Data noise reduction and signal detection method, device, system and storage medium

Abstract

The present application discloses a data denoising and signal detection method, device, system and storage medium, relating to the field of data and signal processing technology, including: generating simulated data with a low signal-to-noise ratio; preprocessing the mixed signals in the simulated data to obtain training data; fusing a first convolutional neural network, a second convolutional neural network, a self-attention mechanism and a multi-layer fully connected neural network to obtain a fused deep neural network; step-by-step training of the fused deep neural network using the training data to obtain a target model with a denoising function; inputting real data with a low signal-to-noise ratio into the target model so that the target model performs denoising and signal detection on the real data, thereby obtaining a processing result of the denoising and signal detection. The present application can accurately perform denoising and detection processing on low signal-to-noise ratio data.

Description

Technical Field

[0001] The present application relates to the field of data and signal processing technology, and in particular to a data noise reduction and signal detection method, device, system and storage medium. Background Art

[0002] In some application scenarios, such as deep space communications, gravitational wave detection, and seismic exploration, the problem of processing low signal-to-noise ratio data is often faced. In the received signal, because the noise intensity is much higher than the signal corresponding to the valid information, an effective method is needed to reduce the noise of the data. General data signal processing methods often use matched filtering and nonlinear filters. When using such methods to process low signal-to-noise ratio data, there is a problem of poor noise reduction effect, making it difficult to accurately reduce the noise of the data and detect potential signals therein. Therefore, how to process low signal-to-noise ratio data has become a technical problem that needs to be solved urgently. Summary of the Invention

[0003] The present application aims to solve at least one of the technical problems existing in the prior art. To this end, the present application proposes a data denoising and signal detection method, device, system and storage medium that can accurately denoise and detect low signal-to-noise ratio data.

[0004] According to the first aspect of the present application, the data noise reduction and signal detection method includes:

[0005] generating analog data with a low signal-to-noise ratio, wherein the analog data includes a plurality of mixed signals, the mixed signals being obtained by mixing noise and pure signals, the pure signals representing noise-free signals;

[0006] Preprocessing the mixed signal in the simulation data to obtain training data;

[0007] The first convolutional neural network, the second convolutional neural network, the self-attention mechanism and the multi-layer fully connected neural network are integrated to obtain a fused deep neural network;

[0008] The fused deep neural network is trained step by step using the training data to obtain a target model with noise reduction function;

[0009] Real data with a low signal-to-noise ratio is input into the target model, so that the target model performs noise reduction and signal detection on the real data to obtain processing results of the noise reduction and signal detection.

[0010] According to the data denoising and signal detection method of the embodiment of the present application, there are at least the following beneficial effects: first, low signal-to-noise ratio simulation data is generated; secondly, the mixed signal in the simulation data is preprocessed to obtain training data; then, the first convolutional neural network, the second convolutional neural network, the self-attention mechanism and the multi-layer fully connected neural network are fused to obtain a fused deep neural network; then, the fused deep neural network is step-by-step trained through the training data to obtain a target model with a denoising function; finally, the real data with a low signal-to-noise ratio is input into the target model so that the target model performs denoising and signal detection on the real data to obtain the processing results of denoising and signal detection. The data denoising and signal detection method of the present application, on the one hand, can improve the denoising and detection accuracy of the target model when processing real data with a low signal-to-noise ratio through sufficient training based on simulation data, and does not require frequent calculations, thereby improving the speed of denoising and detection; on the other hand, by constructing a fused deep neural network, it can process ultra-long data and extract signal features of real data with a low signal-to-noise ratio from different dimensions, thereby achieving higher denoising and detection accuracy. Therefore, the data denoising and signal detection method of the present application can accurately perform denoising and detection processing on low signal-to-noise ratio data.

[0011] According to some embodiments of the present application, generating analog data with a low signal-to-noise ratio includes:

[0012] Simulate noise generation based on the sensitivity curve of the gravitational wave detector;

[0013] Select corresponding parameter ranges for different types of gravitational wave sources;

[0014] generating, according to the parameter range, a plurality of pure signals corresponding to the plurality of gravitational wave sources, wherein the gravitational wave sources and the pure signals correspond one to one;

[0015] Each of the clean signals is mixed with the noise to obtain analog data including a plurality of mixed signals.

[0016] According to some embodiments of the present application, preprocessing the mixed signal in the analog data includes:

[0017] performing whitening processing on the mixed signal in the analog data so that a noise spectrum of the mixed signal is uniformly distributed;

[0018] Normalization is performed on the mixed signal after the whitening process, so that all data in the simulation data have the same scale.

[0019] According to some embodiments of the present application, the first convolutional neural network, the second convolutional neural network, the self-attention mechanism, and the multi-layer fully connected neural network are integrated to obtain a fused deep neural network, including:

[0020] Building the first convolutional neural network to extract feature information from the mixed signal;

[0021] Building a fusion network of the self-attention mechanism and the second convolutional neural network to perform noise reduction on the mixed signal;

[0022] Build the multi-layer fully connected neural network to detect target signals.

[0023] According to some embodiments of the present application, the step-by-step training of the fused deep neural network using the training data to obtain a target model with a noise reduction function includes:

[0024] Using the mean square error between the mixed signal and the pure signal as a loss function, and performing initial training on the fused deep neural network using the loss function;

[0025] Setting an alternating cycle number, wherein the alternating cycle number includes a first training cycle number and a second training cycle number, wherein the first training cycle number and the second training cycle number each include multiple training cycles, and each training cycle represents a process in which all samples in the training data are trained once;

[0026] Performing noise reduction training on the fused deep neural network during the first number of training cycles, and performing detection training on the fused deep neural network during the second number of training cycles;

[0027] Based on the number of alternating cycles, performing noise reduction training and detection training alternately;

[0028] An optimizer is used to adjust the parameters of the fused deep neural network, and a learning rate adjustment strategy is used to adjust the learning rate of the fused deep neural network.

