42 results about "Compressed sampling" patented technology

The invention discloses a self-adaptive threshold SAMP reconstruction method for a power quality disturbance signal. According to the method, a compression observation value is obtained by constructing a random Gaussian observation matrix, and sparse representation is carried out on an original signal by using discrete Fourier transform. In the iterative reconstruction process, the spectrum amplitude difference is introduced for the first time to serve as an adaptive termination basis, and automatic adaptation of different noise levels and different disturbance characteristics is achieved in combination with a dynamic threshold update function. According to the method, the problems of traditional SAMP sparseness overestimation and redundant iteration are effectively avoided, and the calculation load is remarkably reduced. Compared with an OMP method, an original SAMP method and the like, the method has the advantages that the number of iterations can be reduced by 30%-60%, the reconstruction signal-to-noise ratio is increased by 2-5 dB, the root-mean-square error is reduced by 10%-25%, higher robustness and real-time performance are achieved in power quality disturbance signal reconstruction, and the method is quite suitable for scenes such as compressed sampling, edge calculation and high-speed signal reconstruction in a power quality monitoring system.

The invention discloses a federal compression learning method and system based on knowledge distillation, and the method comprises the steps: enabling each terminal device to adaptively adjust the compression ratio of the terminal device according to a real-time network, and carrying out the compression sampling through the adjusted compression ratio; then, taking a compressed sampling value as a knowledge distillation medium between edge equipment and a cloud server of different levels, and carrying out cooperative training on heterogeneous models deployed by nodes of different levels in combination with a federated learning method and knowledge distillation; the knowledge distillation is introduced in the federated learning process, so that models with different capacities and architectures can be deployed at different hierarchy nodes, and the compressed sampling value with the internal privacy protection characteristic is used for replacing original data or proxy data as a distillation medium, so that the privacy risk caused by the knowledge distillation is eliminated from the data source. And the communication efficiency, privacy protection and computing power utilization optimization of the end-side cloud are realized.

The application discloses a battery capacity estimation method and system based on adaptive inflection point driving compressed sampling, relates to the technical field of lithium battery capacity estimation, acquires original charging data of a battery to be measured, obtains a voltage curve based on voltage data in the original charging data, identifies an inflection point of the voltage curve, divides the voltage curve into different sampling sections according to the inflection point position, performs segmented random sampling in each section based on a sampling ratio, forms an initial sampling set, introduces a dynamic updating function to iteratively update the sampling ratio according to section information entropy and prediction error feedback under the condition that the total number of sampling points is constant, and realizes adaptive compressed sampling; and the sampling set is input into a capacity prediction model for processing to obtain a predicted battery estimated capacity. The adaptive inflection point guided compressed sampling and the deep learning model adapted to sparse data in structure are combined to realize lightweight and data-efficient battery state estimation.

A power spectrum estimation method for compressed sampling of a generalized stationary HDR signal is disclosed. This method is used for processing aero-engine blade tip vibration monitoring signals. The method involves determining the threshold of a self-recovering analog-to-digital converter and the delay scheme for multi-channel compressed sampling based on prior information about the signal amplitude and frequency; compressing the signal using a multi-channel dual-threshold cooperative sampling method; recovering the correct amplitude information from the acquired signal using the dual-threshold information; constructing an indicator sequence and a data sequence based on the sampling delay scheme; and performing Fourier transform, element-square operation, inverse Fourier transform, and element-product calculation sequentially on the indicator sequence and data sequence to obtain the power spectrum estimate of the signal.

This invention discloses a signal enhancement method and apparatus for time-slice compressed sensing 3D imaging radar. First, an M×N dimensional random binary matrix is ​​generated as the measurement matrix for compressed sampling. The target area is illuminated using a laser source, and repeated sampling and modulation are performed to obtain a sampling matrix. The j-th column of the sampling matrix is ​​extracted to generate a compressed sampling vector for the corresponding time slice. The covariance between the compressed sampling vector and each column of the measurement matrix is ​​calculated to obtain the compressed sampling covariance vector. The covariance between each column of the measurement matrix and all columns is calculated to obtain the covariance measurement matrix. Based on a compressed sensing 2D reconstruction algorithm, the 2D image of each time slice is reconstructed using the covariance measurement matrix. The reconstructed 2D images of all time slices are stitched together to form a complete 3D image. This method and apparatus can solve the problems of low reconstruction accuracy and poor image quality in existing compressed sensing 3D imaging technologies under noisy environments.

