Soil heavy metal content estimation method and device

Abstract

The invention discloses a soil heavy metal content estimation method and device, and relates to the technical field of soil heavy metal estimation.The method comprises the steps of sample collection and data acquisition, specifically, based on geographic information of a research area, soil samples are collected through a layered and partitioned sampling method; according to the method, comprehensive and accurate collection and data acquisition of soil samples are realized through a layered and partitioned sampling method in combination with a high-precision positioning and measuring instrument, the sample representativeness and the data accuracy are ensured, and through an improved correlation analysis algorithm and a depth model based on feature fusion, the soil sampling accuracy is improved. Accurate analysis and efficient modeling of correlation between environmental magnetic parameters and heavy metal content are realized, prediction precision is improved, dynamic optimization of model parameters is realized by applying a self-adaptive step length strategy in model verification and optimization, model stability and generalization ability are guaranteed, and all modules of the device work cooperatively, so that the reliability of the device is improved. And deep analysis of heavy metal sources and migration and transformation rules is realized by using a geochemical analysis module and related models.

Description

Technical Field

[0001] The present invention relates to the technical field of soil heavy metal estimation, and specifically provides a method and device for estimating the content of heavy metals in soil. Background Art

[0002] Heavy metals in soil refer to metal elements with a density greater than 4.5 g / cm 3 , such as lead, cadmium, mercury, chromium, arsenic, etc. These elements have a certain natural background content in the soil. However, due to human activities, such as industrial production, mining, the use of agricultural chemical fertilizers and pesticides, and urban waste treatment, the content of heavy metals in the soil has been continuously increasing, exceeding the natural background value and the soil environmental capacity. Soil heavy metal pollution has characteristics such as concealment, long-term nature, and irreversibility. Therefore, it is crucial to estimate the content of heavy metals in the soil. Soil is the basis for the growth of crops. Excessive heavy metal content in the soil will affect the growth and development of crops, leading to reduced crop yields or even crop failures. Soil heavy metal pollution will also disrupt the balance of the soil ecosystem, affect the activity and diversity of soil microorganisms, and further affect the self-purification ability and nutrient cycling function of the soil. Therefore, accurately estimating the content of heavy metals in the soil is of great significance for ensuring the quality and safety of agricultural products, maintaining human health, and protecting the soil ecological environment.

[0003] Currently, in the field of estimating the content of heavy metals in soil, although a variety of technologies and methods have been developed, there are still some technical problems. Traditional laboratory detection methods, such as atomic absorption spectrometry, inductively coupled plasma mass spectrometry, etc., although they can accurately determine the content of heavy metals in the soil, these methods require professional experimental equipment and technical personnel, the sample pretreatment process is complex, the detection cycle is long, and the cost is relatively high, making it difficult to meet the needs of large-scale and rapid detection. Some rapid detection methods based on technologies such as remote sensing and spectroscopy, although they can achieve on-site rapid detection, the detection accuracy is greatly affected by various factors such as soil type, vegetation cover, topography, etc., and the accuracy needs to be improved. In addition, most of the existing methods focus on the determination of the content of heavy metals in the soil, and relatively few studies have been conducted on the sources and migration and transformation laws of heavy metals in the soil, making it difficult to fundamentally provide comprehensive and in-depth scientific basis for soil pollution prevention and control. Therefore, it is of great significance to develop a method and device for estimating the content of heavy metals in soil that can comprehensively achieve the above characteristics. Summary of the Invention

[0004] The purpose of the present invention is to make up for the deficiencies of the existing technology, and provides a method and device for estimating the content of heavy metals in soil. It can improve the accuracy and scientific nature of soil heavy metal content detection through the combination of environmental magnetism and geochemistry, and at the same time deeply study the sources and migration and transformation laws of heavy metals, so as to fundamentally provide comprehensive and in-depth scientific basis for soil pollution prevention and control.

[0005] To solve the above technical problems, the present invention provides the following technical solution: A method for estimating the heavy metal content in soil, the method comprising the following steps:

[0006] Sample collection and data acquisition: Based on the geographical information of the research area, soil samples are collected using the stratified and zoned sampling method, and the positions of the sampling points are recorded using high-precision positioning equipment. At the same time, the environmental magnetic parameters and heavy metal content of the samples are measured. The environmental magnetic parameters include magnetic susceptibility χ and frequency magnetic susceptibility χ fd %;

[0007] Correlation analysis: An improved association analysis algorithm is used to study the correlation between the environmental magnetic parameters and the heavy metal content. The algorithm constructs an adaptive weight matrix W, and the weight w ij is determined according to the distance d ij between sample points and the feature similarity The formula is where σ is dynamically adjusted according to the dispersion degree of the sample data, and the correlation degree between the magnetic parameters and the heavy metal content is calculated through this matrix;

[0008] Establish a magnetic detection model: A deep model M based on feature fusion is constructed. The input layer of the model is the selected significantly correlated magnetic parameters. After multi-layer non-linear transformation, the predicted heavy metal content is output. The inter-layer weights are determined by an improved random search algorithm combined with the dynamic weight α i of the sample. α i is adjusted according to the evaluation index E i of the contribution of the sample to the overall model. E i is calculated through the formula where y ij is the true value, is the predicted value, and the training objective of the model is to minimize the loss function

