38 results about "Space decomposition" patented technology

The invention relates to the technical field of large language models and privacy protection, and discloses a causal intervention-based vertical domain large model privacy knowledge decoupling method and a terminal. The method comprises the following steps: constructing a privacy, anti-fact and reservation data set; computing the causal effect of an intermediate node through activation value replacement, identifying a key node in combination with anti-factual reasoning, and constructing a map to determine a core causal propagation chain; training a discriminator by using a domain knowledge ontology graph, calculating and calibrating a lexical instantiation score, decomposing a representation space into orthogonal instance and concept subspaces, and determining spatial attribution and causal weight of parameters; and constructing a double-constraint optimization target which comprises retention and forgetting loss and regularization constraint, and updating parameters which are located in a real example subspace and belong to a propagation chain. According to the method, the causal propagation chain of privacy leakage in the vertical domain large language model can be accurately identified, selective decoupling of semantic boundary perception is realized, and the domain professional ability of the model is reserved to the maximum extent while specific privacy data is thoroughly eliminated.

The invention relates to the technical field of artificial intelligence, can be applied to the medical field and the financial science and technology field, and discloses a video generation method, device and equipment based on hidden space decomposition and a storage medium, and the method comprises the steps: obtaining an initial video file, and carrying out the data preprocessing and sample construction of the initial video file to generate a training sample pair; performing joint training on the double-branch encoder to generate a target double-branch encoder; training a noise prediction network through the target content code and the target dynamic code output by the target double-branch encoder to obtain a target noise prediction network; training the confidence prediction network through the intermediate data to generate a target confidence prediction network, and constructing a video generation model based on the target double-branch encoder, the target noise prediction network and the target confidence prediction network; and performing video generation processing based on the to-be-edited image and the user editing instruction through the video generation model to generate a target video frame. According to the invention, the control precision and efficiency of video generation are improved.

The invention relates to the field of building three-dimensional reconstruction, in particular to a building three-dimensional reconstruction method based on multi-source data alignment, which comprises the following steps executed in sequence: S1, acquiring building three-dimensional feature data shot by a plurality of unmanned aerial vehicles, and performing multi-source collaborative alignment processing on the plurality of building three-dimensional feature data, the method comprises the following steps: S1, aligning building three-dimensional feature data to obtain aligned building three-dimensional feature data, S2, carrying out recursive space segmentation on the aligned building three-dimensional feature data by using an octree algorithm to obtain a plurality of segmented regions, and S3, defining each segmented region as a local center point by using tile type space decomposition and generating tile units; s4, feature points with the SDF value being 0 are extracted from the global symbol distance field through a moving cube algorithm, then a continuous geometric surface of the building is constructed through linear interpolation, the model is scaled to a real proportion through a distance reference provided by an RTK-GNSS system of the unmanned aerial vehicle, and a reconstructed model is output; the limitation that a traditional three-dimensional reconstruction model only has a visual rendering effect is broken through.

The application discloses a network fault positioning fast response system based on distributed log analysis, relates to the technical field of computer network and operation and maintenance, and comprises the following modules: a data acquisition module, which generates time series data frames based on a sliding window; a state vector construction module, which extracts indexes and constructs standardized state vectors; a covariance matrix updating module, which recursively updates a dynamic covariance matrix by using a forgetting factor; a feature space decomposition module, which constructs a residual projection matrix based on dynamic principal component dimension parameters; an anomaly detection module, which calculates residual anomaly energy and determines a fault candidate set based on contribution degrees; and a parameter self-adaptive module, which calculates residual space entropy values and feeds back to adjust principal component dimension parameters of the next frame; and the application realizes adaptive and accurate bounding of gray faults through entropy feedback closed loop and orthogonal projection.

The invention discloses an interpretable compressed sensing image reconstruction method and system based on value domain-null space decomposition, and belongs to the technical field of digital image processing and artificial intelligence. The invention provides a deep expansion network which is characterized in that in each iteration stage of the network, image features to be solved are explicitly decomposed into a value domain subspace and a null space subspace which are orthogonal to each other; wherein the value domain subspace is used for recovering and keeping global structure information consistent with a measured value and ensuring data fidelity; the null space subspace focuses on adaptively enhancing the edge and fine details of the image through a multi-scale attention enhancement pyramid fusion module MAPF and a cross-stage feature refinement fusion module FRF. According to the method, through a decoupling modeling strategy, the accuracy and visual quality of the reconstructed image are remarkably improved, a better balance is achieved between precision and efficiency, and the method can be widely applied to reconstruction of natural images and medical images.

