A short wave communication frequency selection method, terminal, computer storage medium and system

Abstract

The application provides a short-wave communication frequency selection method, a terminal, a computer storage medium and a system, relates to the technical field of short-wave communication frequency selection, and first acquires original short-wave detection ionization data and short-wave detection environment data, then pre-classifies the original short-wave detection ionization data, mines short-wave detection environment correlation data from the short-wave detection environment data based on an association algorithm, respectively fuses the short-wave detection environment correlation data into each pre-classification of the original short-wave detection ionization data, forms a pre-classification data set, introduces a deep learning neural network model, trains the model by using the pre-classification data set, realizes prediction classification and deviation regression, obtains a final prediction model, inputs real-time received short-wave frequency data into the prediction model, predicts an optimal short-wave communication frequency, and the optimal short-wave communication frequency is predicted, so that the identification accuracy is high.

Description

Technical Field

[0001] The present invention relates to the technical field of shortwave communication frequency selection, and more specifically, to a shortwave communication frequency selection method, terminal, computer storage medium and system. Background Art

[0002] In daily life, radio waves with frequencies between 3MHz and 30MHz are often referred to as shortwave. One of the most significant features of shortwave communication is that it can achieve long-distance communication by utilizing reflection from the ionosphere. The ionosphere's indestructibility also makes shortwave communication highly resistant to destruction, resulting in its widespread application in both civilian and military applications.

[0003] In shortwave communication, shortwave channel detection is crucial. Common methods for shortwave channel detection include Chirp detection method, pulse method, FSK data signal, etc. Among them, the Chirp detection method is a method in which a linear swept continuous wave signal is transmitted at the transmitter end, and the receiver sweeps and accurately tracks the signal at the transmitter end. After analyzing the different frequency components appearing at the receiver end, the corresponding ionization diagram is drawn. The communication frequency obtained in this way is highly reliable.

[0004] Selecting the best available frequency from the scattered ionization map returned in the Chirp detection is one of the keys to the entire frequency sweep detection, and quickly extracting a reliable frequency can also shorten the entire detection process time. At present, many shortwave detection devices read the best communication frequency through human eye recognition or traditional digital image processing, and this is still the mainstream method. The ionosphere is complex and changeable due to the influence of solar surface radiation storms, so the shortwave propagation environment data also changes. It is not a fixed shape on the shortwave detection ionization map. When the changes are large, traditional image processing is not applicable, and relying on the human eye will prolong the prediction time, which has an adverse effect on the military. Figure 1 As shown, in the prior art, in order to extract the frequency points of the hook position close to the X-axis (frequency axis) and the rightmost end, after the image is binarized, a vertical line and a horizontal line are used to scan from right to left and from bottom to top respectively. The number of overlaps between the vertical line, the horizontal line and the white pixel points is compared with the threshold value. If the number exceeds the threshold, it is determined to be an intersection point. The optimal frequency value can be obtained from the X-axis position of the intersection point. This processing method has a certain accuracy rate for "hook" recognition within a certain range. However, in actual short-wave sweep frequency detection, due to the long detection distance of thousands of kilometers and the influence of climate, the instability is relatively large. The "hook" presented in the detected ionogram is of various types, such as Figure 2 As shown in the figure, every time the detection area is changed, it is necessary to rely on experience accumulation, adjustment of image processing algorithms, and communication verification experiments to improve the recognition accuracy. This process is not only time-consuming but also costly.

[0005] Another approach involves placing the transmitter and receiver in two different areas, using the GNSS system to obtain atmospheric and ionosphere parameters in advance, transmitting scanning signals into the air continuously for a period of time, and analyzing the received detection signals based on the detected ionogram data to determine whether they meet communication requirements and thus determine the communication frequency. Alternatively, methods use historical detection frequencies to learn and predict available frequencies. However, this method determines atmospheric and ionosphere parameters before transmitting the scan and then incorporates them into the detection calculations. This only accounts for the spectrum data at both ends of the transmitter and receiver, and does not include the critical real-time changes in the atmosphere and ionosphere involved in each scan within the detection range. Because shortwave communication is often used for long-distance communication transmission, the atmosphere and ionosphere at both ends of the communication area fluctuate erratically, which can easily lead to misjudgment or failure of the communication frequency. Furthermore, selecting the optimal communication frequency from the detection ionogram data does not incorporate atmospheric and ionosphere data into the calculations. Summary of the Invention