[0009] Model verification and optimization: The sample data is divided into a training set, a validation set, and a test set according to the spatial and feature distributions. The model is trained using the training set, the model parameters are adjusted according to the comprehensive indicators on the validation set, and the parameters are updated using an adaptive step size strategy. The step size η is adjusted according to the change rate ΔL of the validation set loss. The formula is where η0 is the initial step size and λ is the adjustment factor, is the average loss change rate. When the validation set indicators are stable, the model is evaluated using the test set;

[0010] Analysis combined with geochemical knowledge: By applying the principle of geochemical equilibrium and combining with the regional geological background, a model for analyzing the sources of heavy metals is established. By analyzing the ratio relationship between heavy metals and associated elements in the soil and combining with isotope characteristics, the contribution rate of heavy metal sources is determined. At the same time, a model for the migration and transformation of heavy metals is constructed based on the soil geochemical properties, which include the pH value of acidity and alkalinity and the redox potential Eh. For any heavy metal element M, its migration rate equation is where k1 - k4 are rate constants determined by experiments combined with theoretical calculations, and [M] ads 、[M] sol 、[M] pre 、[M] dis represent the concentrations of the heavy metal element M in the adsorbed state, dissolved state, precipitated state, and desorbed state, respectively.

[0011] Furthermore, in the sample collection and data acquisition step, the stratified and zoned sampling method first divides the study area into different texture categories according to soil texture, including sandy soil, loam, and clay texture types. Each texture category is further subdivided into different sub - regions according to topographic and geomorphic features, including mountainous, plain, and hilly terrain regions. For each sub - region, it is divided into three main layers according to soil depth, namely the surface layer (0 - 20 cm), middle layer (20 - 50 cm), and deep layer (50 - 100 cm). Sampling points are selected at a predetermined grid spacing within each layer. The environmental magnetic parameters are measured by magnetic measurement instruments, and the heavy metal content is measured using inductively coupled plasma mass spectrometry.

[0012] Even further, in the correlation analysis step, the feature similarity is determined by calculating the cosine of the angle between sample points in the feature space. The specific formula is where x ik and x jk represent the values of the i - th and j - th samples on the k - th feature dimension respectively, d is the number of feature dimensions, and the dynamic adjustment of σ is based on the standard deviation σ data of the sample data. The formula is σ = μ·σ data , where μ is an empirical coefficient initially set to 0.5 and is fine - tuned according to the stability of the results of each correlation analysis in subsequent analyses, with a range between 0.3 - 0.7 and determined by multiple experimental optimizations. When constructing the adaptive weight matrix W, the weight matrix is regularized, an L2 - norm regularization term is added, and the objective function is modified to where λ is the regularization coefficient, and the optimal value is selected through cross - validation.

[0013] Furthermore, in the step of establishing the magnetic detection model, in each iteration of the improved random search algorithm, according to the sample dynamic weight α i Adjust the search range and step size. The adjustment of the search range is determined according to the distribution range of the sample dynamic weight. For samples with a larger dynamic weight, expand the search range; for samples with a smaller dynamic weight, narrow the search range. The search step size is adjusted according to the current iteration number t and the total iteration number T. The step size formula is where s0 is the initial step size, and the multi-layer non-linear transformation of the model uses an improved activation function β is adaptively adjusted according to the mean μ x of the neuron input and the variance The formula is At the same time, add a sparsity constraint to the hidden layer of the model to constrain the output of the neurons.

[0014] Furthermore, in the model verification and optimization step, in the adaptive step size strategy, the adjustment factor λ is adjusted by the fluctuation coefficient ω on the validation set. ω reflects the fluctuation of the loss function. The specific calculation is where L t and L t-1 are the loss function values of the validation set for the t-th and (t - 1)-th iterations respectively. When ω is greater than the threshold ω0, increase λ; otherwise, decrease λ. During the model verification process, introduce an early stopping mechanism. Set a performance metric window size of w. If the performance metric on the validation set does not improve significantly in consecutive w iterations, stop training and save the current optimal model. At the same time, to further optimize the model performance, during the model training process, adopt the model ensemble technology, combine multiple trained models by weighted averaging, and determine the weights according to the performance of the models on the validation set.

[0015] Furthermore, in the step of analyzing by combining geochemical knowledge, the heavy metal source analysis model uses the positive definite matrix factorization technology, combines the regional pollution source information, and determines the contribution rate of each source to the heavy metals. The specific operation is to form a matrix X with the concentration data of heavy metals and related tracer elements in the soil samples, and decompose the matrix X into a source component matrix F, a source contribution matrix G, and an error matrix E, that is, X = FG + E. Solve for F and G by minimizing the objective function where I is the number of samples, J is the number of elements, K is the number of sources, σ ij is the measurement error, and λ and γ are sparsity penalty coefficients. Select the sparsity characteristics considering the pollution sources through cross-validation.

