Frequency modulation broadcast intelligent monitoring method based on deep learning

Abstract

The invention discloses a frequency modulation broadcast intelligent monitoring method based on deep learning, which relates to the technical field of broadcast monitoring, and comprises the following steps: setting a frequency modulation broadcast intelligent monitoring area, and carrying out broadcast spectrum acquisition in the set frequency modulation broadcast intelligent monitoring area; processing the acquired broadcast spectrum data in a data processing mode; after processing is completed, the processed broadcast spectrum data are analyzed through a data analysis method, meanwhile, spectrum classification is conducted on the analyzed broadcast spectrum data through a feature extraction algorithm, finally, the classified broadcast spectrum data are predicted through a model prediction method, and predicted broadcast spectrum data are obtained. Meanwhile, the illegal frequency in the frequency modulation broadcast is dynamically captured and forbidden according to the predicted broadcast spectrum data, so that the accuracy of intelligent monitoring of the frequency modulation broadcast is improved.

Description

Technical Field

[0001] This invention relates to the field of broadcast monitoring technology, and specifically to an intelligent monitoring method for FM broadcasts based on deep learning. Background Technology

[0002] On existing FM radio frequency bands, there are often radio stations that are illegally occupied. These illegally occupied radio stations often use substandard or counterfeit equipment with high power output, which can easily interfere with the communication systems used by aviation and emergency services, posing a great safety hazard.

[0003] The prior art, such as the invention patent application with publication number CN108880718A, discloses a real-time monitoring system, receiver, and real-time monitoring method for FM broadcasting. The method includes: a signal generator that outputs an audio signal; a transmitter that receives the audio signal, modulates it, and then transmits an audio modulation signal; and a receiver that receives the audio modulation signal and demodulates it to obtain an audio demodulated signal. The receiver includes a real-time monitoring module that receives and processes the audio signal and the audio demodulated signal in real time to output transmitter performance parameters.

[0004] As can be seen from the above solutions, current FM broadcast monitoring methods often focus on monitoring the transmitted signal itself, neglecting to monitor the broadcast signal power when the broadcast frequency band is illegally occupied, which has certain limitations. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent monitoring method for FM broadcasts based on deep learning, which solves the problem in the background art that illegally occupied broadcast stations can easily interfere with the communication systems used by aviation and emergency services.

[0006] To address the technical problem that illegally occupied radio stations can easily interfere with communication systems used by aviation and emergency services, the present invention adopts the following technical solution: This invention provides an intelligent monitoring method for FM broadcasts based on deep learning, specifically including the following steps: S1. Set up an intelligent monitoring area for FM broadcasting, collect broadcast spectrum within the set intelligent monitoring area for FM broadcasting, and obtain the collected broadcast spectrum data through data processing methods. S2. The collected broadcast spectrum data is processed using data processing methods to obtain processed broadcast spectrum data; S21. The broadcast spectrum data is processed by data noise reduction to obtain the noise-reduced broadcast spectrum data; S22. After noise reduction, the noise-reduced broadcast spectrum data is divided by channel division to obtain the divided broadcast spectrum data. S23. After summarizing and processing the divided broadcast spectrum data, the broadcast spectrum data is obtained. S3. Analyze the processed broadcast spectrum data using data analysis methods to obtain the analyzed broadcast spectrum data; S4. Perform spectrum classification on the analyzed broadcast spectrum data using a feature extraction algorithm, and output the spectrum-classified broadcast spectrum data. S5. The classified broadcast spectrum data is predicted using a model prediction method to obtain the predicted broadcast spectrum data. At the same time, illegal frequencies in FM broadcasts are dynamically captured and blocked based on the predicted broadcast spectrum data.

[0007] Preferably, the process of setting up an intelligent monitoring area for FM broadcasting, collecting broadcast spectrum data within that area, and simultaneously processing the collected broadcast spectrum data using data processing methods includes the following steps: S11. Based on the frequency band of FM broadcasting, set up multiple sets of spectrum sensor nodes to collect broadcast spectrum within the intelligent monitoring area of ​​FM broadcasting, and obtain broadcast spectrum data; set up one set of spectrum sensor nodes corresponding to one frequency band of FM broadcasting; Set the broadcast spectrum acquisition time interval, and construct a set of acquisition time points based on the set broadcast spectrum acquisition time interval. , Indicates the set number Each data collection time point This indicates the total number of data collection time points set. S12. Summarize the broadcast spectrum data collected at each acquisition time point and from multiple sets of spectrum sensor nodes, and construct the broadcast spectrum data matrix A; The arrangement of multiple sets of spectrum sensor nodes is set as a matrix arrangement, which includes R×H spectrum sensor nodes; The broadcast spectrum data expression for each spectrum sensor node is defined as follows: ; in, This represents the broadcast spectrum data collected by the i-th spectrum sensor node. Indicates valid broadcast spectrum data. Indicates signal noise; The broadcast spectrum data matrix is ​​constructed as follows: ; in, Indicates the first [unit / item] within the FM broadcast intelligent monitoring area. The collection point is at the [number]th [location]. Broadcast spectrum data collected at various time points.

