64results about How to "Accurate reconstruction" patented technology

The invention relates to the field of nuclear waste bin transmission image reconstruction. The invention discloses an SA-ResNeXt-based method for reconstructing a nuclear waste bin detection image of an array detector, which can accurately reconstruct a nuclear waste bin chromatography TGS transmission image under sparse transmission scanning. The method comprises the following steps: S1, performing a simulation experiment on the nuclear waste bin based on Geant4 simulation software to obtain training data, and performing a real experiment to obtain verification data; establishing a database for the training data and the verification data; s2, constructing a ResNeXt neural network model according to the transmission image characteristics of the nuclear waste bin; s3, training data in the database are selected to train the ResNeXt neural network model; an optimized ResNeXt neural network model is obtained; and selecting verification data in the database, inputting the optimized ResNeXt neural network model, and realizing reconstruction of the nuclear waste bin transmission image through the optimized ResNeXt neural network model. By adopting the method, the reconstruction accuracy of the TGS image can be greatly improved, and the reconstruction speed can be improved.

The invention relates to the technical field of spine data processing, in particular to a spine feature recognition method and system based on image processing. The method comprises the following steps: acquiring user optical data and spine image data, and performing external form analysis according to the user optical data to obtain external form data; performing spine support analysis on the external form data to obtain spine support data; performing spine skeleton feature extraction according to the spine image data to obtain spine skeleton feature data; and according to the spine support data and the spine skeleton feature data, carrying out spine multi-modal feature fusion to obtain spine feature data so as to carry out spine feature identification auxiliary operation. According to the method, based on multi-modal data fusion, precise spine evaluation and feature recognition are carried out by combining the geometrical morphology of the spine, supporting force analysis and spine image data.

The invention discloses a non-stationary industrial process fault monitoring method based on a self-adjusting non-stationary converter network. The method comprises the following steps: firstly, dividing a variable into a stationary variable and a non-stationary variable by adopting enhanced Dickey-Fuller test; and a sequence stabilization module is designed for non-stationary variables, and the non-stationary variables are converted into stationary variables, so that the reconstruction capability of the model is improved. Meanwhile, a stationary attention removing module is provided to compensate non-stationary information lost by stationary operation, and it is ensured that the model can retain and make full use of inherent non-stationary information when processing non-stationary data. Secondly, a two-stage training strategy based on weight self-adjustment is provided, in the first stage, input window data are reconstructed, the reconstruction deviation of the first stage is used as priori knowledge of attention weight updating of an attention module in the second stage, the model is made to pay more attention to a time sequence with the large deviation, and therefore the convergence speed of the model is increased; and the accuracy and reliability of fault monitoring are improved.

The invention belongs to the technical field of hydrology and landform, and discloses a riverway terrain reconstruction method and device based on energy loss and genetic algorithm coupling, and the method comprises the steps: collecting a remote sensing image of a riverway, the actual measurement data of a hydrological station, and the elevation data; determining a water surface width sequence of a river channel overflowing section based on the river channel water surface vector of the remote sensing image; analyzing actually measured data near the river channel, and determining hydraulic parameters corresponding to the river channel overflowing section; based on the hydraulic parameters, with the minimum energy loss as the optimization target, determining a target width-depth ratio sequence of the river channel overflowing section by adopting a genetic algorithm; based on the water surface width sequence and the target width-depth ratio sequence of the river channel overflowing section, constructing a river channel underwater terrain; and fusing the underwater topography with the elevation model data to complete reconstruction of the river topography. According to the scheme, on the basis of the measured data and the optical remote sensing image of the river channel, various hydraulic parameter characteristics and energy loss corresponding to the overflowing section are considered, and the more accurate river channel terrain can be reconstructed.

The invention belongs to the technical field of point cloud, and relates to a point cloud attribute coding method based on a graph attention network, which comprises the following steps: acquiring point cloud data, and carrying out block division on the point cloud data; performing detail level generation on the point cloud data subjected to block division, predicting point cloud attributes according to a detail level generation sequence, and inputting a neighbor point set with attribute information into a multi-head image attention layer for feature extraction to obtain a prediction residual error of a current point; encoding the prediction residual error of the current point to obtain an attribute code stream; inputting the attribute code stream into a decoding module to obtain a prediction residual error; searching a neighbor point set of unpredicted points in the prediction point set, and inputting the neighbor point set into the attention layer of the multi-head image to predict the current point; combining the predicted value with the decoded predicted residual error to obtain a reconstruction attribute; according to the method, efficient compression and accurate reconstruction of the point cloud data are realized in a layer-by-layer recursive predictive coding mode, and the method can be applied to a three-dimensional scene needing efficient data transmission and storage.

