phenotypic characteristics and intensity data of controlled-release fertilizers, wherein the data validation includes:

[0018] First, perform statistical calculations on the sample data to calculate the mean, range, and standard deviation of the phenotypic characteristics and intensities of controlled-release fertilizers.

[0019] The Grubbs test was used to test the discrete values ​​of the controlled-release fertilizer sample data. The Grubbs statistic and its maximum value were calculated. The maximum value of the statistic was compared with the critical value in the Grubbs test table. If the maximum value of the statistic was greater than the critical value, the data was discarded as an outlier; otherwise, it was retained.

[0020] Optionally, the dimensionality reduction processing of the sample set data using principal component analysis includes:

[0021] The high-dimensional data is transformed into a low-dimensional representation through linear transformation. The projection direction is calculated to maximize the variance of the projected data. The data is then sorted according to the magnitude of its variance, and n principal component elements are calculated.

[0022] Optionally, constructing a controlled-release fertilizer intensity prediction model based on support vector machines using the dimensionality-reduced data includes:

[0023] Obtain the dimensionality-reduced data, set phenotypic features as input variables and fertilizer intensity as output variables, construct a model dataset consisting of multiple datasets, and establish a support vector machine using the model dataset;

[0024] Support vector machine optimization is performed by introducing a penalty factor and a positive relaxation variable.

[0025] By introducing the objective function, Lagrange multipliers, and the Lagrange equation, the problem is transformed into a dual problem, and the support vector machine regression function is obtained by solving it.

[0026] Optionally, the particle swarm optimization algorithm includes:

[0027] Initialize the random position and velocity of each particle;

[0028] Based on the evaluation criteria of each particle's current position, determine its individual optimal value and the global optimal value of the entire swarm.

[0029] Update the velocity and position of each particle, and iterate until the stopping condition is met.

[0030] Optionally, the optimization of controlled-release fertilizer intensity prediction model parameters using particle swarm optimization and k-fold function includes:

[0031] S1: Use principal component analysis to reduce the dimensionality of the sample set and then standardize the data after dimensionality reduction.

[0032] S2: Set the initial parameters, number of iterations, number of particles, penalty parameters, and kernel function parameters for the particle swarm optimization algorithm;

[0033] S3: Initialize the particle swarm within the specified search range using random parameter values ​​and calculate the fitness of each particle according to the k-fold function;

[0034] S4: Update the velocity and position of particles according to the particle swarm optimization algorithm, and track the global best fitness and particles;

[0035] S5: If the fitness of the particle is better than the global optimal fitness, then update the global optimal fitness and the global optimal particle and fit different kernel functions respectively;

[0036] S6: Stop when the maximum number of iterations is reached and obtain the optimal penalty parameter, optimal kernel function, and optimal kernel function parameter.

[0037] Optionally, it also includes: evaluating prediction performance through indicators, said indicators being: mean absolute error, root mean square error, and correlation coefficient.

[0038] Optionally, a controlled-release fertilizer intensity prediction system based on phenotypic features includes:

[0039] Data acquisition module: used to acquire phenotypic characteristics and intensity data of controlled-release fertilizers and construct a sample set;

[0040] Data processing module: used for dimensionality reduction of sample data using principal component analysis;

[0041] Model building module: used to build a controlled-release fertilizer intensity prediction model based on support vector machine based on the dimensionality-reduced data;

[0042] Model optimization module: used to optimize the parameters of the controlled-release fertilizer intensity prediction model using particle swarm optimization and k-fold function;

[0043] Prediction module: Used to obtain the phenotypic features of controlled-release fertilizer, input them into the optimized controlled-release fertilizer intensity prediction model, and obtain the controlled-release fertilizer intensity prediction results.

[0044] As can be seen from the above technical solution, compared with the prior art, the present invention discloses a method and system for predicting the intensity of controlled-release fertilizer based on phenotypic characteristics, which has the following beneficial effects:

[0045] This invention discloses a method for predicting the intensity of controlled-release fertilizer based on phenotypic features, comprising: acquiring phenotypic features and intensity data of controlled-release fertilizer, and constructing a sample set; performing dimensionality reduction processing on the sample set data using principal component analysis; constructing a controlled-release fertilizer intensity prediction model based on support vector machine based on the dimensionality-reduced data; optimizing the parameters of the controlled-release fertilizer intensity prediction model using particle swarm optimization and k-fold function; acquiring the phenotypic features of controlled-release fertilizer, inputting them into the optimized controlled-release fertilizer intensity prediction model, and obtaining the controlled-release fertilizer intensity prediction result.