The invention discloses a self-adaptive subspace projection suppression method based on optical pump magnetometer external noise, and belongs to the technical field of neural magnetic imaging (MEG) signal processing.The method comprises the steps that brain magnetic signals are collected through a wearable optical pump magnetometer array and preprocessed, and a signal and noise subspace decomposition model is established; integrating a model-driven noise modeling method and a data-driven noise modeling method, and constructing a noise subspace base vector; constructing a signal subspace basis vector in combination with a guide field matrix and event correlation analysis; identifying residual noise components based on event correlation analysis, and perfecting noise subspace modeling; signal noise separation is realized by adopting a hierarchical projection strategy; and analyzing the time correlation of internal and external space signals, and constructing a time projection matrix to remove residual interference to obtain a clean MEG signal. According to the invention, various types of external noise interferences can be effectively suppressed, the signal-to-noise ratio and the reliability of brain magnetic signals are remarkably improved, and the method is particularly suitable for practical application of a wearable optical pump magnetometer system.

This invention relates to the field of 3D building reconstruction, specifically to a method for 3D building reconstruction based on multi-source data alignment, comprising the following steps executed sequentially: S1: Acquire 3D building feature data captured by multiple drones, and perform multi-source collaborative alignment processing on the multiple 3D building feature data to obtain aligned 3D building feature data; S2: Use an octree algorithm to recursively spatially segment the aligned 3D building feature data to obtain multiple segmentation regions; S3: Use tile-based spatial decomposition to define each segmentation region as a local center point and generate tile units; S4: Use a moving cube algorithm to extract feature points with an SDF value of 0 from the global symbolic distance field, and then construct the continuous geometric surface of the building through linear interpolation. Scale the model to the true scale using the distance reference provided by the drone's RTK-GNSS system, and output the reconstructed model; This method overcomes the limitation of traditional 3D reconstruction models that only have visual rendering effects.

The invention discloses a product quantization method based on hierarchical fuzzy clustering, and the method comprises the steps: carrying out the spatial decomposition of a to-be-compressed high-dimensional vector of image or video data into a plurality of low-dimensional subspaces through employing a hierarchical decomposition strategy; in each low-dimensional subspace, a hierarchical fuzzy clustering method is adopted, so that each data point belongs to a plurality of fuzzy clustering centers at different membership degrees; and on the basis of a hierarchical fuzzy clustering result, a product quantization coding mechanism is adopted, so that the fuzzy clustering center of each subspace forms a corresponding codebook, data points are coded and compressed according to membership degrees of the data points in different clustering centers, and data compression of a high-dimensional vector of the image or video data is realized. The invention provides a compression method combining a hierarchical structure and a fuzzy clustering technology, and the efficiency, the accuracy and the robustness of vector compression are remarkably improved.

The invention discloses a method and a device for determining a fault critical state vector based on state space decomposition of a power system, a storage medium and computer equipment. The method comprises the following steps: determining an element set of a target power system; clustering the power generation elements, generating a target power generation subsystem according to at least one clustering cluster, calculating a first fault reduction set of the target power generation subsystem, calculating a first fault critical state vector set according to the first fault reduction set, and generating a second fault critical state vector set according to the first fault critical state vector set; clustering the power transmission elements, generating a target power transmission subsystem according to at least one clustering cluster, calculating a second fault reduction set of the target power transmission subsystem, calculating a third fault critical state vector set according to the second fault reduction set, and generating a fourth fault critical state vector set according to the third fault critical state vector set; and combining the second fault critical state vector set and the fourth fault critical state vector set to obtain a plurality of fault critical state vectors of the target power system.

The invention discloses a spatial decomposition enhanced database tuning method based on large model guided exploration, comprising the following steps: S1, constructing a multi-source knowledge system and completing verification, and providing domain knowledge support for subsequent knob screening; s2, screening performance key knobs through a hybrid expert mechanism to realize static compression of a configuration space; s3, initializing an adjustment and optimization sample set, and defining a relative performance optimization target taking throughput and delay as cores; s4, performing dynamic adjustment and optimization based on a spatial decomposition algorithm, and realizing efficient configuration exploration in combination with a UCB criterion and Bayesian optimization; and S5, repeating the dynamic tuning step in the step S4 to a preset number of iterations, and outputting the configuration with the optimal relative performance in the sample set as the final parameter configuration of the database. According to the method, the knobs are accurately screened through multi-source knowledge and the MoE mechanism, the configuration space is efficiently explored in combination with space decomposition and BO, the tuning efficiency and the database performance can be improved, and the method can quickly adapt to dynamic working load drift.