[0008] Preferably, the step of performing data noise reduction on the broadcast spectrum data to obtain the noise-reduced broadcast spectrum data includes the following steps: S211. Decompose the broadcast spectrum data using singular value decomposition to obtain the decomposed broadcast spectrum data; S212. The decomposed broadcast spectrum data... The data is filtered and denoised, and then aggregated to obtain the denoised broadcast spectrum data. The expression for filtering and noise reduction is as follows: ; Where W represents the total number of filtering iterations; This indicates the j-th guided filter. This represents the filtered broadcast spectrum data; The broadcast spectrum data after multiple rounds of filtering is set as the broadcast spectrum data after noise reduction.

[0009] Preferably, the step of decomposing the broadcast spectrum data using singular value decomposition to obtain the decomposed broadcast spectrum data includes the following steps: The broadcast spectrum data matrix A is multiplied by another identical broadcast spectrum data matrix A through matrix multiplication. The result of the matrix multiplication is then used to obtain the autocorrelation matrix of the broadcast spectrum data through transpose and conjugate calculation. The obtained autocorrelation matrix is ​​decomposed using singular value decomposition (SVD). The obtained autocorrelation matrix is ​​decomposed into three matrices by singular value decomposition. The three matrices involved in the decomposition include: a diagonal matrix, a left unitary matrix, and a right unitary matrix; The values ​​on the diagonal of the diagonal matrix are defined as the singular values ​​of the autocorrelation matrix, and the left and right unitary matrices are orthogonal identity matrices. The singular values ​​in a diagonal matrix are arranged from largest to smallest, and the singular vectors corresponding to these singular values ​​indicate the direction of matrix transformation. The singular vectors corresponding to the singular values ​​in the diagonal matrix obtained after decomposition are defined as the spectral characteristics of the received broadcast spectrum data; The spectral features of each row in the decomposed diagonal matrix are transformed in the direction corresponding to the spectral features by matrix multiplication with the left unitary matrix and the right unitary matrix, and the transformations in other directions are filtered out to obtain the decomposed broadcast spectrum data corresponding to each row of spectral features in the diagonal matrix. The processed broadcast spectrum data corresponding to the spectral features of each row in the diagonal matrix are summarized to obtain the decomposed broadcast spectrum data. .

[0010] Preferably, after noise reduction, the broadcast spectrum data is divided into segments using a channel division method to obtain the segmented broadcast spectrum data, which includes the following steps: The denoised broadcast spectrum data is processed by frame averaging, combining several consecutive frames of broadcast spectrum data into one frame. The formula for frame averaging is shown below: ; in, This represents the averaged broadcast spectrum data. No. The broadcast spectrum data of the frame, where n represents the number of broadcast spectrum data collected and m represents the number of frames of broadcast spectrum data collected; After synthesis, the noise-reduced broadcast spectrum data is divided by setting the bandwidth based on the frequency band range of FM broadcast; The partitioning expression is as follows: (FM broadcast maximum frequency band - FM broadcast minimum frequency band) / set bandwidth = divided broadcast spectrum data; Each group of divided broadcast spectrum data is designated as a channel.

[0011] Preferably, the step of analyzing the processed broadcast spectrum data using data analysis methods to obtain the analyzed broadcast spectrum data includes the following steps: The broadcast spectrum data for each channel in the processed broadcast spectrum data is summarized, and the average broadcast spectrum data for the current channel is calculated. The broadcast spectrum data of each channel is analyzed using peak detection. The expression for peak detection is as follows: ; in, This represents the total number of channels. Indicates the first Broadcast spectrum data collected at specific times, Indicates the noise floor; When the calculated average broadcast spectrum data is greater than the noise floor, it indicates that there is an anomaly in the current channel.

