Automatic auditing and analyzing method for cultivated land soil data

Abstract

The invention discloses an automatic auditing and analyzing method for cultivated land soil data. The method comprises the following steps: S1, acquiring original soil data through multi-spectral imaging of an Internet of Things sensor and an unmanned aerial vehicle to form a data set; s2, using an extended Kalman filter to reduce noise and remove abnormal values, and using time sequence interpolation to complement missing data; s3, analyzing and auditing the preprocessed data based on the time sequence, and identifying an abnormal mode; s4, analyzing and fusing auditing data by using a causal reasoning graph neural network; s5, performing multi-dimensional analysis and prediction on the fused data by applying a big data analysis technology; s6, realizing real-time data visualization and interactive display through a visualization method driven by quantum computing; s7, using a natural language generation technology to generate a report for the soil data set and the analysis result; and S8, implementing encryption and access control in the data processing and transmission process. The invention provides a comprehensive soil data processing solution, and the data auditing efficiency, accuracy and safety are effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of agricultural information management, and in particular to an automatic review and analysis method for cultivated land soil data. Background Art

[0002] In modern agriculture, cultivated land soil data has become an important production factor and strategic resource. With the development of agricultural information technology, the demand for soil data in various fields is increasing, including precision agriculture, environmental monitoring and land management. Efficient data processing and analysis can promote optimal resource allocation, improve agricultural production efficiency and promote scientific and technological innovation. However, soil data management and analysis also face many challenges, especially in data collection, processing efficiency and data security. Traditional soil data processing methods include manual data collection, data cleaning and simple statistical analysis. Although existing technologies have solved some problems in data processing to a certain extent, there are still many defects and challenges, which are specifically manifested in the following aspects:

[0003] First, the limitations of data collection. Traditional soil data collection methods mainly rely on ground sensors and manual sampling, which have obvious limitations in coverage and data update frequency. Ground sensors can only cover a limited area, with few sampling points, and the data cannot fully reflect the soil conditions of large areas of farmland. The laboratory analysis process is cumbersome, the data update frequency is low, and it is impossible to reflect the dynamic changes of soil conditions in real time. These factors lead to the limitations of soil data collection and the inability to obtain comprehensive and accurate soil information.

[0004] Secondly, data processing is inefficient. Existing soil data processing and review mainly rely on manual data cleaning, normalization and outlier detection. Faced with massive amounts of soil data, manual processing is not only time-consuming and labor-intensive, but also prone to human errors, affecting the accuracy and reliability of data review results. This inefficient processing method is difficult to meet the needs of modern agriculture for fast and efficient data processing, resulting in low efficiency and accuracy in soil data review and analysis.

[0005] Third, the singleness of the analysis method. Existing soil data analysis methods mostly use simple statistical analysis, which makes it difficult to deeply explore the complex relationships between data. For example, for multi-source and multi-temporal soil data, there is a lack of effective integration and analysis methods, and it is impossible to fully understand the comprehensive status and development trend of the soil. This single analysis method is difficult to support scientific decision-making in precision agriculture and cannot provide comprehensive and accurate soil status assessment and prediction.

[0006] In addition, data security is insufficient. In the process of data transmission and processing, the existing technology does not have sufficient encryption and access control measures for data, which can easily lead to data leakage and tampering, affecting the reliability and security of data. The lack of data security makes soil data face greater risks during transmission and processing, affecting the credibility and use value of the data.

[0007] Therefore, how to provide a method for planning discrete patch traversal survey paths under the influence of multiple factors is an urgent problem that technicians in this field need to solve. Summary of the invention

[0008] One object of the present invention is to propose an automated audit and analysis method for cultivated land soil data. The present invention realizes efficient collection, accurate processing, multi-dimensional analysis and security management of soil data by combining Internet of Things sensors, drone multispectral imaging technology, extended Kalman filter, causal reasoning graph neural network, big data analysis, quantum computing-driven visualization technology and natural language generation technology. Specific methods include real-time multi-source data acquisition, noise reduction and outlier removal, time series analysis, deep analysis and data fusion, quantum computing visualization and automatic report generation. The method has the following advantages: it improves data audit efficiency, accuracy and security, provides a comprehensive soil data processing solution, overcomes many shortcomings of the prior art, and promotes precise management and scientific decision-making in modern agriculture.

[0009] According to an embodiment of the present invention, a method for automatic review and analysis of cultivated land soil data comprises the following steps:

[0010] S1. Collect raw soil data, collect and transmit real-time multi-source data through IoT sensors and drone multispectral imaging to form a soil data set;

[0011] S2, preprocessing the soil data set, using the extended Kalman filter to reduce data noise and remove outliers, and using the time series interpolation method to complete missing data;

[0012] S3, perform data audit on the preprocessed soil dataset based on time series analysis to identify potential abnormal data patterns;

[0013] S4. Use graph neural network with causal reasoning to perform data analysis and fusion on the audited soil dataset.