[0046] This invention (1) uses the SVM prediction model optimized by particle swarm optimization algorithm and K-fold cross-validation to find the optimal parameters of the SVM model, thereby improving the accuracy of model parameter selection when predicting different controlled-release fertilizers and increasing the prediction accuracy.

[0047] (2) Intensity prediction of the most widely used controlled-release fertilizers was performed based on the PSO-SVM prediction model. The maximum error rates of the three controlled-release fertilizers (A, D, and M) were between -0.087 and 0.400, -0.050 and 0.340, and -0.024 and 0.328, respectively. The total number of error rates for each controlled-release fertilizer was 60, and the number of error intervals around 0 were 50, 41, and 57, respectively. This method of predicting the intensity of controlled-release fertilizers based on phenotypic features and the PSO-SVM prediction model has high accuracy.

[0048] (3) The intensity prediction performance of three controlled-release fertilizers was compared using random forest regression, K-nearest neighbors, BP neural network, LSTM neural network prediction models and PSO-SVM prediction model. The RMSE, MAE, and R of the PSO-SVM prediction model were compared. 2 All four prediction models outperformed the other four. The PSO-SVM prediction model demonstrated superior performance and accuracy, exhibiting simplicity, efficiency, objectivity, and universality, thus providing a theoretical basis for non-destructive testing methods for controlled-release fertilizer intensity. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0050] Figure 1 This is a schematic diagram of a controlled-release fertilizer intensity prediction method based on phenotypic features provided by the present invention.

[0051] Figure 2 This is a schematic diagram of the triaxial features of the controlled-release fertilizer provided by the present invention.

[0052] Figure 3 This is a schematic diagram of the texture features of the controlled-release fertilizer provided by the present invention.

[0053] Figure 4 A schematic diagram of the force-deformation curve provided by the present invention.

[0054] Figure 5 This is a schematic diagram of the phenotypic characteristics distribution of the controlled-release fertilizer provided by the present invention.

[0055] Figure 6 This is a schematic diagram of the texture feature distribution of the controlled-release fertilizer provided by the present invention.

[0056] Figure 7 This is a schematic diagram of the intensity distribution of the controlled-release fertilizer provided by the present invention.

[0057] Figure 8 This is a schematic diagram of the error rate distribution of the controlled-release fertilizer provided by the present invention. Detailed Implementation

[0058] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0059] This invention discloses a method for predicting the intensity of controlled-release fertilizers based on phenotypic features, such as... Figure 1As shown, the process includes: acquiring phenotypic features and intensity data of controlled-release fertilizers to construct a sample set; performing dimensionality reduction on the sample set data using principal component analysis; constructing a controlled-release fertilizer intensity prediction model based on support vector machine based on the dimensionality-reduced data; optimizing the parameters of the controlled-release fertilizer intensity prediction model using particle swarm optimization and k-fold function; acquiring the phenotypic features of controlled-release fertilizers and inputting them into the optimized controlled-release fertilizer intensity prediction model to obtain the controlled-release fertilizer intensity prediction result.

[0060] This invention proposes a method for predicting controlled-release fertilizer intensity based on phenotypic features. Three of the most widely used controlled-release fertilizers are selected, and their phenotypic features (sphericity, granularity, texture features, etc.) are acquired using machine vision. The intensity of the controlled-release fertilizer is obtained using a universal testing machine. Principal component analysis is used to reduce the dimensionality of the dataset, and particle swarm optimization and k-fold cross-validation are employed to obtain the optimal support vector machine parameters, thus constructing a PSO-SVM controlled-release fertilizer intensity prediction model based on the phenotypic features. To verify the efficiency of the PSO-SVM prediction model, random forest regression, K-nearest neighbors, BP neural network, and LSTM neural network are used to predict the intensity of controlled-release fertilizer based on phenotypic features. The mean absolute error, root mean square error, and correlation coefficient of each prediction model are compared.

[0061] In a specific embodiment, the phenotypic characteristics and intensity data of the controlled-release fertilizer include:

[0062] Phenotypic features refer to the geometric shape and surface characteristics of particles, mainly including particle length, width, thickness, sphericity, grain size, and surface texture. Sphericity, grain size, and surface texture are independent attributes; a change in one parameter will not affect the other two, and they can be used as parameters for prediction models.

