An intersection signal transient information extraction method, system, device and medium

Abstract

The application discloses a kind of intersection signal instantaneous information extraction method, system, equipment and medium, it is related to non-stationary signal processing technical field, including steps: obtaining the intersection signal of time-frequency image;And input convolutional neural network, obtain feature map by feature extraction and down-sampling processing, and feature map is compressed into one-dimensional array;One-dimensional array is mapped as vector by convolutional neural network, and the probability value of the number of signal components is obtained;Setting CNN network model, and the intersection signal is input into CNN network model and is convolved and down-sampling processing, and the feature of signal is mapped as the vector of signal length same;Vector is spliced into a matrix, each row represents a signal component, and the instantaneous information of each component of intersection signal is extracted.The application can realize the adaptive extraction of each component instantaneous information of multi-component intersection signal, can effectively determine the number of components of multi-component intersection signal and accurately extract the instantaneous frequency information of multi-component intersection signal.

Description

Technical Field

[0001] This invention relates to the field of non-stationary signal processing technology, and in particular to a method, system, device and medium for extracting instantaneous information from intersecting signals. Background Technology

[0002] With the continuous development of time-frequency analysis research, the analysis and processing of multi-component intersecting signals has attracted widespread attention from researchers. Multi-component intersecting signals are a special type of multi-component non-stationary signals. The instantaneous information of each component does not meet the separation condition, and there are intersecting instantaneous frequencies or group delay information. Therefore, traditional time-frequency analysis methods and post-processing methods for multi-component non-stationary signals are not suitable for multi-component intersecting signals.

[0003] In recent years, the analysis and processing of multi-component intersecting signals have mainly involved the following work. In 2020, Bruni et al. proposed an iterative rearrangement algorithm, which aims to improve the readability of time-frequency representations of inseparable regions; sparse optimization methods were used to obtain sparse time-frequency representations; Li Lin and Charles K. Chui et al. utilized the difference in linear frequency modulation (CR) between components to separate multi-component intersecting signals; a time-frequency-CR rearrangement / decimation method was proposed, which effectively achieved high-precision feature characterization of intersecting signals and provided a theoretical analysis of three-dimensional rearrangement methods; high-precision time-frequency output was obtained by constructing a convolutional neural network and training with a large amount of time-frequency data.

[0004] While the analysis and processing of multi-component intersecting signals have further expanded the theoretical depth and application scope of time-frequency analysis, the identification and extraction of instantaneous features of multi-component intersecting signals still face many challenges. There is a lack of methods for extracting information from multi-component intersecting signals with small differences in CR values. Existing instantaneous information extraction methods lack adaptability and require setting transformation parameters based on the characteristics of the signal itself. Furthermore, existing neural network-based multi-component intersecting signal analysis methods focus on obtaining a high-resolution time-frequency representation of the signal, without extracting its instantaneous information. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of the prior art by providing a method, system, device, and medium for extracting instantaneous information from intersecting signals. This solves the problem that existing neural network-based multi-component intersecting signal analysis methods focus on obtaining high-resolution time-frequency representations of signals but do not extract their instantaneous information.

[0006] This invention specifically provides the following technical solution: a method for extracting instantaneous information of intersecting signals, comprising the following steps:

[0007] Obtain intersecting signals with different time-frequency images;

[0008] The intersecting signals are input into a CNN network model, and feature maps of the intersecting signals are obtained through feature extraction and downsampling processing of multiple convolutional layers. The feature maps are then compressed into a one-dimensional array.

[0009] The one-dimensional array is mapped to a probability value vector representing the number of intersecting signal components by the fully connected layer of the CNN network model, thereby determining the number of intersecting signal components.

[0010] The CNN network model is configured with n branches, and the intersecting signals of the n components are input into the CNN network model with n branches for multiple convolutions and downsampling processes, which are then mapped into a vector with the same length as the signal.

[0011] The vectors of the same length as the signal are concatenated into a matrix in which each row represents a signal component, and the instantaneous information of each component of the intersecting signal is extracted.

