42 results about "Self correlation" patented technology

The invention relates to the technical field of civil aircraft maintenance and repair, in particular to a flap health state prediction method based on transferable trend grouping autocorrelation comparative learning, which comprises the following steps of: firstly, mining a typical degradation trend from total data by utilizing a clustering algorithm; dividing the individual samples with differentiation into a plurality of trend data groups with high consistency; secondly, introducing an autocorrelation comparison mechanism into each group, capturing common time sequence dependence crossing a long period among samples, and constructing a plurality of pre-training models capable of representing a specific decline mode; then, dynamic matching is carried out according to the characteristics of the flap to be detected, and the most adaptive trend category and the pre-training model are selected; and finally, on the basis of a migration fine tuning strategy, a small amount of data is used for carrying out directional fine tuning on the pre-trained model, so that an exclusive prediction model of the flap is efficiently obtained, compared with the prior art, a model does not need to be built for each aircraft from zero, migration can be carried out on the basis of a limited model library, and effective management of the health state of the civil aircraft is achieved.

The invention belongs to the technical field of industrial artificial intelligence, and particularly relates to an unsupervised multi-mode industrial anomaly detection method based on self-correlation learning. The method comprises the following steps: acquiring a to-be-detected RGB image and depth map data, and generating a multi-modal pseudo sample based on Berlin noise and texture filling so as to enhance the discrimination capability of a model for abnormity; rGB and deep modal features are fused by using comparative learning, and interaction information between modals is extracted; calculating local association and global association of the features by using a learnable Gaussian kernel function and a self-attention mechanism respectively; and finally, the association difference between the local association and the global association is used as a discrimination standard, and the reconstruction error is combined to calculate an anomaly score, so that the tiny anomaly on the surface of the industrial product is accurately detected and positioned. According to the method, the self-correlation learning framework which fuses the multi-modal features and uses the global-local correlation difference as the anomaly discrimination basis is constructed, so that high-precision anomaly positioning is realized under the condition of lack of anomaly sample labeling.

A method of processing an electrocardiosignal capable of accurately obtaining heartbeat information, a device for processing an electrocardiosignal, a computer-readable storage medium, and a computer program product. The method of processing an electrocardiosignal includes: a detection signal acquisition step of acquiring a detection signal by detecting an electrocardiosignal of a living body by a sensor; a signal preprocessing step of dividing the detection signal into a plurality of segmented signals with a prescribed time length, calculating an average value of a plurality of signals included in each of the segmented signals, and obtaining a preprocessed signal by subtracting the average value of the segmented signal from each signal sample point included in the segmented signal; and a self-correlation calculation step of calculating a self-correlation function of the preprocessed signal.

The application discloses a water supply network sensor optimization arrangement method based on graph signal processing, and comprises the following steps: acquiring topological structure information of a water supply network, and establishing a user node weighted graph without direction and connection according to the topological structure information to determine a self-correlation localized graph information value of each user node and a cross-correlation graph information redundancy value between different user nodes; wherein the self-correlation localized graph information value of the user node is: the user node v i The cross-correlation graph information redundancy value between the user nodes v j According to the self-correlation localized graph information value of the user node and the cross-correlation graph information redundancy value between different user nodes, a target function is determined: user nodes meeting the target function are screened out as sensor nodes one by one until the number of sensors reaches the requirement.

The invention provides a time sequence prediction method based on deep learning, and relates to the technical field of time sequence prediction, and the method reduces the length of a long sequence through multi-scale decomposition, replaces the traditional full-connection attention through combining with an autocorrelation attention mechanism, prevents the calculation overhead from increasing along with the square of the length of the sequence, remarkably improves the processing efficiency of the long sequence, and improves the accuracy of time sequence prediction. A multi-scale decomposition and bidirectional fusion strategy can simultaneously capture fine-grained fluctuation (the finest scale) and a macroscopic trend (the coarsest scale), the non-stationary data processing capability is superior to that of an existing model, and sequential decomposition (trend-seasonal decoupling) and self-correlation attention (long-range dependence modeling) have a synergistic effect, so that loss of key time sequence information is reduced; the method does not need to depend on manually designed prior hypotheses (such as a fixed period and coarse scale initialization), can adaptively process periodic and non-periodic data, and is wider in application scene.

