69 results about "Eigenvalue analysis" patented technology

The invention discloses a voltage source type energy storage small signal stability analysis method based on a virtual synchronous machine. A small signal state space model including active control, reactive control, double closed-loop control and a lithium battery energy storage module is established, a characteristic value analysis method is used for solving a state matrix characteristic root, system stability is evaluated, and an oscillation mode is identified. By drawing characteristic root trajectories of key parameters such as virtual rotational inertia, virtual damping coefficients and double-closed-loop PI parameters, an influence rule of the parameters on system stability is analyzed. According to the method, the dynamic response characteristics of the system can be accurately quantified, theoretical support is provided for control strategy optimization and parameter design of the voltage source type energy storage system, and the stability and reliability of a power grid in a renewable energy source grid-connected scene are effectively improved.

The invention discloses a rural power distribution network-oriented network tracking / constructing type source storage stability evaluation and planning method, and relates to the technical field of rural power distribution networks, and the method comprises the steps: constructing an inverter control model, and building a quantitative relation between an inverter capacity ratio and a system short-circuit ratio; according to the control model, an inverter hybrid system is established, and stability weak nodes are identified through characteristic value analysis of a system state matrix; and on the basis of the quantitative relationship and the stability weak nodes, calculating the energy storage capacity of the network to be constructed, and performing iterative configuration according to a preset priority until the stability requirement of the system is met. According to the method, accurate identification of the stability weak nodes and accurate configuration of the constructed energy storage are combined, accurate processing of the risk of the stability weak nodes of the rural power distribution network is realized, the disturbance resistance and the new energy consumption level of the rural power distribution network are improved, and stable, safe and efficient operation of the rural power distribution network is realized.

The invention discloses a forestry surveying and mapping method based on remote sensing big data, and relates to the technical field of image processing, and the method comprises the following steps: S1, obtaining an initial remote sensing image of a to-be-surveyed area, and carrying out the processing through a feature extraction model, and obtaining a standard remote sensing image; s2, generating a global matrix for the to-be-mapped area according to feature row vectors extracted from the initial remote sensing image and the standard remote sensing image; and S3, determining a final vegetation coverage rate according to the global matrix of the to-be-mapped area. According to the method, the concentration ratio of the vegetation features is analyzed and quantified through the feature values of the global matrix, interference of isolated pseudo vegetation pixels is avoided, the stability of a coverage calculation result is improved, and the method is suitable for large-scale and normalized forestry surveying and mapping scenes.

The invention provides a flutter analysis method and device for a photovoltaic flexible support system, equipment and a medium, and the method comprises the steps: building a finite element model of the photovoltaic flexible support system, obtaining the wind field parameters of all model nodes corresponding to a photovoltaic module in the finite element model in a three-dimensional wind field, and calculating the flutter of the photovoltaic flexible support system according to the wind field parameters. Calculating target effective wind attack angle distribution of each model node, determining a static wind load balance position of the photovoltaic module in the three-dimensional wind field according to the target effective wind attack angle distribution of each model node, and performing complex mode characteristic value analysis on the photovoltaic module according to the static wind load balance position to obtain each order of mode complex characteristic value corresponding to the photovoltaic module; and carrying out flutter analysis on the photovoltaic flexible support system according to the modal complex feature values of all orders. According to the method, the flutter analysis is performed by combining the wind field parameters of the model nodes corresponding to the photovoltaic module, so that the non-uniform factors of the wind field are considered, and the flutter analysis accuracy is improved.

The invention belongs to the technical field of lattice tower structure health detection, and relates to a lattice tower rod piece missing intelligent detection method based on multi-view three-dimensional reconstruction. According to the method, an adaptive feature matching algorithm fusing geometric features and texture descriptors is provided, camera pose estimation is optimized by constructing a global depth consistency scoring mechanism, and high-precision sparse point cloud reconstruction is realized; designing a depth map optimization strategy based on energy minimization, and effectively overcoming the interference of complex illumination and a low-texture environment on three-dimensional reconstruction in combination with density threshold filtering and geometric structure feature analysis; an ordered / disordered point cloud adaptive noise reduction method is innovatively provided, and noise robustness processing is realized through gradient constraint and covariance matrix eigenvalue analysis; and finally, an improved neighborhood radius adaptive registration algorithm is adopted, and a rod piece missing region is accurately identified through a significance scoring mechanism. The method remarkably improves the automation level and reliability of rod piece missing detection, and can be widely applied to structural health detection of the lattice tower.

