51 results about "Cross correlation matrix" patented technology

The invention relates to the technical field of zero-carbon equipment fault early warning, and discloses a zero-carbon equipment fault early warning method and system based on ESG scheduling, and the method comprises the steps: collecting multi-source ESG parameter data, such as carbon footprint intensity, clean energy consumption and energy storage system charging and discharging efficiency, extracting equipment state features through employing a graph convolution network in combination with a gating circulation unit, and carrying out the early warning of the zero-carbon equipment fault; and calculating an ESG correlation degree and an inter-time-segment cross-correlation matrix, and generating a fault early warning weight through a ReLU activation function after principal component analysis dimension reduction. And dynamically adjusting the reference threshold based on the weight, correcting the deviation by means of an extended Kalman filter, and generating an early warning parameter to trigger an early warning process. Meanwhile, a multi-dimensional equipment state space is constructed to judge a fault state, a joint optimization model of ESG parameters and an early warning threshold value is solved through a non-dominated sorting genetic algorithm, collaborative optimization of ESG benefits and fault early warning is achieved, and early warning accuracy and dynamic adaptability are improved.

The invention discloses a robust radio frequency fingerprint identification method based on Barlow Twins domain adaptation. Aiming at the problems of training and deployment environment distribution offset and cross-domain identification performance reduction caused by wireless channel multipath and time-varying characteristics in the prior art, the invention provides a domain adaptation framework combined with feature decoupling. The method comprises the following steps: firstly, converting a received time domain signal into a short-time Fourier transform spectrogram to reserve a time-frequency structure; then, a double-end convolutional encoder is constructed, bottom layer features are extracted through a shared backbone network, and device fingerprints and channel features are obtained in an identity branch and a channel branch respectively; in order to realize thorough decoupling of the two types of features, Barlow Twins-based independence regularization is introduced, and the identity features and the channel features are statistically orthogonal by minimizing a cross-correlation matrix of the identity features and the channel features, so that purer fingerprint features are obtained, channel interference is effectively eliminated, and generalization and recognition precision of the model under an unknown channel are remarkably improved.

The embodiment of the invention relates to the technical field of computer security authentication, in particular to a security authentication method and system based on a secure computer. The method comprises the following steps: carrying out sensing array deployment and data acquisition on a secure computer to obtain calibrated time-space synchronization data; performing temperature and current fusion processing on the calibrated time-space synchronization data to obtain a thermoelectric coupling characteristic spectrum; carrying out point location time sequence correlation analysis on the thermoelectric coupling characteristic spectrum to obtain a point location cross-correlation matrix; performing nonlinear feature extraction on the point location cross-correlation matrix to obtain a nonlinear dynamic feature set; performing dynamic behavior analysis on the nonlinear dynamic feature set to obtain a security calculation behavior dynamic phase spectrum; and performing phase difference calculation on the dynamic phase spectrum of the security calculation behavior to obtain a phase difference vector. According to the method, the early detection capability of threats which are difficult to reproduce and are in a novel hardware level is improved through real-time monitoring and anomaly recognition of computer microcosmic physical characteristics.

The invention discloses a mining area load spectrum anomaly detection method based on standard variable analysis, and belongs to the technical field of hydraulic pump anomaly detection.The mining area load spectrum anomaly detection method comprises the steps that S1, standardized data are obtained through data preprocessing, and a typical working condition data set of a hydraulic pump is constructed in combination with working conditions of an excavator; s2, constructing a historical vector and a future vector, and constructing a historical observation matrix and a future observation matrix according to the historical vector and the future vector; s3, constructing a Hankel matrix according to the autocorrelation matrix and the cross-correlation matrix, decomposing the Hankel matrix and determining a model order; s4, mapping the original data to a standard variable space and a residual space, and respectively evaluating the total variable quantity of the standard variable in the state space and the sum of squares of change errors in the residual space; and S5, determining an evaluation threshold value, and if the evaluation threshold value exceeds the control line, judging that the hydraulic pump operates abnormally. The method is based on standard variable analysis, adopts the pressure pulsation data of the hydraulic pump to perform anomaly detection, is sensitive to the internal running state of the pump, is not easily influenced by the external environment, and can perform early warning on faults of the hydraulic pump.