[0006] In order to solve the problem that the area where the optimal frequency point in the detection ionization map is located presents different shapes due to unstable factors such as climate and ionosphere at different sites and time periods during the optimal frequency identification of shortwave sweep detection, the traditional digital image processing algorithm is prone to instability, reduced accuracy or long adjustment time. The present invention proposes a shortwave communication frequency selection method, terminal, computer storage medium and system, which integrate the real-time data of the environment and ionization during shortwave detection and improve the accuracy of optimal frequency identification.

[0007] In order to achieve the above technical effects, the technical solutions of the present invention are as follows:

[0008] A shortwave communication frequency selection method, the method comprising the following steps:

[0009] S1. Obtaining raw shortwave detection ionization data and shortwave detection environment data;

[0010] S2. Pre-classify the original shortwave detection ionization data, and mine shortwave detection environment associated data from the shortwave detection environment data based on the association algorithm. The shortwave detection environment associated data is respectively fused and added to each pre-classification of the original shortwave detection ionization data to form a pre-classified data set;

[0011] S3. Annotate the pre-classified categories with the best frequency to be identified in the pre-classified dataset, and use the annotated pre-classified dataset as a training set;

[0012] S4. Construct a deep learning neural network model and use the training set to train the deep learning neural network model for prediction and classification, obtaining a classification sample of shortwave detection ionization data integrated with shortwave detection environmental data;

[0013] S5. Take the best frequency value in the samples corresponding to each class as the label frequency value, take the label frequency value as the center value, calculate the difference from each classification, and use the deep learning neural network model regression training to optimize the difference, and obtain the final deep neural network model as the prediction model;

[0014] S6. Acquire shortwave detection ionization data in real time, input it into the prediction model, and predict the optimal shortwave communication frequency through the prediction model.

[0015] This technical solution is aimed at shortwave communication frequency selection. First, the original shortwave detection ionization data and shortwave detection environment data are obtained, and then the original shortwave detection ionization data is pre-classified. Based on the association algorithm, shortwave detection environment associated data is mined from the shortwave detection environment data, and the shortwave detection environment associated data are respectively integrated and added to each pre-classification of the original shortwave detection ionization data to form a pre-classified data set. The pre-classified category of the optimal frequency to be identified in the pre-classified data set is marked, that is, the pre-classified category where the optimal frequency is located is clear, and the approximate value of the optimal frequency is known. A deep learning neural network model is introduced, and the pre-classified data set is used for model training to realize prediction classification and deviation regression to obtain the final prediction model. The shortwave frequency data received in real time is input into the prediction model to predict the optimal shortwave communication frequency. The optimal shortwave communication frequency is predicted and the optimal adaptability of shortwave detection in more environments is enhanced.

[0016] Preferably, in step S1, raw shortwave detection ionization data and shortwave detection environment data are acquired using a frequency sweeping shortwave detection device.

[0017] Preferably, in step S2, the upper left corner of the scattering ionization map corresponding to the original shortwave detection ionization data is set as the coordinate origin, and a rectangular coordinate system is established. The horizontal axis direction of the rectangular coordinate system corresponds to the frequency of the scattering ionization map. It is assumed that the scattering ionization map has t pixels in the horizontal axis direction, and the vertical axis direction of the rectangular coordinate system corresponds to the time delay of the scattering ionization map. It is assumed that the scattering ionization map has q pixels in the vertical axis direction; the t pixels in the horizontal axis direction are pre-classified according to the spacing p of the scattering ionization map to obtain K pre-classified categories, wherein:

[0018] K = t / p.

[0019] Here, considering that the features of detecting "gyrations" in the optimal shortwave communication frequency in the scattered ionogram are complex and changeable and appear and disappear at times, the scattered ionogram is divided into K regions. Instead of dividing each pixel point in the horizontal axis direction as a category, a certain interval is used as a category, which is conducive to the comprehensive learning of the "gyration" feature area. Then, combined with regression, precise positioning is achieved, avoiding the disadvantages of too many pixel features under each pixel point and shallow convolution calculation of single-column pixel points in the deep neural network model.