Disclosed are a pose decoupling kinematics calibration method, device and apparatus, and a storage medium. The method includes the following steps: constructing a pose decoupling motion model based on an equivalent rotation vector and a rigid body motion transformation rule; introducing a pose error, and carrying out pose space decomposition through the pose decoupling motion model to obtain an orientation error expression and a position equivalent expression; analyzing a coupling influence of an orientation change on a position according to the orientation error expression and the position equivalent expression to obtain a pose coupling increment; constructing a target decoupling calibration model according to the pose coupling increment and an initial decoupling calibration model; and solving geometric errors in different degrees of freedom for the target decoupling calibration model according to a preset axis pose error by a least square method to obtain a calibration error.

The invention discloses a network fault positioning quick response system based on distributed log analysis, and relates to the technical field of computer networks and operation and maintenance, and the system comprises a data acquisition module which generates a time sequence data frame based on a sliding window; the state vector construction module is used for extracting indexes and constructing standardized state vectors; the covariance matrix updating module is used for recursively updating the dynamic covariance matrix by using the forgetting factor; the feature space decomposition module is used for constructing a residual projection matrix based on the dynamic principal component dimension parameters; the anomaly detection module is used for calculating residual anomaly energy and determining a fault candidate set based on the contribution degree; the parameter self-adaptive module is used for calculating a residual spatial entropy value and feeding back and adjusting a principal component dimension parameter of a next frame; through entropy feedback closed loop and orthogonal projection, adaptive accurate delimitation of the gray scale fault is realized.

A pose-decoupled kinematic calibration method, apparatus and device, and a storage medium. The method comprises: constructing a pose-decoupled motion model on the basis of an equivalent rotation vector and a rigid body motion transformation rule (101); introducing a pose error, and performing pose space decomposition by means of the pose-decoupled motion model to obtain an orientation error expression and a position equivalent expression (102); analyzing the coupling effect of an orientation change on a position on the basis of the orientation error expression and the position equivalent expression so as to obtain a pose coupling increment (103); constructing a target decoupled calibration model on the basis of the pose coupling increment and an initial decoupled calibration model, wherein the initial decoupled calibration model is constructed on the basis of the pose-decoupled motion model (104); and using the least squares method to perform geometric error calculation of different degrees of freedom on the target decoupled calibration model on the basis of preset axial pose errors, so as to obtain a calibration error (105). The method can solve the technical problem in the prior art that the influence of an orientation on a position in a multi-degree-of-freedom motion system is not considered, making it is difficult to eliminate an error difference between different degrees of freedom, and thus resulting in poor calibration and compensation effects.

The invention relates to the technical field of medical image processing, in particular to an ultrasonic image processing system based on image enhancement, which comprises a field construction step: constructing a pixel-level structure tensor matrix, and mapping original ultrasonic image data into tensor field data; a flow direction locking step: performing feature space decomposition on the tensor field data to generate a diffusion guide graph containing a main feature vector; a nonlinear evolution step: constructing an anisotropic diffusion tensor based on characteristic value information, and executing time step iterative solution based on a partial differential equation; a reunion output step: performing boundary continuity reunion on the enhanced image data; according to the method, the direction of the tissue is accurately locked through the Riemannian metric tensor field, smooth texture smoothing and inverse texture sharpening are achieved, speckle noise is greatly restrained, and meanwhile the microstructure features are reserved.

The invention discloses a financial complex query statement generation method based on a search algorithm and action space decomposition, and belongs to the technical field of computers. The method comprises the following steps: acquiring user query data; screening a candidate atomic function set from the atomic function library according to user query data; constructing a function combination according to the candidate atomic function set; and verifying the function combination to obtain an optimal function combination as a query statement. According to the method, through action space decomposition, the difficulty of generating complex query statements is reduced, and common cascade errors of a single-step generation model are effectively avoided by utilizing the global planning capability of Monte Carlo tree search. Compared with the background technology, experimental data shows that the method improves the accuracy and robustness of generating the complex query statement to a certain extent. And meanwhile, simulation execution and consistency verification are assisted, so that the generated function combination is ensured to be correct in grammar and parameter.