[0014] S5. Use big data analysis technology to perform multidimensional data analysis and prediction on the fused soil data set;

[0015] S6. Use quantum computing-driven visualization methods to perform real-time data visualization and interactive display of analysis and prediction results;

[0016] S7. Use natural language generation technology to generate reports based on the latest data and analysis results of the soil dataset, provide data summaries, anomaly detection results, analysis conclusions and suggestions, and support multiple formats of output;

[0017] S8. During the processing and transmission of soil datasets, encryption and access control strategies are applied to each data block.

[0018] Optionally, the S2 specifically includes:

[0019] S21, perform preliminary noise reduction on the soil data set, use an improved adaptive extended Kalman filter, initialize the state vector and covariance matrix, and combine the multivariate Student distribution to perform noise modeling, set the state transfer equation and the Jacobian matrix of the state transfer function;

[0020] S22. Use Gaussian process regression to predict the state and estimate the noise, and optimize the state prediction through variational inference:

[0021]

[0022] Among them, P k|k-1 The prediction error covariance matrix, F k-1 is the Jacobian matrix of the state transfer function, P k-1 The error covariance matrix of the previous moment, ν is the degree of freedom parameter, w k-1 is the process noise vector, I is the identity matrix, α is the weight coefficient, is the Gaussian basis function, N is the number of basis functions, (x k-1 ) T variable x k-1 The transpose of

[0023] S23, performing measurement update through multimodal particle filtering, calculating measurement residual and adaptive Kalman gain, and fusing the measurement result with the prediction result;

[0024] S24, using adaptive Kalman gain to correct the state vector and covariance matrix to further reduce the impact of noise;

[0025] S25. Remove outliers from the denoised and corrected soil dataset, adopt an outlier detection method based on a deep generative model, train the model using a variational autoencoder, build a generative model and a discriminant model, calculate the anomaly score for each data point, set a threshold, and mark and remove outliers;

[0026] S26. Complete the missing data using sparse Bayesian learning method. Through sparse Bayesian matrix decomposition and deep generative adversarial network optimization, define the observation data matrix and dictionary matrix, establish a sparse Bayesian matrix decomposition model, and jointly optimize the generative adversarial network and sparse Bayesian learning:

[0027]

[0028] Where, ln p(Y|X,A,E) is the log-likelihood of the observed data Y given the input X, matrix A and noise E, is the expected value of Y from the true data distribution, ln D(Y) is the log probability of the discriminator D for the true data Y, λ is the weight parameter, γ and δ are regularization parameters, E X～pX(X) is the expected value of X from the input data distribution, ln(1-D(G(X))) is the logarithmic probability of the discriminator D for the generated data G(X), ∥·∥1 represents the L1 norm, D and G represent the discriminator and generator of the generative adversarial network respectively, and the optimal sparse coefficient matrix X is obtained by jointly optimizing the above complex loss function. optimized And the dictionary matrix A, finally reconstruct the missing data.

[0029] Optionally, the S3 specifically includes:

[0030] S31, performing time series decomposition on the preprocessed soil data set, decomposing the time series into trend components, seasonal components and residual components;

[0031] S32. Perform advanced trend analysis on trend components, using logarithmic transformation and nonlinear filtering methods to convert trend components:

[0032]

[0033] Where T(t) is the original trend component, ∈ is a small constant used to avoid calculation errors in logarithmic transformation, and coeff n is the filter coefficient of the nth item, and α is the weight used to enhance the nonlinear processing capability of the model;

[0034] S33. Model the seasonal component S(t) using non-parametric Bayesian method and use Gaussian process regression for prediction:

[0035]

[0036] Where m(t) is the mean function of the Gaussian process, is the function variance, l is the length scale, is the noise variance, p is the period length, corresponding to the seasonal variation period, and δ(t-t') is the Diracdelta function, which handles the autocorrelation problem;

[0037] S34. Perform nonlinear modeling on the residual component R(t), use variational autoencoder for residual generation and prediction, and optimize the objective function as follows:

[0038]

[0039] in, The loss function of the variational autoencoder, E represents the expected value, is the encoder, and its parameters are -lnp θ (x|z) is the negative of the log-likelihood, which indicates the probability of generating data x given the latent variable z; p θ (x|z) is the decoder, p(z) is the prior distribution with parameter θ; z is the latent variable; N is the Gaussian distribution; KL is the Kullback-Leibler divergence;

[0040] S35, performing abnormal data detection based on the decomposed components, marking the trend component and the seasonal component as outliers, determining the outlier threshold using the isolation forest method, and marking and removing the abnormal data based on the variational autoencoder model of the residual component;

[0041] S36. Reconstruct the data after marking and removing outliers, combine the trend component, seasonal component and corrected residual component, and reconstruct the complete time series data.

