Frequency hopping frequency set estimation method and system based on depth pattern recognition model

Abstract

The invention discloses a frequency-hopping frequency set estimation method and system based on a depth pattern recognition model, and relates to the technical field of frequency-hopping communication, and the method comprises the steps: collecting in-phase orthogonal signal data containing frequency-hopping signals, carrying out the preprocessing, obtaining a point-like frequency-hopping graph, and dividing the point-like frequency-hopping graph into a training set and a test set; constructing a frequency hopping periodic pattern depth pattern recognition network model based on an attention mechanism, and performing training by using the training set; extracting a frequency hopping period pattern of the verification set by adopting the trained deep pattern recognition network model; and extracting a frequency hopping frequency set after matching and aligning the frequency hopping period pattern. The method comprises the following steps of: generating a dotted frequency hopping graph, performing mode analysis and identification on the dotted frequency hopping graph by adopting a frequency hopping period pattern depth mode identification network model based on an attention mechanism, extracting a frequency hopping period pattern in the dotted frequency hopping graph, and extracting a frequency hopping frequency set after matching and aligning the frequency hopping period pattern. And under the condition of relatively low algorithm complexity, the accuracy and robustness of frequency hopping frequency set estimation are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of frequency hopping communication, and in particular to a method and system for estimating a frequency hopping frequency set based on a deep pattern recognition model. Background Art

[0002] There are many types of existing technical methods and systems for estimating frequency hopping frequency sets, but from the perspective of estimation accuracy, complexity and robustness, the existing methods and systems have their own shortcomings. Traditional frequency hopping frequency set estimation methods and systems are mostly to obtain time-frequency domain data by processing the collected IQ signals using time-frequency analysis techniques (such as short-time Fourier transform STFT, Wignerville transform, wavelet transform, etc.), detect and extract the position information of the frequency hopping signal in the time and frequency domain based on energy detection, image processing and other methods, and then use probability statistics to estimate the frequency hopping frequency table. Among them, AI-based computer vision model technology is an image processing method that detects and identifies frequency hopping signals in time-frequency images by using deep image target detection algorithms (such as the YOLO series), outputs the start and end time and frequency coordinates of the frequency hopping signal, and finally statistically estimates the frequency hopping frequency table. The time-frequency analysis method based on AI deep image target detection has higher estimation accuracy than traditional non-AI time-frequency analysis methods. However, due to the lack of a complete training data set that can cover all complex environmental conditions, the model training used to learn to capture the characteristics of complex frequency hopping signals and their frequency variation patterns with channel environment and time leads to weak model generalization and poor robustness under actual conditions. How to achieve high estimation accuracy while ensuring low algorithm complexity and high robustness has become the focus and difficulty of the frequency hopping frequency set estimation problem. Deep pattern recognition is an important branch in the field of pattern recognition. It combines deep learning technology to realize the recognition and analysis of complex patterns. Deep pattern recognition uses deep neural networks (such as convolutional neural networks, recurrent neural networks, etc.) to process and analyze these patterns. It has strong nonlinear modeling capabilities. Deep learning models can automatically learn features from data and process complex patterns. Therefore, there is an urgent need for a frequency hopping frequency set estimation method and system based on a deep pattern recognition model. Summary of the invention

[0003] The purpose of the present invention is to overcome the defects of the prior art in common frequency hopping communication scenarios in practical applications, that is, it is not possible to maintain high estimation accuracy while maintaining low algorithm complexity, strong model generalization ability and high robustness, and to provide a frequency hopping frequency set estimation method and system based on a deep pattern recognition model.

[0004] The technical solution of the present invention is as follows: To achieve the above object, the first aspect of the present invention provides a method for estimating a frequency hopping frequency set based on a deep pattern recognition model, comprising the following steps: Collecting in-phase and quadrature signal data including frequency hopping signals and performing preprocessing to obtain a point-shaped frequency hopping graph and dividing the point-shaped frequency hopping graph into a training set and a test set; Constructing a frequency hopping periodic pattern deep pattern recognition network model based on an attention mechanism and training it using the training set; The trained frequency hopping periodic pattern deep pattern recognition network model based on the attention mechanism is used to extract the frequency hopping periodic pattern of the verification set; The frequency hopping frequency set is extracted after matching and aligning the frequency hopping period pattern.

