An intelligent method for suppressing noise in optical fiber sensing signals

Abstract

The present invention belongs to the field of signal denoising, and specifically is an intelligent method for suppressing noise of optical fiber sensor signals. The method comprises signal acquisition, Fourier transform, amplitude coding, phase coding, fusion coding, signal reconstruction and loss calculation and optimization. The present invention constructs a spectrum diagram of a time-series optical fiber sensor signal, performs amplitude coding and phase coding on the spectrum diagram, cross-connects and exchanges information, improves feature representation capability, and accurately and efficiently denoises and reconstructs the optical fiber sensor signal. The present invention adopts dual-stream feature fusion coding based on frequency feature conversion to obtain global correlation and local correlation of amplitude, performs lightweight and accurate reconstruction of phase, increases processing capability for low signal-to-noise ratio, and improves denoising effect. The present invention designs a branch independent processing and information sharing mechanism to realize feature transmission, obtains comprehensive and detailed feature representation, enhances feature learning capability and signal processing accuracy, and increases the naturalness and coherence of the reconstructed signal.

Description

Technical Field

[0001] The invention belongs to the field of signal denoising, and in particular is an intelligent method for suppressing noise of optical fiber sensing signals. Background Art

[0002] In the processing of optical fiber sensing signals, the signal may be affected by various interferences and noises, which will reduce the quality of the signal and make it difficult to accurately analyze and process the signal. Signal noise suppression is an important task, which aims to eliminate or reduce the noise components in the signal, so as to extract the effective information of the signal; however, the existing signal noise suppression methods require a large amount of data for training or learning, and have poor generalization to unknown noise, resulting in serious signal denoising distortion. The existing neural network methods have weak processing capabilities for low signal-to-noise ratios and poor denoising effects. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the existing signal noise suppression method requires a large amount of data for training or learning, resulting in serious signal denoising distortion. The present invention creatively designs an intelligent fiber optic sensor signal noise suppression method, constructs a spectrum diagram of the time-series fiber optic sensor signal, and performs amplitude encoding and phase encoding on the spectrum diagram respectively, exchanges information through cross-connections, improves the feature representation capability, enhances the recognition of data, and denoises and reconstructs the fiber optic sensor signal more accurately and efficiently; in view of the problem that the existing neural network method has weak processing capability for low signal-to-noise ratio and poor denoising effect, the present invention creatively adopts dual-stream characteristic fusion encoding based on frequency characteristic conversion to encode the amplitude The frequency characteristics of the convolution features between channels are transformed, the global correlation of the amplitude in the spectrum graph is obtained, multiple convolution layers are used to obtain the local correlation of the amplitude in the spectrum graph, and layer normalization and convolution coding are used to perform lightweight and accurate reconstruction of the phase in the spectrum graph. The low signal-to-noise ratio processing capability of the model is increased through optimized amplitude coding and optimized phase coding, and the denoising effect is improved. The present invention creatively designs a branch independent processing and information sharing mechanism, and realizes feature transmission through information sharing. Compared with the simple and direct amplitude and phase coding fusion method, each information sharing branch can obtain a more comprehensive and detailed feature representation, enhance the feature learning ability and signal processing accuracy, and make the denoised reconstructed signal more natural and coherent.

[0004] The technical solution adopted by the present invention is as follows: an intelligent method for suppressing noise of optical fiber sensing signals, specifically comprising the following steps:

[0005] Step S1: signal acquisition, collecting optical fiber sensing signals, wherein the optical fiber sensing signals are timing signals;

[0006] Step S2: Fourier transform, windowing the optical fiber sensing signal and performing Fourier transform to obtain a spectrum diagram;

[0007] Step S3: Amplitude coding, performing two-layer convolution activation processing on the spectrum graph to obtain amplitude coding, wherein the first layer of convolution activation includes a convolution layer Conv(1×7) and a ReLU activation function, and the second layer of convolution activation includes a convolution layer Conv(7×1) and a ReLU activation function;

[0008] Step S4: Phase encoding, performing two-layer convolution processing on the spectrum graph to obtain phase encoding, wherein both layers of convolution are convolution layers Conv(5×3);