[0029] According to some embodiments of the present application, performing noise reduction training on the fused deep neural network within the first number of training cycles and performing detection training on the fused deep neural network within the second number of training cycles include:

[0030] Performing noise reduction training on the fused deep neural network during the first number of training cycles to minimize a mean square error between the mixed signal and the clean signal;

[0031] Performing exploration training on the fused deep neural network during the second number of training cycles to minimize a cross entropy loss for binary classification, wherein the number of cycles in the first number of training cycles is equal to the number of cycles in the second number of training cycles.

[0032] According to some embodiments of the present application, the data noise reduction and signal detection method further includes:

[0033] The processing results are evaluated to determine the noise reduction accuracy and detection accuracy of the target model.

[0034] According to the second aspect of the present application, the data noise reduction and signal detection device includes:

[0035] A generating module, configured to generate analog data with a low signal-to-noise ratio, wherein the analog data includes a plurality of mixed signals, the mixed signals being obtained by mixing noise and pure signals, the pure signals being noise-free signals;

[0036] a processing module, configured to pre-process the mixed signal in the analog data to obtain training data;

[0037] A fusion module is used to fuse the first convolutional neural network, the second convolutional neural network, the self-attention mechanism, and the multi-layer fully connected neural network to obtain a fused deep neural network;

[0038] A training module, configured to perform step-by-step training on the fused deep neural network using the training data to obtain a target model with a noise reduction function;

[0039] The noise reduction and detection module is used to input real data with low signal-to-noise ratio into the target model so that the target model performs noise reduction and signal detection on the real data to obtain processing results of noise reduction and signal detection.

[0040] According to the third aspect of the present application, the data noise reduction and signal detection system includes:

[0041] at least one memory;

[0042] at least one processor;

[0043] at least one program;

[0044] The program is stored in the memory, and the processor executes at least one of the programs to implement the data noise reduction and signal detection method as described in the embodiment of the first aspect.

[0045] According to the computer-readable storage medium of the fourth aspect embodiment of the present application, the computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute the data noise reduction and signal detection method as described in the first aspect embodiment.

[0046] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become obvious from the description below, or will be learned through practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The present application is further described below with reference to the accompanying drawings and embodiments, wherein:

[0048] Figure 1 A flow chart of the data noise reduction and signal detection method provided in an embodiment of the present application;

[0049] Figure 2 A waveform diagram of low signal-to-noise ratio data provided in an embodiment of the present application;

[0050] Figure 3 A schematic diagram of a model of a fusion deep neural network provided in an embodiment of the present application;

[0051] Figure 4 A schematic diagram of the structure of the data noise reduction and signal detection device provided in an embodiment of the present application;

[0052] Figure 5 This is a structural diagram of the data noise reduction and signal detection system provided in an embodiment of the present application.

[0053] Reference numerals:

[0054] Generation module 100 , processing module 110 , fusion module 120 , training module 130 , noise reduction and detection module 140 , memory 200 , processor 300 . DETAILED DESCRIPTION

[0055] The following describes in detail embodiments of the present application. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application and are not to be construed as limiting the present application.

[0056] It should be noted that although the system diagrams illustrate functional module divisions and the flowcharts illustrate logical sequences, in certain circumstances, the steps shown or described may be performed in a sequence that differs from the module divisions in the system or the sequence in the flowcharts. Terms used in the specification, claims, and drawings are intended to distinguish similar items and are not necessarily intended to describe a specific sequence or chronological order.

[0057] In the description of this application, "several" means more than one, "plurality" means more than two, "greater than," "less than," and "exceed" are understood to exclude the number itself, while "above," "below," and "within" are understood to include the number itself. The use of "first" and "second" in the description is solely for the purpose of distinguishing technical features and should not be construed as indicating or implying relative importance, implicitly specifying the number of the indicated technical features, or implicitly specifying the order of the indicated technical features.

[0058] In the description of this application, unless otherwise clearly defined, terms such as setting, installing, and connecting should be understood in a broad sense, and technicians in the relevant technical field can reasonably determine the specific meanings of the above terms in this application based on the specific content of the technical solution.

[0059] In the description of this application, reference to the terms "one embodiment," "some embodiments," "illustrative embodiments," "examples," "specific examples," or "some examples" means that the specific features, structures, materials, or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of this application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in any appropriate manner in any one or more embodiments or examples.

[0060] Glossary:

[0061] Low signal-to-noise ratio: refers to a signal strength below 0dB.

[0062] Low signal-to-noise ratio data: usually refers to data with a signal strength below 0dB, where the signal is completely submerged by noise;

[0063] Noise reduction: suppressing noise in data.

[0064] Signal detection: Determine whether there are signals of interest in the data, such as gravitational wave signals, communication signals, seismic wave signals, etc.

[0065] Deep space communication: refers to the communication between communication entities on Earth and spacecraft that leave the Earth's satellite orbit and enter the solar system. The distance can reach millions of kilometers, tens of millions of kilometers, or even hundreds of millions of kilometers.

[0066] Gravitational waves: Gravitational waves are ripples in space-time caused by accelerating mass.