The invention discloses a scene-adaptive multipath time delay alignment method and device for a wireless communication system, and the method comprises the steps: employing an SISO channel detection system to actually measure channel data, employing rectangular or RRC shaping filtering to generate a CAZAC oversampling sequence as a template function, calculating a cross-correlation matrix of an IQ signal and the template function, and enabling the CAZAC oversampling sequence to serve as the template function; a noise threshold is calculated based on the median power and the false alarm rate, dynamic threshold filtering is executed, a cross-correlation matrix after denoising is generated, and preliminary estimation of the time domain impulse response is completed; extracting multipath time delay and complex gain from the cross-correlation matrix by using an SAGE algorithm; drawing a power delay spectrum PDP for the channel impulse response; and performing alignment operation on the PDP in combination with the geographical location information of the transceiver during actual measurement. According to the method, the defect of traditional alignment in a severe scene is overcome, the statistical modeling accuracy is improved, the multipath time delay can be effectively aligned through experimental verification, and reliable support is provided for channel modeling.

The invention discloses a near-field rotation invariant parameter estimation method based on linear fitting, and the method comprises the steps: calculating the cross-correlation between the receiving data of an array element in a first array and the receiving data of an array element in a second array on a signal source three-dimensional space precise propagation model based on a symmetric cross array, and obtaining a cross-correlation matrix; obtaining a virtual receiving data matrix by vectorizing the cross-correlation matrix; performing eigenvalue decomposition on the covariance matrix of the virtual receiving data matrix to obtain a signal subspace matrix; estimating parameters of a narrowband signal source by using an ESPRIT algorithm, specifically, determining a virtual uniform area array according to a covariance matrix, and partitioning a signal subspace matrix into blocks; the rotation invariance is recovered through the subspace relation of the adjacent block matrixes; obtaining the phase difference between the adjacent sub-arrays through the rotation invariance matrix, building a linear observation model, and obtaining the parameter estimation value of the narrowband signal source; the method has the advantages of low calculation complexity and high universality.

The invention provides a model acceleration method based on a face recognition model application. The method comprises the following steps: sampling a face picture; two different data enhancement modes are adopted for the face image, and a first distortion image and a second distortion image are obtained respectively; performing feature extraction on the first distortion graph through an original model, and outputting a first embedded vector; performing feature extraction on the second distortion graph through a simplified model, and outputting a second embedded vector; measuring a cross-correlation matrix of the first embedded vector and the second embedded vector by adopting a contrastive learning thought; and constraining the distance between the cross-correlation matrix and the unit matrix through the relative loss function, repeating until the output of the relative loss function meets the preset stability condition, and replacing the original model with the simplified model. According to the method, knowledge distillation from a heavy-weight network to a lightweight-weight network is completed, and the function of replacing a heavy-weight original model by a lightweight-weight simplified model is realized, so that hardware facilities of embedded equipment are adapted, and the real-time detection efficiency of face recognition is improved under limited computing resources.

The invention discloses an image semantic segmentation method and system based on cross-modal hierarchical knowledge distillation, and the method comprises the steps: inputting an image to a multi-modal teacher model and a multi-modal student model, obtaining teacher features and student features of each stage, and outputting a prediction map; the teacher features of all the stages are fused, common modal features are enhanced, fusion features corresponding to all the stages are obtained, and the fusion features and the student features are divided into high-level features and low-level features according to different stages; the low-level teacher fusion features and the low-level student features are converted into frequency domain graphs respectively, and high-frequency parts are taken out to transmit detail knowledge; respectively calculating a cross-correlation matrix and an autocorrelation matrix by using the high-level teacher fusion features and the high-level student features, and transmitting structural knowledge; and aligning the teacher output prediction map and the student output prediction map, and transmitting response knowledge. According to the invention, the multi-modal features are fused in a mode of modal enhancement, the modal gap between the teacher fusion features and the student features is reduced, and the single-modal student model segmentation precision is improved.