[0012] Preferably, the step of performing spectrum classification on the analyzed broadcast spectrum data using a feature extraction algorithm and outputting spectrum-classified broadcast spectrum data includes the following steps: The analyzed broadcast spectrum data is summarized and transformed into frequency domain data through Fourier transform, and the corresponding frequency domain image is saved. The Fourier transform converts the analyzed broadcast spectrum data into frequency domain data, which is an existing technology. S41. Divide the corresponding frequency domain image into K image data blocks of uniform size, and use the divided image data blocks as the input of the convolutional neural network. Furthermore, the convolutional neural network includes: convolutional layers, pooling layers, and fully connected layers; S42. Set the kernel size and stride of each convolutional layer in the convolutional neural network; and based on the divided image data blocks, move on the input image data blocks according to the set kernel size and stride, and perform convolution calculation on the data in the corresponding region during the movement, so as to realize feature extraction of each image data block. The formula for calculating convolution is as follows: ; in, This represents the input image data block. represents the weights of the corresponding convolution kernel, b represents the bias value, and F represents the output image data block features; Since convolutional and pooling layers are stacked continuously, and the output of the upper layer is the input of the lower layer, after the input image data block has completed all the convolution and pooling operations in the convolutional neural network, the features of the output image data block are passed into the fully connected layer. S43. Expand and combine the output image data block features through a fully connected layer to obtain broadcast spectrum feature data, and save it. By setting a similarity threshold through data similarity comparison, broadcast spectrum feature data that exceeds the similarity threshold is classified as a type of broadcast spectrum data, and the classified broadcast spectrum data is then aggregated.

[0013] Preferably, the step of predicting the classified broadcast spectrum data using a model prediction method to obtain predicted broadcast spectrum data, and then dynamically capturing and blocking illegal frequencies in FM broadcasts based on the obtained predicted broadcast spectrum data, includes the following steps: S51. Construct an LSTM network model; LSTM networks include input gates, output gates, and forget gates; S52. Input the obtained real-time collected broadcast spectrum data as the input threshold into the LSTM network, and input the classified broadcast spectrum data as candidate values ​​into the LSTM network. S53. Process the current input and the broadcast spectrum data from the previous time step using the forget gate, and then update them; S54. The broadcast spectrum data processed by the forget gate is output through the output gate to obtain the predicted broadcast spectrum data for the next time step. S55. Iteratively use the LSTM network to determine and obtain the trained LSTM network; Set an error threshold between the broadcast spectrum data predicted by the LSTM network and the real-time acquired broadcast spectrum data. When the error between the broadcast spectrum data predicted by the LSTM network and the real-time acquired broadcast spectrum data is within the set error threshold range, the iteration stops, and the trained LSTM network is obtained. When the error between the broadcast spectrum data predicted by the LSTM network and the real-time acquired broadcast spectrum data is not within the set error threshold, the weights in the LSTM network are updated. Each time the weights are updated, the current network weights are increased or decreased by 0.1. S56. Dynamically capture and block illegal frequencies in FM broadcasts based on the predicted broadcast spectrum data.

[0014] Preferably, the step of dynamically capturing and blocking illegal frequencies in FM broadcasts based on the predicted broadcast spectrum data includes the following steps: Set a threshold range for broadcast spectrum data management. When the predicted broadcast spectrum data is lower than the set threshold range, it is determined to be an illegal frequency. Increase the monitoring frequency of broadcast spectrum data and capture and monitor broadcast spectrum data that is lower than the set threshold range. After increasing the monitoring frequency of broadcast spectrum data, illegal frequency data can be located based on the area and time point of broadcast spectrum acquisition; When the predicted broadcast spectrum data is higher than the set threshold range, it is determined to be an illegal frequency. The broadcast spectrum feature data of the corresponding illegal frequency is saved, and the saved broadcast spectrum feature data is used as a reference sample for dynamic capture. Simultaneously, for broadcast spectrum data exceeding the set threshold range, illegal frequency data is located and blocked based on the region and time point of the broadcast spectrum collection.

[0015] The present invention also discloses an intelligent monitoring method for FM broadcasts based on deep learning, comprising: a data acquisition module, a data processing module, a data analysis module, a feature extraction and classification module, and a prediction module; The data acquisition module is used to collect broadcast spectrum data in the FM broadcast intelligent monitoring area in real time; The data processing module is used to process the collected broadcast spectrum data to obtain processed broadcast spectrum data; The data analysis module is used to analyze the processed broadcast spectrum data to obtain the analyzed broadcast spectrum data; The feature extraction and classification module is used to classify the analyzed broadcast spectrum data using a feature extraction algorithm. The prediction module is used to predict the classified broadcast spectrum data using model prediction methods.

[0016] The beneficial effects of this invention are as follows: (1) This invention sets up an intelligent monitoring area for FM broadcasts, collects broadcast spectrum within the set area, and processes the collected broadcast spectrum data through data processing. After processing, the processed broadcast spectrum data is analyzed through data analysis methods, and the analyzed broadcast spectrum data is classified through feature extraction algorithms. Finally, the classified broadcast spectrum data is predicted through model prediction methods to obtain the predicted broadcast spectrum data. At the same time, illegal frequencies in FM broadcasts are dynamically captured and blocked based on the predicted broadcast spectrum data, thereby improving the accuracy of intelligent monitoring of FM broadcasts.