[0067] A gravitational-wave observatory is a device used in gravitational-wave astronomy to detect gravitational waves. Detecting gravitational waves allows for experimental verification of general relativity. Commonly used detectors include rod-shaped detectors and laser interferometers. These detectors operate by measuring the effect of a passing gravitational wave on the distance between two distant locations. Since the 1960s, numerous gravitational-wave observatories have been built and commissioned, with continuous improvements in their sensitivity. Today, these detectors are capable of detecting gravitational wave sources both within and beyond the Milky Way, making them the primary detection tools in gravitational-wave astronomy.

[0068] Seismic exploration is a geophysical exploration method that uses artificially excited elastic waves to exploit differences in the elasticity and density of underground media. By observing and analyzing the propagation patterns of these waves underground, the properties and morphology of subsurface rock formations can be inferred. Seismic exploration is the most important and effective method in geophysical exploration for solving oil and gas exploration problems. It is a crucial tool for pre-drilling oil and gas resource surveys and is also widely used in coalfield and engineering geological surveys, regional geological studies, and crustal research.

[0069] Scientific applications such as deep space communications, gravitational wave detection, and seismic exploration often face the challenge of processing data with low signal-to-noise ratios (SNRs). Because signals are often overwhelmed by noise, traditional signal processing methods, including matched filtering and nonlinear filtering, suffer from poor noise reduction and low computational efficiency when processing low-SNR data. This makes it impossible to accurately reduce noise in the data and detect potential signals.

[0070] Specifically, processing data with low signal-to-noise ratios has always been a research challenge in science, especially in physics and communications. Because the noise intensity is much higher than the signal, effective methods are needed to reduce the noise in the data and detect whether there are useful or interesting signals in the data. Current methods have the following main drawbacks:

[0071] Methods such as nonlinear filters have slow calculation speeds, low noise reduction and detection accuracy, and cannot produce accurate results;

[0072] The current neural network algorithm cannot process extremely long data, and the noise reduction and detection speed and accuracy need to be improved.

[0073] Based on the above, the present application proposes a data noise reduction and signal detection method, which can effectively improve the noise reduction and detection accuracy, and can simultaneously achieve data noise reduction and signal detection.

[0074] Below, according to Figure 1-3 The data noise reduction and signal detection method according to the embodiment of the present application is described.

[0075] It is understandable that if Figure 1 As shown, a data noise reduction and signal detection method is provided, comprising:

[0076] Step S100, generating analog data with a low signal-to-noise ratio, wherein the analog data includes a plurality of mixed signals, the mixed signals being obtained by mixing noise and pure signals, and the pure signals being noise-free signals;

[0077] Step S110, preprocessing the mixed signal in the simulation data to obtain training data;

[0078] Step S120, fusing the first convolutional neural network, the second convolutional neural network, the self-attention mechanism, and the multi-layer fully connected neural network to obtain a fused deep neural network;

[0079] Step S130, training the fused deep neural network step by step using the training data to obtain a target model with noise reduction function;

[0080] In step S140 , real data with a low signal-to-noise ratio is input into the target model, so that the target model performs noise reduction and signal detection on the real data to obtain processing results of the noise reduction and signal detection.

[0081] First, generate low signal-to-noise ratio simulation data, wherein the simulation data includes multiple mixed signals, the mixed signals are obtained by mixing noise and pure signals, and the pure signals represent noise-free signals; secondly, preprocess the mixed signals in the simulation data to obtain training data; then, fuse the first convolutional neural network, the second convolutional neural network, the self-attention mechanism and the multi-layer fully connected neural network to obtain a fused deep neural network; then, train the fused deep neural network step by step through the training data to obtain a target model with noise reduction function; finally, input the real data with low signal-to-noise ratio into the target model so that the target model performs noise reduction and signal detection on the real data to obtain the processing results of noise reduction and signal detection. The data denoising and signal detection method of the present application, on the one hand, can improve the noise reduction and detection accuracy of the target model when processing real data with low signal-to-noise ratio through sufficient training based on simulation data, and does not require frequent calculations, thereby improving the speed of noise reduction and detection; on the other hand, by constructing a fused deep neural network, it can process ultra-long data and extract signal features of real data with low signal-to-noise ratio from different dimensions, thereby achieving higher noise reduction and detection accuracy. Therefore, the data denoising and signal detection method of the present application can accurately perform denoising and detection processing on low signal-to-noise ratio data.

[0082] Understandably, simulation data with low signal-to-noise ratios are generated, including:

[0083] Simulate noise generation based on the sensitivity curve of the gravitational wave detector;

[0084] Select corresponding parameter ranges for different types of gravitational wave sources;

[0085] generating, according to the parameter range, a plurality of pure signals corresponding to the plurality of gravitational wave sources, wherein the gravitational wave sources correspond to the pure signals in a one-to-one manner;

[0086] Each clean signal is mixed with noise respectively to obtain analog data including multiple mixed signals.

[0087] It should be noted that in the gravitational wave detection scenario, low signal-to-noise ratio simulated data can be generated by the following steps:

[0088] Step S101: simulating and generating noise according to the sensitivity curve of the gravitational wave detector;

[0089] Step S102: selecting appropriate parameter ranges for different types of gravitational wave sources to generate corresponding signals; wherein the gravitational wave source can be an extreme mass ratio precession system, a supermassive binary black hole, a binary white dwarf, or a random gravitational wave background;

[0090] Step S103: Mix the data generated in step S101 and step S102 to obtain data with different signal-to-noise ratios. In this step, it is crucial to set a specific signal-to-noise ratio (SNR) according to the following formula.

[0091]

[0092] Here, s|s is the inner product, which can be a|b and is calculated by the following formula:

[0093]

[0094] Where * represents complex conjugate, S n is the sensitivity curve.