[0009] Step S5: dual-stream characteristic fusion coding based on frequency characteristic conversion is performed on amplitude coding and phase coding to obtain denoising coding, which specifically includes the following steps:

[0010] Step S51: frequency characteristic conversion based on the attention mechanism is performed on the amplitude coding to obtain optimized amplitude coding, which specifically includes the following steps:

[0011] Step S511: Perform two layers of reshape convolution processing on the amplitude code to obtain intermediate parameters. Each layer of reshape convolution operation includes convolution and Reshape function processing in sequence:

[0012] ;

[0013] In the formula, represents amplitude coding, represents convolution processing, and the subsequent numbers are the convolution processing order identifiers. Reshape function processing, is the intermediate parameter;

[0014] Step S512: feature fusion, perform element-wise multiplication of the intermediate parameters and the amplitude code, and perform full connection layer processing to obtain the amplitude fusion feature:

[0015] ;

[0016] In the formula, represents element-wise multiplication, represents the fully connected layer processing, represents the amplitude fusion feature;

[0017] Step S513: Attention mechanism processing, element-wise addition of the amplitude fusion feature and the amplitude coding, and convolution and activation processing are performed to obtain the preliminary optimized amplitude coding:

[0018] ;

[0019] In the formula, represents row-level addition, represents the activation function, represents preliminary optimization of amplitude coding;

[0020] Step S514: Perform convolution processing on the preliminary optimized amplitude code and perform two-layer convolution activation processing to obtain the secondary optimized amplitude code:

[0021] ;

[0022] ;

[0023] In the formula, represents the secondary optimized amplitude encoding, is the intermediate parameter;

[0024] Step S515: normalize and activate the secondary optimized amplitude code, and repeat steps S511 to S513 to obtain the optimized amplitude code:

[0025] ;

[0026] In the formula, It is an abbreviation of the process from step S511 to step S513. Represents standardized processing, stands for Optimized Amplitude Coding;

[0027] Step S52: Lightweight phase coding optimization, performing two-layer layer normalization and convolution processing on the phase coding to obtain lightweight phase coding:

[0028] ;

[0029] ;

[0030] In the formula, The representative layer is normalized. is an intermediate parameter, stands for lightweight phase encoding, stands for phase encoding;

[0031] Step S53: dual-stream information sharing, using element-level multiplication and Tanh activation to share information between optimized amplitude coding and lightweight phase coding, to obtain dual-stream optimized amplitude coding and dual-stream optimized phase coding:

[0032] ;

[0033] ;

[0034] In the formula, Represents Tanh activation function processing, Represents dual-stream optimized amplitude coding, stands for Dual-Stream Optimized Phase Coding;

[0035] Step S54: split-stream decoding, split-stream decoding is performed on the dual-stream optimized amplitude coding and the dual-stream optimized phase coding to obtain amplitude decoding and phase decoding:

[0036] ;

[0037] ;

[0038] ;

[0039] ;

[0040] ;

[0041] In the formula, Represents Sigmoid activation function processing, represents the fully connected layer processing, ReLU activation function processing, , and is the intermediate parameter, represents amplitude decoding, represents the amplitude normalization processing, stands for Phase Decoding;

[0042] Step S55: extracting spectrum features, extracting spectrum features from the spectrum graph to obtain spectrum features;

[0043] Step S56: shunt feature fusion, using element-level multiplication and Reshape function to fuse the spectrum feature with the amplitude decoding to obtain the spectrum amplitude feature, and using element-level multiplication to fuse the spectrum amplitude feature with the phase decoding to obtain the denoising code:

[0044] ;

[0045] In the formula, stands for Denoising Coding;

[0046] Step S6: signal reconstruction, activating the denoising code to obtain a reconstructed spectrum map, and performing an inverse Fourier transform on the reconstructed spectrum map to obtain a reconstructed signal;

[0047] Step S7: loss calculation and optimization, loss evaluation and optimization of the actual denoising effect of the reconstructed signal.

