201 results about "Linear discriminant analysis" patented technology

The invention discloses a TSFS sample generation method based on an improved generative adversarial network. The TSFS sample generation method comprises the following steps: step 1, acquiring TSFS data of a sample by using a three-dimensional fluorescence spectrometer; step 2, constructing a CA-WGAN-GP optimization model with a convolution attention mechanism, and performing data enhancement on an original TSFS sample; step 3, in a model training stage, storing generated data once after epoch training for a certain number of times, and enabling a generated sample coverage to have higher diversity by utilizing generator models of a plurality of training stages; 4, extracting enhanced sample features by using a pre-trained CNN model, projecting the sample features by using linear discriminant analysis LDA, and preliminarily screening samples according to a Mahalanobis distance and a set threshold value; and 5, performing secondary screening by using a K-means algorithm to obtain a final training sample set. According to the method, the time-consuming and labor-consuming problem of TSFS sample acquisition is effectively solved, the model over-fitting phenomenon is reduced, and the robustness of unknown data is improved.

The invention discloses a tire force estimation method based on in-tire sensing information and a Bayesian neural network. An acceleration type intelligent tire finite element model is built based on a finite element and coordinate system conversion theory; the method comprises the following steps of: determining an optimal mounting position of an acceleration sensor aiming at the three-dimensional force of the tire by adopting a Fourier amplitude sensitivity test (FAST) method, and determining the optimal mounting position of the acceleration sensor aiming at the three-dimensional force of the tire; the method comprises the following steps of: acquiring acceleration signal-tire force data sets under different test working conditions, extracting acceleration signals of a grounding imprinting area, and screening out optimal input characteristics by adopting a linear discriminant analysis (LDA) method; processing the acceleration signal-tire force data set after optimal feature screening by adopting a linear normalization theory; according to the characteristics of training data and test data, a tire force estimation algorithm based on a Bayesian neural network is established, and finally a predicted value and a predicted variance of tire force are output. The method is high in prediction precision, good in stability and high in generalization performance, and a more reliable technical scheme is provided for tire force estimation.

The invention discloses a Raman spectrum training and prediction algorithm SPP-1D of a multichannel one-dimensional convolutional neural network based on spatial pyramid pooling. According to the method, local features of spectral data are extracted through 1D-CNN, and features are pooled on multiple scales by using spatial pyramid pooling (SPP), so that global feature representation is generated. Compared with the traditional principal component analysis (PCA) and linear discriminant analysis (LDA) methods, the method can effectively capture the complex nonlinear relationship of the spectral data and enhance the anti-noise performance. The SPP-1D supports multi-channel input, a plurality of characteristic channels of the Raman spectrum can be processed at the same time, correlation and complementarity between the channels are fused, and more comprehensive characteristic description is achieved. The algorithm is suitable for application scenes such as single substance identification, complex mixture component analysis, spectral imaging and the like, and the precision, robustness and generalization ability of Raman spectrum data processing are remarkably improved. According to the invention, deep learning and a signal processing technology are creatively combined, and an efficient spectral analysis tool is provided for the fields of material science, chemical analysis and biomedicine.

Disclosed in the present invention are a cumulative pressure stress effect evaluation system and method. The system comprises: an animal modeling module, configured to construct various pressure stress level models by means of applying different durations of pressure to a target model animal; a spontaneous behavior capture module, configured to capture spontaneous behaviors of the target model animal corresponding to different pressure stress levels to obtain corresponding action recognition data, and construct a first data set according to a corresponding relationship between the action recognition data and the pressure stress levels; a behavior phenotype analysis module, configured to use linear discriminant analysis to perform dimension reduction on the first data set, so as to obtain a second data set in a discrimination space; and a pressure degree evaluation module, configured to use the second data set to train a machine learning model, so as to obtain an optimized cumulative pressure stress effect evaluation model. The present invention can be used for cumulative pressure stress effect evaluation, is suitable for precise medicine, and can achieve behavior phenotype difference recognition induced by the same pressure stress source and different pressure degrees.