[0095] It should be noted that in the field of deep space communication, noise can also be simulated and generated based on the signal curve of communication between the communication entity and the earth satellite; in the field of seismic exploration, noise can be simulated and generated based on the curve corresponding to the seismic wave.

[0096] It is worth emphasizing that this application uses the matched filter signal-to-noise ratio, and when simulating the actual extremely low signal-to-noise ratio environment, the signal-to-noise ratio of the training data is deliberately set to 50, which is approximately equal to -40dB. After the above steps, a large amount of low signal-to-noise ratio simulation data is obtained. Among them, the simulation data sample example is as follows Figure 2 As shown, in Figure 2 In the figure, orange is pure signal and blue is signal + noise.

[0097] It is understandable that preprocessing of mixed signals in analog data includes:

[0098] Performing whitening on the mixed signal in the simulation data to make the noise spectrum of the mixed signal uniformly distributed;

[0099] The mixed signal after whitening is normalized so that all data in the simulation data have the same scale.

[0100] It should be noted that the data preprocessing part mainly consists of two steps: whitening and normalization, which specifically include the following steps:

[0101] Step S111: Data whitening. The purpose of whitening is to remove redundancy from the input data and improve its statistical properties. The specific calculation method is as follows: the noise spectrum of the original data is converted into a uniformly distributed form, so that all frequency components have a uniform distribution on the power spectrum. This can effectively reduce the complexity of the noise and provide more homogeneous data for the subsequent fusion deep neural network.

[0102]

[0103] Step S112: Normalize the whitened data. Normalization ensures that the data has the same scale in the entire data set, making the training of the fusion deep neural network more stable. In this step, the simulated data is adjusted to [-1, 1] to avoid numerical overflow, inaccurate calculations, or instability in the learning process due to excessively large or small values.

[0104] It can be understood that the first convolutional neural network, the second convolutional neural network, the self-attention mechanism and the multi-layer fully connected neural network are integrated to obtain a fused deep neural network, including:

[0105] Build the first convolutional neural network to extract feature information from the mixed signal;

[0106] Build a fusion network of the self-attention mechanism and the second convolutional neural network to perform noise reduction on the mixed signal;

[0107] Build a multi-layer fully connected neural network to detect target signals.

[0108] It should be noted that in the model construction, a fusion deep neural network algorithm is provided, which is mainly composed of a convolutional neural network (CNN), a self-attention mechanism and a multi-layer fully connected neural network (MLP); the fusion deep neural network model constructed in this application is an end-to-end design, which can directly reduce noise and detect the input low signal-to-noise ratio data at the same time; this design avoids manual feature selection and threshold setting, and realizes automatic processing and analysis of data.

[0109] in, Figure 3 An example of the overall structure of a model is given.

[0110] The specific steps for building a fusion deep neural network model are as follows:

[0111] Step S121: Build the first convolutional neural network. Use the convolutional neural network to perform feature extraction. The data obtained in step S100 is first passed through the CNN network. CNN has excellent performance in processing data such as images, speech, and time series. In the model of this application, the role of CNN is to extract useful feature information from input data with low signal-to-noise ratio. More importantly, through the stacking of convolutional layers, the model has a larger receptive field and can process very long sequence data, thereby enhancing the accuracy of noise reduction and signal detection.

[0112] Step S122: Build a self-attention mechanism and a second convolutional neural network: This application designs an encoder module that combines a self-attention mechanism with a convolutional neural network, specifically for noise reduction. The self-attention mechanism works as shown in the following equation, where Q, K, and V are trainable parameters. This mechanism helps the neural network capture long-range dependencies in the sequence, while the CNN focuses on extracting local features. This unique combination enables the fused deep neural network to consider both global and local information when handling noise reduction tasks.

[0113]

[0114]

[0115] Step S123: Build a multi-layer fully connected neural network. The multi-layer fully connected neural network is used as a classifier in the fusion deep neural network model to detect target signals. The MLP has the ability to learn nonlinear mapping relationships and can accurately detect potential target signals from the noise-reduced data. The classifier formula is as follows:

[0116]

[0117] Among them, ω ij is the training parameter of the classifier, x jis the output of step S123, and logistic is the logistic regression function.

[0118] It is understandable that the fusion deep neural network is trained step by step through the training data to obtain the target model with noise reduction function, including:

[0119] The mean square error between the mixed signal and the pure signal is used as the loss function to perform initial training on the fusion deep neural network;

[0120] Setting the number of alternating cycles, wherein the number of alternating cycles includes a first training cycle number and a second training cycle number, each of the first training cycle number and the second training cycle number includes multiple training cycles, and each training cycle represents a process of training all samples in the training data once;

[0121] Performing noise reduction training on the fused deep neural network during a first training cycle, and performing detection training on the fused deep neural network during a second training cycle;

[0122] Based on the number of alternating cycles, noise reduction training and detection training are performed alternately;

[0123] Use the optimizer to adjust the parameters of the fused deep neural network, and use the learning rate adjustment strategy to adjust the learning rate of the fused deep neural network.

[0124] It is understandable that performing noise reduction training on the fused deep neural network in the first training cycle number and performing detection training on the fused deep neural network in the second training cycle number include:

[0125] Performing denoising training on the fused deep neural network during a first training epoch to minimize the mean square error between the mixed signal and the pure signal;

[0126] Performing exploration training on the fused deep neural network during a second number of training epochs to minimize a cross entropy loss for binary classification, wherein the number of epochs in the first number of training epochs is equal to the number of epochs in the second number of training epochs.