[0020] Preferably, when the original shortwave detection ionization data is divided into K pre-classified categories, each pre-classified category is divided by frequency interval. The shortwave detection environment data is dynamic and has different changes in the K pre-classified categories. Considering that the atmospheric density and composition of the shortwave detection environment data are different, the detection environment data includes: atmospheric density ρ, atmospheric temperature σ, atmospheric composition Atmospheric density ρ, atmospheric temperature σ, and atmospheric composition in shortwave detection environmental data at a certain altitude As data, it is integrated into each pre-classification category. During the integration process, shortwave detection environment related data is mined from the shortwave detection environment data based on the association algorithm, and the shortwave detection environment related data is fused and added to each pre-classification of the original shortwave detection ionization data. That is, in addition to the ionospheric detection data, the data obtained by other equipment are also placed in one image. The data can be placed in a unified range through normalization to form a pre-classification data set, which is conducive to improving the accuracy of identifying the "cyclotron" feature and making the obtained optimal shortwave communication frequency more reliable.

[0021] Preferably, in step S4, in the process of training the deep learning neural network model using the training set, a prediction classification loss function is constructed:

[0022]

[0023] Among them, L i Represents the prediction classification loss function, s j is the predicted classification of the current output of the deep learning neural network model, is the optimal frequency y in each training data i The classification area where the classification is located is the one where the loss is generated in the process of training the deep learning neural network model using the training set. The gradient descent method is used to train the deep learning neural network model until the prediction classification loss function converges. The deep learning neural network model is used for prediction and classification, and the recognition accuracy is improved in a "regression" manner.

[0024] Preferably, the best frequency value in the sample corresponding to each class is used as the label frequency value, and the label frequency value is used as the center value. The process of calculating the difference from each classification satisfies the formula:

[0025]

[0026] Among them, i represents the category order, t (i) represents the i-th category, D (i) Using the label frequency value in the i-th category, m represents the total number of categories, and the least squares method is used to find the optimal solution for the parameter θ; when using the deep learning neural network model regression training to optimize the difference, the overall optimization formula is:

[0027]

[0028] Among them, p(θ) represents the probability of the output frequency, Indicates the integrated atmospheric density ρ, atmospheric temperature σ, and atmospheric composition

[0029] Preferably, in step S6, when the prediction model predicts the optimal shortwave communication frequency, the prediction model outputs the probability corresponding to each frequency, and the frequency corresponding to the maximum probability is used as the optimal shortwave communication frequency.

[0030] The present application also proposes a terminal, comprising a processor, a memory and a computer program stored in the memory, wherein the processor executes the computer program stored in the memory to implement the shortwave communication frequency selection method.

[0031] The present application also proposes a computer storage medium, wherein the computer-readable storage medium stores computer program instructions, and when the instructions are executed by a processor, the steps of the shortwave communication frequency selection method are implemented.

[0032] This application also proposes a shortwave communication frequency selection system, comprising:

[0033] A data acquisition unit, used to acquire original shortwave detection ionization data and shortwave detection environment data;

[0034] The category pre-classification unit pre-classifies the original shortwave detection ionization data, mines shortwave detection environment related data from the shortwave detection environment data based on the association algorithm, and fuses the shortwave detection environment related data into each pre-classification of the original shortwave detection ionization data to form a pre-classified data set;

[0035] A data set preparation unit marks the pre-classification categories with the best frequency to be identified in the pre-classification data set, and the marked pre-classification data set is used as a training set;

[0036] A neural network prediction and classification unit is used to construct a deep learning neural network model and train the deep learning neural network model using a training set to perform prediction and classification to obtain classification samples of shortwave detection ionization data integrated with shortwave detection environmental data;

[0037] The regression training unit takes the best frequency value in the samples corresponding to each class as the label frequency value, takes the label frequency value as the center value, calculates the difference from each classification, and uses the deep learning neural network model regression training to optimize the difference, and obtains the final deep neural network model as the prediction model;

[0038] The test unit is used to obtain shortwave detection ionization data in real time, input it into the prediction model, and predict the optimal shortwave communication frequency through the prediction model.