[0063] (1) Triaxial characteristics

[0064] The macroscopic profile of a particle is described by three mutually perpendicular axes, called the major axis, the median axis, and the minor axis, which correspond to the particle's dimensions in the length, width, and thickness directions, respectively. In a stable state under natural placement, the particle's length (a) corresponds to the maximum dimension in its planar projection, its width (b) refers to the maximum dimension perpendicular to the length direction, and its thickness (c) refers to the longest straight-line distance of the particle perpendicular to both the length and width directions.

[0065] (2) Sphericity

[0066] Sphericity represents the degree of difference between the actual shape of a particle and that of a sphere. Waddell defines particle sphericity φ as: v s Let be the volume of the smallest sphere circumscribed by the particle.

[0067] Based on Waddell's definition of sphericity, Krumbein considers particles as equivalent to ellipsoids. Therefore, the equivalent volume v of a particle can be expressed as...

[0068] v = (π / 6)abc;

[0069] v s =(π / 6)a 3 ;

[0070] The overall sphericity φ of the particles is obtained as follows:

[0071]

[0072] In the formula: a, b, and c represent the length, width, and thickness of the particle, respectively.

[0073] (3) Particle size

[0074] Particle size (d) is a parameter used to describe particle size, representing the size of a single particle or the average diameter of a swarm of particles. For a single spherical particle, the particle size d is...

[0075]

[0076] (4) Texture features

[0077] Texture features are attributes that describe the texture structure and organization within an image region on the surface of an object. The main texture features within an image region on the surface of an object include: Angular Second Moment (ASM), which reflects the uniformity of grayscale variation and distribution in the image; the ASM eigenvalue is positively correlated with roughness; Contrast (CON), which reflects the depth and sharpness of the texture grooves in the image; increased contrast leads to greater texture groove depth and increased sharpness; Entropy (ENT), which measures the amount or richness of information in the image texture; a small entropy value indicates ordered texture, while a large entropy value indicates less regularity; and Inverse Difference Matrix (IDM), which measures the smoothness of the image texture; a larger IDM value indicates higher texture smoothness and stronger regularity. Figure 3 As shown.

[0078] (5) Strength

[0079] In production practice, strength(s) is a comprehensive indicator, encompassing aspects such as fracture resistance, abrasion resistance, and compressive strength. Fracture resistance is typically used to assess overall strength level. Fracture resistance refers to the maximum pressure an object or particle can withstand when subjected to vertical force.

[0080] In a specific implementation, the phenotypic characteristics and intensity data of controlled-release fertilizer are obtained, a sample set is constructed, and the sample set data is dimensionality reduced using principal component analysis, including:

[0081] (1) Image Acquisition and Preprocessing: Three of the most widely used controlled-release fertilizers were selected for prediction. The fertilizers included: Devodo controlled-release fertilizer (D fertilizer) produced by Hebei Devodo Fertilizer Co., Ltd., Aolv No. 1 controlled-release fertilizer (A fertilizer) produced by ICL Group Ltd., and Milak universal controlled-release fertilizer (M fertilizer) produced by Scotts Miracle-Gro. The image acquisition equipment mainly consisted of a DMC-GH4GK camera, a LUMIX zoom lens, a camera bracket, and a white background board. 100 controlled-release fertilizers of each of the three types were randomly selected and neatly arranged on the background board. The first shot was taken using a DSLR camera, ensuring that the controlled-release fertilizers remained in place. All the controlled-release fertilizers on the background board were flipped in turn for the second shot. This process was repeated three times for each type of controlled-release fertilizer to obtain three images, for a total of nine images of the three types of controlled-release fertilizers.

[0082] After image acquisition, each image was cropped into a single 100×100 pixel image containing only one controlled-release fertilizer seed. This resulted in 100 individual controlled-release fertilizer images per large image, 300 images for each of the three types of controlled-release fertilizers, for a total of 900 individual controlled-release fertilizer images. The hardware environment for data processing and algorithm development consisted of an Intel(R) Core(TM) i7-6700 CPU @ 3.40GHz and 16GB of memory, while the software environment was Python 3.7.