This invention discloses a short-time compressed sampling method and system based on a modulation wideband converter (MWT) to solve the problem of real-time acquisition of short-time burst wideband sparse signals. The method reduces the compression and reconstruction unit of the MWT to multiple short-time segments of the uncovered signal's complete time domain. For each segment, a short-time reconstruction algorithm is used for independent reconstruction. This algorithm integrates principal component analysis for dimensionality reduction, a modified atom matching strategy for selecting conjugate atom pairs, recursive prospect orthogonal matching pursuit for determining the support set, and time-domain support extraction to suppress noise. Finally, the complete signal is recovered through segment reconstruction. This invention significantly reduces reconstruction complexity, improving the signal-to-noise ratio of reconstructed 5GHz wideband signals at a 2.5GHz sampling rate by more than 4dB compared to traditional methods, and exhibits stronger robustness to changes in sparsity and channel number, achieving low-power, high-precision sub-Nyquist sampling.

The application discloses a single light guide antenna pulse terahertz time-domain spectroscopy system and a detection method, and relates to the technical field of terahertz. The system comprises a laser pulse generation module, a single light guide antenna module, a compressed sampling module, a signal reconstruction module and a control unit. The application integrates a random light delay modulation unit and a single-pixel detector, realizes random time interval sampling of a terahertz signal, effectively breaks through the limitation of a traditional sampling mode, improves the flexibility and efficiency of signal acquisition, and combines a signal reconstruction module of an FPGA+GPU heterogeneous computing architecture. The system can adaptively construct a sparse base, accurately sparsely represent a measurement vector, and then efficiently reconstruct an original terahertz time-domain signal through an optimization objective function and an adaptive iterative solution algorithm. The application not only improves the accuracy of signal reconstruction, but also reduces the complexity of data processing, and provides strong support for the practical application of terahertz technology.

The invention discloses an inverse synthetic aperture radar imaging method and system with phase distortion and two-dimensional compression sampling, terminal equipment and a medium, and relates to the technical field of radar imaging. The method comprises the following steps: constructing a sparse recovery model containing phase error correction based on a low-rank and sparsity principle; wherein the sparse recovery model is adapted to an inverse synthetic aperture radar echo signal which is subjected to two-dimensional compression sampling and has phase distortion; and based on a unified alternating direction multiplier method framework, solving the sparse recovery model to complete sparse recovery and phase error correction of the two-dimensional compressed sampling signal, and outputting an inverse synthetic aperture radar image. According to the method, the problems that phase errors are ignored in existing inverse synthetic aperture radar imaging and phase correction fails during down-sampling are solved, a clear image can be generated, and the calculation complexity is reduced.

The invention relates to a wireless communication technology, and discloses a multi-area RIS cascade channel estimation method, device and equipment and a medium, and the method comprises the steps: constructing an RIS-assisted multi-area uplink communication system model according to a base station entity, an RIS entity and user entities in different areas; modeling a channel of each entity node to obtain a system signal model; establishing a base station entity-RIS entity-user entity cascade channel matrix, mapping the cascade channel matrix to a virtual angular domain to obtain a virtual angular domain channel matrix, and performing angular domain sparse processing to obtain an angular domain sparse channel matrix; utilizing a structure sparse compression sampling matching pursuit algorithm to obtain a common row support set and a column support set shared by user entities in different areas; and calculating a non-zero element value of the angular domain sparse channel matrix according to the common row support set and the column support set, and inversely transforming the non-zero element value to a spatial domain to obtain a cascade channel estimation matrix. According to the method, the sparse channel estimation precision of the multi-region user channel structure can be improved.

This invention provides an anti-interference electronic signal acquisition system, comprising: an anti-interference random sampling front-end module for randomly modulating and low-speed integrating and sampling an analog electrical signal to output a compressed measurement vector; a dual-domain joint sparse reconstruction processing module for performing joint sparse decomposition of the compressed measurement vector using an augmented dictionary matrix to simultaneously obtain a pure electrical signal estimate and an interference component estimate; and an industrial internet communication interface module for encapsulating the pure electrical signal estimate into an industrial internet of things protocol data frame for transmission, or directly transmitting the compressed measurement vector. This invention aims to accurately extract pure current or voltage signals from a strongly interfering environment at a low sampling rate through random compressed sampling and dual-domain joint sparse reconstruction, thereby solving the problems of high resource consumption and insufficient anti-interference capability in existing measurement systems.