[0042] Optionally, S4 includes the following steps:

[0043] S41. Extract features from the audited soil data set, build a causal graph model of soil data using the graph neural network method, and generate the hidden state vector of each node by aggregating and updating node features:

[0044]

[0045] in, is the hidden state vector of node v in the k+1th layer, σ is the activation function, W (k is the weight matrix of the kth layer, AGG is the aggregation function, is the hidden state vector of neighbor node u at layer k, e vu and e vv is the edge weight vector between nodes, b (k) is the bias vector of the kth layer, ⊙ represents element-wise multiplication;

[0046] S42, perform causal reasoning on the constructed causal graph model and use the causal reasoning framework to evaluate the causal relationship in the soil dataset;

[0047] S43, perform feature selection on the causal reasoning results, use the attention mechanism to calculate the weight of each feature, and select key features;

[0048] S44. Perform cluster analysis on the selected key features, adopt a deep embedding clustering method based on graph neural network, and optimize the objective function as follows:

[0049]

[0050] Among them, z i is the representation of data point i in the embedding space, μ j is the representation of cluster center j, α is the degree of freedom parameter of Student’s t distribution, is the frequency of cluster center j, j' is the index variable when summing the cluster centers;

[0051] S45. Based on the cluster analysis results, combined with feature selection and cluster analysis for causal reasoning, the soil dataset is updated, including soil types in different clusters and the main characteristics of each type of soil;

[0052] S46. Combine the updated soil data set with the time series analysis results to perform data fusion.

[0053] Optionally, the S44 includes the following steps:

[0054] S441, preprocessing the selected key features, embedding the features into a high-dimensional space using a graph neural network method, and generating an embedded representation of the node and an initial cluster center representation;

[0055] S442, initialize cluster center μ j , select the initial cluster center in the embedding space;

[0056] S443, calculate the adaptive soft assignment q of each data point i belonging to cluster center j ij :

[0057]

[0058] in, is the variance of cluster center j, z i is the embedding representation of data point i, μ j is the embedding representation of cluster center j;

[0059] S444. Calculate target distribution p ij , further optimize the soft allocation results:

[0060]

[0061] in, is the frequency of cluster center j;

[0062] S445. Define the regularized KL divergence loss function to measure the soft assignment q ij With the target distribution p ij The difference between , and adding information entropy regularization:

[0063]

[0064] Among them, λ is the regularization parameter;

[0065] S446, based on the adaptive gradient descent method, minimize the regularized KL divergence loss function and optimize the objective function Update node embedding representations and cluster centers through back-propagation:

[0066]

[0067] Where η is the learning rate, and are the cluster center and node embedding representation at the tth iteration, respectively. and are the cumulative sum of squared gradients, respectively;

[0068] S447, introduce adaptive neighborhood graph regularization, and update the loss function to:

[0069]

[0070] Among them, β and δ are regularization parameters, is the edge set of the graph, cos(z i ,z j 0 represents the cosine similarity between node embeddings;

[0071] S448. Combine contrastive learning to construct contrastive loss and update the loss function as follows:

[0072]

[0073] Among them, γ is the weight parameter of contrastive learning, 1 is the indicator function, and y i and j are the category labels of data points i and j respectively;

[0074] S449, repeat steps S443 to S448 until the objective function Converge and get the final clustering result.

[0075] Optionally, the S6 includes the following steps:

[0076] S61. Build quantum visualization algorithms and design quantum circuits to process and map soil datasets.

[0077] S62, define the encoding and decoding strategy of quantum state, encode the soil data set into quantum state, and decode it after the quantum computation is completed;

[0078] S63, execute quantum computing tasks, input analysis and prediction data into the quantum processor, and realize data processing through quantum state evolution:

[0079]

[0080] Among them, ψ evolved is the quantum state after evolution, β j is the evolution coefficient, is the quantum operator, M is the number of evolution steps, γ(x) is the weight function, and exp(iHx) is the quantum coherent operation involving the Hamiltonian H;

[0081] S64, converting the output data of the quantum calculation into classical data, obtaining the classical data through quantum state measurement, and reconstructing the classical data;

[0082] S65. Convert the classical data output by quantum computing into visual graphics, use classical computers to post-process the quantum results, and generate charts and visual interfaces;

[0083] S66, realize real-time interactive display, through which users can dynamically adjust and view analysis and prediction results;

[0084] S67, introduces the visualization of quantum state entanglement and superposition to show the correlation between different quantum states of the soil dataset;

[0085] S68. Using quantum state coherence and decoherence analysis, we show how soil datasets change under different environments:

[0086]

[0087] in, is the Leveille operator, ρ is the density matrix, L i is the environmental operator, YesL i The conjugate transpose operator, γ jk is the decoherence rate, A j and A k is the coherence operator.

[0088] The beneficial effects of the present invention are:

[0089] (1) The present invention significantly improves the efficiency and accuracy of cultivated land soil data review and analysis by combining efficient data preprocessing technology, anomaly detection methods based on time series analysis, causal reasoning graph neural networks, and big data analysis technology. In particular, in terms of data fusion and prediction, the present invention overcomes the shortcomings of traditional methods, can quickly and accurately process large-scale, multi-source soil data, and provide a unified data representation.