The application discloses a 3D Gaussian splash optimization method based on iterative spatial decomposition, first data acquisition and preprocessing, then iterative spatial decomposition and hierarchical geometric prior sampling; hierarchical geometric perception initialization; 3D Gaussian model training and hierarchical geometric perception densification; finally, iterative optimization and output. The application solves the problem that the random downsampling or curvature-based downsampling method in the prior art is difficult to balance between efficiency and spatial structure reservation.

The application provides a medical image cross-modality generation method based on double-contrast frequency domain decomposition and adaptive diffusion, which comprises the following steps: acquiring MRI samples and preprocessing, constructing a k-space decomposition and enhancement process, constructing a k-space double-flow adaptive degradation network, and constructing a k-space double-flow adaptive prediction network. Through data enhancement technology and an adaptive Gaussian filter set, different frequency components are separated and enhanced, common edge features are highlighted, and the quality of generated information is improved. By using a residual noise diffusion model, high-resolution images with rich details and high diversity are generated while maintaining smooth areas and large-scale structures of images. By introducing anatomical structure consistency constraints and multi-modality fusion optimization strategies, the complementary information of T1 and T2 weighted MRI is fully utilized, and the generation effect and visual authenticity of the medical image cross-modality generation in a complex scene are significantly improved.

The present application relates to the technical field of computer vision, and provides a mirror reflection scene rendering method based on multi-space decomposition, which can perform high-quality new view rendering on a scene containing mirror reflection interference, implicitly decomposes a three-dimensional real scene into multiple parallel subspaces through a neural network, obtains a feature map through rendering, finally maps each sub-space feature map into an RGB image and a weight image through a decoding network and a gating network, and finally obtains an image of the scene at a specified view angle through weighted summation, and uses the corresponding view angle RGB image for supervision, which not only can render a higher quality image, but also avoids the multi-view inconsistency interference model fitting process caused by mirror reflection.

The invention discloses an unmanned aerial vehicle detection path planning method for low-altitude security and protection, and relates to an unmanned aerial vehicle. The method comprises the steps that geographic space data and flight mission parameters of a target area are acquired; constructing a three-dimensional grid map model according to the geographic space data; based on the three-dimensional grid map model, establishing a multi-constraint flight model comprising a height safety constraint, an airspace limitation constraint and a sensor coverage constraint; carrying out two-stage decomposition on the flight space by adopting a hierarchical set division method to obtain a hierarchical space decomposition graph; constructing an optimization problem of unmanned aerial vehicle detection path planning based on the spatial decomposition graph; solving the optimization problem by adopting a hybrid heuristic algorithm to obtain an unmanned aerial vehicle detection path covering the target area; according to the method, a large-scale complex three-dimensional urban space path planning problem is converted into a hierarchical small-scale sub-problem through hierarchical spatial decomposition, efficient solving is performed in combination with a hybrid heuristic algorithm, and the solving efficiency is improved.

The application discloses an array inter-coupling channel estimation method based on an OFDM waveform, and comprises the following steps: S1, a signal transmission model based on an OFDM waveform is given; S2, assuming that a self-interference channel is static, multiple times of changing a transmitting and receiving beamforming vector are given, and multiple groups of r(f k ) are obtained, so that a calculation formula of an H(f k ) estimation value is obtained; S3, zero space decomposition is carried out on a transmitting beam pointing direction vector q t and a receiving beam pointing direction vector q r , and a W and V selection strategy is given; S4, the transmitting beamforming vector is sequentially configured as a column vector in W, and the receiving beamforming vector is sequentially configured as a column vector in V, so that multiple groups of r(f k ) are obtained; S5, the step S4 is performed K*J times, so that an R(f k ) matrix is obtained, and an estimation value of H(f k ) is obtained, that is, a self-interference channel matrix on each subcarrier is obtained. The application can be performed when a transmitting and receiving array surface normally works, transmission efficiency of array service is improved, and changes of a channel state can be timely responded.

The invention discloses a machine forgetting method based on gradient decomposition, which comprises the following steps of: determining a forgetting sample set and a reserved sample set from a training sample set of a trained model, and respectively inputting training samples in the forgetting sample set and the reserved sample set into the trained model; respectively calculating to obtain a forgetting gradient and a retention gradient corresponding to the forgetting loss and the retention loss, carrying out orthogonal decomposition of the forgetting gradient on the retention gradient, extracting components of the forgetting gradient in an orthogonal subspace, and then carrying out gradient symbol space decomposition to obtain a positive space and a negative space of dimensions in the gradient; and fusing the forgetting gradient and the reserved gradient in different modes according to the space to which the gradient dimension belongs to obtain a final gradient, updating parameters of the trained model according to the final gradient, and repeating the process until an end condition is reached. According to the method, forgetting information and reserved knowledge are explicitly decoupled through gradient decomposition, and the interpretability and robustness of machine forgetting are improved.