[0017] (2) The present invention decomposes the broadcast spectrum data by singular value decomposition. After the decomposition is completed, the broadcast spectrum data is denoised by multiple filtering and denoising methods, which improves the effectiveness of broadcast spectrum data processing.

[0018] (3) The present invention divides the noise-reduced broadcast spectrum data by using a channel division method and calculates the channel data in the noise-reduced broadcast spectrum data by recording the frequency band range and bandwidth of FM broadcast, thereby improving the reliability of broadcast spectrum data division.

[0019] (4) The present invention analyzes the processed broadcast spectrum data by using data analysis methods and analyzes the broadcast spectrum data of each channel by using peak detection to determine whether there is any abnormality in each channel, thereby improving the security of broadcast spectrum data monitoring.

[0020] (5) The present invention converts the analyzed broadcast spectrum data into frequency domain data through Fourier transform and saves the corresponding frequency domain image. At the same time, the corresponding frequency domain image is input into a convolutional neural network, and the frequency domain image is extracted and classified by the convolutional neural network, thereby improving the accuracy of broadcast spectrum data classification.

[0021] (6) The present invention uses a model prediction method to predict the classified broadcast spectrum data. By constructing an LSTM network model, the real-time collected broadcast spectrum data is used as the input gate to the LSTM network, and the classified broadcast spectrum data is used as the candidate value to be input to the LSTM network. The prediction of the real-time collected broadcast spectrum data is completed by iterative processing, which improves the real-time performance of broadcast monitoring. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a schematic diagram of the intelligent monitoring method for FM broadcasting of the present invention. Detailed Implementation