The invention belongs to the technical field of geological exploration, and particularly discloses a symmetrical excitation anti-explosion geological exploration system and method, and the method comprises the steps: receiving array multi-channel signals at two sides, and carrying out the time alignment correction of all channel signals, so as to eliminate the time delay; a multi-channel cross-correlation matrix between the left array and the right array is constructed based on the aligned signals, and matrix elements are obtained through integral cross-correlation operation of array channel signals on the two sides; performing low-rank decomposition on the cross-correlation matrix, and extracting the first k main components as public noise modes; carrying out inversion calculation on the contribution of each noise mode in each acquisition channel, and reconstructing a channel noise component; and subtracting the reconstructed noise component from the original signal, and outputting a purified geological reflection signal. The method does not need special hardware, can accurately suppress the homologous low-frequency noise of the tunnel, improves the signal-to-noise ratio of the signal and the geological imaging resolution, and is suitable for geological detection of closed spaces such as the tunnel.

An isometric self-supervised line-of-sight estimation method and device, the method establishes a camera coordinate system and a world coordinate system, calibrates the camera, and estimates the extrinsic parameters of the world coordinate system; adjust the camera posture, rotate and scale the camera coordinate system; a virtual camera coordinate system is established, and after the head of the human body moves in three directions for two-dimensional projection, a homography matrix corresponding to the image transformation is obtained; the transformed picture is sent into the encoder for feature extraction, and after the features are extracted, they are projected to the vector space through the multilayer perceptron network, and the Barlow Twins loss is calculated through the hidden space feature mutual correlation matrix; the obtained encoder is connected with the multilayer perceptron network, and in the line-of-sight gaze data environment, the 3D line-of-sight direction is fitted by using transfer learning. The application can efficiently utilize unlabeled data for 3D line-of-sight estimation; it is invariant to appearance transformation and isometric to geometric transformation.

The application belongs to the technical field of transformer partial discharge identification, and discloses a transformer partial discharge intelligent identification method based on deep learning, which constructs an electroacoustic thermal three-mode acquisition architecture, captures 1ns level discharge pulses through a high-frequency current sensor, avoids sound field superposition interference by reasonably arranging an ultrasonic sensor on the oil tank wall, focuses on the easy discharge area to collect thermal signals through an infrared thermal imager, and synchronously obtains working condition parameters such as load rate and oil temperature; subsequently, the effective signal segments are retained through a multi-modal signal cross-correlation matrix, and the interference signals with large deviations are removed, then different networks are used to extract electroacoustic thermal characteristics, the coupling characteristics are calculated by combining the physical constraint layer weight and the cross-field interaction layer, and the key characteristics are retained; the feature overlap of air gap and surface discharge is effectively reduced, the influence of high-frequency interference and signal distortion is reduced, the partial discharge identification precision is improved, the samples are expanded and the fault and normal sample ratio is reasonable through time stretching and additive noise processing.

The invention provides a large-scale MIMO-OFDM (Multiple Input Multiple Output-Orthogonal Frequency Division Multiplexing) channel acquisition method based on multiple groups of adjustable phase shift pilot frequencies. According to the multi-group adjustable phase shift pilot frequency method, users are divided into a plurality of groups, each group uses the same basic pilot frequency sequence to generate a plurality of adjustable phase shift pilot frequencies, and different groups use different basic pilot frequency matrixes; wherein the autocorrelation matrix of the basic pilot matrix is a unit matrix, the sequence of diagonal elements of the cross-correlation matrix of different basic pilot matrixes after FFT transformation is sparse, and each non-zero complex element has the same argument. In the channel acquisition method, each user terminal sends a plurality of groups of known and scheduled adjustable phase shift pilot signals to the base station, and the base station performs preprocessing according to the received signals so as to complete channel estimation. The method provided by the invention can greatly improve the spectrum efficiency and the channel information acquisition precision of a large-scale MIMO-OFDM system, and has excellent performance especially in a communication scene with a large number of users and strong mobility.