[0012] Preferably, the step of acquiring intersecting signals with different time-frequency images is wherein the specific expression of the intersecting signal is:

[0013]

[0014] Where P is the number of components of the intersecting signal; a p (t), These are respectively called components x p The instantaneous amplitude and instantaneous phase of signal x(t); the instantaneous frequency of the p-th component of signal x(t) is

[0015] If there exist t0, n, m such that f m (t0)=f n If m ≠ n, then the signal x(t) is called a multi-component intersecting signal.

[0016] Preferably, when obtaining the feature map of the intersecting signal through feature extraction and downsampling processing of multiple convolutional layers, features are extracted through pooling operations. The feature map obtained after each pooling operation is retained. Then, the feature map after the last pooling is convolved to obtain a heatMap. The heatMap is then upsampled using transposed convolution and convolved to supplement detailed features. The previously retained feature maps are then stitched together for feature fusion. The fused feature map is then convolved to further supplement details, and finally, high-resolution time-frequency representation data of the signal with the same data size as the original data is obtained.

[0017] Preferably, when inputting the intersecting signals into the CNN network model, the cross-entropy loss function is used, and Adam optimization is employed, setting the initial learning rate, batch size, and number of training epochs.

[0018] Preferably, when determining the number of components of the intersecting signals, the number of components output by the CNN network model is determined based on the probability value and the number of components according to the number of nodes in the fully connected layer.

[0019] Preferably, when setting n branches for the CNN network model, the n branches correspond to the n components of the intersecting signals.

[0020] Preferably, the number of branches is added to the CNN network model according to the number of components of the discriminant intersecting signal, and the branch model of the CNN network model is constructed by the number of branches.

[0021] This invention provides a system for extracting instantaneous information from intersecting signals, comprising:

[0022] Acquisition module for intersecting signals with different time-frequency images;

[0023] The feature extraction module is used to input the intersecting signal into the CNN network model, obtain the feature map of the intersecting signal through feature extraction and downsampling processing of multiple convolutional layers, and compress the feature map into a one-dimensional array;

[0024] The probability value acquisition module is used to map the one-dimensional array into a probability value vector representing the number of intersecting signal components through the fully connected layer of the CNN network model, and to determine the number of intersecting signal components.

[0025] The mapping module is used to set n branches for the CNN network model and input the intersecting signals of the n components into the CNN network model with n branches for multiple convolution and downsampling processes, mapping them into a vector with the same length as the signal.

[0026] The instantaneous information extraction module is used to concatenate the vectors with the same length as the signal into a matrix in which each row represents a signal component, and extract the instantaneous information of each component of the intersecting signals.

[0027] The present invention provides a computer device, including a memory and a processor. The memory stores a program, and when the program is executed by the processor, the processor performs the steps of the method for extracting instantaneous information of intersecting signals.

[0028] The present invention provides a storage medium storing a computer program thereon, wherein the computer program, when executed by a processor, implements the steps of the method for extracting instantaneous information of intersecting signals.

[0029] Compared with the prior art, the present invention has the following significant advantages:

[0030] This invention inputs intersecting signals into a CNN network model, extracts feature maps and compresses them into a one-dimensional array, maps the one-dimensional data through a fully connected layer, and determines the number of components of the intersecting signal. It can adaptively extract the instantaneous information of each component of a multi-component intersecting signal, and the extracted instantaneous information meets the accuracy requirements well. It can effectively determine the number of components of a multi-component intersecting signal and accurately extract the instantaneous frequency information of the multi-component intersecting signal. According to actual use needs, after the scale of the model group of this method is expanded, its application fields will be greatly increased. Attached Figure Description

[0031] Figure 1 This is the component number discrimination network diagram of the present invention;

[0032] Figure 2 This is the time-frequency diagram of the three-component signal of the present invention;

[0033] Figure 3 This is a diagram of the single-branch information extraction model of the present invention;

[0034] Figure 4 This is a diagram of the three-component separation model of the present invention;