[0024] 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 embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] In a specific embodiment of the present invention, Reference Figure 1 As shown, this invention provides an intelligent monitoring method for FM broadcasts based on deep learning, comprising the following steps: S1. Set up an intelligent monitoring area for FM broadcasting, collect broadcast spectrum within the set intelligent monitoring area for FM broadcasting, and obtain the collected broadcast spectrum data through data processing methods. S2. The collected broadcast spectrum data is processed using data processing methods to obtain processed broadcast spectrum data; S21. The broadcast spectrum data is processed by data noise reduction to obtain the noise-reduced broadcast spectrum data; S22. After noise reduction, the noise-reduced broadcast spectrum data is divided by channel division to obtain the divided broadcast spectrum data. S23. After summarizing and processing the divided broadcast spectrum data, the broadcast spectrum data is obtained. S3. Analyze the processed broadcast spectrum data using data analysis methods to obtain the analyzed broadcast spectrum data; S4. Perform spectrum classification on the analyzed broadcast spectrum data using a feature extraction algorithm, and output the spectrum-classified broadcast spectrum data. S5. Predict the classified broadcast spectrum data using model prediction methods to obtain the predicted broadcast spectrum data. At the same time, dynamically capture and block illegal frequencies in FM broadcasts based on the obtained predicted broadcast spectrum data. Furthermore, referring to Figure 1As shown, a smart monitoring area for FM broadcasting is set up, and broadcast spectrum is collected within the set smart monitoring area. Simultaneously, the collected broadcast spectrum data is processed using data processing methods, including the following steps: S11. Based on the frequency band of FM broadcasting, set up multiple sets of spectrum sensor nodes to collect broadcast spectrum within the intelligent monitoring area of ​​FM broadcasting, and obtain broadcast spectrum data; set up one set of spectrum sensor nodes corresponding to one frequency band of FM broadcasting; Furthermore, a broadcast spectrum acquisition time interval is set, and a set of acquisition time points is constructed based on the set broadcast spectrum acquisition time interval. , Indicates the set number Each data collection time point This indicates the total number of data collection time points set. S12. Summarize the broadcast spectrum data collected at each acquisition time point and from multiple sets of spectrum sensor nodes, and construct the broadcast spectrum data matrix A; The arrangement of multiple sets of spectrum sensor nodes is set as a matrix arrangement, which includes R×H spectrum sensor nodes; Furthermore, the broadcast spectrum data expression for each spectrum sensor node is defined as follows: ; in, This represents the broadcast spectrum data collected by the i-th spectrum sensor node. Indicates valid broadcast spectrum data. Indicates signal noise; Furthermore, the broadcast spectrum data matrix is ​​constructed as follows: ; in, Indicates the first [unit / item] within the FM broadcast intelligent monitoring area. The collection point is at the [number]th [location]. Broadcast spectrum data collected at various time points; Furthermore, referring to Figure 1 As shown, the process of denoising broadcast spectrum data to obtain denoised broadcast spectrum data includes the following steps: S211. Decompose the broadcast spectrum data using singular value decomposition to obtain the decomposed broadcast spectrum data; The broadcast spectrum data matrix A is multiplied by another identical broadcast spectrum data matrix A through matrix multiplication. The result of the matrix multiplication is then used to obtain the autocorrelation matrix of the broadcast spectrum data through transpose and conjugate calculation. Furthermore, the obtained autocorrelation matrix is ​​decomposed using singular value decomposition (SVD). The obtained autocorrelation matrix is ​​decomposed into three matrices by singular value decomposition. The three matrices involved in the decomposition include: a diagonal matrix, a left unitary matrix, and a right unitary matrix; Furthermore, the values ​​on the diagonal of the diagonal matrix are set to be the singular values ​​of the autocorrelation matrix, and the left unitary matrix and the right unitary matrix are orthogonal identity matrices. The singular values ​​in a diagonal matrix are arranged from largest to smallest, and the singular vectors corresponding to these singular values ​​indicate the direction of matrix transformation. The singular vectors corresponding to the singular values ​​in the diagonal matrix obtained after decomposition are defined as the spectral characteristics of the received broadcast spectrum data; The spectral features of each row in the decomposed diagonal matrix are transformed in the direction corresponding to the spectral features by matrix multiplication with the left unitary matrix and the right unitary matrix, and the transformations in other directions are filtered out to obtain the decomposed broadcast spectrum data corresponding to each row of spectral features in the diagonal matrix. The processed broadcast spectrum data corresponding to the spectral features of each row in the diagonal matrix are summarized to obtain the decomposed broadcast spectrum data. ; S212. The decomposed broadcast spectrum data... The data is filtered and denoised, and then aggregated to obtain the denoised broadcast spectrum data. The expression for filtering and noise reduction is as follows: ; Where W represents the total number of filtering iterations; This indicates the j-th guided filter. This represents the filtered broadcast spectrum data; The broadcast spectrum data after multiple rounds of filtering is set as the noise-reduced broadcast spectrum data; Furthermore, referring to Figure 1 As shown, after noise reduction, the broadcast spectrum data is divided into sections using a channel-based method to obtain the divided broadcast spectrum data, which includes the following steps: The denoised broadcast spectrum data is processed by frame averaging, combining several consecutive frames of broadcast spectrum data into one frame. The formula for frame averaging is shown below: ; in, This represents the averaged broadcast spectrum data. No. The broadcast spectrum data of the frame, where n represents the number of broadcast spectrum data collected and m represents the number of frames of broadcast spectrum data collected; After synthesis, the noise-reduced broadcast spectrum data is divided by setting the bandwidth based on the frequency band range of FM broadcast; The partitioning expression is as follows: (FM broadcast maximum frequency band - FM broadcast minimum frequency band) / set bandwidth = divided broadcast spectrum data; Furthermore, each group of divided broadcast spectrum data is designated as a channel; Furthermore, referring to Figure 1 As shown, the processed broadcast spectrum data is analyzed using data analysis methods. The analyzed broadcast spectrum data includes the following steps: The broadcast spectrum data for each channel in the processed broadcast spectrum data is summarized, and the average broadcast spectrum data for the current channel is calculated. Furthermore, the broadcast spectrum data of each channel is analyzed using peak detection. The expression for peak detection is as follows: ; in, This represents the total number of channels. Indicates the first Broadcast spectrum data collected at specific times, Indicates the noise floor; Furthermore, when the calculated average broadcast spectrum data is greater than the noise floor, it indicates that there is an anomaly in the current channel; Furthermore, referring to Figure 1 As shown, the analyzed broadcast spectrum data is classified using a feature extraction algorithm, and the output of the classified broadcast spectrum data includes the following steps: The analyzed broadcast spectrum data is summarized and transformed into frequency domain data through Fourier transform, and the corresponding frequency domain image is saved. S41. Divide the corresponding frequency domain image into K image data blocks of uniform size, and use the divided image data blocks as the input of the convolutional neural network. Furthermore, the convolutional neural network includes: convolutional layers, pooling layers, and fully connected layers; S42. Set the kernel size and stride of each convolutional layer in the convolutional neural network; and based on the divided image data blocks, move on the input image data blocks according to the set kernel size and stride, and perform convolution calculation on the data in the corresponding region during the movement, so as to realize feature extraction of each image data block. The formula for calculating convolution is as follows: ; in, This represents the input image data block. represents the weights of the corresponding convolution kernel, b represents the bias value, and F represents the output image data block features; Furthermore, since convolutional and pooling layers are stacked continuously, and the output of the upper layer is the input of the lower layer, after the input image data block completes all convolution and pooling operations in the convolutional neural network, the output image data block features are passed into the fully connected layer. S43. Expand and combine the output image data block features through a fully connected layer to obtain broadcast spectrum feature data, and save it. Furthermore, by setting a similarity threshold through data similarity comparison, broadcast spectrum feature data that exceeds the similarity threshold is classified as a type of broadcast spectrum data, and the classified broadcast spectrum data is obtained by summarizing the data. Furthermore, referring to Figure 1 As shown, the classified broadcast spectrum data is predicted using a model prediction method to obtain the predicted broadcast spectrum data. Simultaneously, based on the obtained predicted broadcast spectrum data, illegal frequencies in FM broadcasts are dynamically captured and blocked, including the following steps: S51. Construct an LSTM network model; LSTM networks include input gates, output gates, and forget gates; S52. Input the obtained real-time collected broadcast spectrum data as the input threshold into the LSTM network, and input the classified broadcast spectrum data as candidate values ​​into the LSTM network. S53. Process the current input and the broadcast spectrum data from the previous time step using the forget gate, and then update them; S54. The broadcast spectrum data processed by the forget gate is output through the output gate to obtain the predicted broadcast spectrum data for the next time step. S55. Iteratively use the LSTM network to determine and obtain the trained LSTM network; Set an error threshold between the broadcast spectrum data predicted by the LSTM network and the real-time acquired broadcast spectrum data. When the error between the broadcast spectrum data predicted by the LSTM network and the real-time acquired broadcast spectrum data is within the set error threshold range, the iteration stops, and the trained LSTM network is obtained. When the error between the broadcast spectrum data predicted by the LSTM network and the real-time acquired broadcast spectrum data is not within the set error threshold, the weights in the LSTM network are updated. Each time the weights are updated, the current network weights are increased or decreased by 0.1. S56. Dynamically capture and block illegal frequencies in FM broadcasts based on the predicted broadcast spectrum data; Set a threshold range for broadcast spectrum data management. When the predicted broadcast spectrum data is lower than the set threshold range, it is determined to be an illegal frequency. Increase the monitoring frequency of broadcast spectrum data and capture and monitor broadcast spectrum data that is lower than the set threshold range. After increasing the monitoring frequency of broadcast spectrum data, illegal frequency data can be located based on the area and time point of broadcast spectrum acquisition; When the predicted broadcast spectrum data is higher than the set threshold range, it is determined to be an illegal frequency. The broadcast spectrum feature data of the corresponding illegal frequency is saved, and the saved broadcast spectrum feature data is used as a reference sample for dynamic capture. At the same time, for broadcast spectrum data exceeding the set threshold range, illegal frequency data will be located and blocked based on the region and time point of broadcast spectrum collection; In one specific embodiment, the deep learning-based intelligent monitoring system for FM broadcasts is used to implement a deep learning-based intelligent monitoring method for FM broadcasts. The system includes: a data acquisition module, a data processing module, a data analysis module, a feature extraction and classification module, and a prediction module. The data acquisition module is used to collect broadcast spectrum data in the FM broadcast intelligent monitoring area in real time; The data processing module is used to process the collected broadcast spectrum data to obtain processed broadcast spectrum data; The data analysis module is used to analyze the processed broadcast spectrum data to obtain the analyzed broadcast spectrum data; The feature extraction and classification module is used to classify the analyzed broadcast spectrum data using a feature extraction algorithm. The prediction module is used to predict the classified broadcast spectrum data using model prediction methods.