[0083] (2) Acquisition of triaxial features

[0084] To determine the length (a), width (b), and thickness (c), and subsequently calculate the sphericity and particle size parameters, the specific process is as follows: 1) Convert a single controlled-release fertilizer image to a grayscale image; 2) Perform edge detection using the Canny operator; 3) Locate and obtain the image contour; 4) Draw and calculate the minimum circumcircle and maximum incircle based on the contour; 5) For each controlled-release fertilizer, take three images and combine the three sets of data for comparison. Since the minimum circumcircle diameter represents the maximum size in the planar projection image, and the maximum incircle diameter represents the maximum size perpendicular to the length direction, the maximum minimum circumcircle diameter is equivalent to the fertilizer particle length 'a', the maximum maximum incircle diameter is equivalent to the fertilizer particle width 'b', and the minimum maximum incircle diameter is equivalent to the fertilizer particle thickness 'c'. The sphericity and particle size values ​​are then calculated. An example image showing the length, width, and thickness of a single controlled-release fertilizer is shown below. Figure 2 As shown.

[0085] (3) Acquisition of texture features

[0086] Texture features of controlled-release fertilizer are obtained using the gray-level co-occurrence matrix (GLCM) algorithm. The GLCM obtains texture features by statistically analyzing the frequency or probability of gray levels in pixel pairs. GLCM is non-invasive, does not modify the original image, captures global image information, is multi-scale and multi-directional, and has a simple and efficient computation process, providing the four texture features required in this paper. The four texture features—angular second moment (ASM), contrast (CON), entropy (ENT), and inverse difference (IDM)—are obtained using the GLCM algorithm. The specific acquisition process is as follows: 1) Define the maximum gray level as 16 levels; 2) Calculate the maximum number of gray levels in the image; 3) Calculate the GLCM of the image; 4) Read the image and calculate its texture feature values.

[0087] (4) Acquisition of strength

[0088] The crushing force of controlled-release fertilizer was obtained using a WDW-5E microcomputer-controlled electronic universal testing machine. The process for obtaining the crushing force is as follows: First, place the labeled fertilizer on a stage lined with acid-free paper. Move the indenter downwards to approximately 2 cm above the controlled-release fertilizer using the control handle. Use the software to zero all values, adjust the feed speed (0.5 mm / s), and click "Start." Wait for the fertilizer to be crushed. When a distinct crushing sound is heard and a clear peak point appears on the "force-deformation curve" on the display, click "Stop." The force (kN) at the peak point is the crushing force, and thus the strength is obtained. The force-deformation curve and its peak point are shown below. Figure 4 As shown.

[0089] In a specific implementation, a controlled-release fertilizer intensity prediction model based on support vector machine is constructed based on the dimensionality-reduced data; the parameters of the controlled-release fertilizer intensity prediction model are optimized using particle swarm optimization and k-fold function; the phenotypic features of the controlled-release fertilizer are obtained, specifically including:

[0090] Phenotypic characteristics and intensity of three controlled-release fertilizers were determined using images and a universal testing machine. After dimensionality reduction through principal component analysis combined with iterative particle swarm optimization, and based on a k-fold cross-validation function using support vector machines, the optimal hyperparameters and kernel function were obtained, thus constructing a predictive model for the intensity of the controlled-release fertilizers.

[0091] (1) Data collection and preprocessing

[0092] One hundred granules of three controlled-release fertilizers were randomly sampled. Phenotypic characteristic parameters were determined by photographing, cropping, and inputting the data into a calculation program. Strength parameters were also determined by crushing the granules using a universal testing machine. The phenotypic characteristic values ​​of the controlled-release fertilizers are as follows: Figure 5 , Figure 6 As shown, the intensity distribution is as follows Figure 7 As shown.

[0093] Depend on Figure 5 , Figure 6It can be seen that the distribution of phenotypic characteristic values ​​of controlled-release fertilizers shows a pattern of being thicker in the middle and thinner at both ends, which conforms to a normal distribution according to calculations. Fertilizer A has a sphericity distribution between 0.789 and 1.078 mm, a particle size distribution between 2.554 and 4.792 mm, and texture characteristic values ​​(ASM, CON, ENT, and IDM) distributed between 0.440 and 0.818, 0.342 and 0.932, 0.496 and 1.312, and 0.928 and 0.969, respectively. Fertilizer D has a sphericity distribution between 0.753 and 1.143 mm, a particle size distribution between 2.675 and 4.539 mm, and texture characteristic values ​​(ASM, CON, ENT, and IDM) distributed between 0.753 and 1.143 mm, and a particle size distribution between 2.675 and 4.539 mm. T and IDM are distributed between 0.448–0.783, 0.287–0.895, 0.569–1.396, and 0.920–0.970, respectively. The sphericity of fertilizer M is distributed between 0.673–0.990 mm, and the particle size between 2.479–4.539 mm. The texture feature values ​​ASM, CON, ENT, and IDM are distributed between 0.473–0.805, 0.347–0.964, 0.533–1.272, and 0.929–0.967, respectively. Among the same type of controlled-release fertilizer, the particle size and sphericity distributions are relatively dispersed, while the texture feature values ​​ASM, CON, ENT, and IDM are relatively concentrated. The texture feature values ​​of different types of controlled-release fertilizers are also relatively dispersed in terms of particle size and sphericity. Sphericity, particle size, and texture features all affect the predictive performance of the model and can be used as prediction input values.

