31 results about "Bispectrum" patented technology

The invention discloses an electric energy quality disturbance positioning identification method based on phase perception and multi-mode fusion, and relates to the technical field of electric power system monitoring and fault diagnosis. The electric energy quality disturbance positioning identification method based on phase perception and multi-modal fusion comprises the following steps: acquiring an original voltage signal of a three-phase electric power system; extracting a sample embedding vector of each phase based on the original voltage signal, comparing the sample embedding vector with a normal sample vector, and determining a disturbance phase with power quality disturbance in combination with an adaptive threshold value; performing time sequence feature vector extraction and bispectrum image feature extraction on the disturbance phase, and performing alignment and collaborative fusion to obtain fusion features; the electric energy quality disturbance type is identified based on fusion features, phase-level positioning of the three-phase PQD is achieved, through an innovative phase perception embedding and prototype comparison mechanism, accurate and rapid positioning of abnormal phases in a three-phase system is achieved for the first time under the condition of not depending on disturbance phase labels, and the core pain point of a traditional method is solved.

The invention relates to the technical field of refrigeration equipment state monitoring and data analysis, and discloses a freezer remote monitoring and fault prediction system, which comprises the steps of intercepting operation data of a freezer in a door closing silence period, calculating an environment dew point temperature sequence and extracting a unit temperature difference load fluctuation ratio sequence of a refrigeration system; based on the time change rate of the environment dew point temperature sequence and the unit temperature difference load fluctuation ratio sequence, an instantaneous cross bispectrum coupling potential energy sequence representing the nonlinear modulation relation of the time change rate and the unit temperature difference load fluctuation ratio sequence is constructed; then, a two-dimensional phase space hysteresis track is constructed with the environment dew point temperature sequence as the horizontal axis and the instantaneous cross bispectrum coupling potential energy sequence as the longitudinal axis, and the cross bispectrum hysteresis entropy of the dew point driving type load volatility is calculated based on the probability density distribution of the track and the directed circulation area; and finally, judging an operation state and outputting a fault prediction result by verifying whether the entropy value is within a historical reference interval.

The invention relates to the technical field of monitoring and early warning, and discloses a charging monitoring and early warning protection system, which comprises a signal acquisition module used for acquiring a bus voltage signal of a charging bus and performing windowing preprocessing; the feature extraction module is used for positioning carrier frequency and difference frequency from the preprocessed signal, estimating a bispectrum phase curve and calculating a correlation index, a scale curvature, time drift and an endothermic time constant estimation value; the early warning quantity generation module is used for combining and generating an early warning quantity based on the correlation index, the scale curvature, the time drift, the endothermic time constant estimation value and a preset parameter; and the early warning judgment module is used for carrying out standardization processing on the early warning quantity and outputting an alarm or non-alarm instruction in combination with the signal-to-noise ratio and a preset threshold value.

The invention discloses an unknown radiation source individual identification system, method and equipment. The invention particularly relates to an unknown radiation source classification and identification method based on electromagnetic signal SLIB, time-frequency grid-energy feature fusion and entropy penalty K-means clustering. According to the method, time-frequency grid-energy features and contour integral bispectrum features divided by a time-frequency spectrum grid are subjected to PCA + t-SNE dual dimension reduction and splicing fusion to form a fusion feature vector, so that nonlinear high-order statistical features of signals are included, local fine-grained energy change information of a time-frequency domain is reserved, and the fusion feature vector is more accurate and efficient. Therefore, the essential characteristics of the signal can be represented more comprehensively. Secondly, a K-means clustering method based on entropy penalty is constructed, an unsupervised adaptive strategy is adopted, dynamic cluster center adjustment and an information entropy weight updating mechanism are combined, and the number of clusters is adaptively adjusted. Meanwhile, a Gamma dynamic adjustment factor and a Beta weight mechanism are introduced, sample attribution calculation is optimized, and the stability and robustness of clustering are improved.

