Neural network-based direction finding method, apparatus and device, and medium
The neural network-based direction finding method addresses the limitations of traditional methods by pre-training and fine-tuning with simulated and real-world data, resulting in faster and more accurate direction finding with reduced storage requirements.
Patent Information
- Application Number
- CN202411185378.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-27
- Publication Date
- 2024-12-13
AI Technical Summary
Traditional radio direction finding methods are limited by the lack of pre-training with prior electromagnetic environment data, sensitivity to interference, complex and time-consuming calculations, large storage requirements, and high costs associated with spatial spectrum estimation.
A neural network-based direction finding method that utilizes simulated electromagnetic environment data, noise-injected training sets, and real-world data to pre-train and fine-tune a neural network model, optimizing it for complex environments and reducing interference.
The method achieves faster response speeds, improved accuracy, and reduced storage needs while enhancing adaptability to complex electromagnetic environments.
Smart Images
Figure CN119126007A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of radio direction finding, and in particular relates to a direction finding method, device, equipment and medium based on a neural network. Background Art
[0002] Radio direction finding is a method of passively measuring the azimuth of a radio radiation source in a certain frequency band. According to the frequency band, it can be divided into long wave, short wave, ultra short wave, microwave, etc.; according to the propagation path of radio waves, it can be divided into underwater, sky wave, ground wave direction finding, etc.; according to the ratio of the antenna array to the wavelength of the radio wave, it can be divided into large-based, medium-based, and small-based direction finding; according to the direction finding principle system, it can be divided into amplitude comparison lateral direction finding, Doppler direction finding, interferometer direction finding, time difference direction finding, spatial spectrum estimation direction finding, etc.; according to whether the antenna rotates, it can be divided into handheld direction finder, mechanical scanning direction finder, fully automatic direction finder, etc. The principle is basically based on the characteristics of linear propagation of radio waves, combined with the antenna's directional diagram or the difference in the electrical characteristics of the signal received by different antenna arrays (amplitude difference, Doppler frequency shift, phase difference, time difference) to determine the direction of the incoming wave. Radio direction finding is widely used in military and civilian fields, such as using one or more direction finding stations to check for radio interference sources, direction finding and positioning of specific communication stations, and radio navigation.
[0003] The commonly used direction finder interferometer system direction finder generally uses a circular antenna array (five arrays, nine arrays, etc.), that is, the direction is determined by detecting the phase difference of the specific frequency signal on different antenna arrays. The specific method is to use one of the arrays in the antenna array as the reference array, detect the difference in phase between the remaining arrays and the reference array signal, and perform direction finding by the method of least squares of phase difference or correlation calculation. Typical correlation interferometer direction finders include the DDFxxx series products of the German RS company. Its principle is based on the following a priori facts: theoretically, for a wave of a specific frequency in a specific direction, the phase difference of each antenna array relative to the reference array can be calculated in advance (a antenna arrays, a-1 pairs of phase differences), assuming that the resolution (direction finding accuracy) of the direction finder is 1°, then 360 sets of theoretical phase differences can be pre-stored as a comparison benchmark. In actual work, the measured phase difference is correlated with the above 360 sets of reference phase differences, and the ordinal number corresponding to the group with the largest correlation coefficient is used as the output result of the direction estimation.
[0004] The principles of direction finding by other systems such as Doppler and time difference are similar to those of the direction finder interferometer system, except that the phase difference detection between arrays is changed to frequency shift detection, time difference detection, etc.
[0005] The existing direction finding methods have at least the following defects: 1. Traditional interferometer, Doppler, and time difference direction finding cannot use prior electromagnetic environment data for pre-training and fine-tuning; 2. Parameter estimation algorithms such as least squares use little prior information and are easily affected by interference, which affects the accuracy of direction finding. 3. The calculation process is time-consuming and complicated, with poor stability and slow response speed; 4. Correlation direction finding requires pre-storage of a large amount of theoretical data for each frequency point in the working frequency band. As the requirements for frequency accuracy increase, a large amount of storage space is occupied. For example, if the frequency accuracy is 1 khz and the direction finding accuracy is 1°, then the theoretical value data of the phase difference required to be stored in the 3 GHz frequency band must reach about 360*3*10 6 *(a-1) yes; 5. Although spatial spectrum estimation direction finding can be used to find the direction of multiple targets at the same time, each array needs to be equipped with a channel receiver, which is not only costly and expensive, but also has a complex algorithm. In addition, mutual inductance coupling, array adaptation, and array position error lead to large direction finding errors. It is generally used for shortwave direction finding. Summary of the invention
[0006] The purpose of the present invention is to provide a direction finding method, device, equipment and medium based on a neural network to solve the above-mentioned problems existing in the prior art.