[0016] Furthermore, in the heavy metal migration and transformation model, the rate constants k1 - k4 are determined by combining indoor simulation experiments and theoretical derivations, considering factors such as soil organic matter content and clay mineral type. In the indoor simulation experiments, microenvironments with different soil conditions are constructed, and factors such as soil pH, Eh, organic matter content, and clay mineral type are changed. The change of heavy metal concentration over time is monitored, and the rate constants are obtained by fitting the experimental data using the non - linear least - squares method. At the same time, considering the influence of microorganisms on heavy metal migration and transformation, the activity and species of microorganisms are added as influencing factors to the migration and transformation model to further refine the model. For any heavy metal element M, its migration and transformation model can be expressed as where f(S) is a function related to the activity and species of microorganisms, and f(S)=k6·M A ·[M] sol , M A represents the microbial activity, and k6 is the influence coefficient of microorganisms on the migration and transformation of this heavy metal, which is determined by microbial culture experiments and soil microenvironment simulation experiments.

[0017] An estimation device for soil heavy metal content, which includes a sample collection module, a magnetic parameter detection module, a data processing and modeling module, a model verification and optimization module, a geochemical analysis module, and a result output module;

[0018] The sample collection module includes a positioning mechanism and a sampling mechanism. The positioning mechanism is guided by a geographic information system and uses a positioning chip to obtain the precise location of the sampling point. The sampling mechanism consists of an adjustable robotic arm and a sampling probe, which automatically collects soil samples at different depths and positions according to a preset plan and records environmental parameters;

[0019] The magnetic parameter detection module is equipped with magnetic measurement instruments to measure the magnetic parameters of soil samples and pre - process the data;

[0020] The data processing and modeling module takes a high - performance processor as the core and runs customized algorithm software. This module receives sample location, environmental parameters, and magnetic parameters, performs standardization processing, and then conducts correlation analysis and modeling using an improved association analysis algorithm and a deep model based on feature fusion;

[0021] The model verification and optimization module has the functions of intelligent data partitioning and dynamic evaluation. It divides the data set according to sample characteristics and adopts an adaptive step - size strategy to optimize model parameters during model training;

[0022] The geochemical analysis module has a built - in geochemical database. Combining the results of the magnetic detection model, it uses geochemical principles and models to analyze the sources and migration and transformation laws of heavy metals;

[0023] The result output module supports multiple output methods, and displays the estimation results and analysis conclusions in the form of charts and reports.

[0024] Further, the sampling probe of the sample collection module includes a spiral type and a piston type. The probe type is automatically selected according to the soil texture. The resistance during sampling is monitored by a pressure sensor, and the sampling depth is automatically adjusted according to the resistance magnitude. The environmental parameter sensor is integrated on the sampling head, including a temperature sensor, a humidity sensor, a barometric pressure sensor, and a soil moisture sensor, which monitors and stores the environmental parameters at the sampling site in real time. The stored data also includes timestamp information, which is used to analyze the impact of the spatio-temporal changes of the soil environment on the heavy metal content subsequently.

[0025] Furthermore, the high-performance processor of the data processing and modeling module adopts a multi-core architecture, has parallel computing capabilities, supports multi-threaded processing. The customized algorithm software adopts a modular design, including a correlation analysis module, a model construction module, a model training module, and a model evaluation module. The module supports remote data transmission and cloud computing, and uses cloud resources to accelerate data processing and model training. When performing data processing, data encryption technology is used to encrypt the data during transmission and storage.

[0026] Compared with the prior art, the method and device for estimating the heavy metal content in soil have the following

[0027] Beneficial effects:

[0028] First, through the hierarchical and zonal sampling method, combined with high-precision positioning and measurement instruments, the present invention realizes the comprehensive and accurate collection of soil samples and data acquisition, ensures the representativeness of the samples and the accuracy of the data. Through the improved correlation analysis algorithm and the deep model based on feature fusion, the present invention realizes the accurate analysis and efficient modeling of the correlation between environmental magnetic parameters and heavy metal content, and improves the prediction accuracy.

[0029] Second, in the model verification and optimization, the present invention uses an adaptive step size strategy, etc., to realize the dynamic optimization of model parameters, ensure the stability and generalization ability of the model. Each module of the device works in coordination, and uses the geochemical analysis module and related models to realize the in-depth analysis of the sources and migration and transformation laws of heavy metals, provide a comprehensive scientific basis for soil pollution prevention and control, effectively solve the deficiencies of the prior art, and improve the comprehensive level of soil heavy metal detection.

[0030] Other advantages, objectives, and features of the present invention will be described to some extent in the subsequent specification, and to some extent, will be obvious to those skilled in the art based on the study of the following text, or can be taught from the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on these drawings.

[0032] Figure 1 It is a schematic flowchart of a method for estimating the heavy metal content in soil;

[0033] Figure 2 It is a schematic structural diagram of a device for estimating the heavy metal content in soil. Detailed implementation manners

[0034] To further elaborate on the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following, in combination with the accompanying drawings and preferred embodiments, will detail the specific implementation manners, structures, features, and their effects according to the present invention.

[0035] Embodiment 1

[0036] Around a large mining area, due to long-term ore mining and smelting activities, the surrounding soil is at extremely high risk of heavy metal pollution. These heavy metals may not only affect the growth of surrounding crops but also endanger human health through the food chain. To effectively monitor and control the soil pollution in this area, an accurate and efficient soil heavy metal content estimation scheme is urgently needed to comprehensively understand the soil heavy metal pollution situation and provide a scientific basis for subsequent treatment measures.