[0026] It should be noted that: The above content is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined by the present invention, and all such modifications or additions should fall within the protection scope of the present invention.

Claims

1. A method for intelligent monitoring of FM broadcasts based on deep learning, characterized in that, Includes the following steps: S1. Set up an intelligent monitoring area for FM broadcasting, collect broadcast spectrum within the set intelligent monitoring area for FM broadcasting, and obtain the collected broadcast spectrum data through data processing methods. S2. The collected broadcast spectrum data is processed using data processing methods to obtain processed broadcast spectrum data; S21. The broadcast spectrum data is processed by data noise reduction to obtain the noise-reduced broadcast spectrum data; S22. After noise reduction, the noise-reduced broadcast spectrum data is divided by channel division to obtain the divided broadcast spectrum data. S23. After summarizing and processing the divided broadcast spectrum data, the broadcast spectrum data is obtained. S3. Analyze the processed broadcast spectrum data using data analysis methods to obtain the analyzed broadcast spectrum data; S4. Perform spectrum classification on the analyzed broadcast spectrum data using a feature extraction algorithm, and output the spectrum-classified broadcast spectrum data. S5. The classified broadcast spectrum data is predicted using a model prediction method to obtain the predicted broadcast spectrum data. At the same time, illegal frequencies in FM broadcasts are dynamically captured and blocked based on the predicted broadcast spectrum data.