The invention provides an unmanned aerial vehicle rapid autonomous exploration method based on spatial layering. The method comprises the steps that S1, point cloud information around an unmanned aerial vehicle is collected to construct a three-dimensional voxel map, and an area in a boundary is recognized; s2, dividing an area in the boundary into a plurality of sub-areas by adopting an online space decomposition algorithm, constructing a target cost matrix, combining path length cost and speed optimization cost, solving a non-closed traveling salesman problem with boundary constraint, and determining a traversal sequence of each sub-area; s3, generating a sphere safety space in each sub-region according to a traversal sequence, and generating a candidate viewpoint set by adopting polar coordinate Gaussian distribution sampling; and S4, screening an optimal viewpoint from the candidate viewpoint set based on the global loss function, optimizing the target function by adopting a B-spline trajectory to generate a local trajectory, and controlling the unmanned aerial vehicle to execute a flight exploration task along the local trajectory. The method has the beneficial effects that the application scene of the unmanned aerial vehicle can be expanded, optimal exploration path planning is realized, and the collision risk of unknown obstacles is avoided.

The application discloses a structure settlement visual measurement method and system based on deep learning compensation, and relates to the technical field of building settlement measurement. The method comprises the following steps: according to a mobile detection platform, driving a detection component to collect a structure region and determining structure data; constructing a settlement measurement component by spatial decomposition and implicit causal analysis of settlement dimensions, motion dimensions and interference dimensions; performing background domain segmentation on the structure data, executing subspace decomposition feature recognition and implicit causal coupling analysis, and determining settlement data pairs; performing directional projection and three-dimensional mapping through a three-dimensional projector, realizing three-dimensional distance measurement, and generating and updating a settlement curve display. The technical problems of the existing structure settlement measurement, such as the difficulty in effectively correcting the spatial displacement measurement error caused by attitude change and environmental interference, and the insufficient distance measurement accuracy, are solved, the technical effect of realizing high-precision non-contact distance measurement and dynamic compensation of structure settlement displacement is achieved, and the stability of spatial distance measurement and the environmental adaptability are improved.

The invention discloses a three-dimensional scene reconstruction method in a complex environment, which has the functions of capturing scene information and constructing a three-dimensional model by adopting a neural network, and is characterized in that the scene information capturing step comprises the following steps of: firstly, capturing a scene high-resolution image from multiple angles, performing standardization processing such as color correction and noise removal and the like; and then depth information of each pixel is extracted through geometric layout, and consistency and integrity of data are ensured. According to the step of constructing the three-dimensional model through the neural network, according to collected scene data information, through spatial decomposition and assumption that the intersection part of light and a sampling point is a sphere, the sampling process is optimized, input information is simplified, and the network modeling efficiency is improved; the network comprises a density neural network and a color neural network which output volume density and color information of a scene respectively, and then the density and color information are integrated through a volume rendering technology to generate a high-quality three-dimensional model. According to the invention, the accuracy and efficiency of three-dimensional reconstruction are obviously improved.

The invention discloses a multi-radiation source target direct tracking method, which is characterized in that a plurality of spatially distributed receiving arrays perform data acquisition on an unknown number of targets, each receiving array is composed of a plurality of array elements, and snapshot data are synchronously acquired and received, so that a receiving signal model under a multi-array collaborative observation condition is established. For the model, a plurality of spatial covariance matrixes are constructed, noise subspace information of the spatial covariance matrixes is extracted, and a MUSIC spatial spectrum is formed in combination with a subspace decomposition theory. According to the method, a spatial spectrum value is used as an observation likelihood function, and a generalized Labeled Multi-Bernoulli (GLMB) filtering framework is embedded, so that joint estimation of the number and the state of multiple targets is realized. And finally, performing propagation and weighted updating on the state particle set by combining a sequential Monte Carlo method to obtain the position of the target.