[0037] Using the sample collection module, based on the geographic information system data of the area, first divide the area around the mining area into different regions according to soil types, and then further subdivide according to topography and landforms. Use an adjustable robotic arm and a variety of sampling probes to conduct stratified and zoned sampling at different depths. The robotic arm is made of lightweight and high-strength alloy materials, has multiple joints, can operate flexibly in three-dimensional space, achieve 360° rotation and multi-angle extension, and can automatically collect soil samples according to a preset plan. The sampling probe is automatically replaced according to the soil texture, and the sampling resistance is monitored in real time through a pressure sensor to ensure accurate sampling depth. At the same time, the environmental parameter sensor integrated on the sampling head records the environmental parameters at the sampling site in real time, and together with the sample position information, is recorded and stored by a high-precision positioning chip.

[0038] The collected soil samples are quickly sent to the magnetic parameter detection module, which is equipped with professional magnetic measurement instruments, such as a superconducting quantum interference magnetometer and an alternating gradient magnetometer, to quickly and accurately measure the magnetic susceptibility χ and frequency magnetic susceptibility χ of the samples fdEnvironmental magnetic parameters such as %, and at the same time, using professional analysis instruments, the inductively coupled plasma mass spectrometry method is used to determine the contents of heavy metals such as lead, cadmium, and mercury in soil samples.

[0039] Transmit the environmental magnetic parameters and heavy metal content data obtained in the sample collection and data acquisition steps to the data processing and modeling module. This module uses an improved correlation analysis algorithm to conduct a correlation study on the data. First, calculate the distance d i between sample points j for the feature similarity φ ij According to the formula Calculate, where x i k and x j k represent the values of the i-th and j-th samples on the k-th feature dimension respectively, and d is the number of feature dimensions.

[0040] According to the standard deviation σ data of the sample data, dynamically adjust σ. The formula is σ = μ·σ data , initially set μ to 0.5, and then fine-tune it between 0.3 - 0.7 according to the stability of the results of each correlation analysis. Through the formula Construct an adaptive weight matrix W to calculate the correlation degree between magnetic parameters and heavy metal content. To avoid overfitting, regularize the weight matrix and add an L2 norm regularization term to the objective function Select the optimal regularization coefficient λ through cross-validation.

[0041] Construct a deep model M based on feature fusion in the data processing and modeling module. Use the magnetic parameters significantly related to the heavy metal content screened out as the input layer. The multi-layer non-linear transformation of the model uses an improved activation function where β is adaptively adjusted according to the mean μ x of the neuron input and the variance . The formula is

[0042] The inter-layer weight is determined by an improved random search algorithm combined with the dynamic weight α i of the sample. In each iteration of the improved random search algorithm, according to the dynamic weight α i of the sample, adjust the search range and step size. The search range is adjusted according to the distribution range of the sample dynamic weight, and the step size is adjusted according to the formula Adjust, where s0 is the initial step size, t is the current iteration number, T is the total iteration number, and the sample dynamic weight α i is adjusted according to the evaluation index E i of the contribution of the sample to the overall model. E i is calculated through the formula Calculate, yij is the true value is the predicted value, and the model training objective is to minimize the loss function Meanwhile, a sparsity constraint is added to the hidden layer of the model to constrain the output of neurons. By adding a sparsity penalty term to the loss function, the model is guided to automatically learn more representative features.

[0043] The model validation and optimization module divides the sample data into a training set, a validation set, and a test set according to the spatial and feature distributions. The model is trained using the training set, and the model parameters are adjusted based on comprehensive indicators such as the root mean square error and mean absolute error on the validation set. The parameters are updated using an adaptive step size strategy, and the step size η is adjusted according to the change rate ΔL of the validation set loss. The formula is where η0 is the initial step size and λ is the adjustment factor is the average loss change rate, and the adjustment factor λ is adjusted by the fluctuation coefficient ω on the validation set. ω is calculated according to the formula where L t and L t -1 are the validation set loss function values for the t-th and (t - 1)-th iterations respectively. When ω is greater than the threshold ω0, λ is increased; otherwise, λ is decreased.

[0044] An early stopping mechanism is introduced. The performance metric window size w = 10 is set. If the improvement in the validation set performance metric is less than e in consecutive w iterations, the training is stopped and the current optimal model is saved. At the same time, the model ensemble technology is adopted, and multiple trained models are combined through weighted averaging. The weights are determined according to the performance of the models on the validation set.

[0045] The data processing and modeling module transmits the model prediction results to the geochemical analysis module. This module has a built-in geochemical database. Combining the regional geological background, a heavy metal source analysis model is established using the geochemical equilibrium principle. The concentration data of heavy metals and related tracer elements in soil samples are formed into a matrix X. Using the positive definite matrix factorization technique, the matrix X is decomposed into a source component matrix F, a source contribution matrix G, and an error matrix E, that is, X = FG + E. By minimizing the objective function to solve for F and G, where I is the number of samples, J is the number of elements, K is the number of sources, and σ ij is the measurement error, and λ and γ are sparsity penalty coefficients selected through cross-validation. Combining the regional pollution source information, the contribution rate of each source to heavy metals is determined.