The invention relates to the technical field of equipment health assessment, and discloses a health assessment method and system based on industrial equipment electrical characteristics and electric fingerprint identification. The method comprises the steps of obtaining an electrical signal of industrial equipment, extracting a time domain statistical feature, a frequency domain harmonic feature, bearing fault feature frequency sideband energy, a wavelet packet energy entropy and a bispectrum amplitude feature in the electrical signal, and constructing an original feature vector; performing standardization processing on the original feature vector to obtain a comprehensive electric fingerprint feature vector, and calculating a distance value from the comprehensive electric fingerprint feature vector to a sphere center through a pre-established health reference model; and mapping the distance value into an equipment health degree score, positioning a fault feature component causing health degree reduction, and generating a fault diagnosis suggestion report. According to the invention, through deep fusion of multi-dimensional electrical fingerprint features, unsupervised health benchmark modeling and continuous quantitative evaluation, accurate evaluation and early fault early warning of the health state of the industrial equipment are realized.

A deep learning-based method for monitoring anesthesia depth based on high time-frequency resolution bispectral density includes the following steps: acquiring single-channel EEG data, preprocessing it, and dividing it into time segments; extracting features from the EEG data, and calculating a high time-frequency resolution wavelet bispectral matrix by combining variational mode decomposition and synchronous squeeze wavelet transform; constructing and training a deep neural network classification model based on partially convolutional neural networks and gated recurrent units; for test data, inputting the extracted features into the model, and selecting the state with the highest output probability (awake, anesthesia maintenance, or awakening) as the final prediction result for the test data; this invention uses single-channel EEG signals as input, greatly reducing the difficulty of acquiring clinical signals, and the proposed high time-frequency resolution bispectral features combined with a dedicated deep learning network have a strong ability to capture complex nonlinear EEG dynamic features and cross-subject generalization adaptability; therefore, this invention can perform high-precision, multi-classification, and real-time automatic monitoring of anesthesia depth.

The invention relates to the technical field of machine vision, and discloses an enameled wire surface defect detection and classification method based on machine vision, which comprises the following steps: firstly, collecting a wire scanning image, extracting a single-pixel time sequence of a fixed transverse position, performing frequency domain transformation after window processing, and obtaining a single-pixel time sequence; determining the position of a main peak according to the power spectrum in a range from zero frequency to Nyquist frequency; calculating a spiral bispectrum locking degree by taking a main peak as a reference; determining a phase sampling resolution according to the locking degree, and performing phase resampling on a single pixel time sequence and an adjacent pixel time sequence to generate phase-aligned two-dimensional blocks; then parameterizing channel gain and channel bias of each layer of the convolutional neural network by using a spiral bispectrum locking degree, and inputting a two-dimensional block into the network to obtain a defect probability; a binary cross entropy loss function is optimized by taking a power function of the spiral bispectrum locking degree as a sample weight; the arithmetic mean value of the defect probabilities of the same axial section is obtained, and when the mean value reaches or exceeds a preset threshold value, it is judged that defects exist.

The application relates to a bearing radial clearance state monitoring method and device in a gear-shaft-bearing system. The method comprises the following steps: arranging an acceleration sensor at a position close to the bearing of a gear box of a gear-shaft-bearing-box fourteen-degree-of-freedom nonlinear coupling vibration system, continuously collecting vibration time domain signals of the system at different running moments, carrying out detrending and filtering processing on the vibration time domain signals, and obtaining pretreated signals; carrying out discrete Fourier transform on the pretreated signals, obtaining modulation signal bispectrum based on traditional bispectrum expansion; carrying out carrier frequency normalization processing on the modulation signal bispectrum, obtaining nine groups of MSB-SE amplitudes, superimposing the nine groups of MSB-SE amplitudes on (1, 2, 3) x GMF on (1, 2, 3) x BPFO, obtaining MSB-SE comprehensive indexes of each moment for monitoring the bearing radial clearance, drawing a curve of the indexes changing with running time, and determining a bearing radial clearance monitoring result. The accurate and online monitoring of the bearing radial clearance is realized.