[0005] Preferably, the pretreatment comprises: The collected in-phase orthogonal signal data containing the frequency hopping signal Perform time-frequency transformation to obtain a two-dimensional amplitude spectrum time-frequency diagram:

[0006] in, , T represents the number of time frames, , L represents the number of frequency points, represents the short-time Fourier transform, Indicates that the signal data x is an M-dimensional complex vector, represents the complex field, M is the vector dimension, represents the field of real numbers, represents a real matrix of dimension T×L; The obtained time-frequency diagram Perform noise reduction, specifically, perform corrosion and expansion operations on the time-frequency graph according to the time dimension to remove mixed nonlinear structural signals and noise points; The expression of the erosion operation is as follows:

[0007] After the erosion operation is completed, the expansion operation is performed. The expansion operation expression is as follows:

[0008] in, Indicates that P takes a positive integer of the same magnitude as the time length of the single-frequency signal in the time-frequency diagram, represents a set of positive integers; After the expansion operation is completed, the denoised time-frequency graph is obtained , edge detection, contour detection and rectangle center extraction are used to transform the time-frequency graph Convert to a dotted frequency hopping diagram.

[0009] Preferably, the time-frequency graph is sequentially detected by edge detection, contour detection and rectangle center extraction. Converting to a dotted frequency hopping diagram includes the following steps: The time-frequency diagram after noise reduction Normalize element by element and linearly map to [0,255] to get ,right Perform binarization judgment to obtain a binary image:

[0010] Wherein, U is the preset noise reduction threshold; The edge detection algorithm and contour detection algorithm are used in turn to detect possible frequency hopping signals in the time-frequency graph and obtain the signal contour image:

[0011] in, It is the Canny edge detection function of the open source computer vision library OpenCV. and They are the low threshold and high threshold of the Canny edge detection function respectively;

[0012] in, It is the contour detection function of the open source computer vision library OpenCV. The contour retrieval mode uses , that is, only the external contour is retrieved, and the contour approximation method is adopted , that is, keep all contour points; right Perform rectangular frame analysis and extraction to obtain a rectangular set :

[0013] in, It is a rectangular contour fitting function in the open source computer vision library OpenCV, which is used to detect and fit the set of circumscribed rectangular contours in the image:

[0014] in, is the coordinate of the upper left corner of the rectangle, , are the width and height of the rectangle respectively, and V is the number of detected rectangular boxes; use Calculate the center coordinates of each rectangle to generate a dotted frequency hopping pattern: .

[0015] Preferably, the frequency hopping periodic pattern deep pattern recognition network model FHPR based on the attention mechanism adopts an encoder-attention module-decoder structure; the encoder includes 3 convolution blocks, each convolution block includes a 3x3 convolution layer, batch normalization and ReLU activation function, and a 2x2 maximum pooling layer is added after each convolution block. The encoder part finally passes through two consecutive convolution layers to further extract features; the decoder includes 3 deconvolution blocks, each deconvolution block includes a 3x3 deconvolution layer, batch normalization and ReLU activation function, and a 2x2 upsampling layer corresponding to the 2x2 maximum pooling layer in the encoder is added before each deconvolution block, and each layer in the decoder is connected to the corresponding layer in the encoder through a jump connection; an attention module is also arranged between the encoder and the decoder to strengthen the important features of the point-like frequency hopping pattern.

[0016] Preferably, the input of the model is of dimension The dot-shaped frequency hopping graph first passes through the encoder part to extract multi-scale features, and then gradually reduces the size of the feature map of the multi-scale features and increases the number of channels through convolution and pooling operations; The important features of the dot-like frequency-hopping pattern are reinforced by an attention module between the encoder and decoder; Through the decoder part, the feature map size is gradually upsampled, the number of channels is reduced, and the output is reconstructed. At the same time, the features of the encoder are fused using jump connections to retain detail information. Finally, a 1x1 convolutional layer is used to map the decoder features to the final output, and the frequency hopping periodic pattern recognition result with the same dimension as the dot-like frequency hopping map is output. After that, segmentation is performed to independently separate each periodic pattern pattern area in the frequency hopping periodic pattern.

[0017] Preferably, the attention module divides the input feature map into groups, each group size is , the grouped input feature map passes through two parallel branches of the attention module; The first branch uses average pooling along the X and Y directions to extract global information, and then performs feature fusion through splicing and 1×1 convolutional layers. Then, the attention maps in the X and Y directions are generated by the Sigmoid function, and then weight adjustment, group normalization and average pooling operations are performed. Finally, the attention map of the first branch is generated through the Softmax function. The second branch uses a 3×3 convolutional layer to capture local spatial information, followed by an average pooling operation and then normalization using the Softmax function to generate the attention map of the second branch; The attention maps generated by these two branches are subjected to matrix multiplication operations to generate the final feature representation, and then the final feature map is generated through the Sigmoid function. The final feature map is combined with the input feature map after addition and multiplication operations, that is, weight adjustment, to generate the final output features of the attention module.