[0046] Based on the geochemical properties of soil such as the soil pH value and redox potential Eh, a heavy metal migration and transformation model is constructed. For any heavy metal element M, its migration rate equation is where k1 - k4 are rate constants, which are determined by combining experiments with theoretical calculations, [M] ads 、[M] sol 、[M] pre 、[M] dis represent the concentrations of heavy metal element M in the adsorbed state, dissolved state, precipitated state, and desorbed state respectively. Considering factors such as soil organic matter content and clay mineral type, the rate constants are determined by combining indoor simulation experiments with theoretical derivation.

[0047] In the indoor simulation experiment, microenvironments with different soil conditions are constructed, and factors such as soil pH, Eh, organic matter content, and clay mineral type are changed. The changes in heavy metal concentrations over time are monitored, and the rate constants are obtained by fitting the experimental data using the non - linear least - squares method. At the same time, considering the influence of microorganisms on the migration and transformation of heavy metals, the activity and species of microorganisms are added as influencing factors to the migration and transformation model to further refine the model. For any heavy metal element M, its migration and transformation model can be expressed as where f(S) is a function related to the activity and species of microorganisms, f(S)=k6·M A ·[M] sol ,M A represents the microorganism activity, and k6 is the influence coefficient of microorganisms on the migration and transformation of this heavy metal, which is determined by microorganism culture experiments and soil microenvironment simulation experiments.

[0048] The result output module displays the results of soil heavy metal content predicted by the magnetic detection model and the analysis results of heavy metal sources and migration and transformation laws obtained by the geochemical analysis module on a high - resolution display screen in the form of intuitive charts and detailed reports. The charts include bar charts, line charts, and contour maps, and the reports include text descriptions, data analysis results, and conclusion suggestions. At the same time, a paper report can be output through a printer, or the results can be transmitted to external devices through wired or wireless networks, facilitating viewing and analysis by relevant personnel.

[0049] In summary, through the complete technical solution of this embodiment, the accurate estimation of soil heavy metal content around the mining area is achieved. Compared with the traditional method, the root - mean - square error and absolute error of model prediction are both reduced, significantly improving the detection accuracy. It is clear that the main sources of soil heavy metals in this area are ore mining and smelting activities, providing a strong basis for controlling pollution sources. At the same time, the migration and transformation laws of heavy metals in the soil are clearly understood, which helps to formulate more targeted soil pollution remediation plans and is of great significance for protecting the surrounding ecological environment and the health of residents.

[0050] Example Two

[0051] In a large-scale agricultural planting area, the long-term unreasonable use of chemical fertilizers and pesticides and the discharge of nearby industrial wastewater have led to a decline in soil quality, affecting the yield and quality of crops. To ensure the quality and safety of agricultural products and achieve sustainable agricultural development, it is necessary to accurately understand the status of soil heavy metal content in order to take scientific and reasonable soil improvement and planting management measures.

[0052] With the help of the sample collection module, based on the detailed geographical information of the agricultural planting area, it is stratified and partitioned according to different planting areas and soil fertility levels. Different planting areas include grain planting areas, vegetable planting areas, etc. The flexible robotic arm in the sample collection module is used, equipped with a variety of sampling probes suitable for different soil conditions, to sample at different depth levels. The robotic arm can achieve multi-degree-of-freedom movement and flexibly extend in three-dimensional space. It automatically adjusts the sampling action according to the feedback of soil texture to ensure that representative samples are collected. At the same time, the position of the sampling point is accurately recorded through a high-precision positioning chip, and the environmental sensors integrated on the sampling head continuously monitor and record environmental parameters such as temperature, humidity, and light intensity, providing comprehensive data support for subsequent analysis.

[0053] The collected soil samples are immediately sent to the magnetic parameter detection module. Using professional magnetic measurement instruments, such as high-precision alternating gradient magnetometers and frequency susceptibility meters, the magnetic susceptibility χ, frequency magnetic susceptibility χ fd % and other environmental magnetic parameters of the soil samples are quickly measured. At the same time, using advanced analytical instruments, inductively coupled plasma mass spectrometry is used to accurately measure the content of heavy metal elements such as lead, cadmium, and mercury in the soil samples to ensure the high precision and reliability of the data.

[0054] The obtained environmental magnetic parameters and heavy metal content data are transmitted to the data processing and modeling module. An improved correlation analysis algorithm is used to conduct a correlation study, and the distance d i between sample points is calculated. For feature similarity According to the formula where x ik and x jk represent the values of the i-th and j-th samples on the k-th feature dimension respectively, d is the number of feature dimensions, and according to the standard deviation σ data of the sample data, σ is dynamically adjusted. The formula is σ = μ·σ data . The initial μ is set to 0.5, and then it is fine-tuned between 0.3 - 0.7 according to the stability of the results of each correlation analysis. Through the formula An adaptive weight matrix W is constructed to calculate the correlation degree between magnetic parameters and heavy metal content. To avoid overfitting, the weight matrix is regularized, and an L2 norm regularization term is added to the objective function The optimal regularization coefficient λ is selected through cross-validation.