[0048] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0049] (1) In view of the problem that existing signal noise suppression methods require a large amount of data for training or learning, resulting in serious signal denoising distortion, the present invention creatively designs an intelligent fiber optic sensor signal noise suppression method, constructs a spectrum diagram of the time-series fiber optic sensor signal, and performs amplitude encoding and phase encoding on the spectrum diagram respectively, exchanges information through cross-connection, improves the feature representation capability, enhances the recognition of data, and denoises and reconstructs the fiber optic sensor signal more accurately and efficiently;

[0050] (2) In view of the problem that the existing neural network methods have weak processing capabilities for low signal-to-noise ratios and poor denoising effects, the present invention creatively adopts dual-stream feature fusion coding based on frequency characteristic conversion, performs frequency characteristic transformation of inter-channel convolution features on amplitude features, obtains the global correlation of amplitudes in the spectrum graph, uses multiple convolutional layers to obtain the local correlation of amplitudes in the spectrum graph, and uses layer normalization and convolution coding to perform lightweight and accurate reconstruction of the phase in the spectrum graph. Through optimized amplitude coding and optimized phase coding, the low signal-to-noise ratio processing capability of the model is increased, and the denoising effect is improved;

[0051] (3) The present invention creatively designs a branch independent processing and information sharing mechanism to realize feature transmission through information sharing. Compared with the simple and direct amplitude and phase coding fusion method, each information sharing branch can obtain a more comprehensive and detailed feature representation, enhance the feature learning ability and signal processing accuracy, and make the denoised reconstructed signal more natural and coherent. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 A flow chart of an intelligent method for suppressing noise in optical fiber sensing signals provided by the present invention;

[0053] Figure 2 Flowchart of dual-stream feature fusion encoding based on frequency feature conversion.

[0054] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0055] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0056] Example 1, see Figure 1 The present invention provides an intelligent method for suppressing noise of optical fiber sensing signals, which specifically comprises the following steps:

[0057] Step S1: signal acquisition, collecting optical fiber sensing signals, wherein the optical fiber sensing signals are timing signals;

[0058] Step S2: Fourier transform, windowing the optical fiber sensing signal and performing Fourier transform to obtain a spectrum diagram;

[0059] Step S3: Amplitude coding, performing two-layer convolution activation processing on the spectrum graph to obtain amplitude coding, wherein the first layer of convolution activation includes a convolution layer Conv(1×7) and a ReLU activation function, and the second layer of convolution activation includes a convolution layer Conv(7×1) and a ReLU activation function;

[0060] Step S4: Phase encoding, performing two-layer convolution processing on the spectrum graph to obtain phase encoding, wherein both layers of convolution are convolution layers Conv(5×3);

[0061] Step S5: dual-stream characteristic fusion coding based on frequency characteristic conversion is performed on amplitude coding and phase coding to obtain denoising coding;

[0062] Step S6: signal reconstruction, activating the denoising code to obtain a reconstructed spectrum map, and performing an inverse Fourier transform on the reconstructed spectrum map to obtain a reconstructed signal;

[0063] Step S7: loss calculation and optimization, loss evaluation and optimization of the actual denoising effect of the reconstructed signal.

[0064] By performing the above operations, in view of the problem that the existing signal noise suppression methods require a large amount of data for training or learning, resulting in serious signal denoising distortion, the present invention creatively designs an intelligent fiber optic sensor signal noise suppression method, constructs a spectrum diagram of the time-series fiber optic sensor signal, and performs amplitude encoding and phase encoding on the spectrum diagram respectively, exchanges information through cross-connections, improves feature representation capabilities, enhances data recognition, and more accurately and efficiently denoises and reconstructs the fiber optic sensor signal.