The application discloses a multispectral image registration method based on multiscale deep feature map fusion, which can be used for registering spectral images between different wave bands. The method first acquires consistent feature maps of a target image and a source image by using a multiscale consistent feature map mapper, performs self-correlation feature modification on the multiscale consistent feature maps by using a transformer structure to obtain multiscale matching information feature maps, and then processes the multiscale matching information feature maps based on a self-attention mechanism to output registration parameter residuals. Under the supervision of an image registration true value and consistent feature map content consistency loss function, the network is iteratively run until a preset iteration number is reached, and finally, registration parameters are obtained. Experimental results and analysis on a multispectral dataset show that the multispectral image registration method based on multiscale deep feature map fusion effectively improves the registration accuracy of multispectral images.

The embodiment of the invention discloses a data transmission method, device and system. The method comprises: acquiring a data frame comprising a plurality of subframes, the subframes comprising 226 pilot symbols in one polarization direction, the 226 pilot symbols being generated by a target polynomial and a seed, the target polynomial being x11 + x10 + x9 + x7 + 1; and sending the data frame. The super-frame architecture provided by the embodiment of the invention is relatively low in frame redundancy, and relatively good in sequence self-correlation and cross-correlation characteristics of the pilot symbols, so that the recovery signal of a receiving end is improved, and the quality of the recovered signal is improved.

The invention relates to the technical field of data processing, and particularly provides a method, a system, equipment and a medium for restoring an unknown signal constellation diagram, and the method comprises the steps: carrying out the self-correlation calculation of an in-phase I-path component of a received signal, carrying out the spectrum analysis of a self-correlation result, so as to enhance a spectrum component corresponding to a symbol rate, estimating the code element rate of the signal from the frequency spectrum; based on the code element rate, analyzing energy change characteristics of I-path and orthogonal Q-path components of the received signal, and detecting and identifying a code element boundary position to obtain a code element boundary sequence; and for each code element interval divided by the code element boundary sequence, aggregating I-path and Q-path sampling values in a stable region of each code element interval to obtain constellation point coordinates corresponding to each code element, and generating a constellation diagram by all constellation points. According to the invention, clear and accurate constellation diagram recovery of various modulation signals, especially high-order QAM signals, under the condition of low signal-to-noise ratio is realized.

The invention discloses a period-phase optimized robot tactile signal segmentation and alignment method, which can quickly extract effective signal segments when a robot contacts an object, and solves the problems of unstable tactile signal form, baseline drift, uncertain period length and the like. The method provides a joint optimization normal form: firstly, estimating the average period of a tactile signal through self-correlation analysis; then, unified modeling is carried out on the segmentation tangency points and the fragment duration under structural constraints to form variables to be optimized, and an aggregation similarity objective function with normalized mutual measurement standards is constructed; and finally, iteratively solving the optimal segmentation point and phase of the time sequence signal in the contact process by maximizing the objective function. Compared with a traditional non-global optimization method, the method can be suitable for non-fixed-length quasi-periodic signals with unstable forms, has high robustness for amplitude drift and periodic jitter, is higher in segmentation precision and universality, and can provide accurate contact segment effective signals for robot touch recognition.

The invention discloses a monitoring method and device for quality data of a self-correlation mixing process, and relates to the technical field of disease monitoring. The method comprises the following steps: obtaining a group of multivariate self-correlation data in a historical controlled state and a group of multivariate self-correlation data in a historical out-of-control state; the method comprises the following steps: respectively calculating a sample mean value and a covariance matrix, carrying out decorrelation on samples through a Cholesky decomposition method, constructing an MEWMA sequence by utilizing standard independent data after decorrelation, training a classifier model of a random forest, designing a control chart method based on the random forest, and monitoring a linear autocorrelation process; on the basis of data decorrelation, according to the random forest theory, a novel control chart is generated, the modeling and monitoring problems of quality data in the self-correlation mixing process are effectively solved, the trained random forest model automatically learns and captures complex and nonlinear modes of data in controlled and out-of-control states, and the quality of the self-correlation mixing process is improved. And effective modeling and monitoring can be carried out no matter the data is in skewed, multi-peak or completely unknown distribution.

The invention relates to the technical field of geophysical exploration, and particularly discloses a while-drilling seismic drill string reference signal feature extraction method, which comprises the following steps: obtaining a drill string reference signal; performing self-correlation processing on the drill string reference signal to obtain a correlation domain drill string reference signal; obtaining a correlation domain drill string reference signal frequency spectrum; according to the correlation domain drill string reference signal frequency spectrum, the correlation domain drill string reference signal is decomposed into a low and medium frequency mode and a high frequency mode, and a low and medium frequency signal and a high frequency signal are obtained; performing sparse decomposition on the low and medium frequency signals, further introducing a regularization parameter optimization strategy based on spectrum energy physical constraint, and extracting low frequency signals; and subtracting the low-frequency signal from the low and medium-frequency signal to obtain an intermediate-frequency signal. According to the method, the random noise is suppressed by using self-correlation, and then frequency band separation and sparse reconstruction are performed, so that the target frequency band characteristics of the drill stem reference signal are effectively extracted.