[0035] Figure 5 The output results of the component separation network at a signal-to-noise ratio of 35dB of the present invention are shown in the figure; where (a) is the original STFT image, (a1)-(a3) are the instantaneous information images of three different simulated signals respectively, (b) is the instantaneous information separation and extraction result (the blue solid line represents the separation value; the red dashed line represents the actual value), and (b1)-(b3) are the instantaneous information extraction results of three different simulated signals respectively. Detailed Implementation

[0036] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0037] To address the challenges of multi-component intersecting signals, this paper leverages the superior feature extraction capabilities of convolutional neural networks (CNNs) to achieve the separation and extraction of instantaneous frequency information for each component of such signals through time-frequency data generation, CNN construction, and training. The extraction of instantaneous information from multi-component intersecting signals involves two steps: first, a classification network is constructed to determine the number of signal components; then, based on the number of components, a tree-structured network with a specific number of branches is used to extract the instantaneous information for each component.

[0038] The embodiments of the present invention are as follows Figure 1As shown; embodiments of the present invention provide a method for extracting instantaneous information of intersecting signals, comprising the following steps:

[0039] Step S1: Obtain intersecting signals with different time-frequency images.

[0040] The specific expression for the intersecting signal is:

[0041]

[0042] Where P is the number of components of the intersecting signal; a p (t), These are respectively called components x p The instantaneous amplitude and instantaneous phase of signal x(t); the instantaneous frequency (IF) of the p-th component of signal x(t) is...

[0043] If there exist t0, n, m such that f m (t0)=f n If m ≠ n, then the signal x(t) is called a multi-component intersecting signal.

[0044] Step S2: Input the intersecting signals into the CNN network model, obtain the feature map of the intersecting signals through feature extraction and downsampling processing of multiple convolutional layers, and compress the feature map into a one-dimensional array.

[0045] Set up a CNN model to determine the number of components in a multi-component intersecting signal, and thus determine the number of branches in the separation model to use.

[0046] The input to this model is a 128×256 time-frequency image. After a series of feature extractions and downsampling processes through three convolutional layers, the resulting feature map is compressed into a one-dimensional array. This array is then passed through a fully connected layer to finally map the features into a vector representing the probability value of the number of components. The network structure of this invention is as follows: Figure 1 As shown in Table 1, the number and size of convolutional kernels used in the convolutional layers of this model are the same.

[0047] Table 1 Convolutional Layer Parameters

[0048]

[0049]

[0050] In this step, when obtaining feature maps through feature extraction and downsampling, features are extracted through pooling operations. The feature maps obtained after each pooling operation are retained. Then, the feature map after the last pooling operation is convolved to obtain a heatmap. The heatmap is then upsampled using transposed convolution and convolved to supplement detailed features. The previously retained feature maps are then stitched together for feature fusion. The fused feature map is then convolved to further supplement details, ultimately obtaining high-resolution time-frequency representation data of the signal with the same scale as the original data.

[0051] Step S3: Map the one-dimensional array into a probability vector representing the number of intersecting signal components through the fully connected layer of the CNN network model, and determine the number of intersecting signal components.

[0052] When inputting time-frequency images into a convolutional neural network, the cross-entropy loss function is used and Adam optimization is employed. The initial learning rate is set to 0.01, the batch size to 32, and the training duration to 500 epochs.

[0053] The model outputs probability values ​​for signals with 3 and 2 components, indicating the probability of the signal having 3 or 2 components. The number of nodes in the final fully connected layer can be adjusted to accommodate more components, depending on the actual needs. For example, in this embodiment, the signal has three components, and the network output is [9.4286555e-015.7134442e-02], meaning the probability of having 3 components is much greater than the probability of having 2 components, therefore the signal is determined to have 3 components.

[0054] Step S4: Set up n branches for the CNN network model, and input the intersecting signals with n components into the CNN network model with n branches for multiple convolutions and downsampling processes, mapping them into a vector with the same length as the signal.

[0055] When setting up a CNN network model with n branches, the n branches correspond to the n components of a multi-component intersecting signal (the number of branches corresponds to the number of signal components being processed), and the structure of each branch is determined by... Figure 3 The given information is provided. Based on the number of components, the model determines the number of components and automatically adds the number of branches, then constructs the CNN network model using the number of branches.

