Radio monitoring method and system

Abstract

The present invention discloses a radio monitoring method and system, which includes signal acquisition, signal preprocessing, establishing a radio monitoring model, and radio monitoring. The present invention belongs to the field of signal monitoring, and specifically refers to a radio monitoring method and system. This solution uses inverse Fourier transform to obtain the output of the filter, thereby reducing the steady-state error; the gradient of the error function is calculated by the derivative form of the exponential function, thereby improving the stability of the error back propagation; the SE module is integrated into the residual block unit, and the SE module adaptively learns and adjusts the weights of each channel in the feature map through global average pooling and multi-layer perceptron, thereby effectively enhancing the recognition ability of the network; by designing adaptive factors and mobile strategies, the global and local search capabilities are dynamically adjusted; a dynamic population strategy is adopted to manage the population size to achieve more effective parameter optimization and model establishment, thereby achieving superior model performance.

Description

Technical Field

[0001] The present invention relates to the technical field of signal monitoring, and in particular to a radio monitoring method and system. Background Art

[0002] Radio monitoring methods enable efficient and accurate signal monitoring and classification in complex radio signal environments and are widely used in fields such as telecommunications security, spectrum management, and radio reconnaissance. However, typical radio monitoring methods suffer from poor raw signal quality, severe noise interference, and low signal processing accuracy and stability, resulting in poor monitoring effectiveness. They also suffer from poor expressiveness and robustness of model features, weak network recognition capabilities, and low population stability and diversity in parameter optimization methods, leading to poor monitoring accuracy. Summary of the Invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a radio monitoring method and system. In view of the problems that general radio monitoring methods have poor original signal quality, serious noise interference, low signal processing accuracy and stability, resulting in poor monitoring effect, this scheme processes signals at all time steps, uses Fourier transform to convert time series data into frequency domain data, uses Hadamard product and inverse Fourier transform to obtain the output of the filter, reduces steady-state error, and improves the accuracy and stability of signal processing; applies the Catmull-Rom spline interpolation method to construct an interpolation matrix, making signal reconstruction more accurate and efficient; calculates the output of the nonlinear part based on the Catmull-Rom spline interpolation method, improves the model's sensitivity and accuracy to signal changes; calculates the gradient of the error function through the derivative form of the exponential function, and uses FFT and inverse FFT to calculate the gradient of the weight vector and the spline node matrix, thereby improving the stability and convergence of error back propagation speed, thereby improving the radio monitoring effect; in view of the problems that general radio monitoring methods have poor expression ability and robustness of model features, weak network recognition ability, and low population stability and diversity of parameter optimization methods, which lead to poor monitoring accuracy, this scheme adopts sliding window technology to segment time series data, and each sample is decomposed by multi-wavelet transform to extract rich feature information; the SE module is integrated into the residual block unit. The SE module adaptively learns and adjusts the weights of each channel in the feature map through global average pooling and multi-layer perceptron, effectively enhancing the recognition ability of the network; by designing adaptive factors and movement strategies, the global and local search capabilities are dynamically adjusted to optimize the convergence speed and stability of the model; a dynamic population strategy is used to manage the population size, and the individual position is dynamically updated according to the population changes to achieve more effective parameter optimization and model establishment. The number of individuals is dynamically adjusted according to the changes in population size to ensure the diversity and stability of the population, thereby achieving superior model performance.

[0004] The technical solution adopted by the present invention is as follows: The present invention provides a radio monitoring method, which comprises the following steps:

[0005] Step S1: signal acquisition;

[0006] Step S2: signal preprocessing;

[0007] Step S3: establishing a radio monitoring model;

[0008] Step S4: Radio monitoring.

[0009] Furthermore, in step S1, the signal collection is to collect historical radio monitoring data; the historical radio monitoring data includes radio signal spectrum data, signal parameters, environmental noise, timestamp, signal strength, modulation mode, signal bandwidth, signal quality index and signal category; and the signal category is used as a data label.