[0090] (2) By introducing quantum computing-driven visualization methods and natural language generation technology, the present invention optimizes the display of data analysis results and report generation. The quantum computing-driven visualization method realizes real-time data visualization and interactive display, allowing users to intuitively understand and analyze data results. Natural language generation technology automatically generates detailed reports based on the latest data and analysis results, including data summaries, anomaly detection results, analysis conclusions and suggestions, which improves the efficiency and accuracy of users' understanding and application of data.

[0091] (3) The present invention adopts advanced data encryption and access control strategies to enhance data security and effectively prevent unauthorized access and data leakage. Through these measures, it is possible to cope with ever-changing security threats and ensure the security of soil data during processing and transmission. BRIEF DESCRIPTION OF THE DRAWINGS

[0092] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0093] Figure 1 This is a general framework diagram of an automated audit and analysis method for cultivated land soil data proposed by the present invention;

[0094] Figure 2 This is a flow chart of extended Kalman filter denoising and outlier removal for an automated audit and analysis method for cultivated soil data proposed by the present invention;

[0095] Figure 3 A schematic diagram of a causal reasoning graph neural network deep analysis and fusion module of an automated audit and analysis method for cultivated soil data proposed in the present invention;

[0096] Figure 4 This is a quantum computing-driven real-time data visualization and interactive display block diagram of the automated review and analysis method for cultivated soil data proposed in the present invention. DETAILED DESCRIPTION

[0097] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0098] refer to Figure 1-4 , a method for automatic review and analysis of cultivated land soil data, comprising the following steps:

[0099] S1. Collect raw soil data, collect and transmit real-time multi-source data through IoT sensors and drone multispectral imaging to form a soil data set;

[0100] S2, preprocessing the soil data set, using the extended Kalman filter to reduce data noise and remove outliers, and using the time series interpolation method to complete missing data;

[0101] S3, perform data audit on the preprocessed soil dataset based on time series analysis to identify potential abnormal data patterns;

[0102] S4. Use graph neural network with causal reasoning to perform data analysis and fusion on the audited soil dataset.

[0103] S5. Use big data analysis technology to perform multidimensional data analysis and prediction on the fused soil data set;

[0104] S6. Use quantum computing-driven visualization methods to perform real-time data visualization and interactive display of analysis and prediction results;

[0105] S7. Use natural language generation technology to generate reports based on the latest data and analysis results of the soil dataset, provide data summaries, anomaly detection results, analysis conclusions and suggestions, and support multiple formats of output;

[0106] S8. During the processing and transmission of soil datasets, encryption and access control strategies are applied to each data block.

[0107] In this implementation, S2 specifically includes:

[0108] S21, perform preliminary noise reduction on the soil data set, use an improved adaptive extended Kalman filter, initialize the state vector and covariance matrix, and combine the multivariate Student distribution to perform noise modeling, set the state transfer equation and the Jacobian matrix of the state transfer function;

[0109] S22. Use Gaussian process regression to predict the state and estimate the noise, and optimize the state prediction through variational inference:

[0110]

[0111] Among them, P k|k-1 The prediction error covariance matrix, F k-1 is the Jacobian matrix of the state transfer function, P k-1The error covariance matrix of the previous moment, ν is the degree of freedom parameter, w k-1 is the process noise vector, I is the identity matrix, α is the weight coefficient, is the Gaussian basis function, N is the number of basis functions, (x k-1 ) T variable x k-1 The transpose of

[0112] S23, performing measurement update through multimodal particle filtering, calculating measurement residual and adaptive Kalman gain, and fusing the measurement result with the prediction result;

[0113] S24, using adaptive Kalman gain to correct the state vector and covariance matrix to further reduce the impact of noise;

[0114] S25. Remove outliers from the denoised and corrected soil dataset, adopt an outlier detection method based on a deep generative model, train the model using a variational autoencoder, build a generative model and a discriminant model, calculate the anomaly score for each data point, set a threshold, and mark and remove outliers;

[0115] S26. Complete the missing data using sparse Bayesian learning method. Through sparse Bayesian matrix decomposition and deep generative adversarial network optimization, define the observation data matrix and dictionary matrix, establish a sparse Bayesian matrix decomposition model, and jointly optimize the generative adversarial network and sparse Bayesian learning:

[0116]

[0117] Where, ln p(Y|X,A,E) is the log-likelihood of the observed data Y given the input X, matrix A and noise E, is the expected value of Y from the true data distribution, ln D(Y) is the log probability of the discriminator D for the true data Y, λ is the weight parameter, γ and δ are regularization parameters, E X～pX(X) is the expected value of X from the input data distribution, ln(1-D(G(X))) is the logarithmic probability of the discriminator D for the generated data G(X), ∥·∥1 represents the L1 norm, D and G represent the discriminator and generator of the generative adversarial network respectively, and the optimal sparse coefficient matrix X is obtained by jointly optimizing the above complex loss function. optimized And the dictionary matrix A, finally reconstruct the missing data.