The invention discloses a signal frequency parameter estimation method and device, and relates to the technical field of signal processing. The method comprises the following steps: acquiring a first center frequency, a first bandwidth and a signal-to-noise ratio, and determining a center frequency distribution range and a bandwidth distribution range in combination with a frequency interval; carrying out band-pass filtering processing and self-correlation processing on the original signal, and then carrying out Fourier transform and frequency domain correction processing to obtain a target frequency domain signal; based on the target frequency domain signal, determining a second center frequency in the center frequency distribution range, and then determining a corresponding bandwidth lower bound / upper bound distribution range; and based on the target frequency domain signal, respectively determining bandwidth starting / ending frequencies in the bandwidth lower bound / upper bound distribution range to obtain a second bandwidth. According to the method, the influence of noise interference is reduced through filtering and self-correlation processing, the distribution range is adaptively limited to avoid the limitation of the Fourier change length, accurate positioning of the frequency parameters is realized through the difference method, and the precision of frequency parameter estimation is remarkably improved.

The invention provides a communication method, a communication device and a storage medium, relates to the technical field of communication, and can reduce the self-correlation function sidelobe level of a signal. The method comprises the steps that first information is received, second information is sent, the first information is used for indicating parameters needed for determining at least one coefficient set, the second information is used for indicating at least one coefficient set determined based on the first information, and each coefficient set in the at least one coefficient set comprises at least one set of coefficients; the at least one group of coefficients are coefficients required for filtering the amplitude of the first sequence to obtain the amplitude of the second sequence, the phase of an element with an index of i in the first sequence is the same as the phase of an element with an index of i in the second sequence, the first sequence is a sequence obtained by preprocessing information bits, i is an integer smaller than or equal to M, and M is an integer smaller than or equal to 1; the number of elements included in the first sequence and the number of elements included in the second sequence are both M, and M is a positive integer. The first signal is a signal generated based on the second sequence.

The present application relates to the technical field of civil aircraft maintenance and repair, in particular to a flap health state prediction method based on migratory trend grouping self-correlation contrast learning, first, typical degradation trends are mined from the full data using a clustering algorithm, and individual samples with differentiation are divided into several high-consistency trend data groups; second, a self-correlation contrast mechanism is introduced within each group to capture the common time sequence dependence between samples across long periods, and multiple pre-training models that can represent specific degradation patterns are constructed; then, the most suitable trend category and pre-training model are selected by dynamically matching the characteristics of the flap to be tested; finally, based on the transfer fine-tuning strategy, the pre-training model is fine-tuned with a small amount of data, thereby efficiently obtaining the exclusive prediction model of the flap. Compared with the prior art, it is not necessary to build a model from scratch for each aircraft, and migration can be performed based on a limited model library to effectively manage the health state of a civil aircraft.

The application provides a self-correlation instrument optical system and a laser pulse width measuring device. The self-correlation instrument optical system comprises an optical path difference generating module, a self-correlation signal generating module and a self-correlation signal receiving module. The optical path difference generating module comprises an optical path adjusting sheet and a Fresnel double prism, which are used for dividing the laser to be measured into two beams of pulse laser with different delays and intersecting at a non-collinear included angle. The self-correlation signal generating module comprises a frequency doubling crystal and a filter sheet, which are used for frequency doubling conversion of the pulse laser, generation of a self-correlation signal and filtering. The self-correlation signal receiving module comprises a CCD module, which is used for receiving an intensity signal of the self-correlation signal and converting the intensity signal into a self-correlation function image for output. The self-correlation function image is used for calculating the pulse width of the laser to be measured through image analysis. The problems that the existing technology is difficult to build and the size of the non-collinear included angle is not easy to adjust due to the use of multiple mirrors are solved.