[0018] Preferably, the loss function of the frequency hopping periodic pattern deep pattern recognition network model FHPR based on the attention mechanism is:

[0019]

[0020]

[0021] in, , is the weight parameter, , usually taken , is the binary cross entropy loss, which computes the cross entropy between the true label and the predicted value for each pixel. The Dice loss is used to directly measure the overlap between the predicted frequency hopping cycle pattern area and the actual frequency hopping cycle pattern area to judge the segmentation effect of the model. is the total number of pixels in the image, is the true label of the ith pixel, when the ith pixel belongs to the frequency hopping periodic pattern area is 1 when the i-th pixel belongs to the background area is 0, is the predicted value of the ith pixel, indicating the probability that the pixel belongs to the frequency hopping period pattern area, is a constant that prevents the denominator from being zero.

[0022] Preferably, the matching and alignment comprises: Use the first frequency hopping pattern As a template, the remaining frequency hopping patterns The pattern to be matched is used for template matching and alignment. The template matching and alignment rule adopts the normalized square difference matching method:

[0023] Among them, Q is the number of frequency hopping patterns extracted by the model, To make With template Matches the alignment offset.

[0024] Preferably, the step of extracting the frequency hopping frequency set is specifically: Perform statistical averaging on the matched and aligned frequency hopping patterns by pixel:

[0025] Using the non-maximum suppression method, the frequency hopping pattern after template alignment To find the local maximum point, first, filter The pixel value is greater than the given threshold The pixel coordinates are sorted according to the pixel values ​​from large to small. The pixels are sorted, and then the sorted pixels are traversed in order, and the positions in the surrounding neighborhood are marked to ensure that only the pixel coordinates with the largest pixel value are retained in the neighborhood, and the coordinates of the surrounding pixel points with smaller pixel values ​​are suppressed, and finally the processed frequency hopping pattern pixel coordinate set is obtained:

[0026] in, is the number of pixel coordinates finally retained, The frequency points corresponding to all pixels in Collection That is the extracted frequency hopping frequency set, where the frequency hopping frequency estimate , is the sampling frequency, is the number of Fourier transform FFT points of short-time Fourier transform STFT.

[0027] To achieve the purpose of the invention, the second aspect of the present invention provides a frequency hopping frequency set estimation system based on a deep pattern recognition model, applying a frequency hopping frequency set estimation method based on a deep pattern recognition model described in the above technical solution, the system comprising: A data acquisition and preprocessing module, used for collecting and preprocessing in-phase and quadrature signal data including frequency hopping signals, obtaining a point-shaped frequency hopping diagram and dividing the point-shaped frequency hopping diagram into a training set and a test set; A deep pattern recognition model construction and training module, used to construct a frequency hopping periodic pattern deep pattern recognition network model based on an attention mechanism and to perform training using the training set; A frequency hopping cycle pattern extraction module, used to extract the frequency hopping cycle pattern of the verification set by using the trained frequency hopping cycle pattern deep pattern recognition network model based on the attention mechanism; The frequency hopping frequency set extraction module is used to extract the frequency hopping frequency set after matching and aligning the frequency hopping period pattern.

[0028] Compared with the prior art, the present invention has the following beneficial effects: The present invention converts the frequency hopping signal image in the time-frequency diagram into frequency hopping points through image preprocessing to obtain a point-like frequency hopping diagram, and then uses a frequency hopping periodic pattern deep pattern recognition network model based on an attention mechanism to perform pattern analysis and recognition on the point-like frequency hopping diagram, extract the frequency hopping periodic pattern in the point-like frequency hopping diagram, and finally extract the frequency hopping frequency set after matching and aligning the frequency hopping periodic pattern, which significantly improves the accuracy and robustness of the frequency hopping frequency set estimation under the condition of low algorithm complexity. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 A flowchart of a method for estimating a frequency hopping frequency set based on a deep pattern recognition model according to Embodiment 1 of the present application; Figure 2 This is a structural diagram of a frequency hopping periodic pattern deep pattern recognition network model based on an attention mechanism according to Example 1 of the present application; Figure 3 This is a schematic diagram of the structure of the attention module in the frequency hopping periodic pattern deep pattern recognition network model based on the attention mechanism of Example 1 of the present application; Figure 4 A schematic diagram of pattern recognition training and reasoning process of a frequency hopping periodic pattern deep pattern recognition network model based on an attention mechanism according to Example 1 of the present application; Figure 5 This is a comparison diagram of average errors of frequency hopping frequency set estimation using different algorithms in Example 3 of the present application; Figure 6 This is a graphical visualization result diagram of the frequency-hopping signal time-frequency diagram of Example 3 of the present application; Figure 7 This is a graphical visualization result diagram of the contour detection stage of Example 3 of the present application; Figure 8 This is a graphic visualization result diagram of the rectangle center extraction stage of Example 3 of the present application; Fig. 9 This is a graphical visualization result diagram of the frequency hopping period pattern deep pattern recognition stage of Example 3 of the present application; Fig.10 This is a graphic visualization result diagram of the template matching pixel superposition and averaging stage of Example 3 of the present application; Fig.11 This is a graphical visualization result diagram of the frequency hopping frequency set extraction result of Example 3 of the present application. DETAILED DESCRIPTION

[0030] The specific implementation of the present invention is further described in detail below in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0031] Example 1 See also Figure 1 , Figure 1 This is a flow chart of the steps of a frequency hopping frequency set estimation method based on a deep pattern recognition model in Embodiment 1 of the present application. Embodiment 1 provides a frequency hopping frequency set estimation method based on a deep pattern recognition model, comprising the following steps S1-S4: Step S1: collecting in-phase and quadrature signal data including frequency hopping signals and performing preprocessing to obtain a point-shaped frequency hopping graph and dividing the point-shaped frequency hopping graph into a training set and a test set.