A kind of time domain ground penetrating radar intelligent forward method based on physical parameter guidance, comprising the following steps: step one: build underground medium model and generate supervision data;Step two: the ground penetrating radar GPR forward simulation is expressed as the mapping problem of " physical parameter field to radar profile ";Step three: build PhyTANet network based on encoder decoder;Step four: the PhyTANet network obtained by step three is trained, fine-tuned;From the above steps, finally, after training fine-tuning ends, the dielectric constant and conductivity distribution of any given underground medium are input, and the corresponding prediction result is obtained.The purpose of the present application is to solve the technical problems of the existing ground penetrating radar forward technology, such as insufficient accuracy, insufficient modeling of key physical parameter coupling relationship and limited time domain feature expression in complex heterogeneous medium, and the present application is proposed.

This invention discloses a hip joint prosthesis for bone defects, comprising a crescent-shaped acetabular rim filler assembly, a bone-metal defect block, an acetabular wall defect support block, a cylindrical support block, an angle pad, a locking screw, a locking nut, an assembly screw, an acetabular sealing screw, and an acetabular support block. The crescent-shaped acetabular rim filler assembly includes two sets of crescent-shaped acetabular rim fillers, which are assembled by engaging pins through pin holes. The outer circumference of the bone-metal defect block of the acetabular support block is connected to the crescent-shaped acetabular rim filler assembly via bone cement. This hip joint prosthesis for bone defects effectively fills irregular defect areas of the acetabular rim. The various screw holes and Kirschner wire holes on its surface provide multi-dimensional fixation options. The overall structure, through the synergistic action of multiple components, achieves precise reconstruction while ensuring biomechanical performance, significantly reducing the risk of postoperative prosthesis loosening and wear.

The invention discloses a face gear on-machine measurement comprehensive error compensation method based on sequential decoupling. Specifically, a sequential decoupling compensation strategy is provided for solving the problem that the measurement precision is limited due to mutual coupling of a measuring head pre-stroke error and a workpiece installation coordinate system error in the prior art. The method comprises the following steps: firstly, establishing a measuring head pre-stroke error model through standard ball calibration, and performing anisotropic compensation on a tooth surface measuring point by utilizing Delou inner triangulation and a weighted interpolation method; secondly, based on the compensated data, the outer cylindrical surface and the tooth top plane of the face gear are accurately fitted through an iterative optimization algorithm, an accurate measurement coordinate system is established, and the initial phase of the C axis is optimized; and finally, reconstructing an actual tooth surface by adopting a two-stage NURBS curved surface fitting method, and calculating a normal deviation. According to the method, an error propagation chain is effectively cut off through a strict execution sequence, the overall precision and reliability of on-machine measurement are remarkably improved, and a key technical support is provided for face gear closed-loop manufacturing.

This invention provides a physical information neural network modeling method for the hysteresis characteristics of piezoelectric fast-acting servo systems, belonging to the technical field of hysteresis characteristic modeling for piezoelectric fast-acting servo systems. The method includes the following steps: based on the input voltage of the piezoelectric fast-acting servo system at multiple sampling times up to the current time, an expert feature vector for the current time is constructed through an expert feature layer based on a memory path; static and dynamic representations are extracted through static and dynamic branches respectively; the static and dynamic representations are fused through a feature representation layer to obtain a fused representation for the current time; and the fused representation for the current time is mapped to a scalar displacement prediction of the piezoelectric fast-acting servo system through a regression head. The advantage of this invention is that it improves the accuracy and robustness of modeling the hysteresis characteristics of piezoelectric actuators.

This invention discloses a broadband oscillation disturbance source localization method based on compressed sensing and graph convolutional neural networks (GCNs), comprising two stages: offline training and online localization. In the offline training stage, electrical quantities under subsynchronous / supersynchronous broadband oscillation modes are obtained using measurement data from actual systems or simulation examples, constructing a broadband oscillation offline sample library. Compressed sensing is used to encode and compress the electrical quantities in the offline sample library, obtaining a training set and a test set. The constructed GCN localization model is trained using the training set until the test set reaches the target localization accuracy, resulting in a broadband oscillation localization model. In the online localization stage, a substation collects electrical quantity data and compresses and encodes it using compressed sensing technology; the substation data is uploaded to the master station; at the master station, the feature matrix and system adjacency matrix are input into the trained GCN localization model, and the oscillation source location is output. This invention's method is applicable to broadband oscillation disturbance source localization under various operating conditions.