[0056] Step S5: Concatenate vectors of the same length as the signal into a matrix where each row represents a signal component, and extract the instantaneous information of each component of the intersecting signal.

[0057] For the extraction model of instantaneous information of multi-component intersecting signals with three components, such as... Figure 4As shown, the model consists of three single-branch models. In practical applications, the number of branches needs to be manually added based on the number of components identified in the model. The experiments in this invention only consist of two-branch and three-branch models. The parameter settings for the convolutional layers of the three-branch model are as shown in Table 2.

[0058] Table 2 Parameters of the separation model

[0059] conv2d 8 5×5 conv2d_3 8 5×5 conv2d_6 8 5×5 conv2d_1 10 5×5 conv2d_4 10 5×5 conv2d_7 10 5×5 conv2d_2 16 5×5 conv2d_5 16 5×5 conv2d_8 16 5×5

[0060] Numerical Experiment:

[0061] This model uses the mean squared loss function, Adam optimization, an initial learning rate of 0.01, a training batch size of 32, and 300 training epochs.

[0062] Using three example signals, the input and output of the component separation network are shown in the diagram. Figure 5 In the middle. Each column of the graph corresponds to the instantaneous information of each simulated signal. Figure 5 (b) represents the extracted instantaneous frequency information, where the blue solid line represents the output of the separation model, and the red dashed line is the actual instantaneous frequency curve (label) of the component. Furthermore, the number of plots (b1)-(b3) in (b) corresponds to the number of components in (a1)-(a3).

[0063] Simulation results show that the method designed in this invention can effectively determine the number of components of a multi-component intersecting signal and accurately extract the instantaneous frequency information of the multi-component intersecting signal.

[0064] Based on the above methods and statements, the present invention provides a system for extracting instantaneous information of intersecting signals, comprising: an acquisition module, a feature extraction module, a probability value acquisition module, a mapping module, and an instantaneous information extraction module.

[0065] The system comprises the following modules: an acquisition module for intersecting signals with different time-frequency images; a feature extraction module for inputting the intersecting signals into a CNN network model, obtaining feature maps of the intersecting signals through feature extraction and downsampling processing via multiple convolutional layers, and compressing the feature maps into a one-dimensional array; a probability value acquisition module for mapping the one-dimensional array into a probability value vector representing the number of components of the intersecting signal through the fully connected layers of the CNN network model, thus determining the number of components of the intersecting signal; a mapping module for setting n branches for the CNN network model, inputting the intersecting signals with n components into the CNN network model with n branches for multiple convolutions and downsampling processing, mapping them into a vector with the same length as the signal; and a transient information extraction module for concatenating the vector with the same length as the signal into a matrix where each row represents a signal component, thus extracting the transient information of each component of the intersecting signal.

[0066] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a program, and when the program is executed by the processor, the processor performs the steps of a method for extracting instantaneous information of intersecting signals.

[0067] According to the disclosed embodiments, the computer device can communicate with one or more external devices (e.g., keyboard, pointing device, Bluetooth communication, etc.) or with any device that enables the computing device to communicate with one or more other computing devices (e.g., router, demodulator, etc.).

[0068] The present invention also provides a storage medium storing a computer program thereon, wherein the computer program, when executed by a processor, implements the steps of a method for extracting instantaneous information of intersecting signals.

[0069] According to the disclosed embodiments, the storage medium can be a non-volatile computer-readable storage medium, such as, but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this invention, the storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0070] The above description, in conjunction with specific preferred embodiments, provides a more detailed explanation of the present invention. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such deductions or substitutions should be considered to fall within the scope of protection of the present invention.