[0039] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0040] The present invention proposes a shortwave communication frequency selection method, terminal, computer storage medium and system. First, original shortwave detection ionization data and shortwave detection environment data are obtained, and then the original shortwave detection ionization data is pre-classified. Shortwave detection environment associated data is mined from the shortwave detection environment data based on an association algorithm. The shortwave detection environment associated data is respectively fused and added to each pre-classification of the original shortwave detection ionization data to form a pre-classified data set. A deep learning neural network model is introduced, and the model is trained using the pre-classified data set to implement prediction classification and deviation regression to obtain a final prediction model. Shortwave frequency data received in real time is input into the prediction model to predict the optimal shortwave communication frequency. The optimal shortwave communication frequency is predicted and the optimal adaptability of shortwave detection in more environments is enhanced. The problem that, when identifying the optimal frequency of shortwave sweep frequency detection, the area where the optimal frequency point in the detection ionization map is located presents different shapes due to unstable factors such as climate and ionosphere at different sites and time periods is easily unstable, with reduced accuracy or long adjustment time when identified by traditional digital image processing algorithms is solved. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 A schematic diagram showing the use of a traditional digital image processing method to identify the optimal frequency value proposed in the background art of the present invention;

[0042] Figure 2 It shows various detection ionization diagrams proposed in the background technology of the present invention;

[0043] Figure 3 A schematic flow chart showing the shortwave communication frequency selection method proposed in Example 1 of the present invention;

[0044] Figure 4 A schematic diagram showing the scattered ionogram proposed in Example 1 of the present invention;

[0045] Figure 5 A schematic diagram showing pre-classification based on a scattered ionogram proposed in Example 1 of the present invention;

[0046] Figure 6 A schematic diagram showing the integration of shortwave detection environment data proposed in Example 1 of the present invention;

[0047] Figure 7 A schematic diagram showing the classification and regression of shortwave detection ionization data proposed in Example 2 of the present invention;

[0048] Figure 8 A block diagram showing the structural composition of the shortwave communication frequency selection system proposed in Example 5 of the present invention. DETAILED DESCRIPTION

[0049] The accompanying drawings are for illustrative purposes only and are not to be construed as limiting this patent;

[0050] In order to better illustrate this embodiment, some parts of the drawings may be omitted, enlarged, or reduced, and do not represent the actual size;

[0051] It is understandable to those skilled in the art that descriptions of certain well-known contents may be omitted in the drawings.

[0052] The positional relationships described in the drawings are for illustrative purposes only and should not be construed as limiting this patent;

[0053] The technical solution of the present invention is further described below with reference to the accompanying drawings and embodiments.

[0054] Example

[0055] like Figure 3 As shown, this embodiment proposes a shortwave communication frequency selection method, see Figure 3 , the method comprises the following steps:

[0056] S1. Obtaining raw shortwave detection ionization data and shortwave detection environment data;

[0057] S2. Pre-classify the original shortwave detection ionization data, and mine shortwave detection environment associated data from the shortwave detection environment data based on the association algorithm. The shortwave detection environment associated data is respectively fused and added to each pre-classification of the original shortwave detection ionization data to form a pre-classified data set;

[0058] S3. Annotate the pre-classified categories with the best frequency to be identified in the pre-classified dataset, and use the annotated pre-classified dataset as a training set;

[0059] S4. Construct a deep learning neural network model and use the training set to train the deep learning neural network model for prediction and classification, obtaining a classification sample of shortwave detection ionization data integrated with shortwave detection environmental data;

[0060] In the process of training the deep learning neural network model using the training set, the prediction classification loss function is constructed:

[0061]

[0062] Among them, L i Represents the prediction classification loss function, s j is the predicted classification of the current output of the deep learning neural network model, is the optimal frequency y in each training data i The classification area where s is located; if s j Greater than or equal to Losses are generated in the process of training the deep learning neural network model using the training set, and the deep learning neural network model is trained using the gradient descent method until the prediction and classification loss function converges. The deep learning neural network model is used for prediction and classification, and the recognition accuracy is improved in a "regression" manner.