[0010] Furthermore, in step S2, the signal preprocessing is to process the signals of all time steps, construct the preprocessed signal features and labels into a data set, and divide it into a test set and a training set; specifically, the following steps are included:

[0011] Step S21: Calculate frequency domain data, construct input vector, and extract current time step from time series data and the previous Q samples as input vector , perform Fourier transform on the input data of the current time step and the previous time step; the formula used is as follows:

[0012] ;

[0013] Where, is frequency domain data; is the input of the previous time step; FFT[·] is the Fourier transform;

[0014] Step S22: Calculate the linear filter output, perform Hadamard product on the frequency domain data and the weight vector of the current time step, and then perform inverse Fourier transform to obtain the last Q elements of the filter output, which are expressed as follows:

[0015] ;

[0016] Where, is the filter output; is the Hadamard product; is the weight vector; IFFT is the inverse Fourier transform;

[0017] Step S23: Construct an interpolation matrix and calculate the interpolation position u for each sample l and node index i l , generate the interpolation vector and spline node matrix P of the current sample, and construct the interpolation matrix U; the formula used is as follows:

[0018] ;

[0019] ;

[0020] ;

[0021] ;

[0022] Where, is the filter at time step The output signal value of is the parameter used to scale the signal; is the rounding operation; M is the node offset; is the interpolation vector used to construct the interpolation matrix; is the column vector used to construct the spline knot matrix;

[0023] Step S24: Calculate the nonlinear output. Calculate the output of the nonlinear part and the error signal using the Catmull-Rom spline interpolation method. The formula used is as follows:

[0024] ;

[0025] ;

[0026] Where, is the nonlinear output vector in the frequency domain; C is the coefficient matrix in the spline interpolation; It is a sum operation on the rows of the matrix; It is expectation; is the error;

[0027] Step S25: Calculate the derivative of the error function to update the weights and spline nodes;

[0028] ;

[0029] Where, is the derivative of the error function; It is a parameter used to adjust the amplitude of the function gradient;

[0030] Step S26: Update the error vector; calculate the sensitivity of the linear filter output to the error and update the error vector; the formula used is as follows:

[0031] ;

[0032] ;

[0033] Where, is the sensitivity of the output vector to error; is the derivative matrix of the interpolation matrix; is the updated error vector;

[0034] Step S27: Update the frequency domain error vector and perform FFT transformation on the error vector; the formula used is as follows:

[0035] ;

[0036] Where, is the updated frequency domain error vector;

[0037] Step S28: Update the weight vector gradient. Calculate the gradient of the weight vector through IFFT and take the first 𝑄 elements. The formula used is as follows:

[0038] ;

[0039] ;

[0040] Where, is the gradient of the weight vector; is the complex conjugate; and are the frequency domain weight vectors before and after updating respectively; is the learning rate of the weight vector;

[0041] Step S29: Update the spline node matrix and update the parameters in the spline node matrix; the formula used is as follows:

[0042] ;

[0043] Where, and P are the spline knot matrices before and after updating, respectively; is the learning rate of the spline knot matrix; T is the matrix transpose.

[0044] Furthermore, in step S3, establishing the radio monitoring model specifically includes the following steps:

[0045] Step S31: Input layer design, taking the preprocessed data set as input;

[0046] Step S32: Design a feature extraction layer, use a sliding window to segment the long time series data, extract short time series signal samples, perform multi-wavelet transform on each short time series signal sample, and decompose the signal into multiple wavelet coefficients;

[0047] Step S33: Dimensionality reduction processing, applying principal component analysis to each wavelet coefficient, and using the SE module to adjust the importance of different channels in the feature map to improve the expressiveness and robustness of the features;

[0048] Step S34: Network optimization, integrating the SE module into the residual block unit; the residual block unit includes a convolutional layer, a batch normalization layer, a ReLU activation function, a global average pooling layer, and a fully connected layer; in the SE module, the weights of each channel in the feature map are adaptively learned and adjusted through global average pooling and a multi-layer perceptron; the weighted features are added to the original features in the shortcut connection to enhance the network's recognition ability;

[0049] Step S35: Design the output layer, using a fully connected layer and softmax activation function for final classification;

[0050] Step S36: Model determination. A determination threshold is pre-set. If the radio monitoring model converges on the training set, the radio monitoring model training is completed. If the prediction accuracy of the trained radio monitoring model on the test set is higher than the determination threshold, the radio monitoring model is established. Otherwise, the process goes to step S37 for parameter optimization.