[0032] First, the collected in-phase quadrature (IQ) data containing the frequency hopping signal is preprocessed by short-time Fourier transform (STFT) and filtering noise reduction to obtain a dot-shaped frequency hopping diagram, as shown below: (1) STFT transformation and preprocessing First, the collected in-phase orthogonal signal data containing the frequency hopping signal Perform time-frequency transformation to obtain a two-dimensional amplitude spectrum time-frequency diagram:

[0033] in, , T represents the number of time frames, , L represents the number of frequency points, represents the short-time Fourier transform, Indicates that the signal data x is an M-dimensional complex vector, represents the complex field, M is the vector dimension, represents the field of real numbers, Represents a real matrix of dimension T × L.

[0034] (2) Then, the obtained time-frequency diagram Perform noise reduction, specifically, perform erosion and dilation operations on the time-frequency graph according to the time dimension to remove mixed nonlinear structural signals and noise points; The expression of the erosion operation is as follows:

[0035] After the erosion operation is completed, the expansion operation is performed. The expansion operation expression is as follows:

[0036] in, Indicates that P takes a positive integer of the same magnitude as the time length of the single-frequency signal in the time-frequency diagram, represents a set of positive integers; After the expansion operation is completed, the denoised time-frequency graph is obtained , edge detection, contour detection and rectangle center extraction are used to transform the time-frequency graph Convert to a dotted frequency hopping diagram.

[0037] (3) Finally, edge detection, contour detection and rectangle center extraction are used to transform the time-frequency graph into Converting to a dotted frequency hopping diagram includes the following steps: The time-frequency diagram after noise reduction Normalize element by element and linearly map to [0,255] to get ,right Perform binarization judgment to obtain a binary image:

[0038] Wherein, U is a preset noise reduction threshold, and in this embodiment 1, U=50.

[0039] The edge detection algorithm and contour detection algorithm are used in turn to detect possible frequency hopping signals in the time-frequency graph and obtain the signal contour image:

[0040] in, It is the Canny edge detection function of the open source computer vision library OpenCV. and are respectively the low threshold and the high threshold of the Canny edge detection function. = 50, = 200;

[0041] in, It is the contour detection function of the open source computer vision library OpenCV. The contour retrieval mode uses , that is, only the external contour is retrieved, and the contour approximation method is adopted , that is, keep all contour points; right Perform rectangular frame analysis and extraction to obtain a rectangular set :

[0042] in, It is a rectangular contour fitting function in the open source computer vision library OpenCV, which is used to detect and fit the set of circumscribed rectangular contours in the image:

[0043] in, is the coordinate of the upper left corner of the rectangle, , are the width and height of the rectangle respectively, and V is the number of detected rectangular boxes; use Calculate the center coordinates of each rectangle to generate a dotted frequency hopping pattern:

[0044] Finally, the dot-shaped frequency hopping pattern is divided into a training set and a test set.

[0045] Step S2: construct a frequency hopping periodic pattern deep pattern recognition network model based on the attention mechanism and train it using the training set.

[0046] The frequency hopping pattern deep pattern recognition network model FHPR (Frequency Hopping Pattern Recognition) based on the attention mechanism constructed in this embodiment 1 is used to identify the frequency hopping pattern and extract the dotted frequency hopping pattern obtained in step S1 The frequency hopping cycle pattern is output as follows:

[0047] Wherein, Q is the number of frequency hopping patterns extracted by the FHPR model.

[0048] See also Figure 2 , Figure 2 This is a schematic diagram of the structure of the FHPR model constructed in this embodiment 1, which adopts an encoder-attention module-decoder structure. Figure 2 The convolutional blocks are represented by blue rectangles. The numbers 1, 32, 64, 128, and 256 are the number of channels. The yellow circle A represents the attention module. The blue rectangle on the left is the convolutional block of the encoder part, and the blue rectangle on the right is the convolutional block of the decoder part. The upward green arrow represents upsampling, the rightward black arrow represents skip connection, the rightward white arrow represents convolution, and the downward red arrow represents maximum pooling.