The invention belongs to the technical field of image processing, and particularly relates to an insulator internal bubble identification method and system based on image enhancement, and the method comprises the steps: obtaining a perspective grayscale image of an insulator, constructing a structure background image and a smooth guidance weight based on the weighted least square fitting of gradient image guidance, and obtaining a background residual image; combining the smooth guide weight, the local information entropy and the background residual image to construct a weighted signal-to-noise ratio energy diagram; generating a bubble significance probability graph based on Hessian matrix eigenvalue analysis; and identifying bubble defects through image binaryzation and connected domain analysis. According to the method, through gradient-guided background fitting and multi-dimensional feature joint constraint, umbrella skirt edge artifacts and random noise interference are effectively suppressed, and the accuracy and robustness of insulator internal micro bubble defect detection are improved.

The present application relates to the field of power system stability analysis and control, in particular to a kind of thermal mass energy storage converter broadband oscillation suppression parameter stable domain identification method, comprising the following steps: establish general electric-thermal coupling mechanism model, construct the dynamic differential equation describing the evolution of input electric power and energy storage medium temperature state of converter;Mechanical dynamics equation is constructed;Establish the small signal state space model of whole system, linearization is carried out at steady state operating point and the state space matrix of whole system is derived;Define practical small disturbance stability domain, set the real part decay rate threshold and minimum damping ratio threshold of whole system eigenvalue;Algebraic singularity constraint equation in high-dimensional parameter space is constructed;Solve algebraic singularity constraint equation, establish parameter stable domain.This method avoids the repeated iteration of traditional eigenvalue analysis, can accurately depict parameter boundary, effectively improve the frequency support capability of high proportion of new energy power grid and system robustness.

The application belongs to the technical field of image processing, and particularly relates to an insulator internal bubble identification method and system based on image enhancement, which comprises the following steps: obtaining a perspective gray image of an insulator, constructing a structure background image and a smooth guide weight based on gradient image guided weighted least square fitting, and obtaining a background residual image; combining the smooth guide weight, local information entropy and the background residual image to construct a weighted signal-to-noise ratio energy image; generating a bubble saliency probability image based on eigenvalue analysis of a Hessian matrix; and identifying bubble defects through image binarization and connected domain analysis. The application effectively suppresses umbrella skirt edge artifacts and random noise interference through gradient guided background fitting and multi-dimensional feature joint constraint, and improves the accuracy and robustness of insulator internal small bubble defect detection.

The invention provides a harmonic suppression method, device, equipment and product based on broadband impedance measurement, and the method comprises the steps: carrying out the measurement and analysis of all elements in a hybrid multi-infeed DC system according to a preset broadband impedance measurement method, and obtaining broadband impedance data, the preset broadband impedance measurement method is composed of an active disturbance method and a wavelet time-frequency analysis method; according to the topological structure of the hybrid multi-infeed direct current system, determining an LCC converter and an MMC converter in the hybrid multi-infeed direct current system; according to an LCC converter and an MMC converter in the hybrid multi-infeed direct current system, carrying out classification processing on the broadband impedance data, and constructing a harmonic interaction impedance analysis model; according to a characteristic value analysis method, the harmonic interaction impedance analysis model is analyzed, harmonic analysis of the hybrid multi-infeed direct current system is completed, the real-time performance and robustness of harmonic suppression are improved, and a theoretical basis is provided for harmonic suppression and stability optimization of the hybrid multi-infeed direct current system.

The application belongs to the technical field of wind power generation, and specifically discloses a wind farm grid-connected system grid-following type and grid-forming type wind turbine proportion optimization method and device, which comprises the following steps: changing the initial proportion of the grid-forming type wind turbine, and changing the grid strength under the updated proportion; calculating the power flow distribution of each node of the wind farm grid-connected system under the updated proportion, determining that the grid-connected system does not reach the capacity threshold based on the power flow distribution; establishing a state space model of the grid-connected system, and judging whether the grid-connected system is small signal stable; under the condition of small signal stability, analyzing the stability margin of the grid-connected system based on the dominant modal eigenvalue, evaluating the adaptability of the wind farm to the change of the grid strength by using a discrete coefficient, and recording the discrete coefficient; cyclically performing analysis on all grid strengths and wind turbine proportions, obtaining the discrete coefficients under all proportions, and selecting the grid-forming type wind turbine proportion corresponding to the minimum discrete coefficient as the optimal proportion. The application can improve the operation stability of the wind farm configuration result.