Embodiments of the present disclosure disclose an image compression sensing method and device based on feature domain deep unfolding. A specific implementation of the method comprises: performing compressed sampling on an original image to be compressed according to a sampling matrix to generate a sampling value; performing sampling matrix deconvolution processing on the sampling value to generate an initial reconstruction feature; for the initial reconstruction feature, performing the following iterative processing steps: performing a feature domain momentum operation on a preset stage image feature of the initial reconstruction feature to generate a momentum update result; generating a multi-scale feature according to the momentum update result; taking the obtained multi-scale feature as the initial reconstruction feature, and continuing to perform the iterative processing steps for a preset number of times to generate a multi-scale feature set; performing a second-level multi-scale proximal mapping operation on the multi-scale feature set to generate a reconstruction image feature; and performing convolution aggregation on the reconstruction image feature to generate a reconstruction image. The implementation improves the quality of image reconstruction.

This invention belongs to the field of distributed large language model inference scheduling technology, and relates to a multi-model dynamic scheduling method for high-concurrency dialogue requests. The method includes: collecting the network card hard interrupt timestamp sequence at the streaming response bus to establish a streaming output time interval vector; extracting the differential fluctuation envelope of the streaming output time interval vector and comparing it with a reference transmission delay line to separate the channel congestion index; calculating the real-time routing suppression coefficient based on the queue backlog length and the channel congestion index and reconstructing the cost-aware routing matrix; compressing the sampling window when concurrent data traffic is overloaded, and sending a redirection signal to shift the concurrent data stream when the real-time routing suppression coefficient exceeds the limit. This invention eliminates the need for active probing, decouples the calculation of congestion and network jitter in situ, effectively eliminates long-tail response latency, and improves throughput.

The application discloses a kind of based on self-attention mask uncertainty hyperspectral image reconstruction method, specifically is: step 1: using CASSI optical system compressed sampling, obtain two-dimensional measurement Y;Step 2: using the self-optimization network of mask uncertainty automatically captures the uncertainty between different mask space structures, to obtain the optimized mask m';Step 3: two-dimensional measurement and optimized mask are input into the residual reconstruction network of transformer, to obtain the final high light spectrum.The application shows good reconstruction quality and reconstruction speed, and has the adaptability of different masks, meets the practical application of CASSI system, and further promotes the in-depth development and popularization and application of hyperspectral compressed snapshot imaging technology.

The present invention discloses an indoor zoning positioning method for streamlining measurement information, belonging to the field of indoor positioning. First, the present invention combines the traditional fingerprint positioning method and the compressed sensing technology to construct a positioning system model; then, it uses the spatial position characteristics to complete the fingerprint division of the fuzzy area and effectively screen APs according to multiple scales; finally, it calculates the target position and sparsely reconstructs the position information of the point to be located. The present invention performs compressed sampling on the measurement information, and then restores a nearly complete signal value through a limited number of measurement values, avoiding the acquisition and storage of samples of the complete signal. At the same time, the reduction of the sample size reduces the overhead during online calculation. Overall, APs outside the area with less influence are filtered out effectively, and individually, an AP subset with intensive information in the area is selected by comprehensively considering the discrimination and measurability. The SBL algorithm is selected to handle the influence of high correlation between fingerprint information, and the sorting of the AP information sequence is introduced to enhance the weight of adjacent RPs, effectively improving the positioning accuracy.

A spectrum analysis method based on a circumferential minimum sparse size compression sampling, in the method, a down-sampling factor is determined according to a frequency upper limit of a measured signal and a sampling rate of a sampling channel; a circumferential minimum sparse size mode is determined through an optimization algorithm; periodic non-uniform sampling is performed on the signal by using the circumferential minimum sparse size mode to obtain signal samples; continuous covariance samples are estimated from the signal samples; and a signal power spectrum is estimated from the continuous covariance samples.

This invention discloses a method and system for geotechnical engineering exploration data analysis based on edge computing, comprising the following steps: collecting and preprocessing geotechnical engineering exploration data; constructing a sparse representation basis; generating an adaptive observation matrix; performing compressed sampling processing on the real-time collected geotechnical exploration data to obtain compressed observation data; using the CoSaMP algorithm to perform sparse reconstruction processing on the compressed observation data to obtain a reconstructed geotechnical exploration data sequence; extracting stratigraphic structure evolution features and constructing a stratigraphic structure evolution feature vector; performing stratigraphic state discrimination processing to obtain stratigraphic state discrimination results; updating the sparse representation basis and the adaptive observation matrix, and uploading the compressed observation data, the reconstructed geotechnical exploration data sequence, and the stratigraphic state discrimination results to a cloud server for collaborative processing, thereby achieving efficient and accurate real-time data analysis and processing, and improving the accuracy and real-time performance of exploration data analysis.