[0051] Step S37: Model parameter optimization, specifically including:

[0052] Step S371: Initialization: establishing a search space based on the initial weights, biases, and learning rate of the radio monitoring model network; randomly initializing individual positions, and using the prediction accuracy of the test set by the radio monitoring model trained based on the individual positions as the individual fitness value;

[0053] Step S372: Design an adaptive factor to dynamically change the global and local search capabilities, as follows:

[0054] ;

[0055] Where l is the adaptive factor; t is the current number of iterations, and maxt is the maximum number of iterations; is a smoothing term;

[0056] Step S373: Design a mobility strategy and perform location update based on the adaptive factor. The formula used is as follows:

[0057] ;

[0058] ;

[0059] Where, and are the positions of the individual j-th dimension at the t+1th iteration and the tth iteration respectively; rand is a random number between 0 and 1; h is the adaptive inertia weight; h max and h min are the maximum inertia weight and the minimum inertia weight respectively; is the optimal individual position of the population; p is the average fitness of the population;

[0060] Step S374: Dynamic population, using a dynamic population strategy. If the updated population increases, the positions of the added individuals are initialized; if the updated population decreases, the last individual is collected and eliminated. The formula used to update the population is as follows:

[0061] ;

[0062] Where, and are the population sizes before and after the update, respectively; round(·) means rounding; is a random number between -0.1 and 0.1; sign is a sign function, which takes 1 if p is greater than 0.7, otherwise takes -1;

[0063] Step S375: Optimization judgment. A fitness threshold is set in advance. If there is an individual fitness value higher than the fitness threshold, the individual position is the optimal parameter setting of the radio monitoring model, and the radio monitoring model is established. If the maximum number of iterations is reached, return to step S371; otherwise, return to step S372.

[0064] Furthermore, in step S4, the radio monitoring is based on the established radio monitoring model; radio signal spectrum data, signal parameters, environmental noise, timestamp, signal strength, modulation mode, signal bandwidth and signal quality indicators are collected in real time; and after preprocessing, they are input into the radio monitoring model, and the signal category based on the model output is used as the radio monitoring result.

[0065] The present invention provides a radio monitoring system, comprising a signal acquisition module, a signal preprocessing module, a radio monitoring model establishment module and a radio monitoring module;

[0066] The signal acquisition module collects historical radio monitoring data and sends the data to the signal preprocessing module;

[0067] The signal preprocessing module processes the signals of all time steps, converts the time series data into frequency domain data using Fourier transform, obtains the filter output using Hadamard product and inverse Fourier transform, constructs the interpolation matrix using Catmull-Rom spline interpolation method, calculates the output of the nonlinear part based on Catmull-Rom spline interpolation method, calculates the gradient of the error function using the derivative form of the exponential function, and calculates the gradient of the weight vector and the spline node matrix using FFT and inverse FFT; and sends the data to the radio monitoring model establishment module;

[0068] The radio monitoring model building module uses a sliding window technique to segment time series data, and each sample is decomposed through a multi-wavelet transform. The SE module is integrated into the residual block unit. The SE module adaptively learns and adjusts the weights of each channel in the feature map through global average pooling and a multi-layer perceptron. By designing adaptive factors and movement strategies, the global and local search capabilities are dynamically adjusted. A dynamic population strategy is used to manage the population size, and individual positions are dynamically updated according to population changes to achieve more efficient parameter optimization and model building. The data is then sent to the radio monitoring module.

[0069] The radio monitoring module predicts the signal type based on the established radio monitoring model to realize radio monitoring.

[0070] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0071] (1) In view of the problems of poor original signal quality, serious noise interference, low accuracy and stability of signal processing in general radio monitoring methods, which lead to poor monitoring effects, this scheme uses Fourier transform to convert time series data into frequency domain data for signal processing at all time steps, and uses Hadamard product and inverse Fourier transform to obtain the output of the filter, which reduces the steady-state error and improves the accuracy and stability of signal processing; uses Catmull-Rom spline interpolation method to construct the interpolation matrix, making signal reconstruction more accurate and efficient; calculates the output of the nonlinear part based on Catmull-Rom spline interpolation method, and improves the sensitivity and accuracy of the model to signal changes; calculates the gradient of the error function through the derivative form of the exponential function, and uses FFT and inverse FFT to calculate the gradient of the weight vector and spline node matrix, which improves the stability and convergence speed of error back propagation, thereby improving the radio monitoring effect.