The application relates to a dimension reduction method, device and storage medium of a model of a new energy grid-connected system, and the method comprises the following steps: S1, a linear dynamic model of a new energy unit grid-connected system is established; S2, an eigenvalue analysis method is used to analyze oscillation modes generated by the model, characteristic roots and participation factors existing in the system are calculated, and state variables mainly participated by the system are obtained; S3, the state variables are removed based on a time scale method, fast decay modes are omitted, dominant eigenvalues are reserved, a state equation of a dimension reduction system is established, and a dynamic response curve is analyzed based on the state equation of the dimension reduction system. Compared with the prior art, the application has the advantages of effectively realizing model dimension reduction of the new energy grid-connected system while guaranteeing model application effects and calculation accuracy, and reducing calculation amount.

The invention discloses a method for analyzing the influence of heterogeneous micro-grid parameters on system stability in a continuous steady state. The method is specifically implemented according to the following steps: step 1, acquiring steady-state working points of a heterogeneous micro-grid through parameter scanning; step 2, acquiring a low-frequency characteristic value of the heterogeneous micro-grid system in a continuous steady state based on the state space model of the micro-grid; and step 3, obtaining a participation matrix and a participation factor of the heterogeneous micro-grid system, determining a key variable according to the participation factor, and analyzing an influence rule of the parameter on the stability of the heterogeneous micro-grid according to the key variable. According to the method, the problem that in the prior art, a fixed working point is used for representing behaviors in the whole period in a traditional characteristic value analysis method, and consequently analysis is not accurate is solved.

The application belongs to the technical field of electrical engineering and automation control, and provides a device and method for improving anti-leakage current and anti-interference, which comprises a first instrument amplification module, a second instrument amplification module, a third instrument amplification module, a static current adjustment module and a microprocessor module; static current data and dynamic current data of a circuit are obtained; static characteristic values and dynamic characteristic values are obtained through analysis and processing; a feedback signal value is obtained through analysis and processing based on the obtained static characteristic values and dynamic characteristic values; the feedback signal value is compared with a feedback signal threshold value, and is analyzed and divided into different regions; the feedback signal value is analyzed and processed into a feedback signal of an electronic delay module, and the electronic delay module is confirmed to be connected correctly through analysis and processing, thereby completing networking; the anti-leakage current and anti-interference are effectively improved, and the data processing of a mixed signal is optimized and analyzed, and potential problems in the circuit are specifically analyzed, thereby improving the reliability and safety of equipment.

Embodiments of the present application disclose a user clustering method and device based on power load, equipment and a storage medium, the method comprising: obtaining power load data set of candidate users for indicating power consumption of the candidate users; performing outlier processing on the power load data set by using annual temperature characteristics and the power load data set of each candidate user, determining target users and target power load data set; performing eigenvalue analysis processing on the target users, the target power load data set and preset power consumption characteristic parameters, determining eigenvalues of the power consumption characteristic parameters corresponding to each target user; and performing clustering processing on each target user according to the eigenvalues, determining power consumption categories corresponding to each target user for reflecting power consumption habits of the target users. Through the above manner, the power consumption categories can be better classified, which helps the power grid company to better understand the load demand and power consumption characteristics of the users, and lays a foundation for individualized demand response scheme.

This application provides an adaptive infrared vascular imaging enhancement method and apparatus based on image entropy feedback. It achieves precise adaptive adjustment of infrared emission power through local information entropy calculation and a proportional-integral-differential controller. An enhancement mechanism is constructed, combining multi-scale Gaussian smoothing and Hessian matrix eigenvalue analysis to establish a reliable vascular response extraction strategy. Structural analysis optimization is introduced, iteratively refining skeleton extraction and cross-point calculation to ensure continuous improvement in vascular cross-point localization. This method effectively addresses the shortcomings of traditional techniques in exposure adaptation, vascular enhancement, and topology analysis, providing technical support for infrared vascular imaging-assisted diagnosis.

In aspect, a computerized method of a genetic algorithm-based adaptive batch selection for hessian quantization in neural networks comprising: with at least one computer processer, computing a Hessian Matrix; performing an Eigenvalue Analysis on the Hessian matrix to generate a Hessian matrix eigenvalue that provides information about the curvature of the loss surface; determining a quantization level based on the Hessian matrix eigenvalue; using the quantization Level to set an appropriate quantization level for a layer weights of a neural network; and applying the quantization level to the layer weights of the neural network. This involves mapping the continuous floating-point values of the weights to discrete levels based on the determined quantization intervals.