The invention relates to a block adaptive measurement wavelet compressed sensing method and system for video or image optical communication, and the method comprises the steps: obtaining a target image, carrying out the preprocessing and sparse blocking of the target image, and generating image blocks; clustering the image blocks, distributing an initial measurement number for each image block, generating a measurement number matrix, traversing each image block by using the measurement number matrix, and carrying out compressed sampling to obtain compressed data; according to the compressed data, establishing a model based on norm minimization as a target function, calculating an initial starting point, taking the initial starting point as a first-round hot start stage, iteratively solving a regularization parameter vector by adopting a near-end gradient descent method, generating a solution vector, performing data reconstruction based on the solution vector, and outputting a reconstructed target image. The invention provides an image and video compressed sensing scheme with higher efficiency and higher adaptability.

The invention relates to a compressed sampling system suitable for high dynamic range broadband signals and a rapid reconstruction method, and the method is used for processing aero-engine blade tip vibration monitoring signals. Based on the signal Nyquist rate fNYQ, the bandwidth B, the effective frequency band number K and the maximum amplitude Amax, hardware parameters of a modulation broadband converter MWC and a threshold value of a self-recovery analog-to-digital converter SR-ADC are set in a priori mode; performing multi-channel frequency band demodulation, low-pass filtering and analog-to-digital sampling on the signal by using a modulation broadband converter and a dual-threshold self-recovery analog-to-digital converter to obtain the output of each channel; correct amplitude information is recovered from the collected signals according to the dual-threshold information, and continuous signals are constructed to a finite field conversion module; adaptively solving a spectrum support set by using variational Bayesian inference, energy aggregation, FDR saliency detection and a BIC model reduction algorithm; and performing up-sampling and frequency band modulation on the low-rate sample sequence based on the frequency spectrum support set, and reconstructing a high-dynamic range time domain signal meeting the Nyquist rate and a complete frequency spectrum located in an original frequency band.

This invention relates to the field of geophysical exploration technology and provides a method and system for compressed sensing reconstruction of magnetotelluric data. The method includes the following steps: acquiring raw magnetotelluric time-series data, performing segmentation and standardization preprocessing to obtain multiple standardized signal segments; transforming each standardized signal segment to obtain a sparse representation in the transform domain, and compressing and sampling the sparse representation to obtain a compressed measurement vector; extracting multidimensional feature vectors from each signal segment, predicting the corresponding regularization parameter scaling factor based on a neural network model, and calculating an adaptive regularization parameter; using the adaptive regularization parameter as the regularization intensity, constructing a convex optimization model containing L2 data fidelity terms and L1 sparse regularization terms, and reconstructing the magnetotelluric time-series data based on a greedy, accelerated iterative threshold shrinkage algorithm. This invention can reduce bandwidth and facilitate real-time monitoring in environments with high transmission requirements, such as field environments.

The application belongs to the technical field of signal processing, and relates to a variable step length compression sampling method of a gearbox oil temperature signal, comprising the following steps: collecting a plurality of historical gearbox oil temperature signals, performing segmentation and adaptive clustering, and obtaining a plurality of signal classes similar in shape; constructing a training matrix of each signal class; extracting an approximate part and a detail part of all signal segments, and constructing a plurality of mixed dictionaries according to the approximate part of the same training matrix and the detail part of each training matrix; calculating the information entropy and the required sampling point number of each signal segment in the training matrix, constructing a measurement matrix with equal time sampling intervals according to the sampling point number, and obtaining a compression sampling model; collecting an online gearbox oil temperature signal and inputting the compression sampling model to obtain a plurality of compression signals with variable step lengths; and reconstructing the compression signals to obtain reconstructed signals. The method can be used for variable step length compression sampling of aperiodic variable condition gearbox oil temperature signals.