The invention belongs to the technical field of motor fault diagnosis, and particularly relates to an alternating current motor fault diagnosis method and device based on modulation bispectrum and a storage medium. In order to solve the problems of signal distortion and the like in motor fault diagnosis, the method comprises the following steps of: acquiring a current signal and a voltage signal of an alternating current motor in real time, performing short-time Fourier transform on the current signal, and extracting a single-peak ridge line R1 of a time-frequency spectrogram TFS1; time-frequency analysis based on modulation bispectrum is carried out on the current signal, and a single-peak ridge R2 of a time-frequency spectrogram TFS2 is extracted; obtaining phase average power PI according to the current signal, the voltage signal and the single-peak ridge line R1; determining a characteristic frequency threshold H1 for judging whether a fault occurs or not according to the ratio of the phase average power PI to the rated power PR of the alternating current motor; acquiring a fault characteristic frequency amplitude HF according to the current signal I1 and the single-peak ridge line R2; and judging whether the AC motor has a motor broken bar fault or not and the severity of the fault according to the ratio of the fault characteristic frequency amplitude HF to the characteristic frequency threshold H1.

The invention provides a multi-dimensional feature fusion winter wheat weed detection method and system, and relates to the field of agricultural information technology and precision agriculture technology, and the method comprises the steps: constructing a deep learning model of WMGNet, the deep learning model of the WMGNet comprises a wavelet feature processing module, a bispectrum Mamba branch module, a graph volume integral branch module, a feature fusion module and a classifier which are connected in sequence; inputting the standardized winter wheat image into a deep learning model of WMGNet; extracting a multi-scale feature map of the standardized winter wheat image through a wavelet feature processing module; through a bispectrum Mama branch module, extracting long-range dependency features of the multi-scale feature map; non-local features of the enhanced features are extracted through a graph volume integral branch module; fusing the long-range dependency feature and the non-local feature to obtain a fused feature; and inputting the fusion features into a classifier, and outputting a detection result of the winter wheat high-resolution hyperspectral image.

The invention discloses an unmanned aerial vehicle low-altitude inspection fire real-time monitoring system and method based on deep learning. The method comprises the steps that S1, a visible light image, an infrared image and environment sensor data are collected; s2, performing preprocessing to form a bispectrum image; s3, inputting a double-branch encoder of the improved PP-LiteSeg model, extracting multi-scale features, and obtaining a time sequence evidence vector and a gas anomaly score according to environmental sensor data; s4, performing weighted fusion on the multi-scale features; s5, outputting a flame and smoke segmentation mask, and calculating the uncertainty of a segmentation result; s6, executing time sequence consistency constraint on adjacent frame mask segmentation results; s7, calculating a fire risk probability, and generating fire alarm information; and S8, when the re-cruise mechanism is triggered, the unmanned aerial vehicle is controlled for supplementary collection. The accuracy, the real-time performance and the robustness of unmanned aerial vehicle fire monitoring are remarkably improved.

The invention discloses a deception jamming identification method and system based on double-frequency-domain multi-feature and dimension reduction processing, mainly relates to the technical field of deception jamming detection, and is used for solving the problems that the feature resolution of a time-frequency analysis-based method is rapidly reduced under the condition of low signal-to-noise ratio; the number of features is limited, and each feature shows obvious linear inseparability in a high-dimensional space. Comprising the following steps: performing dimension reduction processing on signal bispectrum features by using an axis slicing method to obtain dimension-reduced bispectrum features; extracting five-dimensional axis integral bispectrum features of the bispectrum features after dimension reduction; constructing an original feature matrix by using the five-dimensional axis integral bispectrum features corresponding to all the echo data sequences, obtaining category label array vectors corresponding to the five-dimensional axis integral bispectrum features in the original feature matrix, and further calculating to obtain preset dimension reduction features corresponding to the original feature matrix; and obtaining a trained preset classification model by using the preset dimension reduction features.