The invention discloses a multi-angle one-dimensional scattering center three-dimensional reconstruction method, and belongs to the technical field of radar signal processing. According to the method, through a scattering center dimension raising method combining subspace decomposition, RANSAC, spherical projection and visibility screening, a three-dimensional scattering center space structure of a target is recovered on the basis of a multi-angle one-dimensional distance scattering center; and the three-dimensional scattering center space distribution of the target can be deduced only by depending on multi-angle one-dimensional broadband distance domain scattering center estimation. According to the method, the problem that the three-dimensional scattering center cannot be recovered when only small-angle or discrete angular domain measurement is carried out is solved, and the availability of the three-dimensional scattering center in a non-cooperative observation scene is remarkably improved; the method has the advantages of high robustness and strong anti-noise and false scattering center capability. The method does not depend on continuous angle coverage, can only depend on non-missing data to realize three-dimensional scattering center inversion when part of angles are missing, and is especially suitable for non-cooperative targets such as maneuvering aircrafts, ships and satellites.

The application relates to a flow timing model continuous learning method, device and equipment based on orthogonal rotation. The method comprises the following steps: locking a pre-training timing model deep layer parameter, sampling a sub-data set from each time window of historical traffic data, and inputting the sub-data set into a shallow layer multiple times for light gradient updating according to time; after each updating, a key matrix is extracted and spliced, projected to a shallow layer weight matrix row space, decomposed to obtain an input orthogonal matrix; a covariance matrix is constructed, decomposed to obtain an output orthogonal matrix, and shallow layer weight updating is iterated until a condition is met, so that a congestion prediction model is obtained; and real-time traffic data is input to obtain a future road condition result. The method does not damage an original feature space, realizes continuous learning with high efficiency and low consumption, and the model is accurate in prediction.

The application relates to a product quantization nearest neighbor search method based on optimal space decomposition, which comprises the following steps: obtaining a sample data set; solving the optimal solution of space decomposition of the sample data set; decomposing the sample data set into M subspaces according to the optimal solution of space decomposition, and respectively performing cluster analysis in each subspace to obtain a plurality of clusters; constructing an inverted order table by taking the cluster center as an index; obtaining data to be searched; decomposing the data to be searched into M subspaces according to the optimal solution of space decomposition, and respectively searching a plurality of nearest clusters in each subspace; calculating the Cartesian product of the plurality of nearest clusters to obtain a plurality of candidate code sets; and querying the inverted order table according to the plurality of candidate code sets to obtain a search result set. The application converts the space division problem of product quantization into a space decomposition problem of balancing the data distribution volume of each subspace, realizes reasonable division of the space, effectively reduces the quantization error, and improves the search precision.

The invention relates to the technical field of brain nerve data driving, in particular to a complex brain state discrimination method and system and a storage medium, and the method comprises the steps: carrying out the hierarchical processing of the time sequence dynamic characteristics of an acquired original EEG signal, so as to extract the time sequence characteristics of each time window; the time sequence features are fused to construct a space graph structure, multi-layer dynamic graph convolution operation is carried out on the space graph structure to generate a multi-layer space feature tensor used for representing the space dependency relationship of the original EEG signals, and the multi-layer space feature tensor is stacked into a space feature three-dimensional tensor according to the hierarchy dimension of space decomposition; performing cross-level feature fusion on the spatial feature three-dimensional tensor by using a gating circulation unit to obtain cross-level spatial fusion features, and calculating discrimination values of different complex brain states; and finally, by simulating a brain cognitive decision loop, determining the type of the complex brain state based on the discrimination value. According to the method and the device, the technical problem of how to efficiently decode the EEG signal to comprehensively and accurately represent the complex cognitive state of the brain is solved.

The invention provides a power grid large-scale grounding grid state evaluation method, which belongs to the technical field of grounding grid detection, and comprises the following steps of: extracting a resistance component and an inductance component through frequency domain impedance analysis to construct a sparse impedance characteristic matrix, and establishing an aggregation grid distribution matrix by adopting a self-adaptive grid strategy; potential distribution and current density distribution are calculated through a regional decomposition parallel algorithm to obtain a conversion equivalent circuit parameter vector, a solving strategy is dynamically adjusted through nuclear space and image space decomposition, and the conversion equivalent circuit parameter vector and a standard grounding grid parameter vector are compared to calculate an abnormal state evaluation index vector; a corrosion area is marked to generate a corrosion position distribution matrix, a grounding grid state evaluation report is output in combination with an adaptive measurement strategy and an automatic correction mechanism, and the technical problems that the detection cost is high, the workload is large and comprehensive detection cannot be achieved due to the fact that excavation verification is needed for large grounding grid corrosion state detection are solved.