This invention discloses a blind estimation method for frequency-hopping signal parameters based on atomic features. The method involves: first, windowing, compressed sampling, and atomic feature extraction of the frequency-hopping signal; then, roughly estimating the hopping time based on the distribution of atomic features within different windows; next, precisely estimating the hopping time, hopping period, and hopping speed using a block-diagonalized Fourier orthogonal matrix; and finally, estimating the frequency of each hop signal using an orthogonal matching pursuit algorithm, and obtaining the frequency-hopping pattern of the entire sample by concatenating each hop signal. This invention can adjust the search step size and compression ratio to change the computational complexity and estimation accuracy, adapting to different signal-to-noise ratio (SNR) environments. In high SNR environments, increasing the search step size and compression ratio effectively reduces computational complexity, while in low SNR environments, decreasing the search step size and compression ratio effectively improves estimation accuracy.

The present application relates to the technical field of image processing and compressed sensing reconstruction, and discloses an iterative unfolding compressed sensing image reconstruction method and system fusing convolution and attention mechanism, which comprises the steps of compressed sampling, initialization reconstruction and unfolding reconstruction processing; the unfolding reconstruction processing comprises multiple iteration stages, each stage comprising the steps of priori enhancement processing and data consistency updating, wherein the priori enhancement processing is used for enhancing the image structure expression capability through convolution feature transformation and attention mechanism, and the data consistency updating is used for constraining and correcting the current reconstruction result according to the measured data; the data consistency updating contains learnable updating parameters and is kept as a positive value through function constraint, so as to improve the stability of the reconstruction process. Through alternately executing feature enhancement and consistency updating, the present application realizes the fusion of physical constraint and feature expression capability, and improves the image reconstruction quality and stability under the condition of low sampling rate.

The invention discloses a battery capacity estimation method and system based on adaptive inflection point driving compression sampling, and relates to the technical field of lithium battery capacity estimation, original charging data of a to-be-measured battery is obtained, and a voltage curve is obtained based on voltage data in the original charging data; identifying an inflection point of the voltage curve, dividing the voltage curve into different sampling sections according to the position of the inflection point, and executing segmented random sampling in each section based on a sampling proportion to form an initial sampling set; under the condition that the total number of sampling points is kept constant, a dynamic updating function is introduced to iteratively update the sampling proportion according to section information entropy and prediction error feedback, and self-adaptive compressed sampling is achieved; and inputting the sampling set into a capacity prediction model for processing to obtain the predicted battery estimated capacity. And in combination with adaptive inflection point guided compression sampling and a deep learning model of adaptive sparse data in structure, light-weight and efficient-data battery state estimation is realized.

The invention provides a laser ultrasonic signal adaptive compressed sampling method, which comprises the following steps of: extracting learning type attention features from an original signal through a neural network, and obtaining a learning type attention weight through a Softmax function; generating a threshold attention weight according to the amplitude distribution and the energy threshold of the original signal; carrying out fusion to generate a sampling weight; based on the sampling weight, performing weighted modulation on the basic measurement matrix to generate an adaptive measurement matrix; compression sampling is carried out on an original signal. The laser ultrasonic signal self-adaptive compression sampling method is suitable for non-contact nondestructive testing imaging of surface defects constructed by additive manufacturing, and the problem of loss of defect related key information at a low sampling rate is avoided by integrating the sampling weight generated by learning attention and threshold attention; finally, effective compression sampling is carried out on the original signal, the data storage and transmission cost is remarkably reduced, and the large-scale detection efficiency is improved.

The application discloses a CoSaMP pipeline defect positioning method based on NRBO adaptive sparsity, and belongs to the technical field of signal processing and nondestructive testing. The method comprises the following steps: performing pretreatment on the original echo signal collected, performing peak value detection on the echo signal, and adaptively completing window segmentation; a dictionary for sparse representation is constructed; a Newton-Raphson optimization algorithm NRBO is used to adaptively select a sparsity parameter k; based on the sparsity parameter k, a compressed sampling matching pursuit algorithm CoSaMP is used to complete sparse decomposition and reconstruction of the echo signal, and defect positioning is realized according to the reconstructed signal. The application realizes adaptive determination of the sparsity parameter through the NRBO, can reduce the subjectivity of parameter selection and improve the calculation efficiency under the premise of effectively preserving the key information of the echo signal, is suitable for defect echo signal processing and pipeline defect positioning scenes, and has good application prospect and popularization value.