The invention relates to the field of motor performance classification evaluation, and provides a motor performance small sample classification evaluation method based on a multi-mode SPM. Extracting four feature sample sets of original data signal slices, time-frequency diagrams, start-stop morphological parameter combinations and cyclic bispectrum graphic features from the data set; establishing an adaptive proportional decision model to divide a support set and a query set; firstly, support set data is input into a Swinin-T module for feature extraction, and then part of features are randomly selected and input into a multi-mode SPM module for feature enhancement; inputting query set data into a DA module and a multi-modal generative adversarial network module in parallel for data expansion, and after two paths are fused, screening and strengthening through a double-channel attention mechanism; positive and negative sample pairs are screened according to the physical distance of the motor performance grade, a loss optimization model is calculated and compared, and accurate evaluation is realized through a full connection layer and a classification function. The framework solves the problem that a traditional method is low in precision, improves the accuracy of motor performance classification evaluation, and is wide in application prospect.

The invention discloses a radiation source individual open set identification method based on bispectrum characteristics. The method comprises the following steps: firstly, preprocessing an electromagnetic radiation source signal, performing bispectrum estimation to obtain a bispectrum, calculating a contour integral bispectrum value according to an integral path, and inputting the obtained integral bispectrum value into a feature extraction module formed by introducing a one-dimensional residual convolutional network into a space attention module; and the feature extraction module extracts multi-scale individual fingerprint features to obtain an activation vector, Weibull model fitting prediction and OpenMax correction calculation are carried out in combination with a statistical extremum theory, and finally identification of an unknown category of electromagnetic radiation source individual in an open set environment is realized. According to the method, the calculation amount is reduced, the calculation efficiency is improved, and the identification accuracy of the electromagnetic radiation source individuals in the open set environment is improved.

The application discloses a back-stepping optimization method based on vortex detection signal depth features and thermal process parameters, relates to the technical field of vortex signal processing, and comprises the following steps: collecting vortex signals of a valve, and performing denoising processing on the vortex signals; performing principal component analysis on the denoised vortex signals by adopting a multi-scale principal component analysis method to obtain a principal component feature matrix; extracting depth feature characteristics and bispectrum characteristics from the denoised vortex signals; combining the extracted depth features, bispectrum characteristics and the principal component feature matrix into a comprehensive feature vector; constructing a valve parameter inversion model based on multivariate linear regression, taking the comprehensive feature vector as input, and inversely outputting valve parameters; establishing a relationship model based on the inversely output valve parameters and thermal process parameters, and inversely optimizing the thermal process parameters by an optimization algorithm. The application realizes accurate inversion of process parameters through vortex signal processing and a machine learning model.

The application provides a radar radiation source data label screening method based on uncertainty threshold detection, and relates to the technical field of specific radiation source identification, wherein the method comprises the following steps: constructing a learning model and pre-training the learning model, wherein the learning model is pre-trained by extracting signal features through bispectrum and learning feature representation of radiation source data through multi-scale dilated convolution; performing label prediction on to-be-detected radar radiation source data through the pre-trained learning model to obtain predicted labels; removing the predicted labels inconsistent with original labels from the predicted labels, and performing uncertainty threshold detection correction on the remaining predicted labels, and taking the predicted labels meeting the threshold range as reserved data. The application adopting the above scheme realizes screening of error labels of radar radiation source data, and ensures the accuracy of the trained model.

The invention discloses a bearing fault monitoring method based on demodulation bispectrum, and belongs to the field of bearing fault monitoring, and the method comprises the steps: firstly, carrying out the time-frequency analysis of a collected vibration signal, and calculating the demodulation bispectrum, so as to highlight a fault modulation feature; then, bispectrum slices are extracted in the modulation frequency direction, and the correlation spectrum negentropy of each slice is calculated, so that the periodic fault information amount contained in each frequency component is quantified; and finally, identifying a fault characteristic frequency and a harmonic wave thereof according to a peak value in the correlation spectrum negentropy sequence. According to the method, fault features can be clearly and completely extracted under high-noise interference by demodulating the depiction capability of bispectrum to a modulation relation and the sensitivity of correlation spectrum negentropy to periodic pulses. Simulation and experiments show that compared with a traditional method, the rolling bearing fault diagnosis method has higher diagnosis accuracy and robustness and is suitable for early fault and weak fault diagnosis of the rolling bearing.