[0049] The encoder consists of three convolutional blocks, each of which includes a 3x3 convolutional layer, batch normalization, and ReLU activation function. A 2x2 maximum pooling layer is added after each convolutional block to reduce the size of the feature map while retaining important information. The encoder part finally passes through two consecutive convolutional layers ( Figure 2The blue rectangle at the bottom of the middle and the rightward white arrow part) further extracts features to enhance the network's ability to express the features of the input data. The decoder includes 3 deconvolution blocks, each of which includes a 3x3 deconvolution layer, batch normalization, and ReLU activation function. Before each deconvolution block, a 2x2 upsampling layer corresponding to the 2x2 maximum pooling layer in the encoder is added to increase the size of the feature map. Each layer in the decoder is connected to the corresponding layer in the encoder through a jump connection; an attention module is also set between the encoder and the decoder to strengthen the important features of the dot-like frequency hopping pattern.

[0050] The input of the FHPR model is of dimension The dot-like frequency hopping map first passes through the encoder part to extract multi-scale features, and then the size of the feature map of the multi-scale features is gradually reduced and the number of channels is increased through convolution and pooling operations.

[0051] Then an attention module is used between the encoder and decoder to enhance the important features of the dot-like frequency hopping pattern. Figure 3 , Figure 3 is a schematic diagram of the structure of the attention module of the FHPR model of this embodiment 1, Figure 3 In , 1×H×W, 1 / G, 1 / G×1×W, etc. represent the dimension size. The attention module divides the input feature map into G groups according to the channel dimension (G is usually 8, 16, 32, 64), and the size of each group is , the grouped input feature map passes through two parallel branches of the attention module; The first branch uses average pooling along the X and Y directions to extract global information, then passes through splicing and 1×1 convolution layers and performs feature fusion (spinning feature maps at different levels to achieve multi-scale feature fusion). Then, the attention maps in the X and Y directions are generated by the Sigmoid activation function, and then weight adjustment, group normalization and average pooling are performed. Finally, the attention map of the first branch is generated by the Softmax activation function. The second branch uses a 3×3 convolutional layer to capture local spatial information, followed by an average pooling operation and then normalization using the Softmax activation function to generate the attention map of the second branch; The attention maps generated by these two branches are subjected to matrix multiplication operations to generate the final feature representation, and then the final feature map is generated through the Sigmoid function. The final feature map is combined with the input feature map after weight adjustment to generate the final output feature of the attention module, namely the attention map. In this process, the feature maps are fused through multiple steps to generate a global attention map, which captures the pairing relationship at the pixel level.

[0052] Then, through the decoder part, the feature map size is gradually upsampled, the number of channels is reduced, and the output is reconstructed. At the same time, the features of the encoder are fused using jump connections to retain detail information. Finally, through a 1x1 convolutional layer, the decoder features are mapped to the final output, and the frequency hopping periodic pattern recognition result with the same dimension as the dot-like frequency hopping map is output. After that, segmentation is performed to independently separate each periodic pattern mode area in the frequency hopping periodic pattern.

[0053] The loss function L of the FHPR model of this embodiment 1 is:

[0054]

[0055]

[0056] in, , is the weight parameter, , usually taken , is the binary cross entropy loss, which computes the cross entropy between the true label and the predicted value for each pixel. The Dice loss is used to directly measure the overlap between the predicted frequency hopping cycle pattern area and the actual frequency hopping cycle pattern area to judge the segmentation effect of the model. is the total number of pixels in the image, is the true label of the ith pixel, when the ith pixel belongs to the frequency hopping periodic pattern area is 1 when the i-th pixel belongs to the background area is 0, is the predicted value of the ith pixel, indicating the probability that the pixel belongs to the frequency hopping period pattern area, is a small constant that prevents the denominator from being zero.

[0057] After the FHPR model is built, it needs to be trained using the training set obtained in S1. After the model training is completed, it can be used for subsequent test set reasoning.

[0058] See also Figure 4 , Figure 4 This is a schematic diagram of the pattern recognition training and reasoning process of the frequency hopping periodic pattern deep pattern recognition network model based on the attention mechanism of Example 1 of the present application.

[0059] In the training phase, the IQ signal samples in the training set are subjected to STFT transformation and preprocessing to obtain a dot-shaped frequency hopping pattern sample set, which is used as input to train the frequency hopping cycle pattern deep pattern recognition network model based on the attention mechanism. During the learning process, the model is trained using the Adam optimizer in a supervised learning manner, and the label data is used to guide the model to learn the mapping relationship between the frequency cycle hopping law and the frequency hopping cycle pattern mode. In the subsequent inference phase, the trained model is used to perform frequency hopping pattern recognition on the test set samples to obtain a frequency hopping cycle pattern set.

[0060] Step S3: using the trained frequency hopping cycle pattern deep pattern recognition network model based on the attention mechanism to extract the frequency hopping cycle pattern of the verification set.

[0061] See also Figure 4 ,In the inference stage, the trained model is used to perform ,frequency hopping pattern deep pattern recognition on the ,test set samples to obtain the frequency hopping period pattern set.