[0118] In this implementation, S3 specifically includes:

[0119] S31, performing time series decomposition on the preprocessed soil data set, decomposing the time series into trend components, seasonal components and residual components;

[0120] S32. Perform advanced trend analysis on trend components, using logarithmic transformation and nonlinear filtering methods to convert trend components:

[0121]

[0122] Where T(t) is the original trend component, ∈ is a small constant used to avoid calculation errors in logarithmic transformation, and coeff n is the filter coefficient of the nth item, and α is the weight used to enhance the nonlinear processing capability of the model;

[0123] S33. Model the seasonal component S(t) using non-parametric Bayesian method and use Gaussian process regression for prediction:

[0124]

[0125] Where m(t) is the mean function of the Gaussian process, is the function variance, l is the length scale, is the noise variance, p is the period length, corresponding to the seasonal variation period, and δ(t-t') is the Diracdelta function, which handles the autocorrelation problem;

[0126] S34. Perform nonlinear modeling on the residual component R(t), use variational autoencoder for residual generation and prediction, and optimize the objective function as follows:

[0127]

[0128] in, The loss function of the variational autoencoder, E represents the expected value, is the encoder, and its parameters are -lnp θ (x|z) is the negative of the log-likelihood, which indicates the probability of generating data x given the latent variable z; p θ (x|z) is the decoder, p(z) is the prior distribution with parameter θ; z is the latent variable; N is the Gaussian distribution; KL is the Kullback-Leibler divergence;

[0129] S35, performing abnormal data detection based on the decomposed components, marking the trend component and the seasonal component as outliers, determining the outlier threshold using the isolation forest method, and marking and removing the abnormal data based on the variational autoencoder model of the residual component;

[0130] S36. Reconstruct the data after marking and removing outliers, combine the trend component, seasonal component and corrected residual component, and reconstruct the complete time series data.

[0131] In this implementation, S4 specifically includes:

[0132] S41. Extract features from the audited soil data set, build a causal graph model of soil data using the graph neural network method, and generate the hidden state vector of each node by aggregating and updating node features:

[0133]

[0134] in, is the hidden state vector of node v in the k+1th layer, σ is the activation function, W (k is the weight matrix of the kth layer, AGG is the aggregation function, is the hidden state vector of neighbor node u at layer k, e vu and e vv is the edge weight vector between nodes, b (k) is the bias vector of the kth layer, ⊙ represents element-wise multiplication;

[0135] S42, perform causal reasoning on the constructed causal graph model and use the causal reasoning framework to evaluate the causal relationship in the soil dataset;

[0136] S43, perform feature selection on the causal reasoning results, use the attention mechanism to calculate the weight of each feature, and select key features;

[0137] S44. Perform cluster analysis on the selected key features, adopt a deep embedding clustering method based on graph neural network, and optimize the objective function as follows:

[0138]

[0139] Among them, z i is the representation of data point i in the embedding space, μ j is the representation of cluster center j, α is the degree of freedom parameter of Student’s t distribution, is the frequency of cluster center j, and j' is the index variable when summing the cluster centers. This formula combines embedding representation, Student's t distribution, and frequency adjustment. By minimizing the loss function, it finally realizes cluster analysis of key features. It can not only process high-dimensional data, but also effectively identify outliers and abnormal values, ensuring the stability and accuracy of clustering results. This optimization objective function can effectively improve the effect of data analysis and provide a reliable basis for subsequent data fusion and decision-making.

[0140] S45. Based on the cluster analysis results, combined with feature selection and cluster analysis for causal reasoning, the soil dataset is updated, including soil types in different clusters and the main characteristics of each type of soil;

[0141] S46. Combine the updated soil data set with the time series analysis results to perform data fusion.

[0142] In this implementation manner, the S44 specifically includes:

[0143] S441, preprocessing the selected key features, embedding the features into a high-dimensional space using a graph neural network method, and generating an embedded representation of the node and an initial cluster center representation;

[0144] S442, initialize cluster center μ j , select the initial cluster center in the embedding space;

[0145] S443, calculate the adaptive soft assignment q of each data point i belonging to cluster center j ij :

[0146]

[0147] in, is the variance of cluster center j, z i is the embedding representation of data point i, μ j is the embedding representation of cluster center j;

[0148] S444. Calculate target distribution p ij , further optimize the soft allocation results:

[0149]

[0150] in, is the frequency of cluster center j;

[0151] S445. Define the regularized KL divergence loss function to measure the soft assignment q ij With the target distribution p ij The difference between , and adding information entropy regularization:

[0152]

[0153] Among them, λ is the regularization parameter;

[0154] S446, based on the adaptive gradient descent method, minimize the regularized KL divergence loss function and optimize the objective function Update node embedding representations and cluster centers through back-propagation:

[0155]

[0156] Where η is the learning rate, and are the cluster center and node embedding representation at the tth iteration, respectively. and are the cumulative sum of squared gradients, respectively;

[0157] S447, introduce adaptive neighborhood graph regularization, and update the loss function to:

[0158]

[0159] Among them, β and δ are regularization parameters, is the edge set of the graph, cos(z i ,z j ) represents the cosine similarity between node embeddings;

[0160] S448. Combine contrastive learning to construct contrastive loss and update the loss function as follows:

[0161]

[0163] Among them, γ is the weight parameter of contrastive learning, 1 is the indicator function, and y i and j are the category labels of data points i and j respectively;

[0164] S449, repeat steps S443 to S448 until the objective function Converge and get the final clustering result.