The present invention relates to the technical field of zero-carbon equipment fault warning, and discloses a zero-carbon equipment fault warning method and system based on ESG scheduling. The method collects multi-source ESG parameter data such as carbon footprint intensity, clean energy consumption, and energy storage system charging and discharging efficiency, and uses a graph convolutional network combined with a gated cyclic unit to extract equipment state characteristics, calculates ESG correlation and time-segment cross-correlation matrix, and generates fault warning weights through a ReLU activation function after dimensionality reduction through principal component analysis. The benchmark threshold is dynamically adjusted based on the weight and the deviation is corrected with the help of an extended Kalman filter to generate warning parameters to trigger the warning process. At the same time, a multi-dimensional equipment state space is constructed to determine the fault state, and a joint optimization model of ESG parameters and warning thresholds is solved by a non-dominated sorting genetic algorithm to achieve coordinated optimization of ESG benefits and fault warnings, thereby improving warning accuracy and dynamic adaptability.

The invention provides a channel estimation method for a minimum mean square error, which comprises the following steps of: performing Fourier transform on a received baseband signal to obtain OFDM (Orthogonal Frequency Division Multiplexing) frequency domain data, and obtaining frequency domain data of all symbols on a receiving side through solution resource mapping; the method comprises the following steps: processing by utilizing frequency domain data of a pilot symbol at a receiving side and a priori known pilot symbol sequence to obtain channel responses of SNR (Signal to Noise Ratio) estimation and LS (Least Square) channel estimation; processing according to the channel response of the LS channel estimation to obtain root-mean-square delay expansion; calculating a data-pilot frequency cross-correlation matrix and a pilot frequency self-correlation matrix according to root mean square time delay expansion, and calculating to obtain an MMSE filtering channel estimation coefficient according to the data-pilot frequency cross-correlation matrix and the pilot frequency self-correlation matrix; mMSE filtering is carried out according to the MMSE filtering channel estimation coefficient, and finally a final channel estimation value is obtained. According to the method, channel response is accurately estimated while operation is simplified, so that reliable communication and low operation delay are ensured.

The present invention relates to a three-dimensional parameter rough and fine estimation method for near-field sources based on second-order cross-correlation, including: establishing a near-field signal model based on a cross array; constructing virtual received data r(m,n,τ) using the cross-correlation function of the signals received by the m-th sensor and the n-th sensor, and converting r(m,n,τ) into a time-delay cross-correlation matrix R(τ) containing information collected by all sensors in the cross array; obtaining estimated values #imgabs0# and #imgabs1# of the array manifold matrices A and B according to R(τ); extracting amplitude attenuation from #imgabs2# and #imgabs3#; constructing a coefficient matrix for the array x; directly obtaining estimated values #imgabs4# and #imgabs5# of the k-th incident signal according to the coefficient matrix of the array x; constructing a coefficient matrix for the array y to obtain an estimated value of #imgabs6#; the present invention adopts a cross array, and in the case of an accurate signal source-sensor spatial geometric relationship, without using the Fresnel approximation and considering the existence of amplitude attenuation, it can still accurately estimate the angle and distance of the signal, and these position parameters are associated with each signal source, without the need for an additional pairing process.

The invention discloses a three-dimensional reconstruction method, system, equipment and medium for an orbit CT image, and relates to the technical field of image segmentation, and the method comprises the steps: obtaining an unmarked orbit CT image; the method comprises the following steps: carrying out intra-block pixel replacement on a plurality of sub-regions randomly selected from an unlabeled orbit CT image through nonlinear intensity transformation based on a Bezier curve on the basis of two-way contrast learning and random horizontal overturning and rotation, and generating two distorted views; performing coding analysis of the same weight on the two distorted views, generating coding feature representations of fixed dimensions for two analysis results by using a projection network, and generating a cross-correlation matrix through the two coding feature representations to serve as coding features for fusing different levels; and decoding the fused coding features to generate a segmentation result of the orbit CT image, generating three-dimensional body data through the segmentation result, reconstructing an orbit surface grid, and calculating an orbit volume to obtain a three-dimensional reconstructed orbit model. According to the method, the morphological characteristics of the orbits can be automatically learned from a large number of eye CT images without manual annotation, and accurate orbital segmentation and reconstruction are realized.