[0062] Step S4: extracting a frequency hopping frequency set after matching and aligning the frequency hopping cycle pattern.

[0063] Matching alignment includes: Use the first frequency hopping pattern As a template, the remaining frequency hopping patterns The pattern to be matched is used for template matching and alignment. The template matching and alignment rule adopts the normalized square difference matching method:

[0064] Among them, Q is the number of frequency hopping patterns extracted by the FHPR model, To make With template Matches the alignment offset.

[0065] Perform statistical averaging on the matched and aligned frequency hopping patterns by pixel:

[0066] Using the non-maximum suppression method, the frequency hopping pattern after template alignment To find the local maximum point, first, filter The pixel value is greater than the given threshold (In this embodiment 1 ) and sort them according to the pixel values ​​from large to small The pixels are sorted, and then the sorted pixels are traversed in order, and the positions in the surrounding neighborhood are marked to ensure that only the pixel coordinates with the largest pixel value are retained in the neighborhood, and the coordinates of the surrounding pixel points with smaller pixel values ​​are suppressed, and finally the processed frequency hopping pattern pixel coordinate set is obtained:

[0067] in, is the number of pixel coordinates finally retained, The frequency points corresponding to all pixels in Collection That is the extracted frequency hopping frequency set, where the frequency hopping frequency estimate , is the sampling frequency, is the number of Fourier transform FFT points of short-time Fourier transform STFT.

[0068] Example 2 This embodiment 2 is based on embodiment 1, and applies a frequency hopping frequency set estimation method based on a deep pattern recognition model in embodiment 1 to provide a frequency hopping frequency set estimation system based on a deep pattern recognition model, the system comprising: A data acquisition and preprocessing module, used for collecting and preprocessing in-phase and quadrature signal data including frequency hopping signals, obtaining a point-shaped frequency hopping diagram and dividing the point-shaped frequency hopping diagram into a training set and a test set; A deep pattern recognition model construction and training module, used to construct a frequency hopping periodic pattern deep pattern recognition network model based on an attention mechanism and to perform training using the training set; A frequency hopping cycle pattern extraction module, used to extract the frequency hopping cycle pattern of the verification set by using the trained frequency hopping cycle pattern deep pattern recognition network model based on the attention mechanism; The frequency hopping frequency set extraction module is used to extract the frequency hopping frequency set after matching and aligning the frequency hopping period pattern.

[0069] The other steps or technical details of this embodiment 2 are the same as those of embodiment 1 and will not be repeated here.

[0070] Example 3 In order to verify the effectiveness and advancement of the present invention, this embodiment 3 performs simulation results and visual analysis based on embodiments 1 and 2.

[0071] The simulation generates a frequency hopping sequence and a random transmission bit stream, where the frequency hopping signal uses MSK modulation, the number of frequency hopping points is 62, the frequency hopping interval is 200Hz, the frequency hopping rate is 500Hz, and the signal-to-noise ratio range is [-12dB, 6dB]. The FFT points of STFT are 10MHz. Compared with the traditional power spectrum peak detection method and the frequency hopping signal detection method based on the time-frequency graph YOLO-v5, the average estimation error is used as the performance evaluation indicator:

[0072] in, is the true value of the kth frequency, K is the number of frequencies in the true frequency set, is the frequency estimate corresponding to the true value of the kth frequency in the nth experiment, and N is the number of simulation experiments. For the performance comparison results of different frequency hopping frequency sets, please refer to Figure 5 , Figure 5 This is a comparison diagram of the average errors of the frequency hopping frequency set estimation of different algorithms in Example 3 of the present application, Figure 5 It can be seen that the average error of the frequency hopping frequency set estimation method based on the deep pattern recognition model of the present invention is the lowest among several methods under the same signal-to-noise ratio, which reflects the progress of the method. Figure 6-11 , are respectively the visualized simulation results of each stage output of the frequency hopping frequency set estimation method, among which, Figure 6 This is the graphical visualization result of the frequency hopping signal time-frequency diagram. Figure 7 This is a graphical visualization of the contour detection stage. Figure 8 This is a graphical visualization of the results of the rectangle center extraction stage. Fig. 9 This is a graphical visualization of the results of the deep pattern recognition phase of the frequency hopping periodic pattern. Fig.10 This is a graphical visualization of the template matching pixel overlay averaging stage. Fig.11 Result plot for a graphical visualization of the extraction results for the frequency hopping frequency set.