[0127] It should be noted that a unique step-by-step training strategy was introduced during the model training phase. This strategy involves alternating training of different loss functions to achieve both signal denoising and detection. The specific steps are as follows:

[0128] Step S131: Initialization phase. First, in the initialization phase, the mean squared error between the noisy mixed signal and the pure signal without noise is used as the loss function. The main goal is to enable the network to learn to reduce noise. Training in this phase helps the model to initially understand and extract valuable signal features, while also preventing the classifier from capturing too many noise features.

[0129] Step S132: Alternating training: After the initial training is completed, a phased alternating training strategy is adopted, which is carried out every 6 epochs (an epoch represents the process of training once using all samples in the training set) (the following two steps ab); in this training strategy, the model will alternate between noise reduction training and detection training, so that the two tasks can complement each other and improve the overall performance of the model. The specific steps are as follows:

[0130] a) Noise reduction training: In the first three epochs of each alternating cycle, training is focused on the noise reduction task. The training goal of the model in this stage is to minimize the mean square error between the mixed signal and the pure signal, enabling the model to better learn how to reduce noise on the signal.

[0131] b) Detection training: In the last three epochs of each alternating cycle, the model is trained on the detection task. The training goal of the model in this stage is to minimize the cross entropy loss of the binary classification and improve the model's detection accuracy.

[0132] Step S133: Optimizer and learning rate adjustment; During the model training process, the Adam optimizer is used to adjust the network parameters, the initial learning rate is 0.01, and the total training rounds are 100 epochs; At the same time, a learning rate adjustment strategy is introduced. When the training loss stagnates, the learning rate is appropriately reduced to ensure that the model can continue to learn effectively; Through the above steps, the entire model training process not only improves the learning efficiency of the network, but also accelerates the convergence speed of the network, so that the model can achieve ideal performance faster.

[0133] It is understandable that the data noise reduction and signal detection method also includes:

[0134] The processing results are evaluated to determine the noise reduction accuracy and detection accuracy of the target model.

[0135] In this paper, model inference is a step to determine whether the model meets the requirements. In the model inference stage, noise reduction and detection operations are the core steps. The specific process is as follows:

[0136] Step S141: Noise reduction and detection: The low signal-to-noise ratio data to be processed is input into the model trained in step S130. The model will automatically perform noise reduction and signal detection tasks. First, the noise reduction network part uses a convolutional neural network and a self-attention mechanism to perform feature extraction and noise reduction processing. Then, signal detection is performed through a pre-trained fully connected network.

[0137] Step S142: Result evaluation: Evaluate the noise-reduced data and signal detection results output in step S141. The evaluation method mainly depends on the specific application scenario.

[0138] For example, in gravitational wave applications, noise reduction accuracy is evaluated by calculating the overlap between the output noise reduction signal and the true signal, and detection accuracy is evaluated using the ROC curve;

[0139]

[0140] in:

[0141]

[0142] In the above two formulas, h is the network output, s is the real signal, and the overlap value is between [0, 1]. The larger the value, the higher the matching degree and the higher the noise reduction accuracy.

[0143] The data noise reduction and signal detection method of the present application will be further described below in conjunction with the above embodiments.

[0144] It's important to note that traditional signal denoising methods, such as wavelet denoising and modulus maximum denoising, have low generalization capabilities and require certain prior knowledge, such as determining the wavelet transform threshold and selecting the wavelet. Deep learning, based on learning data representations, can effectively extract key features hidden in the data and automatically filter out the effects of interfering noise.

[0145] In deep learning, typical neural networks include convolutional neural networks, recurrent neural networks, and fully connected neural networks. Recurrent neural networks are neural networks whose nodes are connected in a loop. They exhibit dynamic temporal behavior well and utilize internal memory cells to process input sequences of arbitrary temporal order, addressing temporal problems such as natural language processing and speech recognition.

[0146] Long short-term memory (LSTM) networks are a special type of recurrent neural network. Based on recurrent neural networks, they add input, forget, and output thresholds, allowing weights to change during the self-loop. Furthermore, the integration scale at different time points can be dynamically adjusted, cleverly avoiding the problems of vanishing or expanding gradients that arise during the loop.

[0147] Specifically, the wavelet transform is a frequency analysis method for radio signals. Compared to traditional Fourier transform analysis, the wavelet transform offers advantages such as time-domain localization and frequency-domain localization. It can decompose signals at different scales and preserve the characteristics of signals under different modulation types. Signals exhibit a certain degree of continuity in space or time, resulting in larger wavelet coefficients for valid signals in the wavelet domain. Noise signals, on the other hand, are generally discrete in space or time, resulting in smaller wavelet coefficients in the wavelet domain. This property can be exploited to reduce signal noise using the wavelet transform. The original input signal is subjected to wavelet decomposition to calculate different wavelet coefficients. Assuming the noise signal follows a Gaussian distribution, the vast majority of noise coefficients will fall within a certain range. Setting the coefficients within this range to zero maximizes noise suppression. Using thresholded wavelet coefficients, the radio signal is reconstructed to obtain a de-noised signal.

[0148] Specifically, an autoencoder based on a convolutional neural network generally consists of a three-layer network, including an input layer, a hidden layer, and an output layer, where the number of neurons in the input layer and the output layer is equal. During the training process, for each input sample, the autoencoder will generate an output sample of the same size. The optimization goal of autoencoder training is to make the output sample as close as possible to the input sample. Based on the autoencoder, this application proposes a denoising autoencoder (DAE), which adds a noise signal to the input data, so that the trained autoencoder has a noise reduction function and is more robust, thereby improving the generalization ability of the model.