[0072] (2) In view of the problems of poor expression and robustness of model features, weak recognition ability of the network, and low population stability and diversity of the parameter optimization method in general radio monitoring methods, which lead to poor monitoring accuracy, this scheme adopts sliding window technology to segment time series data, and decomposes each sample through multi-wavelet transform to extract rich feature information; the SE module is integrated into the residual block unit. The SE module adaptively learns and adjusts the weights of each channel in the feature map through global average pooling and multi-layer perceptron, which effectively enhances the recognition ability of the network; by designing adaptive factors and mobile strategies, the global and local search capabilities are dynamically adjusted to optimize the convergence speed and stability of the model; a dynamic population strategy is used to manage the population size, and the individual position is dynamically updated according to the population changes to achieve more effective parameter optimization and model establishment. The number of individuals is dynamically adjusted according to the changes in population size to ensure the diversity and stability of the population; thereby achieving superior model performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1 A schematic flow chart of a radio monitoring method provided by the present invention;

[0074] Figure 2 A schematic diagram of a radio monitoring system provided by the present invention;

[0075] Figure 3 Schematic diagram of the process of step S2;

[0076] Figure 4 Schematic diagram of the process of step S3.

[0077] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0078] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0079] In the description of the present invention, it should be understood that terms such as "up", "down", "front", "back", "left", "right", "top", "bottom", "inside" and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the system or element referred to must have a specific direction, be constructed and operated in a specific direction. Therefore, they should not be understood as limiting the present invention.

[0080] Example 1, see Figure 1 The present invention provides a radio monitoring method, which comprises the following steps:

[0081] Step S1: signal acquisition, collecting historical radio monitoring data;

[0082] Step S2: Signal preprocessing: For signal processing at all time steps, the time series data is converted into frequency domain data using Fourier transform, and the filter output is obtained using Hadamard product and inverse Fourier transform; the interpolation matrix is ​​constructed using the Catmull-Rom spline interpolation method; the output of the nonlinear part is calculated based on the Catmull-Rom spline interpolation method; the gradient of the error function is calculated through the derivative form of the exponential function, and the gradient of the weight vector and the spline node matrix is ​​calculated using FFT and inverse FFT;

[0083] Step S3: Establish a radio monitoring model, using a sliding window technique to segment time series data, and decompose each sample using a multi-wavelet transform. The SE module is integrated into the residual block unit. The SE module adaptively learns and adjusts the weights of each channel in the feature map through global average pooling and a multi-layer perceptron. By designing adaptive factors and movement strategies, global and local search capabilities are dynamically adjusted. A dynamic population strategy is used to manage population size, and individual positions are dynamically updated based on population changes to achieve more efficient parameter optimization and model building.

[0084] Step S4: Radio monitoring.

[0085] Example 2, see Figure 1 This embodiment is based on the above embodiment. In step S1, the historical radio monitoring data includes radio signal spectrum data, signal parameters, environmental noise, timestamp, signal strength, modulation mode, signal bandwidth, signal quality index and signal category; the signal category is used as a data label.

[0086] Example 3, see Figure 1 and Figure 3 This embodiment is based on the above embodiment. In step S2, signal preprocessing is to process the signals of all time steps, construct the preprocessed signal features and labels into a data set, and divide it into a test set and a training set. Specifically, the following steps are included:

[0087] Step S21: Calculate frequency domain data, construct input vector, and extract current time step from time series data and the previous Q samples as input vector , perform Fourier transform on the input data of the current time step and the previous time step; the formula used is as follows:

[0088] ;

[0089] Where, is frequency domain data; is the input of the previous time step; FFT[·] is the Fourier transform;

[0090] Step S22: Calculate the linear filter output, perform Hadamard product on the frequency domain data and the weight vector of the current time step, and then perform inverse Fourier transform to obtain the last Q elements of the filter output, which are expressed as follows:

[0091] ;

[0092] Where, is the filter output; is the Hadamard product; is the weight vector; IFFT is the inverse Fourier transform;