The present invention relates to the field of data processing technology, and in particular to a method for analyzing the local spectral properties of a Hamiltonian operator. The method comprises: obtaining the real-time temperature, power consumption, and coolant flow of a grid; determining a high temperature difference grid; determining a temporary grid; determining an analysis grid; inputting data into a Hamiltonian operator model; adjusting a temperature gradient threshold, and generating an abnormal property analysis report. The present invention collects temperature, power consumption, and coolant flow data from a quantum computer cooling unit in real time, and combines this with local spectral energy eigenvalue analysis to accurately identify and adjust temperature control anomalies. This not only improves the stability and efficiency of the cooling unit, but also dynamically optimizes the temperature gradient threshold. Through the generated abnormal property analysis report, potential problems can be discovered and maintained in a timely manner, effectively extending the life of the equipment and optimizing the utilization of cooling resources. This effectively solves the problem of low output accuracy due to over-reliance on historical data.

The invention discloses a structure nonlinear dynamic response incremental modal overlay analysis method and system, and belongs to the field of aerospace structure dynamics calculation. The method comprises the steps of constructing a finite element model and obtaining an initial matrix; dispersing the time-varying temperature field into a plurality of time increment intervals, updating material parameters in each interval, and calculating an instantaneous matrix; carrying out eigenvalue analysis and converting the kinetic equation into a modal coordinate system; solving a modal response increment by adopting a numerical integration method and converting the modal response increment to a physical coordinate system for superposition; and circularly executing until a full-time-history response is output. According to the method, the problem of rigidity nonlinearity caused by the temperature is solved through incremental linearization and modal order reduction, high calculation precision and high efficiency are achieved, and the method is suitable for dynamic design and analysis of aerospace structures in the high-temperature variable-temperature environment.

The invention discloses a risk identification system based on macro-micro percolation feature and multi-modal deep learning fusion, and relates to the technical field of network risk identification, and the system comprises the steps: obtaining real-time dynamic data from a distributed storage unit, constructing a time data curve, calculating a feature value, and carrying out the analysis of the feature value, and generating a corresponding node display signal; identifying abnormal nodes based on node display signals, constructing a traffic network model, performing multi-dimensional comparison with a preset model library, generating a decision instruction packet when the similarity exceeds a preset value, and executing node bearing capacity analysis and an intelligent distribution strategy through the decision instruction packet. According to the method, an intelligent risk identification and management and control system with active early warning, global optimization and autonomous decision making is realized, the robustness and operation efficiency of a network system are greatly improved, and the method is suitable for popularization and application. The problem of a risk response mode of passive response of traditional risk identification is solved.

The invention relates to the technical field of power system stability analysis, and particularly provides a photovoltaic grid-connected inverter interaction identification method based on damping ratio characteristics. The method comprises the following steps: establishing a power grid model comprising a photovoltaic inverter, and setting a power grid operation condition; carrying out load flow calculation according to the step 1; based on the step 2, selective characteristic value analysis configuration is carried out; according to the step 3, extracting a dimension reduction mode; the step 4 is utilized to analyze the participation factors, and small signal stability verification is carried out; and according to the step 5, calculating a reference index, and carrying out inverter interaction strength determination, and the method improves the planning and stable operation of the power grid.

The application discloses a low-frequency oscillation suppression method of a wind power / pumped storage combined grid-connected power system, which is composed of four grid-connected operation modes considering different working conditions of a wind power plant and a pumped storage power station, and comprises the following steps: 1. wind power and pumped storage two kinds of units are added with oscillation suppression links to suppress low-frequency oscillation together; 2. based on a wind speed probability distribution model, steady-state operation points and linearization models representing different grid-connected operation modes are determined; 3. eigenvalue analysis is adopted to define a dangerous mode by using a damping ratio, and a dangerous degree evaluation index is established; 4. eigenvalue sensitivity is calculated to determine key parameters of oscillation suppression and the adjustment ability of the key parameters to the dangerous mode; and 5. an optimization model of the damping ratio of the dangerous mode is established, and an artificial bee colony algorithm is extended with sensitivity information and inequality constraints, a fitness function is improved, and control parameters are coordinated and determined. The application can define the dangerous degree of an oscillation mode and the suppression ability of control parameters to low-frequency oscillation, so that low-frequency oscillation under different grid-connected operation modes can be effectively suppressed.