This invention relates to the field of machine vision technology and discloses a robust extraction method for the center line of structured light in fillet welds. It aims to solve the problem of missegmentation and loss of corner features caused by traditional algorithms relying on geometric connectivity segmentation under strong arc light and high reflectivity interference. The invention first obtains a light stripe backbone mask through multi-scale morphological filtering and clustering screening; secondly, it constructs a scattered data pool by performing global coarse extraction within the mask; subsequently, it uses a sequential random sampling consensus algorithm to adaptively extract the equations of the upper and lower main lines and calculate the theoretical intersection point through a "fit-stripping" iterative mechanism; finally, it extends along the main line towards the intersection point and performs independent pixel grayscale thresholding and sub-pixel fine extraction within a local window of the original image. This invention completely eliminates the risk of missegmentation, accurately repairing large-scale reflective fractures while perfectly preserving and reconstructing the sharp corner features of the fillet weld.

This invention relates to the field of optical detection technology and discloses a method for measuring the three-dimensional topography of an object surface based on optical interferometry. The method includes: acquiring a mixed temporal-domain interferometric signal sequence during axial scanning; extracting seed pixels and constructing a spatial topological connectivity feature matrix; introducing a fractional Fourier transform to construct a joint decoupling objective function containing a frequency-domain nonlinear phase distortion evaluation index and a temporal-domain physical invariant evaluation index; performing multi-level optimization for the seed pixels to lock the globally optimal rotation angle; performing array adaptive decoupling in conjunction with the spatial topological connectivity feature matrix to generate the optimal rotation angle matrix; and performing signal decoupling and reconstruction on the mixed signal to extract the absolute surface height of each pixel to generate three-dimensional topography data. This invention solves the phase decoupling problem caused by thin-film interferometric aliasing through multi-dimensional joint decoupling and spatial prior constraints, improving measurement accuracy while reducing the computational overhead of large pixel arrays.

The application discloses a kind of active implantable medical device in-vivo radio frequency response modeling method based on equivalent lumped element model, by constructing and matching the electromagnetic characteristics of the actual conductive path of AIMD and surrounding tissue, equivalent lumped element model (LEM), the key factors such as the multiscale electrical characteristics of the geometry size of implantable medical device and its in homogeneous, non-uniform even real anatomic tissue environment are parameterized characterization, so as to quickly and accurately reconstruct the radio frequency transfer function of implant.The method of the application effectively breaks through the bottleneck of the existing Tier 3 evaluation method, meets the technical requirements of the latest standard based on Tier 4 evaluation method for modeling the radio frequency transfer function of implant in the environment closely related to clinical tissue, and provides a new technical path and implementation means for efficient and accurate evaluation of the radio frequency heating risk of implantable medical device in the magnetic resonance environment.

The invention relates to the technical field of computer vision and image processing, and discloses an image interpolation method based on high dimensionality. The method comprises the following steps: establishing a video decoder module, a timestamp analysis module, a feature extraction module, a time sequence correlation modeling module, a time sequence correlation modeling module, a convolutional neural network module, an edge smoothing module, an edge detection module, an edge enhancement module, an image output module and a quality evaluation module; the time stamp analysis module records time points and positions; the feature extraction module extracts image edge feature data; the time sequence correlation modeling module captures time and spatial correlation data; the convolutional neural network module performs pixel-level interpolation on image frames; the edge enhancement module is used for enhancing edge pixel points, the image output module is used for generating high-quality interpolation image frames, and the quality evaluation module is used for carrying out quality inspection on the generated image.

The application discloses a motor controller sampling delay error compensation method, a motor controller and a motor driving system. The method comprises the following steps: obtaining operation parameters comprising motor physical model parameters, real-time state parameters, pulse width modulation parameters and a preset sampling delay time; calculating a threshold time based on the pulse width modulation parameters; determining a target calculation model according to the size relationship between the sampling delay time and the threshold time; dividing the sampling delay time into one or more time segments according to voltage vector switching points, and calculating the real-time current change slope of each segment; multiplying each segment by the corresponding slope and accumulating the current error amplitude, thereby correcting the current sampling value. Thus, the real current value lost due to hardware delay is accurately calculated, so as to solve the current sampling delay error problem.