The application discloses a locomotive transmission system diagnosis method based on adaptive wavelet bispectrum, and relates to the technical field of locomotive transmission system diagnosis. The application comprises the following steps: continuously collecting current signals of a locomotive traction motor; performing time-frequency conversion on the current signals, obtaining time-frequency distribution of the signals, and extracting time-frequency ridge lines from the time-frequency distribution; based on the time-frequency ridge lines, calculating the change rate of the dominant frequency in the current signals, and identifying the current operation condition of the locomotive according to the change rate; according to the identified operation condition, processing the current signals to extract the secondary phase coupling features associated with faults in the current signals; and determining the fault type and / or fault position of the locomotive transmission system. The application can relatively effectively extract the secondary phase coupling features caused by transmission system faults under various actual operation conditions by constructing a diagnosis framework that adaptively selects analysis tools according to the signal frequency change rate and introducing time-frequency rearrangement technology to suppress frequency ambiguity.

This invention discloses a radar target recognition method based on pole-symmetric mode decomposition and bispectral estimation, relating to the fields of radar signal processing and intelligent target recognition technology. The method includes acquiring the original echo signal of a radar target at an aircraft, preprocessing the original echo signal including removing DC components and energy normalization, performing pole-symmetric mode decomposition on the preprocessed radar echo signal to obtain multiple intrinsic mode function components and residual components, selecting and reconstructing micro-motion components based on modal characteristics, and filtering out translational components. This radar target recognition method based on pole-symmetric mode decomposition and bispectral estimation can achieve stable and accurate recognition of aircraft targets in complex electromagnetic environments and under multiple interference backgrounds, improving the intelligence level and engineering practical value of radar target recognition systems.

The application discloses a kind of based on multi-transform domain feature fusion's cross-domain radiation source individual identification method, first parallel extraction three kinds of transform domain features to radiation source signal: rectangular integral bispectrum feature, fuzzy function orthogonal slice feature and hilbert marginal spectrum feature, and convert into two-dimensional image.Then, a fusion identification model is constructed by multiple ResNet branches and an MLP.Multiple ResNet branches are used as base learner, and three kinds of feature images are extracted respectively; MLP is used as meta learner, and the extracted high-level feature vector is fused and classified.To cope with cross-domain data distribution difference, the training of the model is first independently pre-trained each ResNet branch on source domain data;When fine-tuning on target domain data, the ResNet backbone network is frozen, and only the adaptation layer at the end of each branch and the MLP parameters are updated, so as to efficiently realize knowledge transfer.The application significantly improves the accuracy and robustness of the radiation source individual identification model in the cross-domain scene.

The application discloses a semi-supervised game model classification method and device based on a bispectrum feature representation, and the method comprises the following steps: acquiring an original received signal and an original received signal sample of a transmitting stage; inputting the original received signal into a trained semi-supervised game model for data signal processing; wherein the trained semi-supervised model is obtained by training the frequency order of the original received signal sample based on a high-order spectrum feature representation; and based on the data signal processing, the original received signal is classified by using a classifier to obtain a classification result. The high-order spectrum feature is used as the frequency order representation of the signal, so that the end-to-end semi-supervised classification is realized, and a higher evaluation score and classification accuracy are shown on a real data set and a simulation data set.