[0149] Existing radio signal recognition methods based on deep learning have good recognition performance for large amounts of radio signal data, but their recognition accuracy remains low in areas with low signal-to-noise ratios. To address this issue, this paper proposes a data denoising and signal detection method based on deep learning.

[0150] The data denoising and signal detection method of the present application is implemented by setting a low signal-to-noise ratio classifier, a denoising autoencoder and a recognition network, wherein the low signal-to-noise ratio classifier is essentially a binary classifier that can identify high signal-to-noise ratio signals and low signal-to-noise ratio signals in the signal-to-noise ratio by setting different signal-to-noise ratio thresholds. The denoising autoencoder can achieve denoising processing for low signal-to-noise ratio radio signals. The modulation type recognition network is composed of a long short-term memory network (LSTM network) that can identify the modulation type of the input radio signal.

[0151] Specifically, inspired by the long short-term memory neural network, this application designs a radio signal modulation type recognition model based on the long short-term memory network, including three layers: LSTM (128), LSTM (32), and FC (11), wherein the size of the original input signal is len*2, len represents the number of sampling nodes, and 2 represents the time dimension of a certain sampling node. LSTM (128) represents mapping the time dimension of the input data to a feature space of size len*128. FC (11) represents a fully connected network, mapping the input data to 11 distribution areas, where 11 is determined by the modulation type of the training data set. The LSTM layer uses tanh as the activation function and dropout is 0.8. The fully connected layer uses softmax as the activation function, cross entropy is used as the loss function for training, Adam with a learning rate of 0.001 is selected as the optimizer, batchsize is 64, and epoch is 20. Top_one is selected as the evaluation index of the model, that is, the model recognition is correct only when the class label corresponding to the highest confidence is the correct class label.

[0152] A long short-term memory (LSTM) network is used to extract characteristic information from radio signal data and classify different signal types based on this characteristic information. The LSTM-based radio signal recognition model's noise reduction effectiveness is determined by comparing its accuracy before and after noise reduction.

[0153] Specifically, for a low signal-to-noise ratio classifier, the recognition model has a good classification effect in the high signal-to-noise ratio area, but the recognition accuracy in the low signal-to-noise ratio interval is extremely low. Therefore, it is necessary to perform noise reduction processing on the radio signals in the low signal-to-noise ratio interval to improve the accuracy of the recognition model in the recognition interval. To perform noise reduction processing on low signal-to-noise ratio signals, it is first necessary to extract low signal-to-noise ratio signals from the signal data. In order to more accurately screen out low signal-to-noise ratio signals, this application designs an LSTM-based radio signal classifier for noise reduction processing of low signal-to-noise ratio signals.

[0154] For example, the classifier can be one with an input signal size of len*2, where len represents the number of sampling nodes and 2 represents the time dimension of a sampling node. LSTM(32) represents mapping the time dimension of the input data to a feature space of size len*32. The LSTM layer uses tanh as the activation function and dropout is 0.8. The FC layer is a fully connected layer and uses softmax as the activation function. Cross entropy is used as the loss function for training, Adam with a learning rate of 0.008 is selected as the optimizer, batch size is 64, and epochs are 20.

[0155] The low signal-to-noise ratio classification network can be used to map radio signals into two intervals: high signal-to-noise ratio and low signal-to-noise ratio, thereby realizing the LSTM-based radio signal binary classification task.

[0156] This application proposes a radio signal denoising and reconstruction model based on autoencoding technology. This model adds Gaussian-distributed white noise to high-SNR signals. These signals containing additive noise and the original high-SNR signals are then fed into the denoising model for training. The trained denoising model is then used to perform denoising and reconstruction of low-SNR signals.

[0157] In general, this application addresses the problem of low recognition accuracy of existing models in low signal-to-noise ratio areas. By combining autoencoder technology, a data denoising and signal detection method using autoencoder technology is proposed. This method can, to a certain extent, improve the recognition accuracy of the modulation type recognition model based on deep learning in the low signal-to-noise ratio range and make more effective use of limited signal resources.

[0158] This application first uses LSTM to construct a radio signal type recognition model to identify the types of radio signals with different signal-to-noise ratios, and uses a low signal-to-noise ratio classifier to implement the LSTM-based radio signal binary classification task. Then, based on autoencoding technology, a radio signal denoising and reconstruction model is constructed to achieve denoising and reconstruction of low signal-to-noise ratio radio signals. Finally, the type recognition model is retrained using the denoised and reconstructed data, and then tested.

[0159] It is understandable that if Figure 4 As shown, the present application also provides a data noise reduction and signal detection device, comprising:

[0160] A generating module 100 is configured to generate analog data with a low signal-to-noise ratio, wherein the analog data includes a plurality of mixed signals, the mixed signals being obtained by mixing noise and pure signals, wherein the pure signals represent noise-free signals;

[0161] The processing module 110 is used to pre-process the mixed signal in the analog data to obtain training data;

[0162] A fusion module 120 is configured to fuse the first convolutional neural network, the second convolutional neural network, the self-attention mechanism, and the multi-layer fully connected neural network to obtain a fused deep neural network;

[0163] A training module 130 is used to train the fused deep neural network in steps using training data to obtain a target model with noise reduction function;

[0164] The noise reduction and detection module 140 is used to input real data with a low signal-to-noise ratio into the target model, so that the target model performs noise reduction and signal detection on the real data to obtain processing results of the noise reduction and signal detection.