The application discloses a kind of physical constraint under aero-magnetic data deep learning forward and inversion method and system. Including: constructing the forward network based on deep operator network DeepONet, the network adopts logarithm-linear dual space mechanism, to log resistivity profile and flight height as input, predicts log quadratic field;Inversion network is constructed, with the input of observed transient electromagnetic response and flight height, and the output of resistivity profile;When training, the forward network is pre-trained to convergence under data-driven loss and Maxwell equation physical constraint loss;Inversion network is trained under the joint supervision of its model space loss and physical data space loss, wherein the physical data space loss is obtained by comparing the re-synthesized response with the original observation after inputting the resistivity profile predicted by the inversion network into the frozen forward network.The application deeply integrates physical laws into data-driven models, and realizes the integrated efficient solution of aero-magnetic forward and inversion.

The invention relates to the technical field of image inversion, and provides a diffusion probability model-based image inversion method, which comprises the following steps of: firstly, carrying out noise addition on a real image based on a deterministic noise addition algorithm of a denoising diffusion implicit model to generate an implicit vector; secondly, regarding a text embedding vector generated by corresponding text description as an optimizable variable; then, inputting the text embedded vector and the implicit vector into a stable diffusion model together, and performing a denoising process based on a denoising algorithm of a denoising diffusion implicit model to obtain a reconstructed image; and finally, adopting a root-mean-square error between the reconstructed image and the real image as an optimization loss function, and optimizing a text embedding vector through a gradient descent algorithm so as to realize high-precision reconstruction of the real image.

The present application provides a method, device, computer equipment and storage medium for constructing a face model, which can be applied to the field of artificial intelligence or intelligent transportation, etc., to solve the problem of low efficiency in constructing a face model. The method includes: based on the multiple target face high-polygon data, adjusting the pre-stored template face data, respectively generating the first face low-polygon data corresponding to each of the multiple target face high-polygon data; based on the pre-stored reference face low-polygon data set, respectively performing noise filtering processing on each of the multiple first face low-polygon data to obtain corresponding second face low-polygon data, wherein the reference face low-polygon data set contains the face features of each reference face; performing data fusion processing on the obtained multiple second face low-polygon data to generate a target face model. By reducing the number of rounds and the amount of data that need to be adjusted, the purpose of improving the efficiency of constructing a face model is achieved.

This invention discloses an online detection method and device for the trajectory of sphere polishing, relating to the field of precision machining online detection. The method includes: S1, acquiring a continuous image sequence during the sphere polishing process at time i using a high-speed camera; S2, preprocessing and feature recognition of the continuous image sequence; S3, a rotation angular velocity measurement module calculating the rotation angle and rotation angular velocity to obtain theoretical values; and a three-dimensional trajectory reconstruction module radially projecting discrete trajectory points onto a sphere surface determined based on the first point, to achieve visualized three-dimensional trajectory reconstruction under geometric constraints, thus completing the online detection of the sphere polishing trajectory. This invention combines a transparent grinding disc structure, a multi-view observation layout, and a dedicated image processing algorithm to achieve real-time, high-precision, and high-efficiency online detection of the sphere's rotational motion parameters and trajectory during the polishing process.

The invention discloses an SAA-DoA high-precision direction finding method based on LSTM time sequence prediction, and the method comprises the following steps: sequentially switching a plurality of antenna units through a sequential sampling mode of a switch antenna array, and obtaining the signal sampling data of a target direction; performing phase processing on the signal sampling data, and extracting an instantaneous frequency time sequence representing carrier frequency drift; inputting the instantaneous frequency time sequence into a pre-trained long short-term memory network model, learning a dynamic change rule of carrier frequency drift through the model, and predicting an instantaneous frequency value of an unsampled time slot; based on the predicted instantaneous frequency value, reconstructing a complete parallel sampling data matrix through phase integral operation; based on the reconstructed parallel sampling data matrix, an array signal processing algorithm is adopted to calculate the arrival angle of the signal; according to the method, the precision of DoA estimation under the SAA architecture and the robustness in a real wireless environment are remarkably improved.