[0094] pass Figure 7 It can be seen that the intensity distributions of different controlled-release fertilizers all conform to normal distribution curves. Specifically, fertilizer A's intensity distribution ranges from 0.013 to 0.169 kN, fertilizer D's from 0.036 to 0.087 kN, and fertilizer M's from 0.014 to 0.078 kN. The intensity distributions of the three controlled-release fertilizers are dispersed and can be used as predictive output values.

[0095] To verify the accuracy of the measurement, the experimental data were first statistically calculated to determine the average, range, and standard deviation of the phenotypic characteristics and intensity of each controlled-release fertilizer. The results are shown in Table 1.

[0096] Table 1

[0097]

[0098] To avoid the impact of outlier data caused by measurement errors on prediction performance, the Grubbs test was chosen to test the discrete values ​​of the raw parameters of the controlled-release fertilizer. First, the mean (Mean) and standard deviation (std) of the data set were calculated. Second, the Grubbs statistic (Gi) and its maximum value (GMax) were calculated. Finally, the maximum value (GMax) was compared with the critical value GPn in the Grubbs test table. If GMax is greater than GPn, then x... i Outliers must be discarded; otherwise, they are retained. The GMax calculation results for each parameter of the three controlled-release fertilizers are shown in Table 2. The formula for calculating the statistic Gi is...

[0099] In the formula, N is the sample size.

[0100] Table 2

[0101]

[0102] By consulting the Grubbs test table, setting α = 0.05 and n = 100, we obtain GPn = 3.207. Table 2 shows that the GMax values ​​for all parameters of the controlled-release fertilizer are less than GPn, proving that there are no outliers for any parameter of the controlled-release fertilizer, and that the data is valid and can be used to create a dataset.

[0103] (2) Support Vector Machine Regression

[0104] Because the relationship between texture features and intensity of controlled-release fertilizer is non-linear and the sample size is relatively small, Support Vector Machine (SVM) is chosen to construct a predictive model for the phenotypic features and intensity of controlled-release fertilizer. For the regression problem, SVM, based on feature mapping and optimization methods, establishes a function in a high-dimensional feature space to approximate the objective function, thereby constructing the optimal boundary and maximizing the boundary margin. When extracting texture features of controlled-release fertilizer, the input variables (phenotypic features) of each experimental group are multiplied by... i Corresponding output (fertilizer intensity) y i To construct a dataset G consisting of n samples, G = {(x i y i y i The expression for the linear hyperplane is:

[0105] Fy(x)=ωTx i +b;

[0106] In the formula, ωT is the weight vector, and b is the bias term. i The expression for the distance to the hyperplane is:

[0107]

[0108] Support vectors are scaled ω and b, and ensure that |ωTxi The vector closest to the hyperplane after +b|≥1 has a distance of 1 / ||ω||, such that 1 / ||ω|| is maximized, i.e., minimized. To improve the accuracy of SVM predictions, a penalty factor C>0 and a positive slack variable ξ are introduced. i The optimized formula is:

[0109]

[0110] Introducing the objective function Lagrange multiplier α i α j The problem is transformed from the Lagrange equation into a dual problem, and the SVM regression function is obtained by solving it. The expression is as follows:

[0111]

[0112] In the formula, K(x) i ,x j ) is the kernel function.

[0113] (3) Dimensionality reduction using PCA algorithm

[0114] Because the dataset has many features and a large range of values, dimensionality reduction is necessary. PCA is a commonly used dimensionality reduction technique and data analysis method. It transforms high-dimensional data into a low-dimensional representation through linear transformation while retaining the most important information. In PCA, principal components represent the main variance directions in the original data. The goal of the algorithm is to find projection directions that maximize the variance of the projected data. These projection directions are called principal components and are ordered according to their corresponding variance magnitudes. The parameters of the PCA algorithm are set to n_components = 2. Two principal component elements, X1 and X2, are obtained after calculation.