The invention discloses a VSC-HVDC sending end converter station control parameter feasible region determination method and a storage medium, and belongs to the technical field of high voltage direct current transmission, and the method comprises the steps: building a small signal model of a VSC-HVDC system; carrying out eigenvalue analysis and sensitivity calculation on a system matrix in the small signal model, and identifying key control parameters influencing the stability of the system; constructing an energy attenuation index reflecting the stability and the dynamic performance of the VSC-HVDC system; an energy attenuation index is solved, an energy attenuation index analytical solution based on positive definite weight matrix weighting is obtained, and a positive definite weight matrix is adjusted according to the key state variables influencing the dynamic performance; and according to the energy attenuation index analytic solution and the key control parameters, determining a sending-end converter station control parameter feasible region suitable for different operation working conditions. According to the method, the control parameter feasible region of the sending-end converter station under different operation working conditions can be quickly and accurately determined.

The invention provides a multi-surface interference source number estimation and harmonic iteration dissociation method, and belongs to the technical field of optical precision measurement and signal processing. In order to solve the problems that in the prior art, dependence on prior information such as the cavity length of a measured piece is high, and the harmonic separation precision is greatly influenced by noise and frequency aliasing, the method provided by the invention firstly adopts an interference source estimation method based on an information theory criterion; the number of effective surfaces participating in interference is determined by performing eigenvalue analysis on a covariance matrix constructed by single-pixel light intensity time sequence signals. And on the basis of the calculated number of the interference surfaces, an iterative optimization narrowband dissociation algorithm is adopted to adaptively decompose a complex superposed interference signal into a group of narrowband component functions with the minimum total bandwidth, so that harmonic signals corresponding to each surface and the center frequency and phase of the harmonic signals are extracted with high precision, and the interference surface number is calculated. And high-precision reconstruction of each surface is realized.

To evaluate, with high accuracy, the soundness of respective stories including a story not provided with an acceleration sensor.SOLUTION: A building soundness monitoring system comprises: an earthquake information acquisition unit 5 that acquires acceleration waveform data from an acceleration sensor 2; a section rigidity calculation unit 6 that calculates a primary characteristic frequency of a building and section rigidity; an analysis model generation unit 7 that calculates and sets story rigidity that is the rigidity of stories on the basis of the section rigidity calculated for sections and the weight of the stories, to generate an analysis model; an analysis model update unit 8 that updates the analysis model until the ratio between the primary characteristic frequency and a primary characteristic frequency calculated by executing characteristic value analysis on the analysis model becomes equal to or more than a lower limit threshold and equal to or less than an upper limit threshold, and repeats executing the characteristic value analysis; a maximum deformation angle estimation unit 9 that calculates a deformation waveform on the basis of a result of the last characteristic value analysis and the acceleration waveform data, and calculates the maximum deformation angle; and a structural performance estimation unit 10 that compares the maximum deformation angle with a deformation angle determination threshold to evaluate the soundness.SELECTED DRAWING: Figure 1

The application discloses a kind of soft measurement methods for furnace temperature of coal-fired unit based on data fusion, steps include: collecting the same time furnace combustion image and the operation data in coal-fired unit distributed control system DCS;Combustion image is pretreated;Establish hybrid deconvolution and interpolation upsampling feature extraction network;Structural similarity index loss is added to network training loss function;Feature extraction network is trained, and feature extraction model is obtained;M×N two-dimensional feature map obtained by the last layer encoder is extracted;Two-dimensional feature map is divided into same S region, and the feature value in single region is added as the representative feature value of the region;DCS operation parameter characteristics are analyzed and selected, and the highest correlation degree Q of furnace temperature is selected;S combustion image features and Q DCS operating parameters are data fused;Bidirectional long short-term memory neural network prediction model is established to predict furnace temperature.The application can improve the precision and robustness of furnace temperature soft measurement model.