The invention discloses a voiceprint characterization method and system for resisting voice conversion and voice synthesis based on deep learning. Relates to the field of speech recognition and biological feature security. Comprising the steps of 1, acquiring an input voice sample signal in an automatic speaker verification system, and extracting an FBANK feature of the voice sample signal; 2, performing depth feature extraction on the FBANK features by using a convolutional neural network (CNN) to obtain a depth feature vector; 3, processing the depth feature vector by using a recurrent neural network (RNN), and generating a forged identity vector of the voice; and 4, classifying the counterfeit identity vectors by using a linear discriminant analysis (LDA) module, and outputting a judgment result. According to the invention, the accuracy of counterfeit detection can be effectively improved in both clean and noise environments.

The invention provides a method for predicting a compound target based on a compound perturbation transcriptome and a protein sequence and application, and the method creatively combines the characteristics of the compound perturbation transcriptome and the characteristics of the protein target sequence as input characteristics and combines a deep learning model to realize prediction of the compound target. Specifically, perturbation transcriptome and active and inactive target spot information data of a compound are collected, transcriptome features are extracted through principal component analysis and linear discriminant analysis, meanwhile, a protein language model is used for representing target spot protein features, the transcriptome and a target spot of the compound are converted into learnable embedding, and the target spot information of the compound is obtained. And the classification model is used for training and learning potential association of the two, and is finally used for compound target prediction. According to the method, an external test set is used for proving high accuracy and practical applicability of the model. The method has the advantages of being low in cost, high in efficiency and wide in range, the compound target spot discovery efficiency can be remarkably improved, and the research and development process of drugs can be accelerated.

A photovoltaic power generation associated physical quantity mining method based on historical time series data analysis, and a related apparatus. The method comprises: acquiring historical time series data of multiple dimensions during photovoltaic power generation; calculating degrees of mutual information between the historical time series data of the multiple dimensions of photovoltaic power and the photovoltaic power; selecting the historical time series data of which the degree of mutual information satisfies a set requirement to serve as data related to photovoltaic data; and using a linear discriminant analysis (LDA) method to perform feature dimension reduction on the selected historical time series data to obtain data having undergone dimension reduction processing. In the present invention, by using a maximal information coefficient (MIC) feature selection method, data most related to photovoltaic power generation is selected from original feature variables, and then by using an LDA-based feature dimension reduction method, high-dimensional data is mapped to a lower-dimensional space. By means of the MIC feature selection method and the LDA-based feature dimension reduction method, the accuracy of photovoltaic power generation prediction is effectively improved.

The invention relates to the technical field of electromyographic signal analysis and processing, in particular to a low-delay electromyographic signal processing method based on multi-channel fusion, which comprises the following steps of: acquiring full-band electromyographic signals by using a surface electrode array, and screening channels by adopting a mode of combining a Pearson's correlation coefficient and mutual information; the method comprises the following steps: setting dual thresholds, combining signal strength and electrode impedance detection, then performing 20-500Hz Butterworth band-pass filtering, Z-Score standardization and normalization processing on screened signals to improve the signal quality, optimizing sliding window parameters through the Parseval theorem and a dynamic time warping algorithm, balancing the real-time performance and the feature extraction effect, and improving the accuracy and the accuracy of feature extraction. And finally, time domain, frequency domain and time-frequency domain multidimensional features are extracted in each time window, and feature vectors are generated through principal component analysis and linear discriminant analysis fusion and dimension reduction. The invention aims to solve the problems of poor real-time performance and low feature identification degree of the existing electromyographic signal processing method.

The invention relates to the technical field of intracranial pressure monitoring, in particular to a craniocerebral injury patient secondary brain injury early warning method based on multi-modal fusion, which comprises the following steps: acquiring intracranial pressure, cerebral blood flow, electroencephalogram signals and brain tissue oxygen partial pressure signals, carrying out denoising and normalization processing, constructing a multi-modal normalized time sequence data set, and carrying out multi-modal fusion on the basis of the multi-modal normalized time sequence data set; and inputting a linear discriminant analysis model to generate a time sequence difference vector, constructing a difference driving map based on a Bayesian dynamic network model, identifying abrupt change nodes, outputting a risk early warning section set, and carrying out trend stability analysis and reconstruction optimization. According to the method, the abnormal state recognition sensitivity is improved through normalized processing of multi-modal signals, enhancement of signal fusion capability, quantitative capture of direction change, reconstruction of spatial relationship between signals, modeling of dynamic transition probability, recognition of abrupt change nodes, combination of time difference vector change, evaluation of signal linkage trend, positioning of risk sections and extraction of parameter features; and the early warning and intervention precision is improved.