The invention provides a clean spectrum analysis method based on an undersampled harmonic signal. The clean spectrum analysis method comprises the steps of acquiring a multi-channel tip timing signal, processing the multi-channel tip timing signal, calculating a cross-correlation matrix, initializing and iterating. According to the method disclosed by the invention, the aliasing effect is inhibited according to the coherence of the frequency components in the frequency spectrum, and the aliasing components are deleted from the actually measured frequency spectrum forming output by iteratively deleting the frequency spectrum part coherent with the peak frequency. The technical scheme of the invention can be widely applied to the technical field of blade vibration detection.

This invention relates to a method for diagnosing composite faults in rolling bearings based on fast eigenmode decomposition (EMD) in the field of rolling bearing fault diagnosis technology. The method includes the following steps: S1, Signal initialization and period estimation: Segmented narrowband filtering of the rolling bearing vibration acceleration signal is performed using a Hanning window; S2, Adaptive filter design: The original signal is filtered using the MOMEDA method to obtain multiple modal signals; S3, Correlation kurtosis calculation and mode selection: The correlation kurtosis value of each filtered signal in each frequency band is calculated, and a cross-correlation matrix between modes is constructed based on cross-correlation theory, retaining a preset number of optimal modes; S4, Fault diagnosis analysis: Hilbert envelope demodulation is performed on the retained optimal mode signals. This invention solves the mode breakage problem that may occur in existing technologies while saving significant computational costs and improving computational efficiency, demonstrating good practicality and engineering application value.

This invention discloses a method for detecting quality faults in a flotation process based on distributed dynamic graph embedding. Step 1: Collect process variables such as concentration, pH value, ore fineness, froth layer thickness, and maximum allowable density during the flotation production process as input variables, and the concentrate grade of the flotation process as the output quality indicator. Step 2: Based on the mutual information method, establish a cross-correlation matrix MI between process variables and quality variables, calculate the average threshold, and select key variables. Step 3: Based on the production process and field experience of the flotation process, divide the key variables into Q quality-related sub-blocks and B quality-unrelated sub-blocks. Step 4: Use the key variable selection and sub-block decomposition results to establish a distributed dynamic graph model to achieve quality-related fault detection. Step 5: Use a Bayesian fusion network to fuse the monitoring results for decision-making. Step 6: Determine whether the fault is quality-related based on the monitoring results.

The invention discloses a multi-dimensional feature self-supervision fan fault diagnosis system and method based on time sequence weight self-adaption, and belongs to the field of wind power generation. Comprising a data acquisition module, a working condition alignment module, a time sequence weight adaptive convolution module and a fan multi-component fault feature cross-correlation calculation module. The method comprises the following steps: collecting real-time and historical data of multiple fans in the same wind field, carrying out working condition alignment and time synchronization processing, and constructing a time sequence weight adaptive convolution model in combination with multi-fan transverse data and single-fan longitudinal historical data; according to the model, time-frequency local features are extracted through Gabor transformation, broadband domain features are obtained through spectral convolution, and multi-dimensional time-frequency joint features are formed. And further calculating a cross-correlation matrix of multi-fan and multi-component fault features, analyzing feature differences between healthy and fault units, dynamically updating model parameters and a loss function according to the feature differences, and realizing self-learning and continuous optimization of the model. The method does not need a large amount of manual annotation, and is high in diagnosis precision, strong in generalization ability and good in adaptability.

The invention provides an SAR scene matching method based on cross-scale fusion enhancement and point clustering correction, and the method comprises the steps: carrying out the multi-scale downsampling processing of an obtained to-be-matched SAR image and a reference SAR image, and correspondingly obtaining a real-time image pyramid and a reference image pyramid; correspondingly obtaining a real-time image weighted fusion enhanced image and a reference image weighted fusion enhanced image by using the real-time image pyramid and the reference image pyramid; carrying out two-dimensional cross-correlation processing on the two obtained weighted fusion enhanced images to obtain a cross-correlation matrix, and generating a rough matching position by adopting the cross-correlation matrix; sequentially carrying out angle feature point detection and clustering processing on the real-time image weighted fusion enhanced image to obtain geometric center coordinates of a plurality of target clusters; and performing fine matching processing by using the geometric center coordinates of the plurality of target clusters and the rough matching positions to obtain a final matching center, and further obtaining a final registration result. In this way, the calculation amount in the SAR scene matching process is reduced, and the matching real-time performance is improved.