[0073] In summary, the present invention proposes a method for estimating a frequency hopping frequency set of a frequency hopping signal for a frequency hopping communication scenario commonly used in practical applications. First, the frequency hopping signal image in the time-frequency diagram is converted into a frequency hopping point through image preprocessing to obtain a point-shaped frequency hopping diagram. Then, a deep network model based on an attention mechanism is used to perform pattern analysis and recognition on the point-shaped frequency hopping diagram to extract the frequency hopping cycle pattern in the point-shaped frequency hopping diagram. Finally, the frequency hopping frequency set is extracted after matching and aligning the frequency hopping cycle pattern. This method significantly improves the accuracy and robustness of the frequency hopping frequency set estimation under the condition of low algorithm complexity.

[0074] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A frequency hopping frequency set estimation method based on a deep pattern recognition model, characterized in that: The steps include: Collecting in-phase and quadrature signal data including frequency hopping signals and performing preprocessing to obtain a point-shaped frequency hopping graph and dividing the point-shaped frequency hopping graph into a training set and a test set; Constructing a frequency hopping periodic pattern deep pattern recognition network model based on an attention mechanism and training it using the training set; The trained frequency hopping periodic pattern deep pattern recognition network model based on the attention mechanism is used to extract the frequency hopping periodic pattern of the verification set; The frequency hopping frequency set is extracted after matching and aligning the frequency hopping period pattern.

2. The method for estimating a frequency hopping frequency set based on a deep pattern recognition model according to claim 1, characterized in that: The pre-processing comprises: The collected in-phase orthogonal signal data containing the frequency hopping signal Perform time-frequency transformation to obtain a two-dimensional amplitude spectrum time-frequency diagram: in, , T represents the number of time frames, , L represents the number of frequency points, represents the short-time Fourier transform, Indicates that the signal data x is an M-dimensional complex vector, represents the complex field, M is the vector dimension, represents the field of real numbers, represents a real matrix of dimension T×L; The obtained time-frequency diagram Perform noise reduction, specifically, perform corrosion and expansion operations on the time-frequency graph according to the time dimension to remove mixed nonlinear structural signals and noise points; Among them, the expression of the corrosion operation is as follows: After the erosion operation is completed, the expansion operation is performed. The expansion operation expression is as follows: in, Indicates that P takes a positive integer of the same magnitude as the time length of the single-frequency signal in the time-frequency diagram, represents a set of positive integers; After the expansion operation is completed, the denoised time-frequency graph is obtained , edge detection, contour detection and rectangle center extraction are used to transform the time-frequency graph Convert to a dotted frequency hopping diagram.

3. The method for estimating a frequency hopping frequency set based on a deep pattern recognition model according to claim 2, characterized in that: The time-frequency graph is sequentially detected by edge detection, contour detection and rectangle center extraction. Converting to a dotted frequency hopping diagram includes the following steps: The time-frequency diagram after noise reduction Normalize element by element and linearly map to [0,255] to get ,right Perform binarization judgment to obtain a binary image: Wherein, U is the preset noise reduction threshold; The edge detection algorithm and contour detection algorithm are used in turn to detect possible frequency hopping signals in the time-frequency graph and obtain the signal contour image: in, It is the Canny edge detection function of the open source computer vision library OpenCV. and They are the low threshold and high threshold of the Canny edge detection function respectively; in, It is the contour detection function of the open source computer vision library OpenCV. The contour retrieval mode uses , that is, only the external contour is retrieved, and the contour approximation method is adopted , that is, keep all contour points; right Perform rectangular frame analysis and extraction to obtain a rectangular set : in, It is a rectangular contour fitting function in the open source computer vision library OpenCV, which is used to detect and fit the set of circumscribed rectangular contours in the image: in, is the coordinate of the upper left corner of the rectangle, , are the width and height of the rectangle respectively, and V is the number of detected rectangular boxes; use Calculate the center coordinates of each rectangle to generate a dotted frequency hopping pattern: 。 4. The method for estimating a frequency hopping frequency set based on a deep pattern recognition model according to claim 1, characterized in that: The frequency hopping periodic pattern deep pattern recognition network model FHPR based on the attention mechanism adopts an encoder-attention module-decoder structure; the encoder includes 3 convolution blocks, each convolution block includes a 3x3 convolution layer, batch normalization and ReLU activation function, and a 2x2 maximum pooling layer is added after each convolution block. The encoder part finally passes through two consecutive convolution layers to further extract features; the decoder includes 3 deconvolution blocks, each deconvolution block includes a 3x3 deconvolution layer, batch normalization and ReLU activation function, and a 2x2 upsampling layer corresponding to the 2x2 maximum pooling layer in the encoder is added before each deconvolution block, and each layer in the decoder is connected to the corresponding layer in the encoder through a jump connection; an attention module is also arranged between the encoder and the decoder to strengthen the important features of the point-like frequency hopping pattern.