The invention discloses a battery internal resistance state monitoring method based on a voltage pulse method, and relates to the technical field of battery detection, and the method comprises the steps: carrying out the current excitation of a battery pack in a frequency sweeping mode; sending the collected battery data to an upper computer to wait for data processing; the upper computer performs data error verification and data preprocessing on the received data; monitoring the characteristic parameters of the internal resistance of each single battery in real time; drawing and displaying an electrochemical impedance spectrogram of the internal resistance of each single battery; performing curve fitting on the electrochemical impedance spectrogram of the internal resistance of the single battery; calculating model parameters of the equivalent circuit model of the battery pack; and monitoring and diagnosing the health state of the battery pack by using a linear discriminant analysis method. Important characteristic parameters of the cell are monitored on line, so that the health state of the fuel cell is monitored and diagnosed, and the service life is prolonged.

The invention discloses a lightweight optical fiber vibration intrusion event identification method and system for perimeter security and protection, and belongs to the technical field of optical fiber sensing technology and mode identification, and the method comprises the steps: collecting an original vibration signal of a perimeter monitoring region through a distributed optical fiber vibration sensing system; performing wavelet threshold de-noising preprocessing on the original vibration signal to obtain a de-noised signal; performing feature extraction on the denoised signal, and constructing a high-dimensional feature vector; performing dimension reduction processing on the high-dimensional feature vector by using a linear discriminant analysis method to obtain a low-dimensional classification feature vector; and inputting the low-dimensional classification feature vector into a pre-trained lightweight convolutional neural network model for classification and identification, and outputting a corresponding intrusion event category. According to the method, through cooperation of front-end LDA dimension reduction and a rear-end lightweight network, the model parameter quantity and calculation overhead are greatly reduced while high recognition precision is guaranteed, efficient real-time deployment on edge equipment is achieved, and the method is suitable for intrusion detection in perimeter security and protection.

The invention provides a signal integrity analysis method and device and electronic equipment, and the method comprises the steps: collecting signal data, and converting the signal data into corresponding quantum state data; key features in the quantum state data are extracted through a linear discriminant analysis algorithm; performing iterative training on a quantum machine learning model by using the key feature value to obtain a quantum channel model; the quantum channel model is used for representing the physical limitation and evolution law of the signal in the process from sending to receiving; and inputting a signal to be analyzed into the quantum channel model to obtain a signal integrity analysis result. According to the invention, accurate quantum state calculation analysis is carried out on a large amount of high-speed and high-frequency signal data through a quantum calculation technology, and the influence of noise, signal attenuation, coherence weakening and other factors on signal transmission is effectively solved.

The invention discloses a method for identifying freshness of different parts of pork and application. The method comprises the following steps: firstly, constructing a PCA-LDA model for identifying fresh pork and frozen-thawed pork, wherein the PCA-LDA model comprises the following steps: respectively obtaining mass spectrum fingerprint data of the fresh pork and the frozen-thawed pork; and based on a principal component analysis method and a linear discriminant analysis method, constructing a PCA-LDA model by using the mass spectrum fingerprint data of the fresh pork and the mass spectrum fingerprint data of the frozen and thawed pork. Secondly, a pork sample to be detected is identified on the basis of the PCA-LDA model, the result shows that the identification accuracy reaches 99% or above, the identification efficiency is high, the single-time determination identification speed is 6 seconds per sample, fresh pork and frozen-thawed pork at different parts can be rapidly and accurately distinguished, and the method has an extremely good application prospect in pork product identification.

The invention provides an improved incremental linear discriminant analysis feature extraction method. The method comprises the following steps: firstly, calculating an initial OWA global mean value, an inter-class divergence matrix and an intra-class divergence matrix of an image training sample to generate an initial model; in the subsequent step, for each newly added sample, the global mean value variable is dynamically updated, and the inter-class divergence matrix and the intra-class divergence matrix are adjusted based on the updated global mean value. And then, a Fisher discriminant criterion is used to calculate a new projection matrix, and a test image is projected to the new projection matrix, so that dimension reduction and feature extraction of data are realized. The core innovation of the method is that an OWA operator is adopted, and a brand new global sample mean value is generated according to different contributions of features among samples. In addition, an improved inter-class divergence matrix is provided, so that the difference between different classes is remarkably enhanced. By improving the classification distinction degree, the method can more comprehensively reserve data information, and provides feature representation with higher quality for subsequent classification tasks.