Claims

1. A method for extracting instantaneous information from intersecting signals, characterized in that, Includes the following steps: Obtain the intersecting signal with different time-frequency images; the specific expression of the intersecting signal is: in, P The number of components of the intersecting signals; They are respectively called components. The instantaneous amplitude and instantaneous phase of the signal; No. p The instantaneous frequency of each component is ; If it exists , , Make ,in ≠ Then it is called a signal. It is a multi-component intersecting signal; The intersecting signals are input into a CNN network model. Feature maps of the intersecting signals are obtained through feature extraction and downsampling processing by multiple convolutional layers, and the feature maps are compressed into a one-dimensional array. When obtaining feature maps of the intersecting signals through feature extraction and downsampling processing by multiple convolutional layers, features are extracted through pooling operations. The feature maps obtained after each pooling operation are retained. The feature maps after the last pooling operation are then convolved to obtain a heatmap. The heatmap is then upsampled using transposed convolution and convolved to supplement detailed features. The previously retained feature maps are then concatenated for feature fusion. The fused feature maps are then convolved to further supplement details, and finally, high-resolution time-frequency representation data of the signal with the same data size as the original data is obtained. The one-dimensional array is mapped to a probability value vector representing the number of intersecting signal components by the fully connected layer of the CNN network model, thereby determining the number of intersecting signal components. The CNN network model is configured with n branches, and the intersecting signals of the n components are input into the CNN network model with n branches for multiple convolutions and downsampling processes, which are then mapped into a vector with the same length as the signal. The vectors of the same length as the signal are concatenated into a matrix in which each row represents a signal component, and the instantaneous information of each component of the intersecting signal is extracted.

2. The method for extracting instantaneous information from intersecting signals as described in claim 1, characterized in that, When the intersecting signals are input into the CNN network model, the cross-entropy loss function is used and Adam optimization is employed, and the initial learning rate, batch size, and number of training epochs are set.

3. The method for extracting instantaneous information of intersecting signals as described in claim 1, characterized in that, When determining the number of components of intersecting signals, the number of components output by the CNN network model is determined based on the probability value and the number of components according to the number of nodes in the fully connected layer.

4. The method for extracting instantaneous information of intersecting signals as described in claim 1, characterized in that, When setting n branches for the CNN network model, the n branches correspond to the n components of the intersecting signals.

5. The method for extracting instantaneous information of intersecting signals as described in claim 1, characterized in that, The number of branches is added to the CNN network model based on the number of components of the discriminant intersecting signal, and the branch model of the CNN network model is constructed based on the number of branches.

6. A system for extracting instantaneous information from intersecting signals, characterized in that, include: Acquisition module for intersecting signals with different time-frequency images; The specific expression for the intersecting signal is: in, P The number of components of the intersecting signals; They are respectively called components. The instantaneous amplitude and instantaneous phase of the signal; No. p The instantaneous frequency of each component is ; If it exists , , Make ,in ≠ Then it is called a signal. It is a multi-component intersecting signal; The feature extraction module is used to input the intersecting signal into the CNN network model, obtain the feature map of the intersecting signal through feature extraction and downsampling processing of multiple convolutional layers, and compress the feature map into a one-dimensional array. When obtaining the feature map of the intersecting signal through feature extraction and downsampling processing of multiple convolutional layers, features are extracted through pooling operations. The feature map obtained after each pooling operation is retained, and the feature map after the last pooling is convolved to obtain a heatmap. Then, the heatmap is upsampled and convolved to supplement detailed features using transposed convolution. The previously retained feature maps are concatenated for feature fusion. The fused feature map is convolved to further supplement details, and finally, high-resolution time-frequency representation data of the signal with the same data size as the original data is obtained. The probability value acquisition module is used to map the one-dimensional array into a probability value vector representing the number of intersecting signal components through the fully connected layer of the CNN network model, and to determine the number of intersecting signal components. The mapping module is used to set n branches for the CNN network model and input the intersecting signals of the n components into the CNN network model with n branches for multiple convolution and downsampling processes, mapping them into a vector with the same length as the signal. The instantaneous information extraction module is used to concatenate the vectors with the same length as the signal into a matrix in which each row represents a signal component, and extract the instantaneous information of each component of the intersecting signals.

7. A computer device, characterized in that, The method includes a memory and a processor, wherein the memory stores a program, and when the program is executed by the processor, the processor performs the steps of the instantaneous information extraction method for intersecting signals as described in any one of claims 1 to 5.

8. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the instantaneous information extraction method for intersecting signals according to any one of claims 1 to 5.

Images