[0093] Step S23: Construct an interpolation matrix and calculate the interpolation position u for each sample l and node index i l , generate the interpolation vector and spline node matrix P of the current sample, and construct the interpolation matrix U; the formula used is as follows:

[0094] ;

[0095] ;

[0096] ;

[0097] ;

[0098] Where, is the filter at time step The output signal value of is the parameter used to scale the signal; is the rounding operation; M is the node offset; is the interpolation vector used to construct the interpolation matrix; is the column vector used to construct the spline knot matrix;

[0099] Step S24: Calculate the nonlinear output. Calculate the output of the nonlinear part and the error signal using the Catmull-Rom spline interpolation method. The formula used is as follows:

[0100] ;

[0101] ;

[0102] Where, is the nonlinear output vector in the frequency domain; C is the coefficient matrix in the spline interpolation; It is a sum operation on the rows of the matrix; It is expectation; is the error;

[0103] Step S25: Calculate the derivative of the error function to update the weights and spline nodes;

[0104] ;

[0105] Where, is the derivative of the error function; It is a parameter used to adjust the amplitude of the function gradient;

[0106] Step S26: Update the error vector; calculate the sensitivity of the linear filter output to the error and update the error vector; the formula used is as follows:

[0107] ;

[0108] ;

[0109] Where, is the sensitivity of the output vector to error; is the derivative matrix of the interpolation matrix; is the updated error vector;

[0110] Step S27: Update the frequency domain error vector and perform FFT transformation on the error vector; the formula used is as follows:

[0111] ;

[0112] Where, is the updated frequency domain error vector;

[0113] Step S28: Update the weight vector gradient. Calculate the gradient of the weight vector through IFFT and take the first 𝑄 elements. The formula used is as follows:

[0114] ;

[0115] ;

[0116] Where, is the gradient of the weight vector; is the complex conjugate; and are the frequency domain weight vectors before and after updating respectively; is the learning rate of the weight vector;

[0117] Step S29: Update the spline node matrix and update the parameters in the spline node matrix; the formula used is as follows:

[0118] ;

[0119] Where, and P are the spline knot matrices before and after updating, respectively; is the learning rate of the spline knot matrix; T is the matrix transpose.

[0120] By performing the above operations, the problems of poor original signal quality, serious noise interference, low signal processing accuracy and stability, which lead to poor monitoring effect in general radio monitoring methods are addressed. For signal processing at all time steps, this scheme uses Fourier transform to convert time series data into frequency domain data, and uses Hadamard product and inverse Fourier transform to obtain the output of the filter, thereby reducing steady-state error and improving the accuracy and stability of signal processing; the Catmull-Rom spline interpolation method is applied to construct the interpolation matrix, making signal reconstruction more accurate and efficient; the output of the nonlinear part is calculated based on the Catmull-Rom spline interpolation method, thereby improving the model's sensitivity and accuracy to signal changes; the gradient of the error function is calculated through the derivative form of the exponential function, and the gradient of the weight vector and the spline node matrix are calculated using FFT and inverse FFT, thereby improving the stability and convergence speed of error back propagation, thereby improving the radio monitoring effect.

[0121] Example 4, see Figure 1 and Figure 4 This embodiment is based on the above embodiment. In step S3, establishing a radio monitoring model specifically includes the following steps:

[0122] Step S31: Input layer design, taking the preprocessed data set as input;

[0123] Step S32: Design a feature extraction layer, use a sliding window to segment the long time series data, extract short time series signal samples, perform multi-wavelet transform on each short time series signal sample, and decompose the signal into multiple wavelet coefficients;

[0124] Step S33: Dimensionality reduction processing, applying principal component analysis to each wavelet coefficient, and using the SE module to adjust the importance of different channels in the feature map to improve the expressiveness and robustness of the features;

[0125] Step S34: Network optimization, integrating the SE module into the residual block unit; the residual block unit includes a convolutional layer, a batch normalization layer, a ReLU activation function, a global average pooling layer, and a fully connected layer; in the SE module, the weights of each channel in the feature map are adaptively learned and adjusted through global average pooling and a multi-layer perceptron; the weighted features are added to the original features in the shortcut connection to enhance the network's recognition ability;

[0126] Step S35: Design the output layer, using a fully connected layer and softmax activation function for final classification;

[0127] Step S36: Model determination. A determination threshold is pre-set. If the radio monitoring model converges on the training set, the radio monitoring model training is completed. If the prediction accuracy of the trained radio monitoring model on the test set is higher than the determination threshold, the radio monitoring model is established. Otherwise, the process goes to step S37 for parameter optimization.