[0007] In order to achieve the above object, the present invention adopts the following technical solutions: In a first aspect, the present invention provides a direction finding method based on a neural network, the method comprising: Configure the direction finding parameters of the direction finder, which include: operating frequency band, frequency resolution, direction finding accuracy, number of antenna arrays and antenna array; Construct the first training set based on the operating frequency band, frequency resolution, direction finding accuracy, number of antenna arrays and antenna array; Adding preset noise to the first training set to construct a second training set; Performing a range test on the direction finder based on the direction finding parameters, collecting measured data of the range test of the direction finder, and constructing a third training set based on the measured data; The neural network model is trained three rounds in sequence using the first training set, the second training set and the third training set. In the first two rounds of training, when the output error of the neural network model reaches the threshold set in each round, the next round of training is performed. In the third round of training, when the output error of the neural network model reaches the threshold set in the third round, a trained neural network model is obtained.
[0008] Preferably, the method further comprises: Obtain sampling data of the direction finder in real environment; Input the sampled data into the trained neural network model and output multiple estimated values; The multiple estimated values are sorted from large to small to obtain an estimated value sequence, and the first L estimated values are selected from the estimated value sequence as direction finding prediction reference values, which are used to characterize the measurement of target orientation and direction finding quality.
[0009] Preferably, the neural network model is a BP neural network model, and the BP neural network model includes an input layer, multiple hidden layers and an output layer; The input layer includes a plurality of input neurons, and the number of the input neurons is determined according to the number of antenna elements; The output layer includes a plurality of output neurons, and the number of the output neurons is determined according to the direction finding accuracy; The number of hidden layers is 2 to 5, and the number of neurons in each hidden layer is determined based on a preset rule.
[0010] Preferably, the preset rule is: The number of neurons in the first hidden layer is not more than twice the number of input neurons in the input layer; The number of neurons in the second hidden layer is less than or equal to the number of neurons in the first hidden layer, and the product of the number of neurons in the second hidden layer and the number of neurons in the first hidden layer reaches a preset number range, and the preset number range is determined according to the number of input neurons and the number of output neurons; The number of neurons in the third hidden layer ~ the number of neurons in the fifth hidden layer are equal or decrease successively, and the total product of the number of neurons in the third hidden layer ~ the number of neurons in the fifth hidden layer reaches a preset number range.
[0011] Preferably, the first training set is constructed based on the working frequency band, frequency resolution, direction finding accuracy, number of antenna arrays and antenna array, including: Taking values from the minimum working frequency to the maximum working frequency in the working frequency band in sequence according to the value taking step, to obtain a number of working frequency values; the value taking step is determined according to the frequency resolution; A plurality of theoretical incoming wave directions under each working frequency value are set, and the azimuth stepping of two adjacent theoretical incoming wave directions is determined according to the direction finding accuracy; For each theoretical incoming wave direction, the theoretical phase difference, Doppler frequency shift or arrival time difference corresponding to each operating frequency between each antenna array and the reference array is calculated based on the number of antenna arrays and the antenna array; Normalizing multiple theoretical incoming wave directions to obtain a first output set; Normalizing the theoretical phase difference, Doppler frequency shift or arrival time difference corresponding to each operating frequency value between each antenna element and the reference element to obtain a first input set; A first training set is constructed according to the first input set and the first output set.
[0012] Preferably, the preset noise is Gaussian white noise, and the preset noise is added to the first training set to construct the second training set, including: Generate Gaussian white noise with zero mean; The first input set is updated based on the Gaussian white noise and the signal-to-noise ratio of the direction finder to obtain a second input set; A second training set is constructed based on the second input set and the first output set.
