366 results about "Smoothing filter" patented technology

The invention relates to the technical field of geothermal drilling monitoring management, in particular to a construction monitoring system and method for geothermal drilling. The method comprises the following steps: constructing a state variable set, quantifying structural disturbance intensity by adopting a disturbance derivative method, normalizing the structural disturbance intensity, and defining a state transition tensor mapping behavior influence coefficient; in combination with the high-order derivative of the disturbance trend function and the behavior collaborative change matrix, an offset risk factor is generated through fusion for early warning; dividing a well drilling area into three-dimensional grids, calculating local stress potential energy, adjusting weight through a dynamic sensitive belt, and generating a risk distribution diagram through smooth filtering; performing closed-loop dynamic regulation and control by using risk potential energy centroid coordinates and gradient field inversion intervention vectors for medium and high risk areas; and through multi-period risk potential energy tensor analysis, calculating the matching degree of a self-healing index and a stable cone region, evaluating the self-healing capability of the system and feeding back a regulation and control strategy. According to the invention, a closed-loop monitoring and active control system for nonlinear risks of geothermal drilling is formed.

The invention discloses a battery health state prediction method based on multi-modal feature fusion and deep learning, and the method comprises the steps: obtaining the charging data, including current, voltage and temperature data, of each charging period during the multi-stage constant current charging of a battery; determining SOH (state of health) data of the battery; performing smooth filtering on the current and voltage data through a Savitzky-Golay filter, reconstructing a current curve, and reconstructing a voltage curve by adopting a linear interpolation method and a KF filter; constructing a comprehensive characteristic pool according to the current curve, the voltage curve and the temperature data, and directly adding the constant current value at the maximum charging amount into the comprehensive characteristic pool; screening features from the comprehensive feature pool and putting the features into a prediction feature pool; completely integrated empirical mode decomposition is carried out on the SOH data, and IMF components obtained through decomposition are put into a prediction feature pool; the method comprises the following steps: establishing a CNN (Content Number)-BILSTM-ATTENTION And performing single-step prediction and multi-step prediction on the SOH of the battery by using the trained model. According to the invention, high-precision prediction of the SOH of the electric vehicle battery can be realized.

The invention discloses a rice spectrum stress index construction method and system, computer equipment and a storage medium, and relates to the technical field of rice stress classification, and the method comprises the steps: regularly shooting different groups of rice leaves to obtain original hyperspectral images, and carrying out black and white correction; extracting regions of interest from rice leaves under different stresses, calculating average spectral data and performing SG smoothing filter processing; using a generative adversarial network to extract 50 characteristic wave bands with the highest contribution degree; inputting the average spectral data corresponding to the characteristic wave band as model input data into a random forest classification model to obtain a model classification result; and meanwhile, screening the first five characteristic wave bands and corresponding weights to construct spectral indexes, obtaining stress indexes, dividing stress categories, and outputting stress reasons of the rice. According to the method, the nonlinear relationship in the data is more efficiently captured, the classification model performance is improved, the calculation efficiency is high, the adaptability is high, the dimensionality of the hyperspectral data is greatly reduced, and the model calculation complexity is effectively reduced.

The invention discloses an image processing algorithm for pupil center real-time extraction and trajectory tracking in an eye closing state. According to the algorithm, from the angle of digital image processing, graying, contrast enhancement and smoothing processing are firstly carried out on an input eye video frame by frame, and then the boundary of a pupil area can be accurately extracted by adopting an adaptive threshold segmentation technology and combining morphological operation and a Canny edge detection operator. And then, accurate fitting of ellipse parameters is performed on the pupil contour by using a geometric distance-based least square method, so that center coordinates and area information of the pupil are obtained, and the accuracy and reliability of pupil monitoring in an eye closing state are remarkably improved. Through recording and analyzing pupil center position changes in a time sequence, real-time tracking of pupil movement tracks is realized. And finally, evaluating pupil light reflex (PLR) in an eye closing state before and after white light excitation by adopting baseline correction and smooth filtering processing. The method has the advantages of high robustness, adaptability to illumination variation, high real-time performance, high precision and the like.