A method includes receiving a multi-channel input signal detected in an environment of a vehicle and, for each zone in the environment, performing speech detection by converting each frame in a sequence of frames of the multi-channel input signal into a plurality of frequency subbands, each having a cross-correlation matrix (CCM). For each subband, the method also includes applying a focusing matrix to the CCM to generate a corrected CCM, extracting eigenvalues ​​from the corrected CCM, and determining an eigenvalue ratio between a highest and a second-highest extracted eigenvalue.The method further comprises calculating a median value of the eigenvalue ratios of the plurality of frequency sub-bands, determining a difference between the respective median values ​​of the zones, and, if an absolute value of the difference between the respective median values ​​of the zones is greater than a threshold value, generating an initial speech detection indication.

The invention discloses a linear pre-correction method, device and equipment, a medium and a program product, and the method comprises the steps: obtaining a cross-correlation matrix of a primary function matrix based on the primary function matrix of a current input signal, and judging whether the primary function matrix meets a preset orthogonalization processing condition or not; when the primary function matrix meets the orthogonalization processing condition, performing LU decomposition on the primary function matrix based on a coarse-grained reconfigurable computing architecture to obtain a decomposition matrix; performing orthogonalization processing on the primary function matrix by using the decomposition matrix to obtain a primary function orthogonal matrix; and on the basis of the primary function orthogonal matrix, a correction parameter used for linearization pre-correction is obtained by using a minimum root mean square algorithm. According to the method, the processing efficiency of orthogonalization of the primary function matrix can be improved by utilizing the coarse-grained reconfigurable computing architecture, so that the real-time performance of linear pre-correction in a large-bandwidth or large-scale MIMO scene can be ensured, and the scene limitation is relatively small.

The application provides a surface electromyogram signal decomposition method based on fast gradient kernel compensation, comprising the following steps: mathematically modeling a multi-channel surface electromyogram signal; extending the constructed multi-channel surface electromyogram signal model; constructing a convolution mixed model of the corresponding surface electromyogram signal based on the extended electromyogram signal, a firing time sequence and a noise sequence; constructing a cross-correlation matrix of the surface electromyogram signal and initializing an activity index; initializing a cross-correlation vector based on the activity index; obtaining a gradient function and a learning rate; introducing an exponential weighted moving average model, offline calculating and iteratively updating the cross-correlation vector; extracting the electromyogram signal in a sliding window and expanding it, calculating the cross-correlation matrix after expansion, and estimating the firing time sequence of each motor unit; recalculating the firing time sequence of each motor unit until all signal decompositions are completed. The application can improve the surface electromyogram signal decomposition efficiency and is suitable for real-time decomposition process.

The present invention relates to the technical field of target detection and recognition, and discloses a target feature extraction method based on a mutual correlation self-attention mechanism, which specifically includes first inputting a feature map Z of size H*W*C; then performing a window division operation on the feature map; then expanding the channel dimension of the linear layer to 2*C, dividing the matrix along the channel dimension into a matrix M and a matrix V; obtaining a mutual correlation matrix and an activation operation; then performing a self-attention calculation and a channel attention calculation; and finally outputting a feature map Y of size H*W*C. The present invention searches for the correlation between elements in the feature map, obtains similar features of the target, and simultaneously enables information sharing between channels to realize the selection of attention areas in the spatial dimension and the channel dimension. The present invention improves the recognition effect of the model on the information to be tested in the image, and improves the recognition accuracy of the model.