[0165] It should be noted that the generation module 100 includes:

[0166] A simulation module, used to simulate and generate noise according to the sensitivity curve of the gravitational wave detector;

[0167] A selection module is used to select the corresponding parameter ranges for different types of gravitational wave sources;

[0168] a signal module, configured to generate, based on a parameter range, a plurality of pure signals corresponding to a plurality of gravitational wave sources, wherein the gravitational wave sources correspond to the pure signals in a one-to-one manner;

[0169] The mixing module is used to mix each pure signal with noise respectively to obtain analog data including multiple mixed signals.

[0170] It should be noted that the processing module 110 includes:

[0171] A whitening processing module is used to perform whitening processing on the mixed signal in the analog data so that the noise spectrum of the mixed signal is evenly distributed;

[0172] The normalization processing module is used to perform normalization processing on the mixed signal after the whitening processing so that all data in the simulation data have the same scale.

[0173] It should be noted that the fusion module 120 includes:

[0174] A first building module is used to build a first convolutional neural network to extract feature information from the mixed signal;

[0175] The second building module is used to build a fusion network of the self-attention mechanism and the second convolutional neural network to perform noise reduction on the mixed signal;

[0176] The third building module is used to build a multi-layer fully connected neural network to detect target signals.

[0177] It should be noted that the training module 130 includes:

[0178] The initial training module is used to perform initial training on the fusion deep neural network by using the mean square error between the mixed signal and the pure signal as the loss function;

[0179] a setting module, configured to set the number of alternating cycles, wherein the number of alternating cycles includes a first training cycle number and a second training cycle number, wherein the first training cycle number and the second training cycle number each include a plurality of training cycles, and each training cycle represents a process in which all samples in the training data are trained once;

[0180] a noise reduction and detection training module, configured to perform noise reduction training on the fused deep neural network during a first number of training cycles, and to perform detection training on the fused deep neural network during a second number of training cycles;

[0181] An alternating training module, used to perform noise reduction training and detection training alternately based on the number of alternating cycles;

[0182] The adjustment module is used to adjust the parameters of the fused deep neural network using an optimizer, and to adjust the learning rate of the fused deep neural network using a learning rate adjustment strategy.

[0183] It should be noted that the noise reduction and detection training module includes:

[0184] a first minimization module for performing noise reduction training on the fused deep neural network for a first number of training cycles to minimize a mean square error between a mixed signal and a pure signal;

[0185] A second minimization module is configured to perform detection training on the fused deep neural network within a second number of training cycles to minimize a cross entropy loss for binary classification, wherein the number of cycles in the first number of training cycles is equal to the number of cycles in the second number of training cycles.

[0186] It should be noted that the data noise reduction and signal detection device also includes:

[0187] The evaluation module is used to evaluate the processing results to determine the noise reduction accuracy and detection accuracy of the target model.

[0188] Refer to the following Figure 5 A data noise reduction and signal detection system according to an embodiment of the present application is described.

[0189] It is understandable that if Figure 5 As shown, the data noise reduction and signal detection system includes:

[0190] at least one memory 200;

[0191] at least one processor 300;

[0192] at least one program;

[0193] Programs are stored in the memory 200, and the processor 300 executes at least one program to implement the above-mentioned data noise reduction and signal detection method. Figure 5 Take a processor 300 as an example.

[0194] The processor 300 and the memory 200 may be connected via a bus or other means. Figure 5 Take the example of connecting via a bus.

[0195] Memory 200, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer-executable programs, and signals, such as the program instructions / signals corresponding to the data noise reduction and signal detection system in the embodiments of the present application. Processor 300 executes the non-transitory software programs, instructions, and signals stored in memory 200 to perform various functional applications and data processing, thereby implementing the data noise reduction and signal detection method of the above-described method embodiment.

[0196] The memory 200 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and application programs required for at least one function; the data storage area may store relevant data of the above-mentioned data noise reduction and signal detection method, etc. In addition, the memory 200 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 200 may optionally include a memory remotely arranged relative to the processor 300, and these remote memories may be connected to the data noise reduction and signal detection system via a network. Examples of the above-mentioned network include, but are not limited to, the Internet of Things, a software-defined network, a sensor network, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0197] One or more signals are stored in the memory 200, and when executed by one or more processors 300, the data noise reduction and signal detection method in any of the above method embodiments is executed. Figure 1 The method in .

[0198] Refer to the following Figure 5 A computer-readable storage medium according to an embodiment of the present application is described.

[0199] like Figure 5 As shown, the computer readable storage medium stores computer executable instructions, which are executed by one or more processors 300, for example, Figure 5 The execution of one of the processors 300 in the embodiment of the present invention can enable the one or more processors 300 to execute the data noise reduction and signal detection method in the embodiment of the present invention. Figure 1 The method in .

[0200] The system embodiment described above is merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected based on actual needs to achieve the objectives of this embodiment.

[0201] From the description of the above embodiments, one skilled in the art will appreciate that all or some of the steps and systems in the methods disclosed above can be implemented as software, firmware, hardware, or any suitable combination thereof. Some or all of the physical components may be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software may be distributed on computer-readable media, which may include computer storage media and communication media. As is well known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disks or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and that can be accessed by a computer. Furthermore, as is well known to those skilled in the art, communication media generally contains computer-readable signals, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transmission mechanism, and may include any information delivery medium.

[0202] The embodiments of the present application have been described in detail above with reference to the accompanying drawings. However, the present application is not limited to the above embodiments. Various modifications can be made within the scope of knowledge possessed by ordinary technicians in the relevant technical field without departing from the purpose of the present application. In addition, the embodiments of the present application and the features of the embodiments can be combined with each other unless there is a conflict.