[0115] (4) Particle Swarm Optimization Algorithm

[0116] Particle swarm optimization (PSO) optimizes the algorithm by simulating the positions and velocities of particles. Initially, each particle is assigned a random position and velocity. Then, based on the evaluation criteria of each particle's current position, its individual best (pbest) and the global best (gbest) of the entire swarm are determined. Next, the velocity and position of each particle are updated according to the following formula, iterating until a stopping condition is met: the position of the i-th particle is pbesti, and the velocity update formula for each particle is as follows:

[0117] v i (t+1)=wv i (t)+c1r1(pbest i (t)-x i (t))+c2r2(gbest-x i(t));

[0118] The position update formula is

[0119] x i (t+1)=x i (t)+v i (t+1);

[0120] In the formula, v(t) is the velocity vector of the particle at time t, x(t) is the position vector of the particle at time t, w, c1, and c2 are weight parameters, and r1 and r2 are random numbers between 0 and 1.

[0121] (5) Optimize SVM parameters based on particle swarm optimization and k-fold function

[0122] The predictive performance of SVM is determined by the penalty parameter C, the kernel parameter gamma, and the kernel function. The penalty parameter C controls the degree of penalty for data that violates the boundary; the kernel parameter gamma controls the curvature of the decision boundary in Gaussian radial basis function and Sigmoid kernel function; the kernel function transforms the data from a low-dimensional feature space to a high-dimensional space through a nonlinear mapping, and different kernel functions are adapted to different data characteristics.

[0123] The specific steps for optimizing SVM parameters using the Particle Swarm Optimization (PSO) algorithm and the k-fold function are as follows: 1) Use the PCA algorithm to reduce the dimensionality of the initial dataset S and standardize the dataset; 2) Set the initial parameters of PSO: 100 iterations and 10 particles. The penalty parameter C ranges from 0.001 to 100, and the kernel function parameter gamma ranges from 0.001 to 100; 3) Initialize the particle swarm using random parameter values ​​within the specified search range and calculate the fitness of each particle according to the k-fold function (k=10); 4) Update the velocity and position of the particles according to the PSO algorithm, and track the global best fitness and the particle; 5) If the fitness of a particle is better than the global best fitness, update the global best fitness and the global best particle and fit different kernel functions respectively; 6) Stop when the maximum number of iterations is reached and obtain the optimal penalty parameter C, the optimal kernel function, and the optimal kernel function parameter gamma.

[0124] In specific implementation methods, the verification and analysis of the controlled-release fertilizer intensity pre-model includes:

[0125] Predictive performance is evaluated using three metrics: mean absolute error (MAE), root mean square error (RMSE), and correlation coefficient (R²). 2 ).

[0126] The Mean Absolute Error (MAE) calculates the average absolute error between predicted and actual values. It represents the average deviation between the predicted and actual values ​​and has an intuitive physical meaning. The mean absolute error is...

[0127]

[0128] RMSE measures the average of the squared differences between predicted and actual values, and is converted to the same units as the original data by taking the square root. The root mean square error is...

[0129]

[0130] R 2 The correlation coefficient measures how well the model fits the observed data. Values ​​range from 0 to 1, with values ​​closer to 1 indicating a better fit.

[0131]

[0132] In a specific implementation, evaluating predictive performance through indicators includes:

[0133] To verify the feasibility of predicting controlled-release fertilizer intensity based on phenotypic features and the accuracy and efficiency of the PSO-SVM prediction model, K-nearest neighbors, BP neural network, random forest regression, and LSTM neural network prediction models were selected to predict the intensity of three types of controlled-release fertilizers, and the mean absolute error (MAE), root mean square error (RMSE), and correlation coefficient (R²) were calculated. 2 The predicted performance metrics are shown in Table 3.

[0134] Table 3

[0135]

[0136] As shown in Table 3, the random forest regression and K-nearest neighbor prediction models exhibit high convergence and good fit, closely matching the evaluation metrics of the PSO-SVM prediction model. The BP neural network and LSTM neural network show larger prediction errors and lower prediction performance.