[0128] Step S37: Model parameter optimization, specifically including:

[0129] Step S371: Initialization: establishing a search space based on the initial weights, biases, and learning rate of the radio monitoring model network; randomly initializing individual positions, and using the prediction accuracy of the test set by the radio monitoring model trained based on the individual positions as the individual fitness value;

[0130] Step S372: Design an adaptive factor to dynamically change the global and local search capabilities, as follows:

[0131] ;

[0132] Where l is the adaptive factor; t is the current number of iterations, and maxt is the maximum number of iterations; is a smoothing term;

[0133] Step S373: Design a mobility strategy and perform location update based on the adaptive factor. The formula used is as follows:

[0134] ;

[0135] ;

[0136] Where, and are the positions of the individual j-th dimension at the t+1th iteration and the tth iteration respectively; rand is a random number between 0 and 1; h is the adaptive inertia weight; h max and h min are the maximum inertia weight and the minimum inertia weight respectively; is the optimal individual position of the population; p is the average fitness of the population;

[0137] Step S374: Dynamic population, using a dynamic population strategy. If the updated population increases, the positions of the added individuals are initialized; if the updated population decreases, the last individual is collected and eliminated. The formula used to update the population is as follows:

[0138] ;

[0139] Where, and are the population sizes before and after the update, respectively; round(·) means rounding; is a random number between -0.1 and 0.1; sign is a sign function, which takes 1 if p is greater than 0.7, otherwise takes -1;

[0140] Step S375: Optimization judgment. A fitness threshold is set in advance. If there is an individual fitness value higher than the fitness threshold, the individual position is the optimal parameter setting of the radio monitoring model, and the radio monitoring model is established. If the maximum number of iterations is reached, return to step S371; otherwise, return to step S372.

[0141] By performing the above operations, in view of the problems of poor expression and robustness of model features, weak recognition ability of the network, and low population stability and diversity of the parameter optimization method in general radio monitoring methods, which lead to poor monitoring accuracy, this scheme adopts sliding window technology to segment time series data, and decomposes each sample through multi-wavelet transform to extract rich feature information; the SE module is integrated into the residual block unit. The SE module adaptively learns and adjusts the weights of each channel in the feature map through global average pooling and multi-layer perceptron, effectively enhancing the recognition ability of the network; by designing adaptive factors and movement strategies, the global and local search capabilities are dynamically adjusted to optimize the convergence speed and stability of the model; a dynamic population strategy is used to manage the population size, and the individual position is dynamically updated according to the population changes to achieve more effective parameter optimization and model establishment, and the number of individuals is dynamically adjusted according to the changes in the population size to ensure the diversity and stability of the population; thereby achieving superior model performance.

[0142] Example 5, see Figure 1 This embodiment is based on the above embodiment. In step S4, radio monitoring is based on the established radio monitoring model; radio signal spectrum data, signal parameters, environmental noise, timestamp, signal strength, modulation mode, signal bandwidth and signal quality indicators are collected in real time; after preprocessing, they are input into the radio monitoring model, and the signal category based on the model output is used as the radio monitoring result.

[0143] Example 6, see Figure 2, this embodiment is based on the above embodiment, and the present invention provides a radio monitoring system, including a signal acquisition module, a signal preprocessing module, a radio monitoring model establishment module and a radio monitoring module;

[0144] The signal acquisition module collects historical radio monitoring data and sends the data to the signal preprocessing module;

[0145] The signal preprocessing module processes the signals of all time steps, converts the time series data into frequency domain data using Fourier transform, obtains the filter output using Hadamard product and inverse Fourier transform, constructs the interpolation matrix using Catmull-Rom spline interpolation method, calculates the output of the nonlinear part based on Catmull-Rom spline interpolation method, calculates the gradient of the error function using the derivative form of the exponential function, and calculates the gradient of the weight vector and the spline node matrix using FFT and inverse FFT; and sends the data to the radio monitoring model establishment module;

[0146] The radio monitoring model building module uses a sliding window technique to segment time series data, and each sample is decomposed through a multi-wavelet transform. The SE module is integrated into the residual block unit. The SE module adaptively learns and adjusts the weights of each channel in the feature map through global average pooling and a multi-layer perceptron. By designing adaptive factors and movement strategies, the global and local search capabilities are dynamically adjusted. A dynamic population strategy is used to manage the population size, and individual positions are dynamically updated according to population changes to achieve more efficient parameter optimization and model building. The data is then sent to the radio monitoring module.