[0063] S5. Take the best frequency value in the samples corresponding to each class as the label frequency value, take the label frequency value as the center value, calculate the difference from each classification, and use the deep learning neural network model regression training to optimize the difference, and obtain the final deep neural network model as the prediction model;

[0064] S6. Acquire shortwave detection ionization data in real time, input it into the prediction model, and predict the optimal shortwave communication frequency through the prediction model.

[0065] Overall, for shortwave communication frequency selection, the original shortwave detection ionization data and shortwave detection environment data are first obtained, and then the original shortwave detection ionization data is pre-classified. Based on the association algorithm, the shortwave detection environment associated data is mined from the shortwave detection environment data, and the shortwave detection environment associated data is fused and added to each pre-classification of the original shortwave detection ionization data to form a pre-classified data set. A deep learning neural network model is introduced, and the pre-classified data set is used for model training to realize predictive classification and deviation regression to obtain the final prediction model. The shortwave frequency data received in real time is input into the prediction model to predict the optimal shortwave communication frequency. The optimal shortwave communication frequency is predicted, and the optimal adaptability of shortwave detection in more environments is enhanced.

[0066] The entire plan is to first pre-classify the detected ionization data and convert it into graph data of several regional categories. Then, atmospheric data needs to be considered and integrated into the converted graph data of several regional categories to form a fusion relationship. Finally, the optimal shortwave communication frequency is extracted through some algorithms. Based on learning classification and regression of these data, the optimal frequency in the atmospheric data changes of some detected ionization data sets can be known in the future.

[0067] In this embodiment, a frequency sweeping shortwave detection device is used to obtain raw shortwave detection ionization data and shortwave detection environment data.

[0068] The scattered ionogram records the scattering of radio waves transmitted through the ionosphere. From it, we can obtain information about the radio wave coverage area and scattered energy intensity, and reflect the available frequency information of short waves in the ionosphere. Figure 4The following figure shows ionization data from a frequency-sweeping shortwave detection device. The horizontal axis represents the frequency (typically 2-30 MHz) and the vertical axis represents the delay of the detection signal (typically 0-10 ms). The figure shows that the abscissa value of the characteristic "spin" point in the lower right corner of the white area is the optimal frequency for shortwave detection. Therefore, given the fluctuating ionosphere, automatically identifying the optimal frequency in a short period of time is of practical engineering significance.

[0069] For the situation where the characteristics of detecting "gyrations" in the best available frequency of the ionization diagram are complex and changeable and appear and disappear at times, a model can be obtained by learning from a large number of data sample sets, and the model can be used to simulate the human eye to judge the horizontal coordinate of the "gyrations". This patent adopts the "deep learning classification + regression" method.

[0070] For classification, in this embodiment, Figure 5 As shown, the upper left corner of the scattering ionization map corresponding to the original shortwave detection ionization data is set as the coordinate origin, and a rectangular coordinate system is established. The horizontal axis of the rectangular coordinate system corresponds to the frequency of the scattering ionization map. It is assumed that the scattering ionization map has t pixels in the horizontal axis direction, and the vertical axis of the rectangular coordinate system corresponds to the time delay of the scattering ionization map. It is assumed that the scattering ionization map has q pixels in the vertical axis direction. The t pixels in the horizontal axis direction are pre-classified according to the spacing p of the scattering ionization map to obtain K pre-classified categories, where:

[0071] K = t / p.

[0072] Here, the horizontal axis has 280 pixels and the vertical axis has 180 pixels. Considering that the characteristics of detecting "gyration" in the optimal shortwave communication frequency in the scattered ionization map are complex and changeable and sometimes weak, the scattered ionization map is divided into K regions. Instead of dividing each pixel point in the horizontal axis direction as a category, a certain interval is used as a category, which is conducive to the comprehensive learning of the "gyration" feature area. Combined with regression to achieve precise positioning, it avoids the disadvantages of too many pixel features under each pixel point (such as the traditional single classification that takes the 280 pixels in the horizontal axis of the figure as 280 categories) and the shallow convolution calculation of single-column pixel points in the deep neural network model. Figure 5 As shown in the figure, one of the categories contains the "hook" that needs to be identified in the ionization map, and the corresponding probability value can be obtained through the deep convolutional neural network.