[0013] Preferably, a range test is performed on the direction finder based on the direction finding parameters, measured data of the range test of the direction finder is collected, and a third training set is constructed based on the measured data, including: Arranging a target signal source at a shooting range, controlling the target signal source to rotate multiple times in azimuth steps around the center of the direction finder, performing a shooting range test according to the working frequency band and frequency resolution of the direction finder during each rotation of the target signal source, and collecting actual measured data of the shooting range test of the direction finder; Normalizing the measured data of the range test of the direction finder after multiple rotations to obtain a third input set; A third training set is constructed according to the third input set and the first output set.
[0014] In a second aspect, the present invention provides a direction finding device based on a neural network, the device is used to implement the above-mentioned direction finding method based on a neural network, and the device comprises: A data generation module is used to generate simulated electromagnetic environment data of the direction finder, wherein the simulated electromagnetic environment data is based on: an operating frequency band, a frequency resolution, a direction finding accuracy and the number of antenna arrays; A first construction module is used to construct a first training set based on the working frequency band, frequency resolution, direction finding accuracy and number of antenna elements of the direction finder; A second construction module is used to add preset noise to the first training set to construct a second training set; A third construction module is used to perform a range test on the direction finder based on the direction finding parameters, collect measured data of the range test of the direction finder, and construct a third training set based on the measured data; The network training module is used to perform three rounds of training on the neural network model using the first training set, the second training set and the third training set in sequence. In the first two rounds of training, when the output error of the neural network model reaches the threshold set in each round, the next round of training is performed. In the third training, when the output error of the neural network model reaches the threshold set in the third round, a trained neural network model is obtained.
[0015] In a third aspect, the present invention provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned neural network-based direction finding method when executing the computer program.
[0016] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned neural network-based direction finding method.
[0017] Beneficial effects: 1. The present invention generates simulated electromagnetic environment data in advance, constructs a first training set by simulating the electromagnetic environment data, introduces noise, and adds noise to the first training set to obtain a second training set. The first training set and the second training set can be used to pre-train the neural network; 2. The present invention constructs a third training set through the measured data of the shooting range test. The third training set can be used as a priori data to fine-tune the neural network. Through pre-training and fine-tuning, the trained neural network model can be more adaptable to complex electromagnetic environments during the direction finding process and has stronger anti-interference ability; 3. The present invention uses a trained neural network model to perform direction finding, which has a faster response speed and can obtain target azimuth information and estimate the quality metric of the direction indication more quickly; 4. Compared with the correlation method for direction finding, the output of the present invention is more stable and requires less storage space. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following specific implementations, they are used to explain the embodiments of the present invention, but do not constitute a limitation on the embodiments of the present invention. In the accompanying drawings: Figure 1 is a flow chart of a direction finding method based on a neural network provided in one embodiment of the present invention; Figure 2 It is a block diagram of a direction finding device based on a neural network provided in one embodiment of the present invention. DETAILED DESCRIPTION
[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be briefly introduced below in combination with the drawings and the description of the embodiments or the prior art. Obviously, the following description of the structure of the drawings is only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. It should be noted that the description of these embodiments is used to help understand the present invention, but does not constitute a limitation of the present invention.
[0020] Embodiment 1 Figure 1 is a flow chart of a direction finding method based on a neural network provided by an embodiment of the present invention, such as Figure 1 As shown, this embodiment provides a direction finding method based on a neural network, and the method includes: Step S10: configuring direction finding parameters of the direction finder, wherein the direction finding parameters include: operating frequency band, frequency resolution, direction finding accuracy, number of antenna elements and antenna array.
[0021] Step S20: constructing a first training set based on the working frequency band, frequency resolution, direction finding accuracy, number of antenna arrays and antenna array.