The invention provides a small signal stability analysis method and analysis system for an optical storage heterogeneous power generation system, and relates to the technical field of photovoltaic grid-connected control. The method comprises the following steps: constructing a light storage heterogeneous series-parallel power generation system model; the optical storage heterogeneous series-parallel power generation system model comprises a photovoltaic direct current capacitor module, a photovoltaic filter inductor module, a photovoltaic network side control system module, an energy storage network side control system module, an energy storage filter inductor module and an energy storage filter capacitor module. Building a power grid equivalent model matched with the optical storage heterogeneous series-parallel power generation system model based on a unit structure in the optical storage heterogeneous series-parallel power generation system model and an input and output relation of each module; connecting the optical storage heterogeneous series-parallel power generation system model with the power grid equivalent model to form a closed-loop series-parallel power generation system model; and based on the characteristic value of the optical storage heterogeneous series-parallel power generation system model, analyzing the influence of the network construction type energy storage control parameter on the stability of the closed-loop series-parallel power generation system model. According to the invention, the modeling efficiency and the analysis accuracy can be improved.

The invention relates to the technical field of machine learning, can be applied to business system platforms of medical health, financial science and technology and the like, and discloses a risk data feature value analysis method, device and equipment and a medium, and the method comprises the steps: obtaining multi-dimensional data of a preset risk type, carrying out the basic feature extraction of the multi-dimensional data, and obtaining data basic features; performing basic risk analysis on the data basic features to obtain initial risk feature values; obtaining a residual feature value according to a preset residual, and combining the initial risk feature value with the residual feature value to generate a risk feature value; extracting key features of the multi-dimensional data by adopting a preset residual analysis model, and performing interval score assignment on the key features to obtain feature interval scores; and constructing an initial risk analysis model according to the feature interval scores and the key features. By integrating the residual error correction logic into the initial risk analysis model, the initial risk analysis model is optimized in a targeted manner to eliminate the deviation of the initial model, and finally the risk scoring accuracy is improved.

The present invention discloses a method for suppressing shaft oscillation of a doubly-fed pumped-storage unit operating at non-optimal speeds, comprising the following steps: 1. dividing the turbine operating region into an S region and a stable operating region; 2. constructing a linear model for the stable operating region; 3. constructing a refined model for the S region; 4. establishing a state equation for the doubly-fed pumped-storage unit; 5. setting an operating range and calculating a steady-state operating point; 6. linearizing the state equation near the steady-state operating point; 7. performing eigenvalue analysis to obtain a shaft oscillation damping ratio; 8. screening all dangerous operating points and formulating a risk assessment index based on the shaft oscillation damping ratio; 9. screening dominant control parameters; 10. establishing a dominant control parameter optimization model; and 11. introducing a risk assessment index to improve the NSGA-II algorithm and solve the optimization model. The present invention can effectively identify and suppress dangerous speed and flow ranges of doubly-fed pumped-storage units operating at non-optimal speeds, thereby improving the effectiveness of suppressing shaft oscillation.

The invention relates to the technical field of medical image analysis, and discloses a three-dimensional medical image data segmentation and reconstruction method and system, which effectively solves the contradiction between medical image denoising and edge preservation through an anisotropic diffusion technology guided by a structure tensor field, remarkably enhances the signal-to-noise ratio of a weak boundary, and improves the image denoising efficiency. By means of Hessian matrix eigenvalue analysis, blood vessel and skeleton interference is automatically eliminated, a focus topology core is accurately positioned, subjective errors of manual interaction are eliminated, a curvature reverse braking mechanism endows level set evolution with automatic truncation capacity at weak boundary adhesion positions, segmentation leakage is thoroughly prevented by means of physical surface tension, and the segmentation accuracy is improved. Finally, the mechanical balance of the grid vertexes under the sub-voxel precision is realized through the extreme value solution of the variational energy functional, a three-dimensional entity model with geometric high precision and topological fairness is generated, and the strict requirements of precise medical treatment on surgical planning and biomechanical analysis are met.

The embodiment of the present application discloses an image defect detection method, device and storage medium for detecting screen defects and improving the defect detection rate. The method of the present application includes: obtaining a first image and a second image and calculating the mean of the first image according to different light sources and light source positions of the camera; calculating a correction coefficient matrix according to the first image, the second image and the mean of the first image; determining the coordinate position of the dust and the coordinate position of the noise in the first image; adjusting the correction coefficient matrix according to the coordinate position of the dust and the coordinate position of the noise, and correcting the first image with the adjusted correction coefficient matrix; obtaining the fitting edge of the display screen in the first image, and obtaining a third image by calculating the homography matrix; performing a convolution operation with the third image according to the adaptive spatial filter kernel matrix; performing threshold control on the third image after the convolution operation to filter out the fourth image; performing eigenvalue analysis on the fourth image to obtain defect features.