[0147] The radio monitoring module predicts the signal type based on the established radio monitoring model to realize radio monitoring.

[0148] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0149] While the embodiments of the present invention have been shown and described, it will be apparent to those skilled in the art that various changes, modifications, substitutions, and alterations can be made to the embodiments without departing from the principles and spirit of the invention.

[0150] The present invention and its embodiments are described above. This description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if a person skilled in the art is inspired by this and, without departing from the purpose of the present invention, designs structures and embodiments similar to this technical solution without inventiveness, they shall fall within the scope of protection of the present invention.

Claims

1. A radio monitoring method, characterized in that: The method comprises the following steps: Step S1: signal acquisition, collecting historical radio monitoring data; Step S2: Signal preprocessing: For signal processing at all time steps, the time series data is converted into frequency domain data using Fourier transform, and the filter output is obtained using Hadamard product and inverse Fourier transform; the interpolation matrix is ​​constructed using the Catmull-Rom spline interpolation method; the output of the nonlinear part is calculated based on the Catmull-Rom spline interpolation method; the gradient of the error function is calculated through the derivative form of the exponential function, and the gradient of the weight vector and the spline node matrix is ​​calculated using FFT and inverse FFT; Step S3: Establish a radio monitoring model, using a sliding window technique to segment time series data, and decompose each sample using a multi-wavelet transform. The SE module is integrated into the residual block unit. The SE module adaptively learns and adjusts the weights of each channel in the feature map through global average pooling and a multi-layer perceptron. By designing adaptive factors and movement strategies, global and local search capabilities are dynamically adjusted. A dynamic population strategy is used to manage population size, and individual positions are dynamically updated based on population changes to achieve more efficient parameter optimization and model building. Step S4: radio monitoring; Step S2 includes step S25: calculating the derivative of the error function for updating the weights and spline nodes; ； Where, is the derivative of the error function; It is a parameter used to adjust the amplitude of the function gradient; is the error; In step S3, establishing the radio monitoring model specifically includes the following steps: Step S31: Input layer design, taking the preprocessed data set as input; Step S32: Design a feature extraction layer, use a sliding window to segment the long time series data, extract short time series signal samples, perform multi-wavelet transform on each short time series signal sample, and decompose the signal into multiple wavelet coefficients; Step S33: Dimensionality reduction processing, applying principal component analysis to each wavelet coefficient, and using the SE module to adjust the importance of different channels in the feature map to improve the expressiveness and robustness of the features; Step S34: Network optimization, integrating the SE module into the residual block unit; the residual block unit includes a convolutional layer, a batch normalization layer, a ReLU activation function, a global average pooling layer, and a fully connected layer; in the SE module, the weights of each channel in the feature map are adaptively learned and adjusted through global average pooling and a multi-layer perceptron; the weighted features are added to the original features in the shortcut connection to enhance the network's recognition ability; Step S35: Design the output layer, using a fully connected layer and softmax activation function for final classification; Step S36: Model determination. A determination threshold is pre-set. If the radio monitoring model converges on the training set, the radio monitoring model training is completed. If the prediction accuracy of the trained radio monitoring model on the test set is higher than the determination threshold, the radio monitoring model is established. Otherwise, the process goes to step S37 for parameter optimization. Step S37: Model parameter optimization.