[0073] Integrating atmospheric ionization detection data into the calculation of the optimal shortwave communication frequency can mine more potential data that is beneficial to the frequency selection of shortwave communication. Figure 6As shown in the figure, when the original shortwave detection ionization data is divided into K pre-classified categories, each pre-classified category is divided by frequency interval. The shortwave detection environment data is dynamic and has different changes in the K pre-classified categories. Considering that the atmospheric density and composition of the shortwave detection environment data are different, the detection environment data includes: atmospheric density ρ, atmospheric temperature σ, atmospheric composition Atmospheric density ρ, atmospheric temperature σ, and atmospheric composition in shortwave detection environmental data at a certain height As data is integrated into each pre-classification category, during the integration process, shortwave detection environment related data is mined from the shortwave detection environment data based on the association algorithm, and the shortwave detection environment related data is fused and added to each pre-classification of the original shortwave detection ionization data to form a pre-classification data set, which is conducive to improving the accuracy of identifying the "cyclotron" feature and making the best shortwave communication frequency more reliable. Figure 6 As shown, three data (atmospheric density ρ (▲), atmospheric temperature σ (●), atmospheric composition (★)) There are different changes in each category, which is conducive to improving the accuracy of identifying hooks. For example, in the same ionization detection data, there are more imaginary spiral shapes, but the atmospheric density and composition in which they are located are somewhat different. The association algorithm can be used to explore the correlation and improve the frequency selection accuracy.

[0074] Association algorithms, in this context, discover frequently occurring patterns, associations, and correlations between sets of items in large amounts of data stored in transactional databases, relational databases, or other information stores. Association algorithms have achieved significant results in fields such as scientific data analysis, radar signal sorting, classification design, bundled marketing, bioinformatics, medical diagnosis, and web mining. Typical association algorithms include the Aprior algorithm, the Frequent Pattern Growth (FP-G) algorithm, the FreeSpan algorithm, and the PrefixSpan algorithm.

[0075] Example 2

[0076] On the basis of classification, such as Figure 7 As shown, "regression" is performed, and the best frequency value in the sample corresponding to each class is used as the label frequency value. The label frequency value is used as the center value, and the difference from each classification is calculated to meet the formula:

[0077]

[0078] Assuming that the value obtained by forward inference of the deep learning neural network model has no deviation from the sample label value, the deviation value from the correct category is 0. The deviation values ​​of other categories can be obtained by subtracting them in turn. For regression training, this deviation needs to be made smaller and smaller. In the above formula, i represents the category order, t (i)represents the i-th category, D (i) Using the label frequency value in the i-th category, m represents the total number of categories, and the least squares method is used to find the optimal solution for the parameter θ;

[0079] The deviation regression prediction result is represented by t (i) -D (i) In this embodiment, it is assumed that the detection environment data atmospheric density ρ, atmospheric temperature σ, atmospheric composition In practice, the factors are not limited to these three. When using deep learning neural network model regression training to optimize the difference, the overall optimization formula is:

[0080]

[0081] Among them, p(θ) represents the probability of the output frequency, Indicates the integrated atmospheric density ρ, atmospheric temperature σ, and atmospheric composition When the optimal shortwave communication frequency is predicted by the prediction model, the prediction model outputs the probabilities corresponding to the various frequencies, and the frequency corresponding to the maximum probability is used as the optimal shortwave communication frequency. In addition, the training classification loss or regression function proposed in this embodiment is not limited to any other function.

[0082] Example 3

[0083] This embodiment provides a terminal including a processor, a memory, and a computer program stored in the memory. The processor executes the computer program stored in the memory to implement the shortwave communication frequency selection method.

[0084] Example 4

[0085] This embodiment further provides a computer storage medium, wherein the computer-readable storage medium stores computer program instructions, and when the instructions are executed by a processor, the steps of the shortwave communication frequency selection method are implemented.