The present application relates to the technical field of power grid super-harmonic, and specifically discloses a power grid super-harmonic online monitoring and identification method and system, which utilizes a preset learning type measurement matrix to perform non-uniform compression sampling on original signals to be measured of a power grid, so as to obtain a low-dimensional observation vector sequence; the sequence is subjected to linear embedding processing and superposition position coding, so as to generate a feature vector containing time sequence information; the feature vector is input into a pre-trained Transformer reconstruction model, a high-dimensional time domain waveform after restoration is output through a decoding layer of the Transformer reconstruction model, so as to realize reconstruction of a super-harmonic signal; the reconstructed waveform is subjected to feature analysis, and monitoring parameters of the power grid super-harmonic or a harmonic source identification result is output. The present application can obtain a high-frequency signal under the condition of lower sampling rate of hardware, can reduce data transmission amount, and thus effectively reduces system cost; compared with a reconstruction algorithm of traditional compression sensing which depends on iterative solution, the present application only needs one-time forward calculation to complete signal reconstruction, and response speed is improved.

The invention discloses a dynamic adjustable curved surface detector imaging system, and relates to the technical field of optical engineering and adaptive optics. Based on an adaptive compression sampling mechanism of neural pulse gradient distribution, the sampling density of a high-dynamic region is improved by one order of magnitude, a mechanical deformation process is directly replaced, and the imaging efficiency is improved. The limitation of hardware delay on the frame rate is eliminated; the millisecond-level transmission process of the neural signal is completely captured, and signal loss caused by insufficient frame rate in a traditional scheme is avoided; a joint optimization framework of space-time tensor modeling and curved surface geometric mapping is adopted, and distortion correction and sequence reconstruction are synchronously completed in single solution; due to the introduction of an alternating direction multiplier method, strict synchronization of geometric deformation compensation and texture recovery is ensured, and motion blur or edge distortion caused by error accumulation in step-by-step processing is avoided; and a dynamic weight mechanism of tensor multimode nuclear norm constraint automatically adjusts the low-rank constraint intensity according to the space-time structure complexity.

The application discloses a kind of compressed imaging block circulation structure measurement matrix optimization design method, belong to image technical field.The method includes: based on the total pixel number of image to be reconstructed constructs particle swarm, based on the current free element vector of each particle determines the current block circulation measurement matrix, calculates the Gram matrix corresponding to the current block circulation measurement matrix;Using the objective function based on Gram matrix constructs to calculate the fitness of each particle;Based on the current free element vector and fitness of each particle determines and updates the local optimal free element vector corresponding to each particle, determines the inertia weight of each particle;From all local optimal free element vectors, determine the global optimal free element vector;Based on global optimal free element vector determines the block circulation measurement matrix corresponding to the compressed sampling, reconstructs image based on block circulation measurement matrix.The application can improve subsequent image reconstruction effect.

This invention relates to the field of optical inspection technology, specifically disclosing a method and system for detecting the flatness of mobile phone glass covers. The method involves projecting a multi-band composite beam onto the glass under test using a confocal optical system, simultaneously receiving mixed spectral signals containing reflections from multiple interfaces; generating key feature signals through compressed sampling and spectral channel optimization; analyzing and separating the reflection components of each layer in real time by matching with a pre-built template library; reconstructing the three-dimensional morphology of the surface based on the surface reflection signals; extracting macroscopic warpage and microscopic defect features; constructing a normalized and weighted comprehensive feature vector; and inputting this vector into a deep neural network model to achieve intelligent flatness assessment.

The invention relates to the technical field of signal compression, in particular to an ultra-wideband pulse waveform space-time synchronous compression method and system, a medium and a product. The method comprises the following steps: executing multi-dimensional orthogonal projection on an original ultra-wideband pulse signal by using a spatio-temporal joint perception dictionary to extract a spatio-temporal characteristic coefficient; adjusting an observation matrix according to the channel state information, and performing compressed sampling on the spatial-temporal characteristic coefficient by using the adjusted observation matrix to generate a compressed observation vector; performing manifold learning dimensionality reduction mapping on the compressed observation vector to construct a space-time manifold topological structure, and obtaining manifold feature coordinates based on the space-time manifold topological structure; introducing weighted least square constraint to carry out nonlinear reconstruction on the flow shape characteristic coordinates to generate a sparse pulse sequence; and performing variational Bayesian iterative reasoning on the sparse pulse sequence to obtain an optimal quantization parameter, and performing entropy coding on the sparse pulse sequence by using the optimal quantization parameter to generate a transmission code stream. The method can guarantee that the details of the pulse signals are complete and distortionless.