[0065] Example 2, see Figure 2 This embodiment is based on the above embodiment, and step S5 specifically includes the following steps:

[0066] Step S51: frequency characteristic conversion based on the attention mechanism is performed on the amplitude coding to obtain optimized amplitude coding;

[0067] Step S52: Lightweight phase coding optimization, performing two-layer layer normalization and convolution processing on the phase coding to obtain lightweight phase coding:

[0068] ;

[0069] ;

[0070] In the formula, The representative layer is normalized. represents convolution processing, and the subsequent numbers are the convolution processing order identifiers. is an intermediate parameter, stands for lightweight phase encoding, stands for phase encoding;

[0071] Step S53: dual-stream information sharing, using element-level multiplication and Tanh activation to share information between optimized amplitude coding and lightweight phase coding, to obtain dual-stream optimized amplitude coding and dual-stream optimized phase coding:

[0072] ;

[0073] ;

[0074] In the formula, Represents Tanh activation function processing, Represents dual-stream optimized amplitude coding, stands for dual-stream optimized phase encoding;

[0075] Step S54: split-stream decoding, split-stream decoding is performed on the dual-stream optimized amplitude coding and the dual-stream optimized phase coding to obtain amplitude decoding and phase decoding:

[0076] ;

[0077] ;

[0078] ;

[0079] ;

[0080] ;

[0081] In the formula, Represents Sigmoid activation function processing, represents the fully connected layer processing, ReLU represents the ReLU activation function processing. , and is the intermediate parameter, represents amplitude decoding, represents the amplitude normalization processing, stands for Phase Decoding;

[0082] Step S55: extracting spectrum features, extracting spectrum features from the spectrum graph to obtain spectrum features;

[0083] Step S56: shunt feature fusion, using element-level multiplication and Reshape function to fuse the spectrum feature with the amplitude decoding to obtain the spectrum amplitude feature, and using element-level multiplication to fuse the spectrum amplitude feature with the phase decoding to obtain the denoising code:

[0084] ;

[0085] In the formula, Stands for Denoising Coding.

[0086] By performing the above operations, in order to solve the problem that the existing neural network methods have weak processing capabilities for low signal-to-noise ratios and poor denoising effects, the present invention creatively adopts dual-stream feature fusion coding based on frequency characteristic conversion, transforms the frequency characteristics of the amplitude features through inter-channel convolution features, obtains the global correlation of the amplitude in the spectrum graph, uses multiple convolutional layers to obtain the local correlation of the amplitude in the spectrum graph, and uses layer normalization and convolutional coding to perform lightweight and accurate reconstruction of the phase in the spectrum graph. Through optimized amplitude coding and optimized phase coding, the low signal-to-noise ratio processing capability of the model is increased, and the denoising effect is improved.

[0087] Embodiment 3: This embodiment is based on the above embodiment, and step S51 specifically includes the following steps:

[0088] Step S511: Perform two layers of reshape convolution processing on the amplitude code to obtain intermediate parameters. Each layer of reshape convolution operation includes convolution and Reshape function processing in sequence:

[0089] ;

[0090] In the formula, represents amplitude coding, represents convolution processing, Reshape function processing, is the intermediate parameter;

[0091] Step S512: feature fusion, perform element-wise multiplication of the intermediate parameters and the amplitude code, and perform full connection layer processing to obtain the amplitude fusion feature:

[0092] ;

[0093] In the formula, represents element-wise multiplication, represents the fully connected layer processing, represents the amplitude fusion feature;

[0094] Step S513: Attention mechanism processing, element-wise addition of the amplitude fusion feature and the amplitude coding, and convolution and activation processing are performed to obtain the preliminary optimized amplitude coding:

[0095] ;

[0096] In the formula, represents row-level addition, represents the activation function, represents preliminary optimization of amplitude coding;

[0097] Step S514: Perform convolution processing on the preliminary optimized amplitude code and perform two-layer convolution activation processing to obtain the secondary optimized amplitude code:

[0098] ;

[0099] ;

[0100] In the formula, represents the secondary optimized amplitude encoding, is the intermediate parameter;

[0101] Step S515: normalize and activate the secondary optimized amplitude code, and repeat steps S511 to S513 to obtain the optimized amplitude code:

[0102] ;

[0103] In the formula, It is an abbreviation of the process from step S511 to step S513. Represents standardized processing, Stands for Optimized Amplitude Coding.