[0165] In this implementation manner, S6 specifically includes:

[0166] S61. Build quantum visualization algorithms and design quantum circuits to process and map soil datasets.

[0167] S62, define the encoding and decoding strategy of quantum state, encode the soil data set into quantum state, and decode it after the quantum computation is completed;

[0168] S63, execute quantum computing tasks, input analysis and prediction data into the quantum processor, and realize data processing through quantum state evolution:

[0169]

[0170] Among them, ψ evolved is the quantum state after evolution, β j is the evolution coefficient, is the quantum operator, M is the number of evolution steps, γ(x) is the weight function, and exp(iHx) is the quantum coherent operation involving the Hamiltonian H;

[0171] S64, converting the output data of the quantum calculation into classical data, obtaining the classical data through quantum state measurement, and reconstructing the classical data;

[0172] S65. Convert the classical data output by quantum computing into visual graphics, use classical computers to post-process the quantum results, and generate charts and visual interfaces;

[0173] S66, realize real-time interactive display, through which users can dynamically adjust and view analysis and prediction results;

[0174] S67, introduces the visualization of quantum state entanglement and superposition to show the correlation between different quantum states of the soil dataset;

[0175] S68. Using quantum state coherence and decoherence analysis, we show how soil datasets change under different environments:

[0176]

[0177] in, is the Leveille operator, ρ is the density matrix, L i is the environmental operator, YesL i The conjugate transpose operator, γ jk is the decoherence rate, A j and A k is the coherence operator.

[0178] Embodiment 1:

[0179] In order to verify the feasibility of the present invention in implementation, the present invention was fully tested on a large farmland in the North China Plain, and an automated audit and analysis network for cultivated land soil data was established. Each farmland area is used as a node, and data is collected through IoT sensors and drone multispectral imaging technology, and interconnected through a data processing and analysis network. The entire process includes data collection, data preprocessing, deep analysis and fusion, data analysis and prediction, visualization and report generation.

[0180] In various areas of the farmland, a network of IoT sensors is deployed to collect real-time data on soil temperature, humidity, pH value, and nutrient content. In order to obtain data with a wider range and higher resolution, the team also uses drones for multispectral imaging to obtain spectral information of the soil through remote sensing technology.

[0181] All collected data is first input into the extended Kalman filter for processing. This filter effectively reduces noise and removes outliers to ensure the accuracy and reliability of the data. Then, the processed data is supplemented by the time series interpolation method to solve the problem of missing data caused by sensor failure or signal loss.

[0182] The pre-processed data is deeply analyzed and fused through a graph neural network with causal reasoning to identify complex associations between the data and reveal the causal relationship between soil health and the crop growth environment. For example, high soil pH in certain areas may be related to the fertilization method in that area, while insufficient soil moisture may be related to a malfunction in the irrigation system.

[0183] The team then used big data analysis technology to conduct multi-dimensional analysis and prediction of the fused data. The system generates a soil health report by analyzing historical and current data and predicts future soil change trends. For example, based on the changing trend of soil nutrient content, the system predicts that some areas may need additional fertilization in the next few months, so as to prepare in advance and avoid crop yield reduction due to insufficient nutrients.

[0184] In order to make the data analysis results more intuitive, the research team introduced visualization technology driven by quantum computing. Through the efficient processing power of quantum computing, the system generates a visualization chart of soil health in real time and displays it on an interactive display platform. Farmers can view the soil conditions, historical change trends and future forecast results of different regions through simple operations, so as to make scientific agricultural management decisions.

[0185] Finally, the team used natural language generation technology to automatically generate a detailed soil health report based on the data analysis results. The report includes data summary, anomaly detection results, analysis conclusions and recommendations. All information is expressed in plain language for farmers to understand and apply.

[0186] Table 1 Test data and result analysis table of the present invention

[0187]

[0188]

[0189] During the six-month test period, a large amount of soil data was processed, verifying the effectiveness and superiority of the invention in practical applications. All data remained encrypted during sharing and calculation, and no data leakage incidents occurred. The system ensured the security of data storage and transmission of each node through regular security audits, and the privacy protection coverage rate reached 100%.

[0190] By using the method of the present invention, the average data processing time is reduced from 2 hours of the traditional method to 30 minutes, and the efficiency is improved by 4 times; the result merging and verification time is reduced from 1 hour to 5 minutes, which is improved by 12 times; the new block generation and broadcasting time is reduced from 30 minutes to 3 minutes, which is improved by 10 times, and the total time is reduced from 3.5 hours to 38 minutes, which is about 5.5 times more efficient. All data operations are recorded on the chain, the operation logs and transaction records are clear, and the smart contract automatically executes the data authorization and sharing rules, avoiding manual intervention and ensuring the fairness and compliance of data operations.