[0055] In the data processing and modeling module, a deep model M based on feature fusion is constructed. The magnetic parameters significantly related to the heavy metal content screened out are used as the input layer, and an improved activation function is adopted for the multi-layer non-linear transformation of the model β is adaptively adjusted according to the mean μ x and variance of the neuron input, and the formula is

[0056] The inter-layer weights are determined by an improved random search algorithm combined with the dynamic weight α of the samples. In each iteration of the improved random search algorithm, according to the dynamic weight α of the samples i the search range and step size are adjusted. The search range is adjusted according to the distribution range of the dynamic weight of the samples, and the step size is adjusted according to the formula i where s0 is the initial step size, t is the current iteration number, T is the total iteration number, and the dynamic weight α of the samples is adjusted according to the evaluation index E of the contribution of the sample to the overall model i E i is adjusted by the formula i through the formula calculated, y ij is the true value is the predicted value, and the model training aims to minimize the loss function At the same time, a sparsity constraint is added to the hidden layer, and by adding a sparsity penalty term to the loss function, the model is guided to learn more representative features.

[0057] In the model validation and optimization module, according to the spatial distribution and feature attributes of the samples, the data is divided into a training set, a validation set, and a test set. The model is trained using the training set, and the model parameters are adjusted based on comprehensive indicators such as the root mean square error and mean absolute error on the validation set.

[0058] An adaptive step size strategy is adopted to update the parameters. The step size η is adjusted according to the change rate ΔL of the loss on the validation set, and the formula is η0 is the initial step size, λ is the adjustment factor is the average loss change rate, and the adjustment factor λ is adjusted according to the fluctuation coefficient ω on the validation set. ω is calculated by the formula calculated, L t and L t-1 are the validation set loss function values for the t-th and (t - 1)-th iterations respectively. When ω is greater than the threshold ω0, increase λ; otherwise, decrease λ. Introduce an early stopping mechanism, set the performance metric window size w = 10. If the improvement in the validation set performance metric is less than e in consecutive w iterations, stop the training, save the optimal model. At the same time, adopt the model ensemble technique, and perform weighted average combination on multiple trained models according to their performance on the validation set.

[0059] The data processing and modeling module transmits the model prediction results to the geochemical analysis module. This module combines the geochemical background of the agricultural planting area and uses the geochemical equilibrium principle to establish a heavy metal source analysis model. It forms a matrix X with the concentration data of heavy metals and related tracer elements in the soil samples. Using the positive definite matrix factorization technique, the matrix X is decomposed into a source component matrix F, a source contribution matrix G, and an error matrix E, that is, X = FG + E. By minimizing the objective function to solve for F and G, where I is the number of samples, J is the number of elements, K is the number of sources, and σ ij is the measurement error, and λ and γ are sparsity penalty coefficients selected through cross-validation. Combining the regional pollution source information, determine the contribution rate of each source to the heavy metals.

[0060] Based on the geochemical properties of the soil such as the soil acidity pH and redox potential Eh, construct a heavy metal migration and transformation model. For any heavy metal element M, its migration rate equation is where k1 - k4 are rate constants determined through experiments combined with theoretical calculations, and [M] ads 、[M] sol 、[M] pre 、[M] dis represent the concentrations of the heavy metal element M in the adsorbed state, dissolved state, precipitated state, and desorbed state respectively. Considering factors such as soil organic matter content and clay mineral type, determine the rate constants through a combination of indoor simulation experiments and theoretical derivations.

[0061] In the indoor simulation experiment, construct microenvironments with different soil conditions, change factors such as soil pH, Eh, organic matter content, and clay mineral type, monitor the change of heavy metal concentration over time, and obtain the rate constants by fitting the experimental data using the non-linear least squares method. At the same time, considering the influence of microorganisms on the heavy metal migration and transformation, add the activity and species of microorganisms as influencing factors to the migration and transformation model to further refine the model. For any heavy metal element M, its migration and transformation model can be expressed as where f(S) is a function related to the activity and species of microorganisms, f(S) = k6·MA·[M] sol ,M AIt represents the microbial activity, and k6 is the influence coefficient of microorganisms on the migration and transformation of this heavy metal, which is determined through microbial culture experiments and soil microenvironment simulation experiments.

[0062] The result output module displays the results of the soil heavy metal content predicted by the magnetic detection model and the analysis results of the heavy metal source and migration and transformation law obtained by the geochemical analysis module on a high-resolution display screen in the form of intuitive and easy-to-understand charts and detailed reports. The charts include a pie chart showing the proportion of heavy metal sources and a line chart showing the change of heavy metal content at different depths, etc. The report includes instructions on planting suggestions and soil improvement measures. At the same time, it supports printing a paper report through a printer or sending the results to the farmer's mobile device via wireless transmission, facilitating the farmer to obtain information in a timely manner and take corresponding measures.

[0063] In summary, by implementing this technical solution, the accurate estimation of soil heavy metal content is achieved, the prediction accuracy of the model is improved compared with the past, it is clear that part of the soil heavy metals in this area come from industrial wastewater discharge and part is related to the unreasonable use of chemical fertilizers, providing a direction for targeted reduction of pollution input. At the same time, the migration and transformation law of heavy metals in the soil of different planting areas is clearly mastered, which helps to formulate personalized soil improvement plans and planting management strategies, such as adjusting the types and methods of fertilization and selecting suitable heavy metal pollution-resistant crop varieties, etc., and has important significance for ensuring the quality and safety of agricultural products and the sustainable development of agriculture.