[0137] The PSO-SVM prediction model outperformed the other four prediction models in predicting the intensity of controlled-release fertilizer. For controlled-release fertilizer A, the RMSE and MAE of the PSO-SVM prediction model were 37.1, 74.1, 87.4, 86.7 and 56.1, 71.0, 87.0, 87.9 percentage points lower than those of the random forest regression, K-nearest neighbors, BP neural network, and LSTM neural network prediction models, respectively. 2The accuracy rates were 5.9, 12.7, 98.4, and 100.6 percentage points higher than the other four prediction models, respectively. For controlled-release fertilizer D, the RMSE and MAE of the PSO-SVM prediction model were 42.9, 78.4, 89.2, 89.7 and 52.2, 84.3, 92.0, 92.2 percentage points lower than those of the random forest regression, K-nearest neighbors, BP neural network, and LSTM neural network prediction models, respectively. 2 The accuracy rates were 5.8, 12.2, 94.8, and 146.5 percentage points higher than the other four prediction models, respectively. For controlled-release fertilizer M, the RMSE and MAE of the PSO-SVM prediction model were 47.1, 86.2, 92.4, 94.0 and 46.7, 82.3, 92.3, 94.1 percentage points lower than those of the random forest regression, K-nearest neighbors, BP neural network, and LSTM neural network prediction models, respectively. 2 They were 3.5, 11.8, 91.1, and 94.0 percentage points higher than the other four prediction models, respectively.

[0138] The PSO-SVM prediction model showed high performance in all three controlled-release fertilizer prediction metrics, with RMSE and MAE both less than 0.005. 2 All values ​​are greater than 0.99, indicating that the PSO-SVM prediction model can predict the intensity of different controlled-release fertilizers by using phenotypic features.

[0139] In a specific implementation, the error analysis of the PSO-SVM prediction model includes:

[0140] The accuracy of the PSO-SVM prediction model is evaluated using the error rate (e). The error rate distributions for controlled-release fertilizer A, controlled-release fertilizer D, and controlled-release fertilizer M are as follows: Figure 8 As shown, the error rate e is

[0141] In the formula, s1 is the actual value and s2 is the predicted value.

[0142] Depend on Figure 8It can be seen that the errors of the three controlled-release fertilizers are mainly concentrated around the 0 range, where 0 represents the same predicted and actual values. The more values ​​around the 0 range, the higher the prediction accuracy. Therefore, the PSO-SVM prediction model has high prediction accuracy for all three controlled-release fertilizers. For each controlled-release fertilizer (A, D, and M), there are 60 error rate data points. The maximum prediction error for controlled-release fertilizer A is between -0.087 and 0.400, with 50 errors (83.3%) falling around the 0 range. The maximum prediction error for controlled-release fertilizer D is between -0.050 and 0.340, with 41 errors (68.3%) falling around the 0 range. The maximum prediction error for controlled-release fertilizer M is between -0.024 and 0.328, with 57 errors (95.0%) falling around the 0 range. Therefore, the prediction model based on phenotypic features and PSO-SVM has high accuracy in predicting the intensity of controlled-release fertilizers.

[0143] In a specific implementation, a controlled-release fertilizer intensity prediction system based on phenotypic features includes:

[0144] Data acquisition module: used to acquire phenotypic characteristics and intensity data of controlled-release fertilizers and construct a sample set;

[0145] Data processing module: used for dimensionality reduction of sample data using principal component analysis;

[0146] Model building module: used to build a controlled-release fertilizer intensity prediction model based on support vector machine based on the dimensionality-reduced data;

[0147] Model optimization module: used to optimize the parameters of the controlled-release fertilizer intensity prediction model using particle swarm optimization and k-fold function;

[0148] Prediction module: Used to obtain the phenotypic features of controlled-release fertilizer, input them into the optimized controlled-release fertilizer intensity prediction model, and obtain the controlled-release fertilizer intensity prediction results.