The application discloses a feature subset selection method and device and a storage medium. The method comprises the following steps: acquiring high-dimensional feature data; obtaining a feature importance list by using a random forest; obtaining a feature cross-correlation matrix by using Spearman correlation; obtaining a fitting degree of each feature according to a target GRU model; performing feature relationship type fusion on the feature importance list, the feature cross-correlation matrix and the fitting degree of each feature to obtain an overall evaluation value of each feature; continuously performing feature sorting on the overall evaluation value of each feature, removing the feature with the lowest evaluation value and determining whether the number of the remaining features is less than a required number of features until the number of the remaining features is not greater than the required number of features; and obtaining a feature subset. The feature subset selection method of the application uses Spearman correlation analysis, a random forest and a GRU for joint evaluation, completes diversified feature evaluation, realizes feature data dimension reduction, improves the feature evaluation processing flow and reduces the time complexity and the space complexity for subsequent work.

This invention discloses an image semantic segmentation method and system based on cross-modal hierarchical knowledge distillation. The method includes inputting an image into a multimodal teacher model and a student model to obtain teacher features and student features at each stage and an output prediction map; fusing teacher features at each stage and enhancing common modal features to obtain corresponding fused features for each stage; dividing the fused features and student features into high-level features and low-level features according to different stages; converting low-level teacher fused features and low-level student features into frequency domain maps respectively, extracting high-frequency components to convey detailed knowledge; calculating cross-correlation matrices and autocorrelation matrices using high-level teacher fused features and high-level student features respectively, and conveying structural knowledge; aligning teacher output prediction maps and student output prediction maps, and conveying response knowledge. This invention fuses multimodal features through modal enhancement, reducing the modal gap between teacher fused features and student features, and improving the segmentation accuracy of the single-modal student model.

This invention discloses a method for tracing the source of generated images based on spatial-frequency cross-domain second-order statistics. This method extracts spatial domain residual anomalous features and frequency domain anomalous features from the generated image, and constructs a cross-domain second-order statistical coupling relationship between them, forming a cross-domain statistical fingerprint for source tracing. The cross-domain statistical fingerprint matrix is ​​obtained by calculating the cross-domain cross-correlation matrix, and then normalized and compressed to obtain a cross-domain second-order statistical feature representation. Based on the similarity matching or distance measurement results between this feature representation and the statistical fingerprint templates of each source category, the source category or confidence score of the generated image is output. This invention can explicitly characterize the coupling relationship between spatial anomalies and frequency anomalies, and still has strong source discrimination ability and good interpretability and generalization performance even when the generation models are highly similar.

The invention provides a three-dimensional dereverberation beam forming method and system based on a cross-correlation matrix, and relates to the technical field of sound source recognition, and the method comprises the steps: collecting sound pressure data through a spherical microphone array; performing fast Fourier transform on the sound pressure data to obtain corresponding frequency domain data; calculating a cross-spectrum matrix corresponding to the spherical array according to the frequency domain data; eliminating reflected sound components contained in the cross-spectrum matrix to obtain a filtered cross-spectrum matrix; and drawing a sound source identification cloud picture based on the filtered cross spectrum matrix. According to the three-dimensional dereverberation beam forming method based on the cross-correlation matrix, the technical problem of insufficient sound source recognition precision of a spherical array in a complex sound field environment in the related technology is solved.

This invention relates to the field of computer security authentication technology, and more particularly to a security authentication method and system based on a secure computer. The method includes the following steps: deploying a sensor array and acquiring data from the secure computer to obtain calibrated spatiotemporal synchronization data; performing temperature and current fusion processing on the calibrated spatiotemporal synchronization data to obtain a thermoelectric coupling feature spectrum; performing point-to-point temporal correlation analysis on the thermoelectric coupling feature spectrum to obtain a point-to-point cross-correlation matrix; extracting nonlinear features from the point-to-point cross-correlation matrix to obtain a nonlinear dynamic feature set; performing dynamic behavior analysis on the nonlinear dynamic feature set to obtain a dynamic phase spectrum of secure computing behavior; and calculating the phase difference of the dynamic phase spectrum of secure computing behavior to obtain a phase difference vector. This invention improves the early detection capability of difficult-to-reproduce and novel hardware-level threats through real-time monitoring and anomaly identification of the computer's microscopic physical characteristics.