2. A radio monitoring method according to claim 1, characterized in that: In step S2, the signal preprocessing is to process the signals of all time steps, construct the preprocessed signal features and labels into a data set, and divide it into a test set and a training set; The specific steps include: Step S21: Calculate frequency domain data, construct input vector, and extract current time step from time series data and the previous Q samples as input vector , perform Fourier transform on the input data of the current time step and the previous time step; the formula used is as follows: ； Where, is frequency domain data; is the input of the previous time step; FFT[·] is the Fourier transform; Step S22: Calculate the linear filter output, perform Hadamard product on the frequency domain data and the weight vector of the current time step, and then perform inverse Fourier transform to obtain the last Q elements of the filter output, which are expressed as follows: ； Where, is the filter output; is the Hadamard product; is the weight vector; IFFT is the inverse Fourier transform; Step S23: Construct an interpolation matrix and calculate the interpolation position u for each sample l and node index i l , generate the interpolation vector and spline node matrix P of the current sample, and construct the interpolation matrix U; the formula used is as follows: ； ； ； ； Where, is the filter at time step The output signal value of is the parameter used to scale the signal; is the rounding operation; M is the node offset; is the interpolation vector used to construct the interpolation matrix; is the column vector used to construct the spline knot matrix; Step S24: Calculate the nonlinear output. Calculate the output of the nonlinear part and the error signal using the Catmull-Rom spline interpolation method. The formula used is as follows: ； ； Where, is the nonlinear output vector in the frequency domain; C is the coefficient matrix in the spline interpolation; It is a sum operation on the rows of the matrix; It is expectation; is the error; Step S25: Calculate the derivative of the error function; Step S26: Update the error vector; calculate the sensitivity of the linear filter output to the error and update the error vector; the formula used is as follows: ； ； Where, is the sensitivity of the output vector to error; is the derivative matrix of the interpolation matrix; is the updated error vector; Step S27: Update the frequency domain error vector and perform FFT transformation on the error vector; the formula used is as follows: ； Where, is the updated frequency domain error vector; Step S28: Update the weight vector gradient. Calculate the gradient of the weight vector through IFFT and take the first Q elements. The formula used is as follows: ； ； Where, is the gradient of the weight vector; is the complex conjugate; and are the frequency domain weight vectors before and after updating respectively; is the learning rate of the weight vector; Step S29: Update the spline node matrix and update the parameters in the spline node matrix; the formula used is as follows: ； Where, and P are the spline knot matrices before and after updating, respectively; is the learning rate of the spline knot matrix; T is the matrix transpose.

3. A radio monitoring method according to claim 1, characterized in that: In step S1, the signal collection is to collect historical radio monitoring data; the historical radio monitoring data includes radio signal spectrum data, signal parameters, environmental noise, timestamp, signal strength, modulation mode, signal bandwidth, signal quality index and signal category; the signal category is used as a data label.

4. A radio monitoring method according to claim 1, characterized in that: In step S4, the radio monitoring is based on the established radio monitoring model; real-time collection of radio signal spectrum data, signal parameters, environmental noise, timestamp, signal strength, modulation mode, signal bandwidth and signal quality indicators; After preprocessing, the signal is input into the radio monitoring model, and the signal category based on the model output is used as the radio monitoring result.

5. A radio monitoring system, configured to implement a radio monitoring method according to any one of claims 1 to 4, characterized in that: It includes signal acquisition module, signal preprocessing module, radio monitoring model building module and radio monitoring module; The signal acquisition module collects historical radio monitoring data and sends the data to the signal preprocessing module; The signal preprocessing module processes the signals of all time steps, converts the time series data into frequency domain data using Fourier transform, obtains the filter output using Hadamard product and inverse Fourier transform, constructs the interpolation matrix using Catmull-Rom spline interpolation method, calculates the output of the nonlinear part based on Catmull-Rom spline interpolation method, calculates the gradient of the error function using the derivative form of the exponential function, and calculates the gradient of the weight vector and the spline node matrix using FFT and inverse FFT; and sends the data to the radio monitoring model establishment module; The radio monitoring model building module uses a sliding window technique to segment time series data, and each sample is decomposed through a multi-wavelet transform. The SE module is integrated into the residual block unit. The SE module adaptively learns and adjusts the weights of each channel in the feature map through global average pooling and a multi-layer perceptron. By designing adaptive factors and movement strategies, the global and local search capabilities are dynamically adjusted. A dynamic population strategy is used to manage the population size, and individual positions are dynamically updated according to population changes to achieve more efficient parameter optimization and model building. The data is then sent to the radio monitoring module. The radio monitoring module predicts the signal type based on the established radio monitoring model to realize radio monitoring.

Images