The invention discloses a spectral gas concentration inversion method based on a CNN-Transform fusion model, and belongs to the technical field of trace gas detection and spectral signal processing. The method comprises the following steps: acquiring second harmonic spectrum data of target gas through a TDLAS (tunable diode laser absorption spectroscopy) system; carrying out abnormal value elimination and smooth filtering processing on the acquired signal, and constructing a standard input vector; a depth model fusing a one-dimensional convolutional neural network (CNN) and a Transform architecture is designed, the CNN is used for extracting local spectrogram features, and the Transform is used for modeling a global dependency relationship; and inputting the training set and the verification set into the model for training optimization, and finally realizing high-precision inversion of the sample gas concentration of the test set. Experimental results show that the method is superior to an existing model in the aspects of fitting precision, robustness and error control, and has good physical consistency and engineering popularization value.

The invention discloses a spaceborne hyperspectral detection laser radar atmospheric particle inversion method. The method comprises the following steps of solving a laser ranging value and an altitude corresponding to a spaceborne hyperspectral resolution laser radar (HSRL) echo signal sampling point; reducing the noise of the echo signal by using a noise removal method; carrying out distance and system constant correction on the de-noised signal; a forecasting model is used for reanalyzing the data set, and S6, a molecular scattering signal transmittance lookup table is established through an atmosphere model and an iodine molecule hyperspectral filter absorption spectrum; calculating back scattering coefficients and scattering ratios of the cloud and the aerosol; performing hierarchical identification by using a scattering ratio and voltage signal-to-noise ratio threshold method; determining an effective inversion region through hierarchical region processing; calculating an initial extinction coefficient to a radar ratio; and performing Gaussian smoothing filtering on the radar ratio to obtain a final radar ratio, and calculating a final extinction coefficient profile. The method is characterized in that cloud and aerosol level identification is carried out on spaceborne HSRL echo signals, a molecular transmittance lookup table of a hyperspectral channel is established, and cloud and aerosol optical parameters are accurately and efficiently solved.

The invention discloses a multi-base-station AOA cooperative low-altitude target rapid positioning system. The system comprises an AOA measurement base station network, an AOA data preprocessing module, an adaptive weighted intersection positioning module, an extended Kalman filtering state estimation module and a data fusion and system integration module. The system synchronously measures the arrival angle of a target signal through multiple base stations, removes noise in combination with smoothing filtering and an anomaly rejection algorithm, and then solves the initial position of a target by using a self-adaptive weighted intersection algorithm based on measurement quality and geometric distribution. And then, fusing the target motion model and the measurement model by adopting an extended Kalman filtering algorithm to realize dynamic estimation and prediction of the position, the speed and the course. The system can realize high-precision and real-time positioning and continuous tracking of targets such as low-altitude unmanned aerial vehicles, small aircrafts and the like in a complex electromagnetic environment and a sight distance limited scene, and has visual display and regional alarm functions.

The invention relates to the technical field of oil and gas exploitation, in particular to a pipeline sand flow monitoring method and system based on ultrasonic signal layered optimization. The pipeline sand flow monitoring method mainly comprises the steps that firstly, a piezoelectric type ultrasonic sensor is used for collecting ultrasonic signals generated when gravel in a pipeline impacts a pipe wall; the method comprises the following steps: sampling a sand flow, carrying out adaptive nonlinear smooth filtering on an ultrasonic signal, carrying out noise suppression through sparsity analysis based on a frequency domain, finally constructing a mathematical model of a relationship between a sampling signal and the sand flow, and monitoring the sand flow according to the mathematical model. The problem that signals are prone to interference and confusion in a traditional ultrasonic monitoring method can be solved to the maximum extent, high-performance real-time measurement and accumulated sand production amount calculation of the sand production flow of the oil and gas pipeline are achieved, when the real-time sand production flow or the accumulated sand production amount exceeds an early warning value, an alarm is given out, a possible serious sand production event is prompted, and the real-time sand production flow of the oil and gas pipeline is monitored. And data support is provided for related decisions.

The invention discloses an automatic control method and system for a bias point of a photoelectric modulator, and relates to the field of signal processing, and the method comprises the steps: generating a composite disturbance voltage signal through a digital signal processor; the bias electrode is injected into the photoelectric modulator and used for collecting the optical signal output by the photoelectric modulator and converting the optical signal into an electric signal; extracting a fundamental component amplitude and phase difference data of a second harmonic component, a third harmonic component and a fundamental wave; dynamic characteristic parameters are calculated through spectral analysis, and the mapping relation between the characteristic parameters and the offset of the offset points is established; outputting the offset direction and the offset level of the offset point through a deep learning model; a bias voltage compensation amount is generated through a fuzzy PID (Proportion Integration Differentiation) controller; and adjusting the bias electrode voltage, smoothing the filter, and eliminating high-frequency jitter. The method has the advantages that high-precision and stable bias point adjustment is realized through the composite disturbance signal, real-time spectrum analysis and deep learning optimization decision in combination with fuzzy PID control and a dynamic closed-loop mechanism, and the system performance and stability are remarkably improved.