[0104] By performing the above operations, the present invention creatively designs a branch independent processing and information sharing mechanism, and realizes the transmission of features through information sharing. Compared with the simple and direct amplitude and phase coding fusion method, each information sharing branch can obtain a more comprehensive and detailed feature representation, enhance the feature learning ability and signal processing accuracy, and make the denoised reconstructed signal more natural and coherent.

[0105] Embodiment 4: This embodiment is based on the above embodiment. In step S2, the spectrum diagram is constructed as follows:

[0106] Acquisition timing signal:

[0107] Collect the required optical fiber sensing timing signal data;

[0108] Time domain signal:

[0109] Represent a timing signal as a waveform in the time domain that changes over time;

[0110] Apply the window function:

[0111] Before Fourier transform, the window function is applied to the time series signal to perform windowing processing to reduce the influence of spectrum leakage;

[0112] Take the Fourier transform:

[0113] Apply fast Fourier transform to the time domain signal to convert the signal into a spectrum in the frequency domain;

[0114] Get the spectrum:

[0115] Extract the amplitude spectrum and phase spectrum from the Fourier transform results to indicate the distribution of the signal in the frequency domain;

[0116] Spectrum plotting:

[0117] The spectrum information is plotted as a spectrogram, with the horizontal axis representing the frequency and the vertical axis representing the amplitude of the signal.

[0118] Embodiment 5: This embodiment is based on the above embodiment. In step S6, the optimization details of the inverse transformation of Fourier transform are as follows:

[0119] Inverse Fourier transform:

[0120] For the spectrogram, we first need to perform an inverse Fourier transform to convert the spectrogram back into a time domain signal;

[0121] Denormalize the spectrogram:

[0122] If normalization was performed when making the spectrogram, an inverse operation is required to restore the spectrogram to the original spectrum values ​​before performing the inverse Fourier transform;

[0123] Inverse Fourier transform operation:

[0124] Perform an inverse Fourier transform operation on the spectrum graph to convert the spectrum graph into a time domain signal and restore the time domain representation of the original signal;

[0125] Reconstruct the time domain signal:

[0126] The result obtained by inverse Fourier transform is the reconstructed time domain signal, which represents the original signal waveform corresponding to the spectrum diagram;

[0127] Time domain signal post-processing:

[0128] The obtained time series signal is subjected to necessary post-processing to remove possible residual noise, smoothing, resampling, etc., to obtain a more accurate and smooth time series signal.

[0129] Embodiment 6. This embodiment is based on the above embodiment. Before performing step S2: Fourier transform, a VMD modal decomposition method can be used to optimize an intelligent optical fiber sensing signal noise suppression method provided by the present invention. The optimized steps are as follows:

[0130] Step Q1: signal acquisition, collecting optical fiber sensor signals, wherein the optical fiber sensor signals are time series signals, and using VMD modal decomposition method to decompose the time series signals to obtain K modal components, specifically including the following steps:

[0131] initialization:

[0132] Initialize the parameters related to the intrinsic mode function;

[0133] Iterative solution:

[0134] Iteratively solve each intrinsic mode function (IMF), and at each iteration, update the frequency bandwidth parameter according to the frequency bandwidth of the current signal;

[0135] Extract the modal components:

[0136] Get K intrinsic mode functions (IMFs);

[0137] Step Q2: Fourier transform, perform windowing processing on the K modal components and perform Fourier transform to obtain K sub-spectrum graphs;

[0138] Step Q3: Repeat steps S3 to S6 to denoise and reconstruct the K sub-spectrum graphs to obtain K reconstructed sub-signals;

[0139] Step Q4: Inverse modal decomposition, performing inverse modal decomposition on the K reconstructed sub-signals to obtain a reconstructed signal;

[0140] Step Q5: Loss calculation and optimization, loss evaluation and optimization of the actual denoising effect of the reconstructed signal.