2. The intelligent monitoring method for FM broadcasting based on deep learning according to claim 1, characterized in that, The process of setting up an intelligent monitoring area for FM broadcasting, collecting broadcast spectrum data within that area, and processing the collected broadcast spectrum data using data processing methods includes the following steps: S11. Based on the frequency band of FM broadcasting, set up multiple sets of spectrum sensor nodes to collect broadcast spectrum within the intelligent monitoring area of ​​FM broadcasting, and obtain broadcast spectrum data; set up one set of spectrum sensor nodes corresponding to one frequency band of FM broadcasting; Set the broadcast spectrum acquisition time interval, and construct a set of acquisition time points based on the set broadcast spectrum acquisition time interval. , Indicates the set number Each data collection time point This indicates the total number of data collection time points set. S12. Summarize the broadcast spectrum data collected at each acquisition time point and from multiple sets of spectrum sensor nodes, and construct the broadcast spectrum data matrix A; The arrangement of multiple sets of spectrum sensor nodes is set as a matrix arrangement, which includes R×H spectrum sensor nodes.

3. The intelligent monitoring method for FM broadcasting based on deep learning according to claim 1, characterized in that, The process of denoising the broadcast spectrum data to obtain denoised broadcast spectrum data includes the following steps: S211. Decompose the broadcast spectrum data using singular value decomposition to obtain the decomposed broadcast spectrum data; S212. The decomposed broadcast spectrum data... The data is filtered and denoised, and then aggregated to obtain the denoised broadcast spectrum data. The expression for filtering and noise reduction is as follows: ； Where W represents the total number of filtering iterations; This indicates the j-th guided filter. This represents the filtered broadcast spectrum data; The broadcast spectrum data after multiple rounds of filtering is set as the broadcast spectrum data after noise reduction.

4. The intelligent monitoring method for FM broadcasting based on deep learning according to claim 3, characterized in that, The process of decomposing the broadcast spectrum data using singular value decomposition to obtain the decomposed broadcast spectrum data includes the following steps: The broadcast spectrum data matrix A is multiplied by another identical broadcast spectrum data matrix A through matrix multiplication. The result of the matrix multiplication is then used to obtain the autocorrelation matrix of the broadcast spectrum data through transpose and conjugate calculation. The obtained autocorrelation matrix is ​​decomposed into three matrices by singular value decomposition. The three matrices involved in the decomposition include: a diagonal matrix, a left unitary matrix, and a right unitary matrix; The values ​​on the diagonal of the diagonal matrix are defined as the singular values ​​of the autocorrelation matrix, and the left and right unitary matrices are orthogonal identity matrices. The singular vectors corresponding to the singular values ​​in the diagonal matrix obtained after decomposition are defined as the spectral characteristics of the received broadcast spectrum data; The spectral features of each row in the decomposed diagonal matrix are transformed in the direction corresponding to the spectral features by matrix multiplication with the left unitary matrix and the right unitary matrix, and the transformations in other directions are filtered out to obtain the decomposed broadcast spectrum data corresponding to each row of spectral features in the diagonal matrix. The processed broadcast spectrum data corresponding to the spectral features of each row in the diagonal matrix are summarized to obtain the decomposed broadcast spectrum data. .

5. The intelligent monitoring method for FM broadcasting based on deep learning according to claim 1, characterized in that, After noise reduction, the broadcast spectrum data is divided into segments using a channel partitioning method to obtain the partitioned broadcast spectrum data, which includes the following steps: The denoised broadcast spectrum data is processed by frame averaging, combining several consecutive frames of broadcast spectrum data into one frame. The formula for frame averaging is shown below: ； in, This represents the averaged broadcast spectrum data. No. The broadcast spectrum data of the frame, where n represents the number of broadcast spectrum data collected and m represents the number of frames of broadcast spectrum data collected; After synthesis, the noise-reduced broadcast spectrum data is divided by setting the bandwidth based on the frequency band range of FM broadcast.

6. The intelligent monitoring method for FM broadcasting based on deep learning according to claim 1, characterized in that, The process of analyzing the processed broadcast spectrum data using data analysis methods to obtain the analyzed broadcast spectrum data includes the following steps: The broadcast spectrum data for each channel in the processed broadcast spectrum data is summarized, and the average broadcast spectrum data for the current channel is calculated. The broadcast spectrum data of each channel is analyzed using peak detection. The expression for peak detection is as follows: ； in, This represents the total number of channels. Indicates the first Broadcast spectrum data collected at specific times, This indicates the noise level at the floor.