The invention discloses an epilepsy detection method combining uncorrelated multi-linear discriminant analysis and deep metric learning, and belongs to the technical field of epilepsy detection.The epilepsy detection method comprises the steps that S1, original electroencephalogram signals are collected and preprocessed; s2, forming a third-order electroencephalogram tensor by using the electroencephalogram signals subjected to the improved S-transform time-frequency decomposition processing, and extracting electroencephalogram features through uncorrelated multi-linear discriminant analysis; s3, performing epilepsy disease classification by using a triple network metric learning classifier; s4, a Softmax function is used for outputting probability distribution of the test data, and a detection label is obtained; s5, a moving average filter is used for eliminating burst noisy points and isolated abnormal values, and a classification result and post-processing are output; according to the epilepsy detection method combining uncorrelated multi-linear discriminant analysis and deep metric learning provided by the invention, the workload of doctors can be effectively reduced, the efficiency and accuracy of epilepsy detection are improved, and the epilepsy detection method has a remarkable promotion effect on clinical diagnosis and treatment.

The invention discloses a power system network security situation awareness method based on an RBF neural network, and the method comprises the following steps: building a feature index system, and carrying out the quantification of a security situation value; sample data acquisition and processing: acquiring network communication message data in the electric power industrial control system, and analyzing the message data; carrying out LDA algorithm data preprocessing, and projecting a high-dimensional feature sample to an optimal discriminant vector space by using an optimized linear discriminant analysis algorithm; constructing an RBF neural network model, taking the data matrix subjected to LDA preprocessing as data input of the model, and taking a category or a safety index corresponding to the sample matrix as network output; training an RBF neural network model; testing and verifying the model, inputting test data, verifying whether a test result accords with an expected result, if so, ending training, and if not, continuing training; and model reasoning and result output: taking to-be-processed data as input, and obtaining a corresponding situation awareness result after the to-be-processed data passes through the projection matrix and the RBF neural network model. According to the scheme, a reference basis can be provided for power information network security monitoring.

The invention discloses a pork safety traceability quality nondestructive testing method and system for livestock breeding. The method comprises the following steps: constructing a stable detection environment by utilizing a constant temperature and humidity device, a dynamic environment monitoring unit and an optical shielding material; a Fourier transform infrared spectrometer is adopted to collect spectral data, and a non-contact texture detection device is combined to measure physical parameters; denoising, standardization processing and feature fusion are carried out on the collected data, and key variables are extracted through principal component analysis; grouping the samples by using a K-means algorithm, and verifying sample classification by using linear discriminant analysis; quantitative prediction and classification of pork quality are realized based on partial least squares regression and a support vector machine model; the model is applied to an actual scene, real-time processing and visual display are achieved through online data collection, and a data sample library is continuously expanded. The method has the advantages of high detection speed, high precision, wide application range and the like, and can be widely applied to the fields of meat quality detection and food safety.

The invention belongs to the technical field of analytical chemistry, and relates to a method for identifying and detecting endogenous mercaptan of a high-activity oxidized nano-enzyme based on machine learning, which comprises the following steps: firstly, preparing a Mnx (DTPMP) nano-enzyme, then constructing a sensor array of one nano-enzyme, four reaction times, three targets and six parallel samples, and detecting the endogenous mercaptan in the Mnx (DTPMP) nano-enzyme. Constructing a sensor array of one nano enzyme, four reaction times, 7 concentrations and 6 parallel samples for each endogenous mercaptan; and finally, analyzing the data through hierarchical clustering analysis, linear discriminant analysis and a support vector machine to obtain an LDA score chart, an HCA map, an SVM confusion matrix and a concentration prediction map of the endogenous mercaptan. After a sample to be detected is subjected to a chromogenic reaction, the absorbance is measured, and the type and concentration of the endogenous mercaptan of the sample to be detected can be distinguished according to the obtained spectrum. The kit has good anti-interference performance, normal cells and cancer cells can be accurately identified according to the content of glutathione in the cells, and the severity of diseases can be judged according to the content of homocysteine in serum.