The application discloses an acceleration flutter speed optimization method based on aerodynamic stiffness coupling modal similarity, which comprises the following steps: firstly, solving an aeroelastic control equation under a preset incoming flow speed condition to obtain a plurality of first-order aerodynamic coupling modal vectors; then, constructing a self-correlation modal execution criterion matrix based on the modal vectors, extracting the maximum value of non-diagonal line elements of the matrix, and defining the maximum value as an aerodynamic stiffness coupling modal similarity index; and finally, analyzing the mapping relationship between the to-be-optimized parameters and the similarity index, determining a parameter combination that makes the index reach a minimum value, and determining that the parameter combination corresponds to a structure or material design scheme with the maximum critical flutter speed. The application significantly reduces the calculation cost, greatly improves the optimization efficiency, is especially suitable for complex structure flutter boundary optimization design containing multiple degrees of freedom and multiple parameters, and provides an efficient calculation tool for aircraft aeroelastic design.

The invention belongs to the technical field of petroleum seismic exploration, and discloses a node seismograph data quality monitoring method and device, electronic equipment and a medium, and the method comprises the steps: carrying out the reading, thinning and preliminary quality control of imported shot gather seismic data, carrying out the self-correlation analysis of the shot gather seismic data, obtaining a detection point dominant frequency amplitude file, and carrying out the detection point dominant frequency amplitude file; the method comprises the following steps of: acquiring a main frequency amplitude file of a detection point, performing normalization processing on the main frequency amplitude file of the detection point, calculating the ellipsoid volume of the detection point, outputting a quality control result text file, reading the quality control result text file, establishing a three-dimensional scatter diagram, setting an ellipsoid volume threshold value, selecting an abnormal detection point, and completing node seismograph data quality monitoring. According to the method, seismic data can be subjected to shot-by-shot and trace-by-trace detailed analysis, and the effect of improving the data quality of the node seismograph is achieved. According to the method, the device, the electronic equipment and the medium, the application is convenient, and the method is suitable for monitoring the quality of the data acquired by the node seismograph and used for the seismic exploration technology.

The present invention relates to a particle size measurement method, a particle size measurement device, and a particle size measurement program, which set an appropriate measurement time corresponding to a particle size as a measurement object, and save unnecessary measurement time. The particle size measurement method includes: a test measurement step of irradiating a sample with light for a test measurement time including a plurality of measurement time points set in advance, and measuring a test measurement intensity of scattered light scattered by the sample; a self-correlation function calculation step of calculating a self-correlation function representing a relationship between a self-correlation of the test measurement intensity and time; a setting step of setting a part of the measurement time points among the plurality of measurement time points set in advance to be used in a formal measurement, based on a time until the self-correlation function falls below a prescribed threshold and a preliminary time set by adding a time to the time; a formal measurement step of measuring a formal measurement intensity of the scattered light for a formal measurement time including the part of the measurement time points; and a particle size calculation step of calculating a particle size of the sample based on the formal measurement intensity.

The application discloses an ultrafast laser time domain shaping and calibration device, and belongs to the field of ultrafast laser processing. The device comprises a time domain shaping module and a self-correlation calibration module. The time domain shaping module divides linearly polarized ultrafast laser into a fixed optical arm and a delay optical arm. The time delay of the double pulses is regulated through an optical delay line, and the laser parameters of the two optical arms are independently controlled to ensure that the polarization states of the double pulses are orthogonal to each other. The self-correlation calibration module is connected with the time domain shaping module through a coaxial optical path. A type II phase matching nonlinear crystal is used to convert the double pulses into a second harmonic (SHG) signal. The SHG signal intensity under different time sequences is collected through a scanning delay line to generate an intensity self-correlation curve. The delay zero point is calibrated, and the pulse width is calculated. The measured parameters are fed back to the time domain shaping module to correct the delay parameters. The application can be applied to an existing laser processing platform, provides the function of time domain shaping double pulse laser processing, effectively reduces the heat accumulation phenomenon, and improves the quality of ultrafast laser processing.

The invention provides a communication method, a communication device and a storage medium, relates to the technical field of communication, and can reduce the self-correlation function sidelobe level of a signal. The method comprises the steps that second information is determined and sent, and the second information is used for indicating at least one coefficient set. Each coefficient set in the at least one coefficient set comprises at least one group of coefficients, the at least one group of coefficients are used for filtering the amplitude of the first sequence to obtain the amplitude of the second sequence, and the phase of the element with the index of i in the first sequence is the same as the phase of the element with the index of i in the second sequence. The first sequence is a sequence obtained by preprocessing information bits. Wherein i is an integer smaller than or equal to M, the number of elements included in the first sequence and the number of elements included in the second sequence are both M, and M is a positive integer.