[0086] Example 5

[0087] This application also proposes a shortwave communication frequency selection system, such as Figure 8 As shown, including:

[0088] Data acquisition unit 1, used to acquire original shortwave detection ionization data and shortwave detection environment data;

[0089] Category pre-classification unit 2 pre-classifies the original shortwave detection ionization data, mines shortwave detection environment associated data from the shortwave detection environment data based on the association algorithm, and fuses the shortwave detection environment associated data into each pre-classification of the original shortwave detection ionization data to form a pre-classified data set;

[0090] Data set preparation unit 3, marking the pre-classification categories with the best frequency to be identified in the pre-classification data set, and the marked pre-classification data set is used as a training set;

[0091] The neural network prediction and classification unit 4 is used to construct a deep learning neural network model and use the training set to train the deep learning neural network model to perform prediction and classification to obtain a classification sample of shortwave detection ionization data integrated with the shortwave detection environment data;

[0092] Regression training unit 5 takes the best frequency value in the samples corresponding to each class as the label frequency value, takes the label frequency value as the center value, calculates the difference from each classification, and uses deep learning neural network model regression training to optimize the difference, and obtains the final deep neural network model as the prediction model;

[0093] The test unit 6 is used to obtain shortwave detection ionization data in real time, input it into the prediction model, and predict the optimal shortwave communication frequency through the prediction model.

[0094] Obviously, the above embodiments of the present invention are merely examples for the purpose of clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. Those skilled in the art will appreciate that other variations or modifications can be made based on the above description. It is not necessary and impossible to enumerate all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the claims of the present invention.

Claims

1. A frequency selection method for shortwave communication, characterized in that: The method comprises the following steps: S1. Obtaining raw shortwave detection ionization data and shortwave detection environment data; S2. Pre-classify the original shortwave detection ionization data, and mine shortwave detection environment associated data from the shortwave detection environment data based on the association algorithm. The shortwave detection environment associated data is respectively fused and added to each pre-classification of the original shortwave detection ionization data to form a pre-classified data set; The original shortwave detection ionization data is obtained K When there are pre-classification categories, each pre-classification category is divided by frequency interval. The shortwave detection environment data is dynamic. K There are different changes in the pre-classification categories. Considering that the atmospheric density and composition of the shortwave detection environment data are different, the detection environment data include: atmospheric density , atmospheric temperature , atmospheric composition , using shortwave detection of atmospheric density in environmental data at a certain height , atmospheric temperature , atmospheric composition As data, it is integrated into each pre-classification category. During the integration process, shortwave detection environment related data is mined from the shortwave detection environment data based on the association algorithm, and the shortwave detection environment related data is fused and added to each pre-classification of the original shortwave detection ionization data to form a pre-classification data set. In step S2, the upper left corner of the scattered ionization map corresponding to the original shortwave detection ionization data is set as the coordinate origin, and a rectangular coordinate system is established. The horizontal axis direction of the rectangular coordinate system corresponds to the frequency of the scattered ionization map. Assume that the scattered ionization map has t pixel points, the vertical axis direction of the rectangular coordinate system corresponds to the time delay of the scattered ionization image, and the scattered ionization image has q pixel points in the vertical axis direction; t Pixels are spaced according to the p Pre-classify the scattered ionogram to obtain K pre-classification categories, of which: ； S3. Annotate the pre-classified categories with the best frequency to be identified in the pre-classified dataset, and use the annotated pre-classified dataset as a training set; S4. Construct a deep learning neural network model and use the training set to train the deep learning neural network model for prediction and classification, obtaining a classification sample of shortwave detection ionization data integrated with shortwave detection environmental data; S5. Take the best frequency value in the samples corresponding to each class as the label frequency value, take the label frequency value as the center value, calculate the difference from each classification, and use the deep learning neural network model regression training to optimize the difference, and obtain the final deep neural network model as the prediction model; The best frequency value in the sample corresponding to each class is used as the label frequency value, and the label frequency value is used as the center value. The process of calculating the difference from each classification satisfies the formula: in, i Indicates the order of categories, Indicates the i categories, Using the label frequency value in the i-th category, m represents the total number of categories, and the least squares method is used to find the optimal solution for the parameter θ; when using the deep learning neural network model regression training to optimize the difference, the overall optimization formula is: in, represents the probability of the output frequency, Indicates the density of the incorporated atmosphere , atmospheric temperature , atmospheric composition ; S6. Acquire shortwave detection ionization data in real time, input it into the prediction model, and predict the optimal shortwave communication frequency through the prediction model.

2. The shortwave communication frequency selection method according to claim 1, characterized in that: In step S1, raw shortwave detection ionization data and shortwave detection environment data are acquired using a frequency sweeping shortwave detection device.