[0022] Specifically, the first training set is constructed based on the working frequency band, frequency resolution, direction finding accuracy, number of antenna arrays and antenna formation, including: Step S201: taking values in sequence from the minimum working frequency to the maximum working frequency of the working frequency band in a value step to obtain a plurality of working frequency values; the value step is determined according to the frequency resolution; Step S202: setting multiple theoretical wave directions under each working frequency value, and the azimuth step of two adjacent theoretical wave directions is determined according to the direction finding accuracy; wherein, a theoretical wave direction is the incident direction of the target signal source when the direction finder receives the target signal source, and the multiple theoretical wave directions start from 0°, and the theoretical wave directions are adjusted according to the azimuth step until 360°; Step S203: For each theoretical incoming wave direction, the theoretical phase difference, Doppler frequency shift or arrival time difference corresponding to each working frequency between each antenna array and the reference array is calculated based on the number of antenna arrays and the antenna array; one array in the antenna is used as the reference array, and the remaining arrays are compared with it to calculate the phase difference, Doppler frequency shift and time difference; Step S204: normalizing multiple theoretical incoming wave directions to obtain a first output set; Step S205: normalizing the theoretical phase difference, Doppler frequency shift or arrival time difference corresponding to each working frequency value between each antenna element and the reference element to obtain a first input set; Step S206: construct a first training set according to the first input set and the first output set.
[0023] In this embodiment, the size of the first input set is: [(f max -f min ) / δ f ]*(360 / δ θ )*(a-1), where f max is the maximum operating frequency of the operating frequency band, f minis the minimum operating frequency of the operating frequency band, δ f is the frequency resolution, δ θ is the direction finding accuracy, and a is the number of antenna elements.
[0024] The corresponding first output set is: ([1,0,...,0] T , ..., [0, ..., 0, 1, 0, ... 0] T , ..., [0, ..., 0, 1] T ), where T is the transpose, and the "1" in the first output set represents the incoming wave direction corresponding to the first input set.
[0025] The dimension of the first output set is: (360 / δ θ )*1 dimension, the complete first training set is composed of the corresponding combination of the first input set and the first output set.
[0026] Step S30: adding the preset noise to the first training set to construct a second training set.
[0027] In this embodiment, the preset noise is Gaussian white noise; adding the preset noise to the first training set to construct the second training set includes: Step S301: Generate Gaussian white noise with a mean of zero; Step S302: updating the first input set based on Gaussian white noise and the signal-to-noise ratio of the direction finder to obtain a second input set; Step S303: construct a second training set according to the second input set and the first output set.
[0028] In this embodiment, Gaussian white noise with a mean of 0 is added to the signal data (theoretical phase difference, Doppler frequency shift, arrival time difference, etc.) in the first input set obtained in step S205 according to the signal-to-noise ratio requirement. The value range of the Gaussian white noise is from the minimum signal-to-noise ratio required by the sensitivity of direction finders of various systems to the maximum signal-to-noise ratio possible in an actual environment. The step of the signal-to-noise ratio can be 1 dB or other values determined according to requirements. The Gaussian white noise is generated by a white noise simulator, and the white noise simulator can use a random number generator with values between (0, 1).
[0029] The signal of each signal data in the first input set after the introduction of Gaussian white noise and the signal-to-noise ratio of the direction finder (that is, each signal data in the second input set) is: θ = s + [1 / (10 snr / 20 )]*randn(1); Where θ is the signal data in the second input set, s is the theoretical phase difference, Doppler frequency shift or arrival time difference corresponding to each working frequency value after normalization, snr is the signal-to-noise ratio, and randn(1) is the random number generated by the white noise simulator between (0,1).
[0030] In this embodiment, the output set corresponding to the second input set is the same as the first output set, so the corresponding combination of the second input set and the first output set constitutes a complete second training set.
[0031] Step S40: Perform a range test on the direction finder based on the direction finding parameters, collect measured data of the range test of the direction finder, and construct a third training set based on the measured data. In this embodiment, enemy simulated signal sources or target signal sources and turntables of various modulation modes are set in the range, and the position of the target signal source is adjusted to simulate the direction finder receiving signals from different directions; in this embodiment, the antenna of the direction finder can rotate according to the azimuth step interval, or the target signal source can rotate around the center of the antenna in an equal step circle; the data obtained from the actual test of the direction finder is sampled, and in order to save test time, the frequency interval can also be increased, and then filled in by interpolation.