The invention provides a bedding structure rock finite element model reconstruction method based on a rock core digital image, and belongs to the technical field of digital rock cores, and the method comprises the following steps: S1, obtaining a gray level image of a rock sample; s2, calculating a transverse mean value of the analysis area to obtain a longitudinal gray level distribution curve; s3, applying moving average filtering and smooth filtering to obtain a smooth gray curve; s4, first-order difference is carried out on the smooth gray level curve, the position with the absolute difference value larger than a threshold value is taken as an initial stripe boundary candidate, and starting and stopping pixels, the width and the average gray level of each stripe are determined; s5, mapping each clustering label interval into a physical interval based on FEM grid partition; and S6, constructing a sample finite element model, and generating a reconstructed finite element model with real partition information. According to the method, the macrostructure features of the rock can be efficiently recognized and extracted by directly utilizing the pictures and combining gray profile analysis, and efficient and low-cost reconstruction of the finite element model of the bedding structure rock is achieved.

The invention discloses a periodogram and incoming wave direction finding fused multi-unmanned aerial vehicle signal identification method and system, and relates to a signal processing technology, and the method comprises the steps: obtaining to-be-identified signals with time domain aliasing and frequency domain non-aliasing; suppressing the noise of the to-be-identified signal by using an average periodogram method and smooth filtering, and calculating a bottom noise value; according to the calculated bottom noise value, setting a judgment threshold to detect a target unmanned aerial vehicle signal; estimating the start-stop frequency and bandwidth of the detected target unmanned aerial vehicle signal; performing signal separation by using an FIR filter according to the estimated starting and ending frequency and bandwidth; and executing incoming wave lateral direction based on the separated target unmanned aerial vehicle signal so as to complete signal identification. According to the method, unmanned aerial vehicle signal detection is completed through an average periodogram method, unmanned aerial vehicle weak signals which are sparsely distributed in space are further obtained through a direction finding method, and weak signal unmanned aerial vehicles which are far in distance are detected.

The invention relates to the technical field of image processing, and particularly discloses an e-commerce image dynamic optimization adjustment method based on market feedback, which removes color noise and inhibits interference in edge detection through gray level conversion and smooth filtering technologies, extracts and analyzes edge curvature features by using Canny edge detection and wavelet transform, accurately identifies a sawtooth region, and improves the image quality. And performing boundary alignment on the pictures of the same commodity under different perspectives by using a shape context descriptor and a Hungary algorithm, calculating a shape difference coefficient to evaluate the consistency, and for further optimization processing, converting the shape difference coefficient and an edge curvature feature into a comprehensive feature vector as the input of a graph convolutional network model, so as to obtain the image convolutional network model. Global features are extracted through multi-layer graph convolution operation, an area with serious jagging is recognized, a hard threshold method is adopted for denoising, and the jagging area is accurately repaired through a cubic spline interpolation method.

The invention discloses a 4D face reconstruction method and device, and relates to the field of face reconstruction, and the method comprises the steps: obtaining a plurality of face image groups in a preset time period; reconstructing a face triangular mesh model for each face image group; performing fitting optimization on each face triangular mesh model by adopting a three-dimensional face deformation model to obtain a corresponding registered face mesh model; for any face image group, performing flicker removal processing, extracting an initial texture image, and mapping the initial texture image to the corresponding registered face mesh model to obtain a mesh with texture; and performing time sequence smooth filtering processing on the grids with textures corresponding to all the face image groups to obtain a 4D dynamic face grid sequence. According to the method, the 4D dynamic face grid sequence with high fidelity and coherent time sequence can be reconstructed end to end.

A method of performing diffusion-weighted magnetic resonance imaging is provided. The method includes a step of: using a neural network to filter a diffusion-weighted image of an object acquired by a magnetic resonance imaging scanner, the neural network being programmed to produce an output image from the acquired image. The neural network improves the signal to noise ratio of the output image relative to the acquired image. The neural network, when applied to a synthetic knife-edge image to which Rician noise providing a signal-to-noise ratio of 13 or more is added, forms a curve of normalised values of modulation-transfer-function against frequency which has a higher area thereunder than the area under the corresponding curve of normalised values of modulation-transfer-function against frequency for a reference Gaussian smoothing filter.