3. The shortwave communication frequency selection method according to claim 1, characterized in that: In step S4, when training the deep learning neural network model using the training set, a prediction classification loss function is constructed: in, represents the prediction classification loss function, The predicted classification of the current output of the deep learning neural network model, The optimal frequency for each training data y i The classification area where Greater than or equal to , then the loss is generated in the process of training the deep learning neural network model using the training set, and the deep learning neural network model is trained using the gradient descent method until the prediction and classification loss function converges, and the deep learning neural network model is used for prediction and classification.

4. The shortwave communication frequency selection method according to claim 1, characterized in that: In step S6, when the prediction model predicts the optimal shortwave communication frequency, the prediction model outputs the probability corresponding to each frequency, and the frequency corresponding to the maximum probability is used as the optimal shortwave communication frequency.

5. A terminal, characterized in that: The invention comprises a processor, a memory and a computer program stored in the memory, wherein the processor executes the computer program stored in the memory to implement the shortwave communication frequency selection method according to any one of claims 1 to 4.

6. A computer storage medium, characterized in that The computer-readable storage medium stores computer program instructions, which, when executed by a processor, implement the steps of the shortwave communication frequency selection method according to any one of claims 1 to 4.

7. A shortwave communication frequency selection system, characterized in that: include: A data acquisition unit, used to acquire original shortwave detection ionization data and shortwave detection environment data; The category pre-classification unit pre-classifies the original shortwave detection ionization data, mines shortwave detection environment related data from the shortwave detection environment data based on the association algorithm, and fuses the shortwave detection environment related data into each pre-classification of the original shortwave detection ionization data to form a pre-classified data set; The original shortwave detection ionization data is obtained K When there are pre-classification categories, each pre-classification category is divided by frequency interval. The shortwave detection environment data is dynamic. K There are different changes in the pre-classification categories. Considering that the atmospheric density and composition of the shortwave detection environment data are different, the detection environment data include: atmospheric density , atmospheric temperature , atmospheric composition , using shortwave detection of atmospheric density in environmental data at a certain height , atmospheric temperature , atmospheric composition As data, it is integrated into each pre-classification category. During the integration process, shortwave detection environment related data is mined from the shortwave detection environment data based on the association algorithm, and the shortwave detection environment related data is fused and added to each pre-classification of the original shortwave detection ionization data to form a pre-classification data set. The upper left corner of the scattering ionization map corresponding to the original shortwave detection ionization data is set as the coordinate origin, and a rectangular coordinate system is established. The horizontal axis direction of the rectangular coordinate system corresponds to the frequency of the scattering ionization map. Suppose the scattering ionization map has t pixel points, the vertical axis direction of the rectangular coordinate system corresponds to the time delay of the scattered ionization image, and the scattered ionization image has q pixel points in the vertical axis direction; t Pixels are spaced according to the p Pre-classify the scattered ionogram to obtain K pre-classification categories, of which: ； A data set preparation unit marks the pre-classification categories with the best frequency to be identified in the pre-classification data set, and the marked pre-classification data set is used as a training set; A neural network prediction and classification unit is used to construct a deep learning neural network model and train the deep learning neural network model using a training set to perform prediction and classification to obtain classification samples of shortwave detection ionization data integrated with shortwave detection environmental data; The regression training unit takes the best frequency value in the samples corresponding to each class as the label frequency value, takes the label frequency value as the center value, calculates the difference from each classification, and uses the deep learning neural network model regression training to optimize the difference, and obtains the final deep neural network model as the prediction model; The best frequency value in the sample corresponding to each class is used as the label frequency value, and the label frequency value is used as the center value. The process of calculating the difference from each classification satisfies the formula: in, i Indicates the order of categories, Indicates the i categories, Using the label frequency value in the i-th category, m represents the total number of categories, and the least squares method is used to find the optimal solution for the parameter θ; when using the deep learning neural network model regression training to optimize the difference, the overall optimization formula is: in, represents the probability of the output frequency, Indicates the density of the incorporated atmosphere , atmospheric temperature , atmospheric composition ; The test unit is used to obtain shortwave detection ionization data in real time, input it into the prediction model, and predict the optimal shortwave communication frequency through the prediction model.

Images