[0032] Specifically, a range test is performed on the direction finder based on the direction finding parameters, measured data of the range test of the direction finder is collected, and a third training set is constructed based on the measured data, including: Step S401: arranging a target signal source at a target range, controlling the direction finding antenna to rotate or the target signal source to rotate multiple times in azimuth steps around the center of the direction finder, performing a target range test according to the working frequency band and frequency resolution of the direction finder during each rotation, and collecting measured data of the target range test of the direction finder; Step S402: normalizing the measured data of the target range test of the direction finder after multiple rotations to obtain a third input set; Step S403: construct a third training set according to the third input set and the first output set.
[0033] In this embodiment, the output set corresponding to the third input set is also the same as the first output set. Therefore, the corresponding combination of the third input set and the first output set constitutes a complete third training set.
[0034] Step S50: Use the first training set, the second training set, and the third training set to perform four rounds of training on the neural network model in sequence. In the first two rounds of training, when the output error of the neural network model reaches the threshold set in each round, the next round of training is performed. In the third round of training, when the output error of the neural network model reaches the threshold set in the third round, a trained neural network model is obtained.
[0035] In this embodiment, the first training set, the second training set, and the third training set are mainly used to pre-train the neural network, and the third training set is mainly used to fine-tune the neural network. Through pre-training and fine-tuning, the trained neural network model can better adapt to complex electromagnetic environments during direction finding and have stronger anti-interference capabilities.
[0036] As a further optimization of this embodiment, the method further includes: Step a10: obtaining sampling data of the direction finder in a real environment; Step a20: input the sampled data into the trained neural network model and output multiple estimated values; Step a30: sort the multiple estimated values from large to small to obtain an estimated value sequence, and select the first L estimated values from the estimated value sequence as direction finding prediction reference values, which are used to characterize the measurement of target azimuth and direction finding quality.
[0037] The present invention uses a trained neural network model to predict and estimate the sampled data in the real environment, has a faster response speed, and can obtain target azimuth information and direction finding quality measurement more quickly.
[0038] As a further optimization of this embodiment, the neural network model is a BP neural network model, and the BP neural network model includes an input layer, multiple hidden layers and an output layer; The input layer includes a plurality of input neurons, the number of which is determined according to the number of antenna elements, and the number of input neurons b is: b=a-1, and each input neuron corresponds to a pair of input signals such as phase difference or Doppler frequency shift, time difference, etc.; The output layer includes a plurality of output neurons, the number of which is determined according to the direction finding accuracy, and the number of output neurons k is: k=360 / δ θ ; During training, the output layer takes the same direction as the input signal (360 / δ θ )*1-dimensional vector data; The number of hidden layers is 2 to 5, and the number of neurons in each hidden layer is determined based on a preset rule.
[0039] In this embodiment, the number of hidden layers and the number of neurons in each hidden layer depend on the number of input neurons in the input layer, the number of output neurons in the output layer, and the number of training set samples (sample data in the first training set, the second training set, and the third training set). Generally, empirical values are taken to prevent underfitting and overfitting. In view of the large number of output neurons required in this method, as well as the coverage frequency range, direction finding accuracy requirements, coverage signal strength and other factors, the training set is relatively large, and the hidden layer should not be too small. Assuming that the direction finding accuracy requirement is 1°, 2-5 hidden layers are constructed between the input layer and the output layer. The number of neurons in each hidden layer is determined according to preset rules.
[0040] In this embodiment, the preset rule is: The number of neurons in the first hidden layer is not more than twice the number of input neurons in the input layer; The number of neurons in the second hidden layer is less than or equal to the number of neurons in the first hidden layer, and the product of the number of neurons in the second hidden layer and the number of neurons in the first hidden layer reaches a preset number range, which is determined according to the number of input neurons and the number of output neurons, that is, the product of the number of neurons in the second hidden layer and the number of neurons in the first hidden layer is ≈(b+k)*2 / 3; The number of neurons in the 3rd hidden layer ~ the number of neurons in the 5th hidden layer are equal or decrease successively, and the total product of the number of neurons in the 3rd hidden layer ~ the number of neurons in the 5th hidden layer reaches a preset number range; that is, the total product of the number of neurons in the 3rd hidden layer ~ the number of neurons in the 5th hidden layer ≈ (b+k)*2 / 3.