[0064] The above is only a preferred embodiment of the present invention and does not impose any form of limitation on the present invention. Although the present invention has been disclosed above with a preferred embodiment, it is not intended to limit the present invention. Any person skilled in the art can make some changes or modifications to the above-disclosed technical content to obtain equivalent embodiments with equivalent changes. However, as long as it does not depart from the technical content of the present invention, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention still fall within the scope of the technical solution of the present invention.

Claims

1. A method for estimating the content of heavy metals in soil, characterized in that: The method comprises the following steps: Sample collection and data acquisition: Based on the geographical information of the study area, soil samples were collected using a stratified and zoned sampling method. The locations of the sampling points were recorded using high-precision positioning equipment. The environmental magnetic parameters and heavy metal content of the samples were measured at the same time. The environmental magnetic parameters included magnetic susceptibility χ, frequency magnetic susceptibility χ fd %; Correlation analysis: The improved correlation analysis algorithm is used to study the correlation between environmental magnetic parameters and heavy metal content. The algorithm constructs an adaptive weight matrix W, weight w ij According to the distance d between sample points ij and feature similarity OK, the formula is Among them, σ is dynamically adjusted according to the discrete degree of sample data, and the correlation between magnetic parameters and heavy metal content is calculated through this matrix; Establish a magnetic detection model: Construct a deep model M based on feature fusion. The model input layer is the screened significant related magnetic parameters. After multiple layers of nonlinear transformation, the predicted heavy metal content is output. The inter-layer weight The improved random search algorithm is combined with the dynamic weight α of the sample i OK, α i Evaluation index E based on the sample's contribution to the overall model i Adjustment, E i By formula Calculate, where y ij is the true value, is the predicted value, and the model training goal is to minimize the loss function Model verification and optimization: Divide the sample data into training set, validation set and test set according to the spatial and feature distribution. Use the training set to train the model, adjust the model parameters according to the comprehensive indicators on the validation set, and use the adaptive step size strategy to update the parameters. The step size η is adjusted according to the loss change rate ΔL of the validation set. The formula is: Where η0 is the initial step size, λ is the adjustment factor, is the average loss change rate. When the validation set indicator is stable, the test set is used to evaluate the model; Combined with geochemical knowledge analysis: Using the principle of geochemical equilibrium and combining with the regional geological background, a heavy metal source analysis model is established. By analyzing the ratio of heavy metals to associated elements in the soil and combining isotope characteristics, the contribution rate of heavy metal sources is determined. At the same time, a heavy metal migration and transformation model is constructed based on the geochemical properties of the soil. The geochemical properties of the soil include pH and redox potential Eh. For any heavy metal element M, its migration rate equation is: Where k1-k4 are rate constants, determined by experimental and theoretical calculations, [M] ads , [M] sol , [M] pre , [M] dis Respectively represent the concentrations of heavy metal element M in adsorbed, dissolved, precipitated and desorbed states.

2. The method for estimating the heavy metal content in soil according to claim 1, characterized in that: In the sample collection and data acquisition steps, the stratified and zoned sampling method first divides the study area into different texture categories according to soil texture, including sandy soil, loam, and clay texture types. Each texture category is further subdivided into different sub-areas according to topographical features, including mountainous, plain, and hilly terrain areas. For each sub-area, it is divided into three main layers: surface, middle, and deep according to soil depth, of which the surface layer is 0-20 cm, the middle layer is 20-50 cm, and the deep layer is 50-100 cm. Sampling points are selected in each layer according to the predetermined grid spacing. Environmental magnetic parameters are measured by magnetic measuring instruments, and the heavy metal content is measured using inductively coupled plasma mass spectrometry.

3. The method for estimating the heavy metal content in soil according to claim 1, characterized in that: In the correlation analysis step, feature similarity It is determined by calculating the cosine of the angle between the sample points in the feature space. The specific formula is: where x ik and x jk Respectively represent the values ​​of the i-th and j-th samples on the k-th feature dimension, d is the number of feature dimensions, and the dynamic adjustment of σ is based on the standard deviation σ of the sample data data , the formula is σ=μ·σ data , where μ is the empirical coefficient, which is initially set to 0.

5. In the subsequent analysis, it is fine-tuned according to the stability of each correlation analysis result, ranging from 0.3 to 0.7, and is determined through multiple experimental optimizations. When constructing the adaptive weight matrix W, the weight matrix is ​​regularized, and the L2 norm regularization term is added. The objective function is modified to Where λ is the regularization coefficient, and the optimal value is selected through cross-validation.

4. The method for estimating the heavy metal content in soil according to claim 1, characterized in that: In the step of establishing the magnetic detection model, the improved random search algorithm searches for the sample dynamic weight α in each iteration. i Adjust the search range and step length. The search range is adjusted according to the distribution range of the sample dynamic weight. For samples with larger dynamic weights, the search range is expanded, and for samples with smaller dynamic weights, the search range is narrowed. The search step length is adjusted according to the current number of iterations t and the total number of iterations T. The step length formula is: Where s0 is the initial step size, and the multi-layer nonlinear transformation of the model uses an improved activation function β is based on the mean μ of the neuron input x With variance Adaptive adjustment, the formula is At the same time, sparsity constraints are added to the hidden layer of the model to constrain the output of neurons.