The present application relates to a kind of double spectrum adaptive threshold feature selection methods based on random forest, comprising: obtaining feature importance score by random forest model;Based on feature importance score, the first derivative of feature importance is calculated, gradient spectrum is constructed, and threshold candidate point of gradient spectrum is obtained;The cumulative importance contribution spectrum of feature is constructed, and the threshold candidate point of cumulative importance contribution spectrum is obtained;The threshold candidate point of gradient spectrum and cumulative importance contribution spectrum is integrated, each candidate threshold point is evaluated by optimization score function, and the optimal feature is determined.The present application first applies double spectrum feature selection in the field of rainfall simulation systematically, realizes the intelligent feature screening of high-dimensional meteorological data, and provides technical support for refined weather forecast;The present application can be directly integrated into existing weather forecast business system, improve the prediction accuracy through accurate feature selection, reduce disaster loss, reduce the consumption of computing resources, save operation cost, and improve the efficiency of model deployment and maintenance.

The present application relates to the field of radar signal processing, and provides a feature extraction method and device for a composite signal. Local detail features and global context information of a to-be-extracted time-frequency graph and a to-be-extracted bispectrum graph of the composite signal are extracted respectively to obtain time-frequency graph features and bispectrum graph features, and channel exchange is performed on the two features to determine effective channel features corresponding to each channel; a first fusion feature is obtained according to all the effective channel features; local detail features of each position in the to-be-extracted time-frequency graph and the to-be-extracted bispectrum graph are extracted respectively to obtain first distinguishing features and second distinguishing features, and feature fusion is performed on the first distinguishing features and the second distinguishing features based on attention weights to obtain a second fusion feature; and feature splicing is performed on the first fusion feature and the second fusion feature to obtain a target feature, thereby solving the problem of low recognition rate of the composite signal in the case of low signal-to-noise ratio.

The invention discloses a Mama-based multi-domain feature fusion unmanned aerial vehicle radio frequency fingerprint identification method and device, and relates to the technical field of wireless communication and target identification, and the method comprises the steps: fusing the in-phase / orthogonal feature, the bispectrum feature and the short-time Fourier spectrum feature of a signal, so as to achieve the enhanced representation of a radio frequency fingerprint. Aiming at the bottleneck of computation complexity and real-time performance brought by multi-domain feature fusion, an efficient fusion framework based on Mama and Vision-Mama is constructed, the framework gives full play to the linear computation complexity advantage of a Mama model in long sequence modeling, and is combined with an algorithm optimization strategy of hardware perception, so that the computation overhead is remarkably reduced while high recognition precision is kept. In order to solve the problem that redundancy and complementarity are difficult to balance in the feature fusion process, a self-adaptive feature interaction module based on a cross attention mechanism is introduced, the module can dynamically model the dependency relationship among multi-source features, self-adaptive distribution of fusion weights is achieved, and therefore the feature degradation phenomenon caused by environmental interference is effectively restrained.

The application relates to the technical field of monitoring and early warning, and discloses a charging monitoring and early warning protection system, which comprises a signal acquisition module, a feature extraction module, an early warning quantity generation module and an early warning judgment module. The signal acquisition module is used for collecting a bus voltage signal of a charging bus and performing windowed pretreatment. The feature extraction module is used for locating a carrier frequency and a difference frequency from the pretreated signal, estimating a bispectrum phase curve, and calculating an association index, a scale curvature, a time drift and an endogenous heat time constant estimation value. The early warning quantity generation module is used for combining the association index, the scale curvature, the time drift, the endogenous heat time constant estimation value and preset parameters to generate an early warning quantity. The early warning judgment module is used for performing standardized processing on the early warning quantity, combining a signal-to-noise ratio and a preset threshold to output an alarm or non-alarm instruction.

The application discloses a gearbox state monitoring method and system based on single-class normal data, and belongs to the field of fault diagnosis, and comprises the following steps: acquiring normal working condition data of a gearbox, performing nonlinear cross-bispectrum processing on the normal working condition data to obtain a training data set; constructing an interpretable differential diagnosis model, training and optimizing the interpretable differential diagnosis model through the training data set, acquiring running data of the gearbox under unknown states, performing nonlinear cross-bispectrum processing on the running data, and judging whether the running data is fault data according to the interpretable differential diagnosis model after processing the running data; and when the judgment result is fault data, performing feature enhancement on the fault data according to the optimized interpretable differential diagnosis model to obtain an interpretable fault diagnosis result. The above scheme can judge only by using normal working condition data, and overcomes the limitation of a lack of fault samples, and exhibits high precision, strong noise resistance and good generalization capability in gearbox fault diagnosis.