The invention discloses a PCB semantic segmentation method fusing ResNet and an attention mechanism. The method comprises the following steps of performing defect segmentation area cutting and data enhancement on a PCB image; carrying out image smoothing filtering on the enhanced PCB defect image data set by adopting a generalization edge preserving and structure preserving image smoothing model, and inputting a UNet model encoder to extract defect area features; performing layer-by-layer down-sampling on the defect area features to obtain initial feature maps of different scales; inputting the initial feature maps of different scales into a CBAM module according to corresponding levels, and adding the feature maps processed by the CBAM attention mechanism with the initial feature maps of different scales by the CBAM module; and the UNet model decoder obtains the feature map of each layer of the CBAM module, recovers the image resolution layer by layer, obtains an image boundary, and outputs a defect segmentation result. According to the invention, the segmentation precision of PCB image detection can be improved.

Techniques are described herein for processing video data using enhanced interpolation filters for intra-prediction. For instance, a device can determine an intra-prediction mode for predicting a block of video data. The device can determine a type of smoothing filter to use for the block of video data, wherein the type of the smoothing filter is determined based at least in part on comparing at least one of a width of the block of video data and a height of the block of video data to a first threshold. The device can further perform intra-prediction for the block of video data using the determined type of smoothing filter and the intra-prediction mode.

Acoustic resonators with piston mode patches are disclosed. In one aspect, an interdigitated acoustic resonator (300) having plural fingers (308(1)-308(N)) or digits includes a modified piston mode rail. In particular, the piston mode rail is replaced with individual piston mode patches (320(1)-320(N); 322(1)-322(N)) that may be uniform or varied in size. Selection of patch size allows spurious modes to be suppressed to have a smooth filter passband with low overall insertion loss and minimal ripple. Further, the patches may be made through a monolithic process, which reduces overall production cost, complexity and cycle time.

The invention discloses a mechanical arm teleoperation control method and system based on incremental mapping, and the method comprises the steps: carrying out the noise reduction processing of the position and posture of a handle through adaptive low-pass filtering, and achieving the position and posture consistency mapping from a handle space to a mechanical arm working space after coordinate system axis rearrangement and similar transformation mapping; the system provides two kinds of control output of a pose tracking mode and a speed tracking mode, real-time inverse kinematics solution is carried out through a Jacobian matrix and a damping least square method in the speed mode, and resolution scaling and exponential moving average / spherical linear interpolation smooth filtering are carried out on a target pose in the pose mode. And then the handle motion instruction is mapped into a Cartesian instruction or a joint space instruction at the tail end of the mechanical arm, closed-loop execution is conducted through a real-time state feedback link, and the method can be used for scenes such as robot service, teaching, remote operation and data collection.

The invention discloses an underground fluxgate sensor signal processing method, which comprises the following steps of: extracting a magnetic field voltage signal mainly comprising a second harmonic component by adopting a signal detection circuit, and further removing other harmonic components through a digital phase-sensitive rectification and smooth filtering algorithm to realize effective extraction of the magnetic field signal. The second harmonic extraction effect of the phase-sensitive rectification algorithm is verified through a time sequence simulation result, and an open-loop fluxgate sensor designed based on a second harmonic method is experimented. Compared with other types of fluxgate signal detection methods, the second harmonic method effectively improves key performance such as resolution and linearity of the sensor, and the fluxgate sensor designed based on the second harmonic method can realize detection and effective modulation of a detected magnetic field signal and shows a relatively high magnetic measurement level.

The invention provides an underwater gravity measurement method, device, equipment and medium based on position observation, the method is realized by gravity measurement based on Hampel and RTS smoothing filtering, and the method comprises the following steps: cleaning and identifying horizontal position data provided by a USBL by using Hampel filtering, and replacing outliers in the horizontal position data to obtain more reliable horizontal position data; the SINS system performs a moving base alignment process by means of the cleaned high-precision horizontal position data; the SINS system performs pure inertial navigation calculation, continuously provides navigation parameters, and suppresses accumulative errors of the inertial navigation system through RTS smooth filtering processing; and an original gravity anomaly result is calculated, and precision evaluation is carried out after filtering processing of an FIR low-pass filter. According to the method, the gravity data is efficiently and accurately processed under the condition that USBL data output is unstable, and the reliability of a gravity data processing result is improved while a high-precision gravity measurement result is obtained.