7. The intelligent monitoring method for FM broadcasting based on deep learning according to claim 1, characterized in that, The step of performing spectrum classification on the analyzed broadcast spectrum data using a feature extraction algorithm and outputting spectrum-classified broadcast spectrum data includes the following steps: The analyzed broadcast spectrum data is summarized and transformed into frequency domain data through Fourier transform, and the corresponding frequency domain image is saved. S41. Divide the corresponding frequency domain image into K image data blocks of uniform size, and use the divided image data blocks as the input of the convolutional neural network. The convolutional neural network includes: convolutional layers, pooling layers, and fully connected layers; S42. Set the kernel size and stride of each convolutional layer in the convolutional neural network; and based on the divided image data blocks, move on the input image data blocks according to the set kernel size and stride, and perform convolution calculation on the data in the corresponding region during the movement, so as to realize feature extraction of each image data block. The formula for calculating convolution is as follows: ； in, This represents the input image data block. represents the weights of the corresponding convolution kernel, b represents the bias value, and F represents the output image data block features; S43. Expand and combine the output image data block features through a fully connected layer to obtain broadcast spectrum feature data, and save it. By setting a similarity threshold through data similarity comparison, broadcast spectrum feature data that exceeds the similarity threshold is classified as a type of broadcast spectrum data, and the classified broadcast spectrum data is then aggregated.

8. The intelligent monitoring method for FM broadcasting based on deep learning according to claim 1, characterized in that, The process of predicting the classified broadcast spectrum data using a model prediction method to obtain predicted broadcast spectrum data, and then dynamically capturing and blocking illegal frequencies in FM broadcasts based on the predicted broadcast spectrum data, includes the following steps: S51. Construct an LSTM network model; LSTM networks include input gates, output gates, and forget gates; S52. Input the obtained real-time collected broadcast spectrum data as the input threshold into the LSTM network, and input the classified broadcast spectrum data as candidate values ​​into the LSTM network. S53. Process the current input and the broadcast spectrum data from the previous time step using the forget gate, and then update them; S54. The broadcast spectrum data processed by the forget gate is output through the output gate to obtain the predicted broadcast spectrum data for the next time step. S55. Iteratively use the LSTM network to determine and obtain the trained LSTM network; Set an error threshold between the broadcast spectrum data predicted by the LSTM network and the real-time acquired broadcast spectrum data. When the error between the broadcast spectrum data predicted by the LSTM network and the real-time acquired broadcast spectrum data is within the set error threshold range, the iteration stops, and the trained LSTM network is obtained. When the error between the broadcast spectrum data predicted by the LSTM network and the real-time acquired broadcast spectrum data is not within the set error threshold, the weights in the LSTM network are updated. Each time the weights are updated, the current network weights are increased or decreased by 0.

1. S56. Dynamically capture and block illegal frequencies in FM broadcasts based on the predicted broadcast spectrum data.

9. The intelligent monitoring method for FM broadcasting based on deep learning according to claim 8, characterized in that, The process of dynamically capturing and blocking illegal frequencies in FM broadcasts based on the predicted broadcast spectrum data includes the following steps: Set a threshold range for broadcast spectrum data management. When the predicted broadcast spectrum data is lower than the set threshold range, it is determined to be an illegal frequency. Increase the monitoring frequency of broadcast spectrum data and capture and monitor broadcast spectrum data that is lower than the set threshold range. After increasing the monitoring frequency of broadcast spectrum data, illegal frequency data can be located based on the area and time point of broadcast spectrum acquisition; When the predicted broadcast spectrum data is higher than the set threshold range, it is determined to be an illegal frequency. The broadcast spectrum feature data of the corresponding illegal frequency is saved, and the saved broadcast spectrum feature data is used as a reference sample for dynamic capture. Simultaneously, for broadcast spectrum data exceeding the set threshold range, illegal frequency data is located and blocked based on the region and time point of the broadcast spectrum collection.

10. A method for intelligent monitoring of FM broadcasts based on deep learning as described in any one of claims 1-9, characterized in that, include: The system includes a data acquisition module, a data processing module, a data analysis module, a feature extraction and classification module, and a prediction module. The data acquisition module is used to collect broadcast spectrum data in the FM broadcast intelligent monitoring area in real time; The data processing module is used to process the collected broadcast spectrum data to obtain processed broadcast spectrum data; The data analysis module is used to analyze the processed broadcast spectrum data to obtain the analyzed broadcast spectrum data; The feature extraction and classification module is used to classify the analyzed broadcast spectrum data using a feature extraction algorithm. The prediction module is used to predict the classified broadcast spectrum data using model prediction methods.

Images