[0191] In summary, the present invention provides a comprehensive soil data processing solution, which significantly improves the efficiency, accuracy and security of data review, overcomes many shortcomings of the existing technology, provides strong support for the precise management and scientific decision-making of modern agriculture, and has broad application prospects.

[0192] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for automatic review and analysis of cultivated land soil data, characterized in that: The steps include: S1. Collect raw soil data, collect and transmit real-time multi-source data through IoT sensors and drone multispectral imaging to form a soil data set; S2, preprocessing the soil data set, using the extended Kalman filter to reduce data noise and remove outliers, and using the time series interpolation method to complete missing data; S3, perform data audit on the preprocessed soil dataset based on time series analysis to identify potential abnormal data patterns; S4. Use graph neural network with causal reasoning to perform data analysis and fusion on the audited soil dataset. S5. Use big data analysis technology to perform multidimensional data analysis and prediction on the fused soil data set; S6. Use quantum computing-driven visualization methods to perform real-time data visualization and interactive display of analysis and prediction results; S7. Use natural language generation technology to generate reports based on the latest data and analysis results of the soil dataset, provide data summaries, anomaly detection results, analysis conclusions and suggestions, and support multiple formats of output; S8. During the processing and transmission of soil datasets, encryption and access control strategies are applied to each data block.

2. The method for automatic review and analysis of cultivated land soil data according to claim 1, characterized in that: The S2 comprises the following steps: S21, perform preliminary noise reduction on the soil data set, use an improved adaptive extended Kalman filter, initialize the state vector and covariance matrix, and combine the multivariate Student distribution to perform noise modeling, set the state transfer equation and the Jacobian matrix of the state transfer function; S22. Use Gaussian process regression to predict the state and estimate the noise, and optimize the state prediction through variational inference: Among them, P k|k-1 The prediction error covariance matrix, F k-1 is the Jacobian matrix of the state transfer function, P k-1 The error covariance matrix of the previous moment, ν is the degree of freedom parameter, w k-1 is the process noise vector, I is the identity matrix, α is the weight coefficient, is the Gaussian basis function, N is the number of basis functions, (x k -1) T variable x k-1 The transpose of S23, performing measurement update through multimodal particle filtering, calculating measurement residual and adaptive Kalman gain, and fusing the measurement result with the prediction result; S24, using adaptive Kalman gain to correct the state vector and covariance matrix to further reduce the impact of noise; S25. Remove outliers from the denoised and corrected soil dataset, adopt an outlier detection method based on a deep generative model, train the model using a variational autoencoder, build a generative model and a discriminant model, calculate the anomaly score for each data point, set a threshold, and mark and remove outliers; S26. Complete the missing data using sparse Bayesian learning method. Through sparse Bayesian matrix decomposition and deep generative adversarial network optimization, define the observation data matrix and dictionary matrix, establish a sparse Bayesian matrix decomposition model, and jointly optimize the generative adversarial network and sparse Bayesian learning: Where, ln p(Y|X,A,E) is the log-likelihood of the observed data Y given the input X, matrix A and noise E, is the expected value of Y from the true data distribution, ln D(Y) is the log probability of the discriminator D for the true data Y, λ is the weight parameter, γ and δ are regularization parameters, is the expected value of X from the input data distribution, ln(1-D(G(X))) is the logarithmic probability of the discriminator D for the generated data G(X), ∥·∥1 represents the L1 norm, D and G represent the discriminator and generator of the generative adversarial network respectively, and the optimal sparse coefficient matrix X is obtained by jointly optimizing the above complex loss function. optimized And the dictionary matrix A, finally reconstruct the missing data.

3. The method for automatic review and analysis of cultivated land soil data according to claim 1, characterized in that: The S3 comprises the following steps: S31, performing time series decomposition on the preprocessed soil data set, decomposing the time series into trend components, seasonal components and residual components; S32. Perform advanced trend analysis on trend components, using logarithmic transformation and nonlinear filtering methods to convert trend components: Where T(t) is the original trend component, ∈ is a small constant used to avoid calculation errors in logarithmic transformation, and coeff n is the filter coefficient of the nth item, and α is the weight used to enhance the nonlinear processing capability of the model; S33. Model the seasonal component S(t) using non-parametric Bayesian method and use Gaussian process regression for prediction: Where m(t) is the mean function of the Gaussian process, is the function variance, l is the length scale, is the noise variance, p is the period length, corresponding to the seasonal variation period, and δ(t-t') is the Diracdelta function, which handles the autocorrelation problem; S34. Perform nonlinear modeling on the residual component R(t), use variational autoencoder for residual generation and prediction, and optimize the objective function as follows: in, The loss function of the variational autoencoder, E represents the expected value, is the encoder, and its parameters are -lnp θ (x|z) is the negative of the log-likelihood, which indicates the probability of generating data x given the latent variable z; p θ (x|z) is the decoder, p(z) is the prior distribution with parameter θ; z is the latent variable; N is the Gaussian distribution; KL is the Kullback-Leibler divergence; S35, performing abnormal data detection based on the decomposed components, marking the trend component and the seasonal component as outliers, determining the outlier threshold using the isolation forest method, and marking and removing the abnormal data based on the variational autoencoder model of the residual component; S36. Reconstruct the data after marking and removing outliers, combine the trend component, seasonal component and corrected residual component, and reconstruct the complete time series data.