The invention relates to a computed tomography reconstruction method and system for a sparse angle of a cylindrical lithium battery, and the method comprises the steps: obtaining a cylindrical lithium battery projection image sequence as original projection data, and carrying out the diagonal interpolation processing of the original projection data, and generating complete sinogram data; performing Radon inverse transformation on the sinogram data to generate an initial iteration reconstruction image and taking the initial iteration reconstruction image as an iteration initial value, and performing iteration optimization by adopting a simultaneous algebraic reconstruction technology as a data fidelity term to obtain an optimized reconstruction image; designing a tangential smoothing filter, and filtering the optimized reconstructed image to suppress sparse artifacts; a radial weighting template is designed, and differential smoothing processing is carried out on the reconstructed image after sparse artifact suppression according to the distance between pixels and the center of the image; and performing reconstruction image optimization, sparse artifact suppression and differential smoothing processing circularly until a preset number of iterations is reached, thereby obtaining a final reconstruction image. According to the method disclosed by the invention, detail information of the image is effectively recovered while sparse artifacts are suppressed.

The invention relates to the technical field of image processing, and discloses an improved image defogging method and device based on an atmospheric scattering model, and the method comprises the steps: carrying out the color channel balance processing of a to-be-processed foggy image, and obtaining a first image; performing atmospheric light intensity value calculation through the first image to obtain a target atmospheric light intensity value; the method comprises the following steps: performing multi-scale edge smoothing filtering processing on a first image by adopting a to-be-processed foggy image to obtain a transmissivity image; and performing segmentation defogging processing on the first image by using the transmissivity image, a first preset image segmentation threshold, a second preset image segmentation threshold and a target atmospheric light intensity value to obtain a first defogging area image, a second defogging area image and a third defogging area image, and performing correction and splicing processing to obtain a target image. The defogging accuracy of the to-be-processed foggy image can be improved.

The invention provides a histogram equalization method based on gray scale redistribution. The method comprises the following steps: S1, carrying out statistics on image histogram information; s2, performing Gaussian smoothing filtering on the histogram information; s3, histogram clipping is carried out, the filtered histogram is clipped by adopting a preset parameter clipRatio, a clipped histogram clipped Hist is obtained, and statistics is carried out on a clipped part count; s4, carrying out histogram redistribution; and S5, according to the obtained histogram reHist, calculating a mapping function cdf, and mapping the input brightness image y. The key point of the method is the realization process of brightness image histogram equalization. The statistical brightness image histogram is cut and redistributed to different regions of interest, so that the influence of dominant gray levels on a mapping function in a traditional histogram equalization method is avoided, and the problem of excessive brightness enhancement of the traditional histogram equalization algorithm is solved; wherein the distribution weights and distribution modes of different regions of interest keep more dark region / bright region details for the image after histogram equalization.

The invention belongs to the field of petroleum drilling, particularly relates to a downloading pulse signal judgment and pulse width adaptive identification method, system and device, and aims to solve the problem that a downloading instruction after the pulse width is changed cannot be automatically identified. The method comprises the following steps: acquiring pulse frequency data according to a set sampling period; filtering the collected pulse frequency data through a first smooth filtering window and a second smooth filtering window in sequence to obtain the average rotating speed and pulse waveform data of the turbine generator; and taking a curve corresponding to the pulse waveform data as a first curve, selecting a set number of points on the first curve, and determining the pulse width according to the distance between every two pulse wave crests. Under the working conditions of different well depths and different mud densities, the high downloading success rate is achieved, the pulse width can be increased at any time on the ground, the situation that a downhole tool is pulled out of the ground for resetting is not needed, and the drilling efficiency is greatly improved.

The invention discloses a thermal power generating unit boiler wall temperature early warning method and device, and the method comprises the steps: determining a steady-state working condition according to boiler operation parameters; the boiler operation parameters comprise coal feeding quantity, water feeding quantity, primary air quantity and secondary air quantity; calculating the wall temperature change rate of each monitoring point in a preset time window; the time window is set according to a data smooth filtering requirement and an early warning response timeliness requirement, and the monitoring points comprise a superheater, a reheater and a water cooling wall; and if the wall temperature change rate of at least one monitoring point is greater than or equal to a preset early warning threshold value and the duration is greater than or equal to a corresponding time threshold value, executing a corresponding early warning strategy. Whether the boiler is in the steady-state operation state or not is recognized before early warning is triggered, so that wall temperature fluctuation interference in the normal dynamic process such as unit load adjustment and combustion working condition fine adjustment is eliminated, it is avoided that the normal working condition is misjudged to be abnormal, the early warning timeliness of the wall temperature of the boiler is improved, meanwhile, the misjudgment rate is reduced, and the early warning accuracy is improved.