[0141] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0142] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

[0143] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. An intelligent method for suppressing noise of optical fiber sensing signals, characterized in that: The specific steps include: Step S1: signal acquisition, collecting optical fiber sensing signals, wherein the optical fiber sensing signals are timing signals; Step S2: Fourier transform, windowing the optical fiber sensing signal and performing Fourier transform to obtain a spectrum diagram; Step S3: Amplitude coding, performing two-layer convolution activation processing on the spectrum graph to obtain amplitude coding, wherein the first layer of convolution activation includes a convolution layer Conv(1×7) and a ReLU activation function, and the second layer of convolution activation includes a convolution layer Conv(7×1) and a ReLU activation function; Step S4: Phase encoding, performing two-layer convolution processing on the spectrum graph to obtain phase encoding, wherein both layers of convolution are convolution layers Conv(5×3); Step S5: dual-stream characteristic fusion coding based on frequency characteristic conversion is performed on amplitude coding and phase coding to obtain denoising coding, which specifically includes the following steps: Step S51: frequency characteristic conversion based on the attention mechanism is performed on the amplitude coding to obtain optimized amplitude coding, which specifically includes the following steps: Step S511: Perform two layers of reshape convolution processing on the amplitude code to obtain intermediate parameters. Each layer of reshape convolution operation includes convolution and Reshape function processing in sequence: ； In the formula, represents amplitude coding, represents convolution processing, and the subsequent numbers are the convolution processing order identifiers. Reshape function processing, is the intermediate parameter; Step S512: feature fusion, perform element-wise multiplication of the intermediate parameters and the amplitude code, and perform full connection layer processing to obtain the amplitude fusion feature: ； In the formula, represents element-wise multiplication, represents the fully connected layer processing, represents the amplitude fusion feature; Step S513: Attention mechanism processing, element-wise addition of the amplitude fusion feature and the amplitude coding, and convolution and activation processing are performed to obtain the preliminary optimized amplitude coding: ； In the formula, represents row-level addition, represents the activation function, represents preliminary optimization of amplitude coding; Step S514: Perform convolution processing on the preliminary optimized amplitude code and perform two-layer convolution activation processing to obtain the secondary optimized amplitude code: ； ； In the formula, represents the secondary optimized amplitude encoding, is the intermediate parameter; Step S515: normalize and activate the secondary optimized amplitude code, take the secondary optimized amplitude code as input, repeat steps S511 to S513, and obtain the optimized amplitude code: ； In the formula, It is an abbreviation of the process from step S511 to step S513. Represents standardized processing, stands for Optimized Amplitude Coding; Step S52: Lightweight phase coding optimization, performing two-layer layer normalization and convolution processing on the phase coding to obtain lightweight phase coding: ； ； In the formula, The representative layer is normalized. represents convolution processing, and the subsequent numbers are the convolution processing order identifiers. is an intermediate parameter, stands for lightweight phase encoding, stands for phase encoding; Step S53: dual-stream information sharing, using element-level multiplication and Tanh activation to share information between optimized amplitude coding and lightweight phase coding, to obtain dual-stream optimized amplitude coding and dual-stream optimized phase coding: ； ； In the formula, Represents Tanh activation function processing, Represents dual-stream optimized amplitude coding, stands for Dual-Stream Optimized Phase Coding; Step S54: split-stream decoding, split-stream decoding is performed on the dual-stream optimized amplitude coding and the dual-stream optimized phase coding to obtain amplitude decoding and phase decoding: ； ； ； ； ； In the formula, Represents Sigmoid activation function processing, represents the fully connected layer processing, ReLU activation function processing, , and is the intermediate parameter, represents amplitude decoding, represents the amplitude normalization processing, stands for Phase Decoding; Step S55: Extract spectrum features, extract spectrum features from the spectrum graph, and obtain spectrum features ; Step S56: shunt feature fusion, using element-level multiplication and Reshape function to fuse the spectrum feature with the amplitude decoding to obtain the spectrum amplitude feature, and using element-level multiplication to fuse the spectrum amplitude feature with the phase decoding to obtain the denoising code: ； In the formula, stands for Denoising Coding; Step S6: signal reconstruction, activating the denoising code to obtain a reconstructed spectrum map, and performing an inverse Fourier transform on the reconstructed spectrum map to obtain a reconstructed signal; Step S7: loss calculation and optimization, loss evaluation and optimization of the actual denoising effect of the reconstructed signal.

Images