The invention provides a data mechanism fusion prediction system and method for loss of an unbalanced transformer of a distribution network, and relates to the technical field of operation and loss reduction of the distribution network. Firstly, electric quantity data such as voltage, current, active power and reactive power of a power distribution network are collected to form a matrix to form an original data set, correlation analysis is carried out on data in the original data set by adopting a mutual information method, and dimensionality reduction is carried out by adopting a linear discriminant analysis method. Taking the output after dimension reduction as the input of transformer loss intelligent prediction, calculating the zero-sequence impedance corresponding to the original data set to obtain the total loss of the transformer under the unbalanced load of the distribution network, and training the transformer loss intelligent prediction model by using the output after dimension reduction and the corresponding total loss of the transformer to obtain the transformer loss intelligent prediction model. And finally, the real-time intelligent prediction of the transformer loss under the unbalanced load of the distribution network can be realized by using the real-time monitoring data of the distribution network.

The invention relates to a battery thermal runaway state monitoring and evaluating method based on linear discriminant analysis. The method comprises the following steps: acquiring running state data in a to-be-monitored battery; acquiring historical operation state data of the to-be-monitored battery in the first time period, and respectively determining a characteristic gas concentration change rate, a pressure change rate, a temperature change rate and a volume change rate in the to-be-monitored battery in the first time period based on the historical operation state data; determining a thermal runaway risk index value of the to-be-monitored battery by using the running state data; inputting the operation state data, the concentration change rate, the pressure change rate and the temperature change rate into a first prediction model to obtain a volume prediction value and a volume change rate prediction value in a second time period; determining a first weight value based on the first mapping relation; and determining the product of the thermal runaway risk index value and the first weight value to obtain a comprehensive risk index value, and outputting corresponding early warning information based on the comprehensive risk index value. According to the scheme, the timeliness of monitoring the thermal runaway state of the battery is improved.

The present invention relates to a technology for rest state-based user authentication by using an electromyogram signal and an inertia measurement device signal and provides a rest state-based user authentication device using an electromyogram signal and an inertia measurement device signal, which extracts a user's unique electromyogram signal and inertia measurement device signal during a rest state in which a specific operation is not performed, and inputs the extracted signal into a linear discriminant analysis and ensemble classification device to identify and authenticate the user, and a method thereof.

The present invention provides a method for classification and identification of gastropods based on shell element analysis, including: S1. Using a laser ablation inductively coupled plasma mass spectrometer to measure the contents of 55 trace elements in gastropod shells; S2. Screening out 24 trace elements with significant differences based on one-way variance analysis; S3. Performing stepwise linear discriminant analysis based on the 24 trace elements, and classifying them using a random forest classification method. In the S‑LDA model, the contents of the 24 trace elements are gradually input as variables, and a discriminant function is established based on the Wilkes λ value standard, achieving a correct discrimination rate of 92.7% for species identification, and a random forest model based on the contents of the 24 trace elements achieves a correct discrimination rate of 97.0%, which is applicable to the classification and identification of gastropods.

The present invention discloses a method and system for on-demand intelligence distribution, comprising the following steps: (1) receiving superior telegraphic information to obtain action tasks and intelligence information; (2) extracting features: commander features, action task features, and intelligence features; (3) error processing and model optimization; and (4) outputting intelligence in order of interest based on the optimized model. The present invention preprocesses task and intelligence text using an improved linear discriminant analysis method, encodes task and intelligence data using a sparse edge denoising autoencoder, and uses a matrix decomposition algorithm to remove blank portions of the matrix for dimensionality reduction, thereby effectively improving the accuracy of the training model, while also increasing the precision of on-demand intelligence distribution and enhancing the command efficiency of commanders.

The invention relates to the technical field of speech recognition, in particular to a long speech dialect recognition method and device based on a deep learning algorithm, and aims to realize dialect recognition and conversion of long speech data through an adaptive segmentation strategy and a pre-trained dialect conversion model. The method comprises the following steps: adaptively segmenting input long voice data, calculating continuity parameters and zero cross points of voice signals through Fourier transform, and determining the segment length; performing high-pass filtering, wavelet transformation and translation denoising processing on the segmented data, and extracting feature vectors of the typical dialect voice data; and calculating a dialect label probability through an encoder and an attention mechanism, selecting a dialect type, and performing dialect conversion processing according to a pre-trained dialect conversion model. The device adopts the technologies of feature extraction, feature fusion, probability linear discriminant analysis and the like to output a dialect conversion result. The method can effectively improve the dialect recognition precision and conversion effect of the long voice data.