This disclosure provides a method, apparatus, electronic device, and storage medium for lung segment reconstruction. It calls a preset algorithm to reconstruct lung segments from lung imaging data, generating initial lung segment reconstruction data. The reconstructed data is then validated to identify deviated lung segments with reconstruction deviations at their junctions. Based on the reconstruction instructions for these deviated lung segments, a lung segment assembly is reconstructed. Based on the data collected from the lung segment assembly, the junctions of each lung segment within the assembly are reconstructed and repaired to obtain the final lung segment reconstruction data. Compared to related technologies, this disclosure, by calling a preset algorithm for lung segment reconstruction, can utilize advanced algorithmic technology to initially outline the lung segment structure. Then, by validating the reconstructed data, it can accurately identify deviated lung segments with reconstruction deviations at their junctions. For these deviated lung segments, a lung segment assembly is reconstructed according to their specific reconstruction instructions, ensuring accurate reconstruction when dealing with problematic areas.

The invention discloses an automatic test system and method based on a relay protection device, and relates to the technical field of relay protection device analysis. In the relay protection test process, electrical time sequence data and device state signals in the test stage are collected in real time, data are processed, and a scene data unit capable of being independently stored and analyzed is generated; constructing a test path transfer directed graph; predicting a subsequent stage candidate set according to the test path transfer directed graph; forming a to-be-executed candidate list according to the predicted subsequent stage candidate set in combination with the current available resource set; analyzing a resource conflict relationship among the test items, and generating an optimized test sequence; when it is detected that the test process is abnormally interrupted, a continuous test execution instruction is generated by conducting mode matching on the context and a historical successful continuous test case, the test is recovered from the breakpoint, and meanwhile the interruption and recovery mode is stored in a historical continuous test case library. And the success rate and the safety of continuous testing are improved.

The invention discloses a hyperspectral anomaly detection method based on spatial spectrum attention enhancement and a blind spot U-Net network, the overall architecture of the network is in a U-Net form, and interference of an abnormal target on subsequent feature extraction is eliminated through mask convolution and down-sampling operation. Firstly, in order to effectively capture global information, a spectrum-space self-attention module is embedded in a network; particularly, for a spatial self-attention module, in order to prevent blind spot information from interfering subsequent feature extraction, a mask is added to an attention map to meet blind spot requirements. In addition, in order to perceive the multi-class attributes of the background, an attention module based on a sequence dictionary is designed in the potential layer space. The multi-class attributes of the background are perceived through an additional dictionary, so that the background expression is effectively enhanced, and the anomaly detection precision is improved. And after the network converges, obtaining an anomaly detection result through a reconstruction error. According to the method, target features can be enhanced for interference of small targets, complex backgrounds and the like, so that a relatively good detection effect is achieved.

The present application provides a kind of microscopic image distortion correction and reconstruction method, system, equipment and storage medium, preset multi-shape calibration slide, including: regular grid slide, concentric circle slide, irregular curve slide and multi-layer depth slide;The image of multi-shape calibration slide under different field of view, different magnification and different focal plane is collected by microscope, and the geometric distortion and blur characteristics of image are quantified by combining edge detection, contour analysis and blur measurement method, and distortion feature vector is generated;Point spread function model is constructed based on distortion feature vector, and point spread function model is optimized using convolutional neural network;The optimized point spread function model is applied to the deconvolution processing and geometric correction of microscopic image, and finally the clear image after reconstruction is obtained.The present application can effectively correct geometric distortion and edge blur, restore the details and clarity of microscopic image.It has wide application prospect in scientific research, clinical diagnosis and industrial detection field.

This invention belongs to the field of blade trajectory monitoring technology, specifically involving a method for dynamic monitoring of the three-dimensional trajectory of wind turbine blades by fusing imagery and laser point cloud data. The steps include: fixing a camera and scanner with a rigid connection device for use as a dynamic monitoring system for the three-dimensional trajectory of the wind turbine blade; calibrating the camera; calibrating the relative position and attitude between the camera and scanner; acquiring image data and laser point cloud data during normal operation of the wind turbine; performing motion compensation correction on the laser point cloud data; calculating the three-dimensional coordinates corresponding to the monitoring point; obtaining a time-series image-side coordinate sequence of the monitoring point; and calculating the three-dimensional coordinate sequence corresponding to the time-series image-side coordinate sequence of the monitoring point, i.e., the three-dimensional motion trajectory of the monitoring point on the wind turbine blade during operation. This invention uses imagery and laser point cloud technology to monitor the three-dimensional trajectory of wind turbine blades, requiring no auxiliary markers, eliminating complex calibration, offering flexible deployment and high accuracy, and efficiently supporting blade safety assessment and maintenance decisions.