The invention relates to the technical field of femtosecond laser, and discloses a method and equipment for diagnosing and correcting dispersion distribution of a chirped fiber grating, and a medium. The method comprises the following steps: building a chirped pulse amplification system; adjusting the pulse compressor to enable the autocorrelator to display the narrowest pulse width of the compressed pulse in the current state; inserting a narrowband filter capable of transversely moving into a diffraction light path of the pulse compressor, sequentially shielding spectral components with different wavelengths by transversely moving the narrowband filter, and monitoring a compression pulse time domain shape after shielding different spectral components in real time by using an autocorrelator; determining that the corresponding spectral component is an error spectral component causing dispersion distribution mismatching according to the compression pulse time domain shape; and performing chirp fiber grating dispersion distribution correction on a grid region corresponding to the error spectrum component in the chirp fiber grating according to the determined error spectrum component. Through spectral component scanning, the problem that the high-order dispersion error can be perceived and is difficult to position is solved.

The feature selection method, device and electronic equipment provided in the application relate to the field of data processing, and include the following steps: obtaining an input feature data set; preprocessing the input feature data set to obtain a plurality of standardized features; obtaining a pre-constructed feature weight coefficient module; inputting each standardized feature into the feature weight coefficient module to obtain a self-correlation score corresponding to each standardized feature; performing operation on each standardized feature according to an attention mechanism to determine a mutual correlation score between each standardized feature; multiplying and fusing the self-correlation score and the mutual correlation score to obtain a feature comprehensive score; and sorting each standardized feature according to the feature comprehensive score to obtain a sorted target feature set. The method provided in the application achieves the effect of improving the accuracy and adaptability of feature selection.

The present application belongs to the technical field of material property analysis, and particularly relates to a prefabricated component in-mold compactness monitoring method and system. The method comprises the following steps: applying an impact on an outer measuring point of a mold and collecting elastic wave time domain response signals of adjacent sensors to construct a self-correlation attenuation-Gaussian composite window truncated signal; performing Fourier transform on the truncated signal, generating a dimension-entropy weight composite spectrum after dividing frequency subbands, and locating an inherent frequency peak; calculating a quality factor of the inherent frequency peak, extracting harmonic frequency peak energy to form a multi-scale spectrum peak energy distribution vector, and calculating a Shannon entropy; matching a temperature-quality factor coupling model from a pre-set model library by using the Shannon entropy, and correcting the quality factor in combination with an in-mold environment temperature; inputting the inherent frequency peak, the temperature-compensated quality factor and the energy distribution vector into a nonlinear mapping model, and outputting the in-mold compactness of the prefabricated component. The present application realizes high-precision and reliable quantitative evaluation of the in-mold compactness of the prefabricated component.

The invention provides a millimeter wave radar fatigue driving detection method based on multiple features, and the method comprises the steps: firstly, transmitting and receiving a chirp signal through an FMCW radar, obtaining an IF signal through frequency mixing, and obtaining the target distance and speed information; then, performing beam forming by adopting an MVDR algorithm, generating an angle spectrum and extracting a target position; a self-correlation human body signal separation model is designed, phase demodulation and differential processing are carried out on a target position, and chest signals are separated. Micro Doppler feature extraction is carried out on the head signals, and fatigue action features are analyzed through short-time Fourier transform. Blink frequency information is obtained in combination with micro-Doppler information and energy reflectivity, eye information is reserved through CFAR, and then eye actions are judged according to the energy change speed. Finally, adopting a dual-mechanism weighted decision, fusing four characteristics of head fatigue action, blinking frequency, breathing frequency and heartbeat frequency to obtain a comprehensive fatigue index F, if Fgt; and if 0.5, determining that the user is fatigue.

The invention is suitable for the technical field of computer vision, and provides a time self-correlation imaging method and system based on an adaptive orientation space window. The speckle contrast imaging and self-correlation imaging method comprises the following steps: acquiring continuous multi-frame speckle images; the method comprises the following steps: presetting a basic space window by taking a target pixel as a center, and giving a plurality of orientations; selecting a certain frame of speckle image, respectively extracting speckle light intensity sequences of corresponding orientations in the basic space window, and calculating the spatial domain variance of the light intensity sequence of each orientation; selecting the orientation with the maximum light intensity variance, and determining the orientation as an effective calculation window of the target pixel; and performing time autocorrelation function calculation based on the effective calculation window so as to solve a blood flow velocity parameter of the target pixel. According to the method, the noise of the blood flow area can be reduced, and the real-time monitoring requirement is met while the high spatial resolution is guaranteed.