5. The method for estimating the heavy metal content in soil according to claim 1, characterized in that: In the model validation and optimization steps, in the adaptive step size strategy, the adjustment factor λ is adjusted by the fluctuation coefficient ω on the validation set. ω reflects the fluctuation of the loss function. The specific calculation is: Where L t and L t-1 are the validation set loss function values ​​of the t-th and t-1-th iterations, respectively. When ω is greater than the threshold ω0, λ is increased, otherwise it is decreased. In the model verification process, an early stopping mechanism is introduced, and a performance indicator window size is set to w. If the validation set performance indicator does not improve significantly in consecutive w iterations, the training is stopped and the current optimal model is saved. At the same time, in order to further optimize the model performance, in the model training process, the model integration technology is used to combine multiple trained models by weighted averaging, and the weight is determined according to the performance of the model on the validation set.

6. The method for estimating the heavy metal content in soil according to claim 1, characterized in that: In the step of combining geochemical knowledge analysis, the heavy metal source analysis model uses positive definite matrix decomposition technology and combines regional pollution source information to determine the contribution rate of each source to heavy metals. The specific operation is to form a matrix X with the concentration data of heavy metals and related tracer elements in soil samples, and decompose the matrix X into the source component matrix F, the source contribution matrix G and the error matrix E, that is, X = FG + E, and minimize the objective function To solve F and G, where I is the number of samples, J is the number of elements, K is the number of sources, σ ij is the measurement error, λ and γ are the sparsity penalty coefficients, and the sparse features considering the pollution source are selected through cross-validation.

7. The method for estimating the heavy metal content in soil according to claim 1, characterized in that: In the heavy metal migration and transformation model, the rate constants k1-k4 take into account the soil organic matter content and clay mineral type factors, and are determined by combining indoor simulation experiments with theoretical deductions. In the indoor simulation experiments, microenvironments with different soil conditions are constructed, the soil pH, Eh, organic matter content, and clay mineral type factors are changed, and the changes in heavy metal concentrations over time are monitored. The rate constants are obtained by fitting the experimental data using the nonlinear least squares method. At the same time, the influence of microorganisms on the migration and transformation of heavy metals is considered, and the activity and type of microorganisms are added as influencing factors to the migration and transformation model to further refine the model. For any heavy metal element M, its migration and transformation model can be expressed as Where f(S) is a function related to microbial activity and species, f(S) = k6·M A ·[M] sol , M A represents the microbial activity, k6 is the influence coefficient of microorganisms on the migration and transformation of heavy metals, which is determined through microbial culture experiments and soil microenvironment simulation experiments.

8. A device for estimating heavy metal content in soil, characterized in that: The device includes a sample collection module, a magnetic parameter detection module, a data processing and modeling module, a model verification and optimization module, a geochemical analysis module and a result output module; The sample collection module includes a positioning mechanism and a sampling mechanism. The positioning mechanism is guided by a geographic information system and uses a positioning chip to obtain the precise position of the sampling point. The sampling mechanism is composed of an adjustable mechanical arm and a sampling probe. It automatically collects soil samples at different depths and positions according to a preset plan and records environmental parameters. The magnetic parameter detection module is equipped with a magnetic measuring instrument to measure the magnetic parameters of the soil sample and pre-process the data; The data processing and modeling module is based on a high-performance processor and runs customized algorithm software. The module receives sample positions, environmental parameters and magnetic parameters, performs standardization, and then uses an improved correlation analysis algorithm and a deep model based on feature fusion to perform correlation analysis and modeling. The model verification and optimization module has the functions of intelligent data division and dynamic evaluation, divides the data set according to sample characteristics, and optimizes the model parameters according to the verification set indicators using an adaptive step size strategy during model training; The geochemical analysis module has a built-in geochemical database, and combines the results of the magnetic detection model to use geochemical principles and models to analyze the sources and migration and transformation rules of heavy metals; The result output module supports multiple output modes, and displays the estimation results and analysis conclusions in the form of charts and reports.

9. The device for estimating heavy metal content in soil according to claim 8, characterized in that: The sampling probes of the sample collection module include spiral type and piston type. The probe type is automatically selected according to the soil texture. The resistance during sampling is monitored by a pressure sensor, and the sampling depth is automatically adjusted according to the resistance. The environmental parameter sensor is integrated on the sampling head, including a temperature sensor, a humidity sensor, an air pressure sensor and a soil moisture sensor. The environmental parameters of the sampling site are monitored and stored in real time. The stored data also contains timestamp information for subsequent analysis of the impact of spatiotemporal changes in the soil environment on the heavy metal content.

10. The device for estimating heavy metal content in soil according to claim 8, characterized in that: The high-performance processor of the data processing and modeling module adopts a multi-core architecture, has parallel computing capabilities, and supports multi-threaded processing. The customized algorithm software adopts a modular design, including a correlation analysis module, a model building module, a model training module, and a model evaluation module. The module supports remote data transmission and cloud computing, and uses cloud resources to accelerate data processing and model training. When processing data, data encryption technology is used to encrypt the transmitted and stored data.

Images