[0149] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0150] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for predicting the intensity of controlled-release fertilizer based on phenotypic features, characterized in that, include: Obtain phenotypic characteristics and intensity data of controlled-release fertilizers, and construct a sample set; Principal component analysis was used to reduce the dimensionality of the sample data. Based on the dimensionality-reduced data, a controlled-release fertilizer intensity prediction model based on support vector machines is constructed, including: Obtain the dimensionality-reduced data, set phenotypic features as input variables and fertilizer intensity as output variables, construct a model dataset consisting of multiple datasets, and establish a support vector machine using the model dataset; Support vector machine optimization is performed by introducing a penalty factor and a positive relaxation variable. By introducing the objective function, Lagrange multipliers, and the Lagrange equation, the problem is transformed into a dual problem, and the support vector machine regression function is obtained by solving it. Parameter optimization of controlled-release fertilizer intensity prediction model was performed using particle swarm optimization and k-fold function. The particle swarm optimization algorithm includes: Initialize the random position and velocity of each particle; Based on the evaluation criteria of each particle's current position, determine its individual optimal value and the global optimal value of the entire swarm. Update the velocity and position of each particle, and iterate until the stopping condition is met; The optimization of controlled-release fertilizer intensity prediction model parameters using particle swarm optimization and k-fold function includes: S1: Use principal component analysis to reduce the dimensionality of the sample set and then standardize the data after dimensionality reduction. S2: Set the initial parameters, number of iterations, number of particles, penalty parameters, and kernel function parameters for the particle swarm optimization algorithm; S3: Initialize the particle swarm within the specified search range using random parameter values ​​and calculate the fitness of each particle according to the k-fold function; S4: Update the velocity and position of particles according to the particle swarm optimization algorithm, and track the global best fitness and particles; S5: If the fitness of the particle is better than the global optimal fitness, then update the global optimal fitness and the global optimal particle and fit different kernel functions respectively; S6: Stop when the maximum number of iterations is reached and obtain the optimal penalty parameter, optimal kernel function, and optimal kernel function parameter; The phenotypic features of controlled-release fertilizer are obtained and input into the optimized controlled-release fertilizer intensity prediction model to obtain the controlled-release fertilizer intensity prediction results.

2. The method for predicting the intensity of controlled-release fertilizer based on phenotypic features according to claim 1, characterized in that, The phenotypic features include: triaxial features, sphericity, granularity, and texture features.

3. The method for predicting the intensity of controlled-release fertilizer based on phenotypic features according to claim 1, characterized in that, The phenotypic characteristics and intensity data of the controlled-release fertilizer are obtained as follows: The strength of controlled-release fertilizer was obtained using a universal testing machine; Acquire images of controlled-release fertilizer, and obtain the triaxial features, sphericity, and particle size of the controlled-release fertilizer after image preprocessing; Texture features in the phenotypic features of controlled-release fertilizer are calculated using the gray-level co-occurrence matrix algorithm.

4. The method for predicting the intensity of controlled-release fertilizer based on phenotypic features according to claim 3, characterized in that, It also includes data validation of the phenotypic characteristics and intensity data of controlled-release fertilizers, the data validation including: First, perform statistical calculations on the sample set data to calculate the mean, range, and standard deviation of the phenotypic characteristics and intensities of controlled-release fertilizers. The Grubbs test was used to test the discrete values ​​of the controlled-release fertilizer sample data. The Grubbs statistic and its maximum value were calculated. The maximum value of the statistic was compared with the critical value in the Grubbs test table. If the maximum value of the statistic was greater than the critical value, the data was discarded as an outlier; otherwise, it was retained.

5. The method for predicting the intensity of controlled-release fertilizer based on phenotypic features according to claim 1, characterized in that, The dimensionality reduction process of the sample set data using principal component analysis includes: The high-dimensional data is transformed into a low-dimensional representation through linear transformation. The projection direction is calculated to maximize the variance of the projected data. The data is then sorted according to the magnitude of its variance, and n principal component elements are calculated.

6. The method for predicting the intensity of controlled-release fertilizer based on phenotypic features according to claim 1, characterized in that, Also includes: The predictive performance is evaluated by indicators, namely: mean absolute error, root mean square error, and correlation coefficient.

7. A controlled-release fertilizer intensity prediction system based on phenotypic features, employing the controlled-release fertilizer intensity prediction method based on phenotypic features according to any one of claims 1 to 6, characterized in that, include: Data acquisition module: used to acquire phenotypic characteristics and intensity data of controlled-release fertilizers and construct a sample set; Data processing module: used for dimensionality reduction of sample data using principal component analysis; Model building module: used to build a controlled-release fertilizer intensity prediction model based on support vector machine based on the dimensionality-reduced data; Model optimization module: used to optimize the parameters of the controlled-release fertilizer intensity prediction model using particle swarm optimization and k-fold function; Prediction module: Used to obtain the phenotypic features of controlled-release fertilizer, input them into the optimized controlled-release fertilizer intensity prediction model, and obtain the controlled-release fertilizer intensity prediction results.

Images