[0041] In this embodiment, it is assumed that the hidden layer of the neural network model has Q layers, the number of nodes (output neurons) in the output layer is N, and d i is the target output, E is the total mean square error, y q i is the output of the i-th node of the q-th layer, is the weight from node i in layer q to node j in layer q+1. There are M nodes in layer q. The value of q is 1, 2, 3, …, Q. F is the activation function. During the training process, we have: ; ; ; In the formula, is the output of the i-th node in the last hidden layer.
[0042] In this embodiment, a set of optimal weights can be obtained by minimizing the mean square error through an iterative algorithm until convergence.
[0043] The present invention has the following advantages: 1. The present invention generates simulated electromagnetic environment data in advance, constructs a first training set by simulating the electromagnetic environment data, introduces noise, and adds noise to the first training set to obtain a second training set. The first training set and the second training set can be used to pre-train the neural network; 2. The present invention constructs a third training set through the measured data of the shooting range test. The third training set can be used as a priori data to fine-tune the neural network. Through pre-training and fine-tuning, the trained neural network model can be more adaptable to complex electromagnetic environments during the direction finding process and has stronger anti-interference ability; 3. The present invention uses a trained neural network model to perform direction finding, which has a faster response speed and can obtain target position information and direction finding quality measurement more quickly; 4. Compared with the correlation method direction finding, the present invention requires less storage space.
[0044] Embodiment 2 Figure 2 is a block diagram of a direction finding device based on a neural network provided by an embodiment of the present invention, such as Figure 2 As shown, this embodiment provides a direction finding device based on a neural network, and the device is used to implement the direction finding method based on a neural network of embodiment 1, and the device includes: A data configuration module is used to configure the direction finding parameters of the direction finder, wherein the direction finding parameters include: operating frequency band, frequency resolution, direction finding accuracy, number of antenna arrays and antenna array; A first construction module is used to construct a first training set based on an operating frequency band, a frequency resolution, a direction finding accuracy, a number of antenna arrays, and an antenna array; A second construction module is used to add preset noise to the first training set to construct a second training set; A third construction module is used to perform a range test on the direction finder based on the direction finding parameters, collect measured data of the range test of the direction finder, and construct a third training set based on the measured data; The network training module is used to perform four rounds of training on the neural network model using the first training set, the second training set, and the third training set in sequence. In the first two rounds of training, when the output error of the neural network model reaches the threshold set in each round, the next round of training is performed. In the third round of training, when the output error of the neural network model reaches the threshold set in the third round, a trained neural network model is obtained.
[0045] This embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the neural network-based direction finding method of the first embodiment is implemented.
[0046] This embodiment provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the neural network-based direction finding method of the first embodiment is implemented.
[0047] The present invention has the following advantages: 1. The present invention generates simulated electromagnetic environment data in advance, constructs a first training set by simulating the electromagnetic environment data, introduces noise, and adds noise to the first training set to obtain a second training set. The first training set and the second training set can be used to pre-train the neural network; 2. The present invention constructs a third training set through the measured data of the shooting range test. The third training set can be used as a priori data to fine-tune the neural network. Through pre-training and fine-tuning, the trained neural network model can be more adaptable to complex electromagnetic environments during the direction finding process and has stronger anti-interference ability; 3. The present invention uses a trained neural network model to perform direction finding, which has a faster response speed and can obtain target azimuth information and direction quality measurement more quickly; 4. Compared with the correlation method direction finding, the present invention requires less storage space.
[0048] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.
[0049] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0050] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.