5. The method for estimating a frequency hopping frequency set based on a deep pattern recognition model according to claim 4, characterized in that: The input of the model is of dimension The dot-shaped frequency hopping graph first passes through the encoder part to extract multi-scale features, and then gradually reduces the size of the feature map of the multi-scale features and increases the number of channels through convolution and pooling operations; The important features of the dot-like frequency-hopping pattern are reinforced by an attention module between the encoder and decoder; Through the decoder part, the feature map size is gradually upsampled, the number of channels is reduced, and the output is reconstructed. At the same time, the features of the encoder are fused using jump connections to retain detail information. Finally, a 1x1 convolutional layer is used to map the decoder features to the final output, and the frequency hopping periodic pattern recognition result with the same dimension as the dot-like frequency hopping map is output. After that, segmentation is performed to independently separate each periodic pattern pattern area in the frequency hopping periodic pattern.

6. The method for estimating a frequency hopping frequency set based on a deep pattern recognition model according to claim 5, characterized in that: The attention module divides the input feature map into groups, each group size is , the grouped input feature map passes through two parallel branches of the attention module; The first branch uses average pooling along the X and Y directions to extract global information, and then performs feature fusion through splicing and 1×1 convolutional layers. Then, the attention maps in the X and Y directions are generated by the Sigmoid function, and then weight adjustment, group normalization and average pooling operations are performed. Finally, the attention map of the first branch is generated through the Softmax function. The second branch uses a 3×3 convolutional layer to capture local spatial information, followed by an average pooling operation and then normalization using the Softmax function to generate the attention map of the second branch; The attention maps generated by these two branches are subjected to matrix multiplication operations to generate the final feature representation, and then the final feature map is generated through the Sigmoid function. The final feature map is combined with the input feature map after addition and multiplication operations, that is, weight adjustment, to generate the final output features of the attention module.

7. The method for estimating a frequency hopping frequency set based on a deep pattern recognition model according to claim 5, characterized in that: The loss function of the frequency hopping periodic pattern deep pattern recognition network model FHPR based on the attention mechanism is: in, , is the weight parameter, , usually taken , is the binary cross entropy loss, which computes the cross entropy between the true label and the predicted value for each pixel. The Dice loss is used to directly measure the overlap between the predicted frequency hopping cycle pattern area and the actual frequency hopping cycle pattern area to judge the segmentation effect of the model. is the total number of pixels in the image, is the true label of the ith pixel, when the ith pixel belongs to the frequency hopping periodic pattern area is 1 when the i-th pixel belongs to the background area is 0, is the predicted value of the ith pixel, indicating the probability that the pixel belongs to the frequency hopping period pattern area, is a constant that prevents the denominator from being zero.

8. The method for estimating a frequency hopping frequency set based on a deep pattern recognition model according to claim 1, characterized in that: The matching alignment includes: Use the first frequency hopping pattern As a template, the remaining frequency hopping patterns The pattern to be matched is used for template matching and alignment. The template matching and alignment rule adopts the normalized square difference matching method: Among them, Q is the number of frequency hopping patterns extracted by the model, To make the remaining frequency hopping patterns With template Matches the aligned displacement.

9. The method for estimating a frequency hopping frequency set based on a deep pattern recognition model according to claim 8, characterized in that: The step of extracting the frequency hopping frequency set is specifically: Perform statistical averaging on the matched and aligned frequency hopping patterns by pixel: Using the non-maximum suppression method, the frequency hopping pattern after template alignment To find the local maximum point, first, filter The pixel value is greater than the given threshold The pixel coordinates are sorted according to the pixel values ​​from large to small. The pixels are sorted, and then the sorted pixels are traversed in order, and the positions in the surrounding neighborhood are marked to ensure that only the pixel coordinates with the largest pixel value are retained in the neighborhood, and the coordinates of the surrounding pixel points with smaller pixel values ​​are suppressed, and finally the processed frequency hopping pattern pixel coordinate set is obtained: in, is the number of pixel coordinates finally retained, The frequency points corresponding to all pixels in Collection That is the extracted frequency hopping frequency set, where the frequency hopping frequency estimate , is the sampling frequency, is the number of Fourier transform FFT points of short-time Fourier transform STFT.

10. A frequency hopping frequency set estimation system based on a deep pattern recognition model, applying a frequency hopping frequency set estimation method based on a deep pattern recognition model according to any one of claims 1 to 9, characterized in that: The system comprises: A data acquisition and preprocessing module, used for collecting and preprocessing in-phase and quadrature signal data including frequency hopping signals, obtaining a point-shaped frequency hopping diagram and dividing the point-shaped frequency hopping diagram into a training set and a test set; A deep pattern recognition model construction and training module, used to construct a frequency hopping periodic pattern deep pattern recognition network model based on an attention mechanism and to perform training using the training set; A frequency hopping cycle pattern extraction module, used to extract the frequency hopping cycle pattern of the verification set by using the trained frequency hopping cycle pattern deep pattern recognition network model based on the attention mechanism; The frequency hopping frequency set extraction module is used to extract the frequency hopping frequency set after matching and aligning the frequency hopping period pattern.

Images