The invention discloses an AUV (Autonomous Underwater Vehicle) pose detection method and system based on underwater light vision, a medium and a program product, and belongs to the field of underwater robots. The method comprises the following steps: judging the turbidity of a water body by analyzing the blue channel characteristics of an image, dynamically adjusting the graying weight, and generating a preprocessed image by adopting blue channel enhancement; constructing triple constraint conditions of contour number, spacing and area, and combining a variable step size strategy to realize threshold adaptive optimization; wick pixel coordinates are extracted through contour circularity screening, a leak detection grading mechanism is established, and feature point compensation is performed based on exponential smoothing filtering prediction and historical frame data; and finally, performing P3P pose calculation according to confidence levels. The method can be suitable for various water areas, blue light source feature extraction and feature point matching tasks are well achieved, and the stability and accuracy of AUV pose detection are improved.

The invention discloses an aerial image defogging method for enhancing details by using transmissivity. The aerial image defogging method consists of three parts of solving an atmospheric scattering model, extracting image details and fusing the atmospheric scattering model and the image details. Specifically, an atmospheric light value is obtained through a proposed region search scoring method, and meanwhile, the transmissivity obtained from a smoother and filtered fog image is substituted into an atmospheric scattering model to solve a defogged image. In the image detail processing, the smoothed image is subjected to weight smoothing, sharpening and filtering operation, the transmissivity is introduced into the regional variance weight in the filtering operation, a filter is controlled by the transmissivity to perform smoothing or sharpening operation, and the filtered image is subtracted from the smoothed image to obtain a detail image. And finally, a defogged image is formed by linearly fusing an atmospheric scattering model inverse solution image and an image after detail processing. According to the method, the details of the original image after defogging are effectively reserved, and the definition of the restored image is improved.

A method for designing a spectacle lens, said method comprising: an approximate curvature map acquisition step for acquiring a distribution map on xy plane coordinates of an approximate curvature (Cp) for each point on a surface (alpha), which is a design surface; a smoothing processing step of performing smoothing filtering processing on the distribution map; and a design step for designing a new surface (alpha ') on the basis of the distribution diagram of the new approximation curvature (Cp') obtained by the smoothing step.

The invention discloses a multi-degree-of-freedom mechanical arm inverse kinematics real-time solving method and device, and the method comprises the following steps: building a forward kinematics model of a mechanical arm, so as to obtain a forward kinematics solution of the mechanical arm; defining a loss function, and calculating a current pose error of the mechanical arm in combination with a forward kinematics solution; when the current pose error is greater than a set error threshold value, calculating the gradient of the pose error; gradient normalization processing is carried out to prevent gradient explosion or disappearance; gradient filtering after gradient normalization is carried out, and interference of noise on a resolving result is reduced; the step length is adaptively adjusted through the PID controller; step length delta is set and limited, impact is prevented, joint variables are updated, and real-time performance is guaranteed; a track is obtained through calculation, and smooth filtering processing is carried out on the track; and recalculating the error until a set threshold condition is met. According to the method, through the improved gradient descent algorithm, the convergence speed and the calculation precision of inverse kinematics solution of the mechanical arm are effectively improved, and robustness and applicability are achieved.

The invention relates to a node representation method with personalized and global features in a social network. The method comprises the steps of global feature capture, personalized feature reservation and community feature extraction. According to the method, a Laplacian smoothing filter is obtained through calculation according to an adjacent matrix of a graph, multi-level feature aggregation is performed on an initial node feature matrix based on the Laplacian smoothing filter, and a node feature matrix after feature aggregation of each level is obtained; calculating a weight matrix of each level based on the node feature matrix after feature aggregation of each level; according to the weight matrix of each level, calculating a combined feature which retains personalized features; and inputting the combined features into an encoder for community feature extraction to obtain a node embedding matrix. According to the method, high-quality node representation can be learned, the performance is excellent in a node clustering task, and the excellent performance can also be shown in a link prediction task; compared with an existing method, the performance of the method is remarkably improved.