This application provides a computer vision-based power equipment operation status recognition system, relating to the fields of intelligent power system inspection and non-contact fault diagnosis. First, it performs rigorous spatiotemporal registration and channel stacking on multimodal sensor data streams, utilizing the material semantic information and illumination brightness characteristics of the visible light channel to accurately locate potential reflection interference areas. Based on this, it constructs an interference confidence map by calculating the spatial correlation of the bispectral gradient field, achieving probability quantification of reflected light spots. Then, it introduces a physics-based radiation inversion algorithm, dynamically correcting the original infrared radiation data with the confidence map as weights, and using neighborhood reference temperatures to eliminate artifact hotspots, reconstructing the true equipment temperature field. Finally, it performs intelligent defect classification based on the corrected thermal matrix, ensuring the accuracy and robustness of equipment operation status diagnosis while eliminating ambient light interference.

The invention discloses a method for enhancing fluctuation characteristic differences of line spectrums of water surface and underwater sound sources. The method comprises the following steps: (1) initializing processing parameters; (2) reading a radiation noise sampling sequence received by the hydrophone, dividing the sampling sequence into a plurality of sub-sampling sequences, and calculating a bispectrum diagonal slice of each sub-sampling sequence; and (3) accumulating the bispectrum diagonal slice energy of each sub-sampling sequence at all frequency points in the line spectrum frequency cluster, and constructing line spectrum frequency cluster energy and sequence. And (4) dividing the line spectrum frequency cluster energy and the sequence into a plurality of segments, calculating a discrete coefficient of each segment, and sequentially forming a fluctuation index sequence based on the discrete coefficients. And (5) constructing intervals for measuring the fluctuation characteristic difference of the water surface and underwater sound source line spectrums under different fluctuation indexes, comparing the interval width corresponding to each fluctuation index, and indicating that the larger the interval width is, the more obvious the enhancement effect of the fluctuation index on the difference is.

This invention relates to the field of radar signal processing, and provides a method and apparatus for extracting multi-level features of radar signals. The method involves convolving the time-frequency map of the original DRFM relay signal at multiple feature extraction scales to extract local detail features at various locations, obtaining a first distinguishing feature; modeling the global information in the first distinguishing feature to obtain a first global feature; fusing the first distinguishing feature and the first global feature to obtain the time-frequency map feature; processing the bispectral map of the original DRFM relay signal using the same method to obtain the bispectral map feature; and fusing the time-frequency map feature and the bispectral map feature to obtain the multi-level features of the original DRFM relay signal. This method solves the problem that the accuracy of DRFM relay signal category recognition drops sharply due to severe aliasing and the lack of obvious artificial features in DRFM relay signals.

The invention discloses a backstepping optimization method based on eddy current detection signal depth characteristics and thermal process parameters, which relates to the technical field of eddy current signal processing, and comprises the following steps: collecting an eddy current signal of a valve, and carrying out de-noising processing on the eddy current signal; performing principal component analysis on the de-noised eddy current signal by adopting a multi-scale principal component analysis method to obtain a principal component characteristic matrix; performing depth feature extraction and bispectrum feature extraction on the de-noised eddy current signal; combining the extracted depth features, bispectrum features and the principal component feature matrix into a comprehensive feature vector; a valve parameter inversion model based on multiple linear regression is constructed, the comprehensive feature vector serves as input, and valve parameters are output through inversion; based on the valve parameters output through inversion, a relation model is established with the thermal process parameters, and the thermal process parameters are reversely deduced and optimized through an optimization algorithm. According to the method, accurate inversion of the process parameters is realized through eddy current signal processing and a machine learning model.