The invention discloses an unmanned aerial vehicle identification method, device and equipment based on periodic frame structure features, and relates to the field of signal identification. The method comprises the following steps of: segmenting a digital image transmission signal to obtain a plurality of segments of digital signals; performing fast Fourier transform on each segment of digital signal to generate a corresponding time-frequency matrix; dividing the time-frequency matrix into a plurality of sub-channels in a frequency domain, and performing frequency band energy integration on frequency domain information of the time-frequency matrix to obtain a time domain energy sequence of each sub-channel; performing self-correlation operation on the time domain energy sequence to obtain a self-correlation function, and searching a frame period of the self-correlation function in a preset period range; performing incoherent accumulation processing on the time domain energy sequence based on a frame period to obtain an enhanced energy sequence; and matching the enhanced energy sequence with a preset feature library to obtain a matching result, and determining identification information of the unmanned aerial vehicle according to the matching result. By implementing the technical scheme provided by the invention, the power consumption during identification of the unmanned aerial vehicle is reduced.

The invention relates to the technical field of Dcoker information system security, in particular to a multi-dimensional self-adaptive container credibility evaluation method and system, and the method comprises the following steps: S1, deploying a test sample container on a host test board system by using a Docker client, and enabling the sample container to execute an instruction of an internal application program; s2, capturing and quantifying performance parameters of the sample container to obtain a signature sequence of the parameters; s3, obtaining an autocorrelation value of the context of the parameter according to the signature sequence of the parameter; s4, obtaining a direct trust value according to the autocorrelation value of the context of the parameter; s5, obtaining an interaction trust value according to the autocorrelation value and the direct trust value of the context of the parameter; and S6, obtaining a direct trust weight and an interactive trust weight according to the direct trust value and the interactive trust value, thereby obtaining a comprehensive trust value of the sample container. Compared with a traditional safety detection method, the method has the advantages that the detection accuracy is high, real-time adjustment can be carried out according to the parameter change of the container, and the adaptability is higher.

The invention relates to the technical field of privacy computing, and discloses a naval vessel fault expert knowledge management system based on a large language model, and the system comprises the steps: firstly collecting continuous monitoring data of a naval vessel fault correlation system, extracting a main harmonic frequency, a physical period and a harmonic energy ratio through frequency spectrum and self-correlation analysis, and generating a period parameter; a privacy loss multiplication factor is calculated according to periodic characteristics and time window parameters, the time step length is optimized, and a safe release rhythm is formed; calculating the total amount of actual effective privacy budget in combination with the nominal privacy budget, and performing budget allocation according to the importance weight of expert questions and answers and the data sensitivity; and finally, carrying out noise adding processing on the statistical result of each time window, and outputting according to a safety rhythm, thereby realizing intelligent fault reasoning of the large language model under privacy protection.

The application provides a small sample image generation method based on lightweight model fine-tuning, a device and a storage medium, and relates to the field of computer vision.The small sample image generation method based on lightweight GAN fine-tuning can relieve the GAN overfitting problem in the small sample scene, and meanwhile, the distillation strategy with key region constraint can better focus on the features of the region of interest of a user.The method comprises the following steps: a model compression method is used to obtain a lightweight model to relieve the overfitting in the model fine-tuning process; a mask operation is introduced to perform key region constraint; and a cross-domain channel self-correlation distillation loss is designed to process the small sample image generation task with deformation.The application proposes a cross-domain spatial self-correlation loss with key region constraint by combining the mask operation and the spatial self-correlation loss.Through the analysis of the experimental results of multiple data sets, it can be concluded that the small sample image generation method based on lightweight GAN model fine-tuning proposed by the application shows competitive results.

The invention discloses a network anomaly detection method and device based on data analysis, and belongs to the technical field of network security. The method comprises the following steps: acquiring historical data of core network parameters, and dividing the historical data into a plurality of continuous periods according to a time sequence; the core network parameter is a parameter capable of reflecting a network operation state; for each period, executing the following steps: generating self-correlation characteristics and cross-correlation characteristics in the period by utilizing historical data of core network parameters in the period; the self-correlation features and the cross-correlation features are fused to obtain fusion features in the period; and constructing a benchmark by using the fusion features of a plurality of continuous periods, and detecting whether the network is abnormal by using the deviation between the fusion features of the current period and the benchmark. According to the method, the time anomaly of a single parameter can be captured, and the linkage anomaly between the parameters can also be captured, so that the detection accuracy is improved for hidden attack and complex anomaly detection.