The application discloses a five-dimensional seismic data interpolation method and system based on coordinates, constructs a point-by-point seismic data interpolation model based on a NeRF theory, modifies an MLP framework by combining an additional convolution layer on the basis of the point-by-point interpolation model, constructs a profile-by-profile seismic data interpolation model, introduces kernel norm regularization to construct an objective function, reads original five-dimensional seismic data D obs and inputs the five-dimensional seismic data with missing traces into the network to realize reconstruction of the five-dimensional seismic data. The application eliminates the need for additional labeled data, utilizes unique characteristics of seismic data, enables a convolution network decoder to output data profile by profile, can significantly improve data processing efficiency by 40 times, and increases kernel norm regularization in the objective function, so that the noise resistance and robustness of the model can be effectively improved.

The present application relates to finite time reduced dimension observer design method and chaotic communication system, belong to information security and chaotic communication field. In view of the defects of traditional asymptotic observer slow, time is uncontrollable, sliding mode method exists chattering and message signal pollution output is difficult to construct observer, the present application establishes the chaotic drive system containing message signal embedding, constructs the equivalent system without unknown signal, designs finite time reduced dimension observer based on observer matching condition and strong observable condition. Compared with the traditional asymptotic convergence observer method, the present application can realize the complete synchronization of chaotic system within the user preset time, adopts continuous control law without chattering, and the message signal can be accurately recovered through algebraic formula. The synchronization time of the present application is controllable, the convergence speed is fast, the recovery precision is high, and the robustness is strong, which is suitable for various chaotic secure communication scenes, effectively improves the communication real-time, security and engineering practicability.

The present application relates to a method and system for electromagnetic tomography reconstruction based on graph convolutional neural network, the method comprising: carrying out grid division on the to-be-measured region of the target image, obtaining a plurality of grid units and constructing a topological graph; constructing an electromagnetic characteristic unit based on the nodes and edges in the topological graph, calculating the conductivity distribution and the corresponding current density distribution of the electromagnetic characteristic unit, and obtaining the node features in the topological graph; constructing the node features into a node feature matrix and inputting the node feature matrix into a pre-trained graph convolutional neural network model to output a feature map; counting the distribution of feature values in the feature map, combining the concentration degree and the change range of the electromagnetic characteristics in different regions, obtaining the mixing state of the multiphase flow in the stirring process, and completing the electromagnetic tomography reconstruction of the target image. The present application effectively mines the complex relationship of non-euclidean data in the to-be-measured region, and provides strong support for state monitoring, flow behavior analysis and process optimization in the mixing and stirring process of multiphase flow.

The invention discloses a hybrid five-level bidirectional active bridge containing an auxiliary branch and a control method thereof, and relates to the technical field of power electronics, and the hybrid five-level bidirectional active bridge comprises the steps: introducing an auxiliary bridge arm and a coupling auxiliary transformer into a primary side circuit, and enabling the primary side of a high-frequency transformer to generate five step levels by controlling the state of the bridge arm. According to the control, real-time voltage gain is calculated, a target switch time sequence mode is automatically selected according to the relation between the real-time voltage gain and a preset reference threshold value, and an optimal phase shift combination is calculated by calling a GOC optimization algorithm with the purpose of minimizing inductive current stress. According to the method, a plurality of voltage matching points are physically constructed through the coupling auxiliary transformer, and more control freedom degrees are provided for current stress optimization; and a GOC optimization equation is utilized to ensure that the device operates at a current stress minimum point under all working conditions. Through collaborative optimization of software and hardware, current stress of all working conditions is reduced, a ZVS region in a full load range is widened, and efficiency and stability of a wide voltage working range are improved.

The application discloses a table structure recognition method and device, a storage medium and an electronic device. The method comprises the following steps: obtaining a target picture containing a target table object; using a predetermined target key point prediction model to predict key points of the target picture by using a key point heat map prediction method, to obtain coordinate values of each key point and type information of each key point; using a predetermined target key point matching model to match each key point based on the coordinate values of each key point and the type information of each key point, to obtain a plurality of key point sets; and performing table reconstruction based on the coordinate values in each key point set, to obtain a recognition result of the target table object. In the application, the key point prediction model and the key point matching model are used, so that a plurality of key point sets corresponding to each cell can be accurately obtained, and the target table object can be accurately reconstructed based on the plurality of key point sets.