Claims

1. A data noise reduction and signal detection method, characterized in that: include: generating analog data with a low signal-to-noise ratio, wherein the analog data includes a plurality of mixed signals, the mixed signals being obtained by mixing noise and pure signals, the pure signals representing noise-free signals; Preprocessing the mixed signal in the simulation data to obtain training data; The first convolutional neural network, the second convolutional neural network, the self-attention mechanism and the multi-layer fully connected neural network are integrated to obtain a fused deep neural network; The fused deep neural network is trained step by step using the training data to obtain a target model with noise reduction function; Inputting real data with a low signal-to-noise ratio into the target model, so that the target model performs noise reduction and signal detection on the real data, and obtains processing results of the noise reduction and signal detection; The first convolutional neural network, the second convolutional neural network, the self-attention mechanism and the multi-layer fully connected neural network are integrated to obtain a fused deep neural network, including: Building the first convolutional neural network to extract feature information from the mixed signal; Building a fusion network of the self-attention mechanism and the second convolutional neural network to perform noise reduction on the mixed signal; Building the multi-layer fully connected neural network to detect target signals; The step-by-step training of the fused deep neural network using the training data to obtain a target model with noise reduction function includes: Using the mean square error between the mixed signal and the pure signal as a loss function, and performing initial training on the fused deep neural network using the loss function; Setting an alternating cycle number, wherein the alternating cycle number includes a first training cycle number and a second training cycle number, wherein the first training cycle number and the second training cycle number each include multiple training cycles, and each training cycle represents a process in which all samples in the training data are trained once; Performing noise reduction training on the fused deep neural network during the first number of training cycles, and performing detection training on the fused deep neural network during the second number of training cycles; Based on the number of alternating cycles, performing noise reduction training and detection training alternately; An optimizer is used to adjust the parameters of the fused deep neural network, and a learning rate adjustment strategy is used to adjust the learning rate of the fused deep neural network.

2. The data noise reduction and signal detection method according to claim 1, characterized in that: The generating of the simulation data with low signal-to-noise ratio comprises: Simulate noise generation based on the sensitivity curve of the gravitational wave detector; Select corresponding parameter ranges for different types of gravitational wave sources; generating, according to the parameter range, a plurality of pure signals corresponding to the plurality of gravitational wave sources, wherein the gravitational wave sources and the pure signals correspond one to one; Each of the clean signals is mixed with the noise to obtain analog data including a plurality of mixed signals.

3. The data noise reduction and signal detection method according to claim 1, characterized in that: The preprocessing of the mixed signal in the analog data includes: performing whitening processing on the mixed signal in the analog data so that a noise spectrum of the mixed signal is uniformly distributed; Normalization is performed on the mixed signal after the whitening process, so that all data in the simulation data have the same scale.

4. The data noise reduction and signal detection method according to claim 1, characterized in that: The performing noise reduction training on the fused deep neural network within the first number of training cycles, and performing detection training on the fused deep neural network within the second number of training cycles, includes: Performing noise reduction training on the fused deep neural network during the first number of training cycles to minimize a mean square error between the mixed signal and the clean signal; Performing exploration training on the fused deep neural network during the second number of training cycles to minimize a cross entropy loss for binary classification, wherein the number of cycles in the first number of training cycles is equal to the number of cycles in the second number of training cycles.

5. The data noise reduction and signal detection method according to claim 1, characterized in that: The data noise reduction and signal detection method further includes: The processing results are evaluated to determine the noise reduction accuracy and detection accuracy of the target model.

6. Data noise reduction and signal detection device, characterized in that: include: A generating module, configured to generate analog data with a low signal-to-noise ratio, wherein the analog data includes a plurality of mixed signals, the mixed signals being obtained by mixing noise and pure signals, the pure signals being noise-free signals; a processing module, configured to pre-process the mixed signal in the analog data to obtain training data; A fusion module is used to fuse the first convolutional neural network, the second convolutional neural network, the self-attention mechanism, and the multi-layer fully connected neural network to obtain a fused deep neural network; A training module, configured to perform step-by-step training on the fused deep neural network using the training data to obtain a target model with a noise reduction function; a noise reduction and detection module, configured to input real data with a low signal-to-noise ratio into the target model, so that the target model performs noise reduction and signal detection on the real data, and obtain processing results of the noise reduction and signal detection; The first convolutional neural network, the second convolutional neural network, the self-attention mechanism and the multi-layer fully connected neural network are integrated to obtain a fused deep neural network, including: Building the first convolutional neural network to extract feature information from the mixed signal; Building a fusion network of the self-attention mechanism and the second convolutional neural network to perform noise reduction on the mixed signal; Building the multi-layer fully connected neural network to detect target signals; The step-by-step training of the fused deep neural network using the training data to obtain a target model with noise reduction function includes: Using the mean square error between the mixed signal and the pure signal as a loss function, and performing initial training on the fused deep neural network using the loss function; Setting an alternating cycle number, wherein the alternating cycle number includes a first training cycle number and a second training cycle number, wherein the first training cycle number and the second training cycle number each include multiple training cycles, and each training cycle represents a process in which all samples in the training data are trained once; Performing noise reduction training on the fused deep neural network during the first number of training cycles, and performing detection training on the fused deep neural network during the second number of training cycles; Based on the number of alternating cycles, performing noise reduction training and detection training alternately; An optimizer is used to adjust the parameters of the fused deep neural network, and a learning rate adjustment strategy is used to adjust the learning rate of the fused deep neural network.

7. Data noise reduction and signal detection system, characterized in that: include: at least one memory; at least one processor; at least one program; The programs are stored in the memory, and the processor executes at least one of the programs to implement the data noise reduction and signal detection method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute the data noise reduction and signal detection method according to any one of claims 1 to 5.

Images