4. The method for automatic review and analysis of cultivated land soil data according to claim 1, characterized in that: The S4 comprises the following steps: S41. Extract features from the audited soil data set, build a causal graph model of soil data using the graph neural network method, and generate the hidden state vector of each node by aggregating and updating node features: in, is the hidden state vector of node v in the k+1th layer, σ is the activation function, W (k is the weight matrix of the kth layer, AGG is the aggregation function, is the hidden state vector of neighbor node u at layer k, e vu and e vv is the edge weight vector between nodes, b (k) is the bias vector of the kth layer, ⊙ represents element-wise multiplication; S42, perform causal reasoning on the constructed causal graph model and use the causal reasoning framework to evaluate the causal relationship in the soil dataset; S43, perform feature selection on the causal reasoning results, use the attention mechanism to calculate the weight of each feature, and select key features; S44. Perform cluster analysis on the selected key features, adopt a deep embedding clustering method based on graph neural network, and optimize the objective function as follows: Among them, z i is the representation of data point i in the embedding space, μ j is the representation of cluster center j, α is the degree of freedom parameter of Student’s t distribution, is the frequency of cluster center j, j' is the index variable when summing the cluster centers; S45. Based on the cluster analysis results, combined with feature selection and cluster analysis for causal reasoning, the soil dataset is updated, including soil types in different clusters and the main characteristics of each type of soil; S46. Combine the updated soil data set with the time series analysis results to perform data fusion.

5. The method for automatic review and analysis of cultivated land soil data according to claim 4, characterized in that: The S44 comprises the following steps: S441, preprocessing the selected key features, embedding the features into a high-dimensional space using a graph neural network method, and generating an embedded representation of the node and an initial cluster center representation; S442, initialize cluster center μ j , select the initial cluster center in the embedding space; S443, calculate the adaptive soft assignment q of each data point i belonging to cluster center j ij : in, is the variance of cluster center j, z i is the embedding representation of data point i, μ j is the embedding representation of cluster center j; S444. Calculate target distribution p ij , further optimize the soft allocation results: in, is the frequency of cluster center j; S445. Define the regularized KL divergence loss function to measure the soft assignment q ij With the target distribution p ij The difference between , and adding information entropy regularization: Among them, λ is the regularization parameter; S446, based on the adaptive gradient descent method, minimize the regularized KL divergence loss function and optimize the objective function Update node embedding representations and cluster centers through back-propagation: Where η is the learning rate, and are the cluster center and node embedding representation at the tth iteration, respectively. and are the cumulative sum of squared gradients, respectively; S447, introduce adaptive neighborhood graph regularization, and update the loss function to: Among them, β and δ are regularization parameters, is the edge set of the graph, cos(z i ,z j ) represents the cosine similarity between node embeddings; S448. Combine contrastive learning to construct contrastive loss and update the loss function as follows: Among them, γ is the weight parameter of contrastive learning, 1 is the indicator function, and y i and j are the category labels of data points i and j respectively; S449, repeat steps S443 to S448 until the objective function Converge and get the final clustering result.

6. The method for automatic review and analysis of cultivated land soil data according to claim 1, characterized in that: The S6 comprises the following steps: S61. Build quantum visualization algorithms and design quantum circuits to process and map soil datasets. S62, define the encoding and decoding strategy of quantum state, encode the soil data set into quantum state, and decode it after the quantum computation is completed; S63, execute quantum computing tasks, input analysis and prediction data into the quantum processor, and realize data processing through quantum state evolution: Among them, ψ evolved is the quantum state after evolution, β j is the evolution coefficient, is the quantum operator, M is the number of evolution steps, γ(x) is the weight function, and exp(iHx) is the quantum coherent operation involving the Hamiltonian H; S64, converting the output data of the quantum calculation into classical data, obtaining the classical data through quantum state measurement, and reconstructing the classical data; S65. Convert the classical data output by quantum computing into visual graphics, use classical computers to post-process the quantum results, and generate charts and visual interfaces; S66, realize real-time interactive display, through which users can dynamically adjust and view analysis and prediction results; S67, introduces the visualization of quantum state entanglement and superposition to show the correlation between different quantum states of the soil dataset; S68. Using quantum state coherence and decoherence analysis, we show how soil datasets change under different environments: in, is the Leveille operator, ρ is the density matrix, L i is the environmental operator, YesL i The conjugate transpose operator, γ jk is the decoherence rate, A j and A k is the coherence operator.

Images