The present disclosure provides an array measuring method and interpretation device for ultrasonic detection of middle ear effusion, including an ultrasonic probe, an ultrasonic receiver, an analog-to-digital converter, and an analysis unit. The surface of the mastoid is divided into a plurality of measurement areas, and when ultrasonic is used for non-invasive detection of middle ear effusion, linear discriminant analysis is used for pre-training to find the best detection position and weighting parameters thereof to obtain the accurate evaluation value.

The invention relates to the technical field of urban rainfall pattern recognition, and aims to solve the problems that an existing method depends on an empirical formula, rainfall classification is unstable, and interpretability is insufficient. The method comprises the following steps: acquiring multi-site long-time actually measured rainfall data, and identifying an independent rainfall event based on a rainfall interval threshold value; multi-dimensional features such as rainfall duration and peak rainfall intensity are extracted and standardized; performing hierarchical clustering by adopting a Ward minimum variance method, and determining an optimal clustering number by combining a Clinski-Harabasz index and the like to obtain a rainfall mode; training a linear discriminant analysis (LDA) model by taking a clustering result as a label, and verifying and quickly classifying a new rainfall event; the result is used for flood control scheduling, flood risk assessment and the like. According to the method, rationality is improved based on measured data, classification stability is improved by fusing clustering and discriminant analysis, expandability is high, and intelligent disaster prevention is supported.

The present disclosure provides an array measuring method and interpretation device for ultrasonic detection of middle ear effusion, including an ultrasonic probe, an ultrasonic receiver, an analog-to-digital converter, and an analysis unit. The surface of the mastoid is divided into a plurality of measurement areas, and when ultrasonic is used for non-invasive detection of middle ear effusion, linear discriminant analysis is used for pre-training to find the best detection position and weighting parameters thereof to obtain the accurate evaluation value.

The invention relates to the field of medical detection, in particular to a microbial marker composition for predicting prognosis of hepatocellular carcinoma and application of the microbial marker composition. Advanced bioinformatics analysis, including linear discriminant analysis effect size and Cox proportional risk regression, is utilized to determine the correlation between specific microorganisms and clinical results of HCC patients, and further explore the relationship between the spectrum of microorganisms in tumor tissues and gene expression so as to clarify the potential approach that the microorganisms may affect HCC progress; the microbe prognosis characteristic model is developed by utilizing the analysis so as to improve the prediction on the survival of the patient, the influence of the related microbiota in the HCC tumor on the immune condition is also explored, and information can be provided for formulating an immunotherapy strategy. According to the invention, by comprehensively analyzing the attributes and prognosis of the microorganisms in the HCC, the understanding of the tumor microenvironment of the HCC is enhanced, so that the formulation of a more accurate and personalized cancer management strategy is promoted.

The application discloses a tire force estimation method based on in-utero sensing information and a Bayesian neural network, builds an acceleration type intelligent tire finite element model based on finite element and coordinate system conversion theory; determines the optimal installation position of the acceleration sensor for the three-way tire force by using a Fourier amplitude sensitivity test (FAST); collects acceleration signal-tire force data sets under different test conditions, extracts the acceleration signal in the footprint area, and selects the optimal input features by using a linear discriminant analysis (LDA); processes the acceleration signal-tire force data set after optimal feature selection by using linear normalization theory; builds a tire force estimation algorithm based on a Bayesian neural network according to the characteristics of the training data and the test data, and finally outputs the predicted value and the prediction variance of the tire force. The application has high prediction accuracy, good stability and strong generalization performance, and provides a more reliable technical solution for tire force estimation.

The present application relates to the field of artificial intelligence, and relates to a model trust region-based copyright protection method for smart grid deep learning models. The method comprises: acquiring a smart grid deep model set; for each model in the smart grid deep model set, searching an exclusive dataset for feature data samples having a model prediction value approximate to a model discrimination boundary, so as to obtain a trust region feature point set; performing dimensionality reduction on gradient vectors of the trust region feature point set on the model discrimination boundary according to a linear discriminant analysis method, so as to obtain perturbation vectors of the trust region feature point set; on the basis of predicted label changes of each model before and after the trust region feature point set is combined with the perturbation vectors, generating a model feature identifier set corresponding to the smart grid deep model set; and, on the basis of the model feature identifier set, training a copyright detection model to be trained, so as to obtain a pre-trained copyright detection model.