Claims
1. A direction finding method based on a neural network, characterized in that: The method comprises: Configure the direction finding parameters of the direction finder, which include: operating frequency band, frequency resolution, direction finding accuracy, number of antenna arrays and antenna array; Construct the first training set based on the operating frequency band, frequency resolution, direction finding accuracy, number of antenna arrays and antenna array; Adding preset noise to the first training set to construct a second training set; Performing a range test on the direction finder based on the direction finding parameters, collecting measured data of the range test of the direction finder, and constructing a third training set based on the measured data; The neural network model is trained three rounds in sequence using the first training set, the second training set and the third training set. In the first two rounds of training, when the output error of the neural network model reaches the threshold set in each round, the next round of training is performed. In the third round of training, when the output error of the neural network model reaches the threshold set in the third round, a trained neural network model is obtained.
2. The direction finding method based on neural network according to claim 1, characterized in that: The method further comprises: Obtain sampling data of the direction finder in real environment; Input the sampled data into the trained neural network model and output multiple estimated values; The multiple estimated values are sorted from large to small to obtain an estimated value sequence, and the first L estimated values are selected from the estimated value sequence as direction finding prediction reference values, which are used to characterize the measurement of target orientation and direction finding quality.
3. The direction finding method based on neural network according to claim 1, characterized in that: The neural network model is a BP neural network model, and the BP neural network model includes an input layer, multiple hidden layers and an output layer; The input layer includes a plurality of input neurons, and the number of the input neurons is determined according to the number of antenna elements; The output layer includes a plurality of output neurons, and the number of the output neurons is determined according to the direction finding accuracy; The number of hidden layers is 2 to 5, and the number of neurons in each hidden layer is determined based on a preset rule.
4. The direction finding method based on neural network according to claim 3, characterized in that: The preset rules are: The number of neurons in the first hidden layer is not more than twice the number of input neurons in the input layer; The number of neurons in the second hidden layer is less than or equal to the number of neurons in the first hidden layer, and the product of the number of neurons in the second hidden layer and the number of neurons in the first hidden layer reaches a preset number range, and the preset number range is determined according to the number of input neurons and the number of output neurons; The number of neurons in the third hidden layer ~ the number of neurons in the fifth hidden layer are equal or decrease successively, and the total product of the number of neurons in the third hidden layer ~ the number of neurons in the fifth hidden layer reaches a preset number range.
5. The direction finding method based on neural network according to claim 1, characterized in that: The first training set is constructed based on the operating frequency band, frequency resolution, direction finding accuracy, number of antenna arrays and antenna arrays, including: Taking values from the minimum working frequency to the maximum working frequency in the working frequency band in sequence according to the value taking step, to obtain a number of working frequency values; the value taking step is determined according to the frequency resolution; A plurality of theoretical incoming wave directions under each working frequency value are set, and the azimuth stepping of two adjacent theoretical incoming wave directions is determined according to the direction finding accuracy; For each theoretical incoming wave direction, the theoretical phase difference, Doppler frequency shift or arrival time difference corresponding to each operating frequency between each antenna array and the reference array is calculated based on the number of antenna arrays and the antenna array; Normalizing multiple theoretical incoming wave directions to obtain a first output set; Normalizing the theoretical phase difference, Doppler frequency shift or arrival time difference corresponding to each operating frequency value between each antenna element and the reference element to obtain a first input set; A first training set is constructed according to the first input set and the first output set.
6. The direction finding method based on neural network according to claim 5, characterized in that: The preset noise is Gaussian white noise. The preset noise is added to the first training set to construct the second training set, including: Generate Gaussian white noise with zero mean; The first input set is updated based on the Gaussian white noise and the signal-to-noise ratio of the direction finder to obtain a second input set; A second training set is constructed based on the second input set and the first output set.
7. The direction finding method based on neural network according to claim 5, characterized in that: A range test is performed on the direction finder based on the direction finding parameters, measured data of the range test of the direction finder is collected, and a third training set is constructed based on the measured data, including: Arrange a target signal source at a shooting range, control the direction finding antenna to rotate or the target signal source to rotate multiple times in azimuth steps around the center of the direction finding antenna, perform shooting range testing according to the working frequency band and frequency resolution of the direction finder during each rotation of the target signal source, and collect actual measured data of the shooting range testing of the direction finder; Normalizing the measured data of the range test of the direction finder after multiple rotations to obtain a third input set; A third training set is constructed according to the third input set and the first output set.
8. A direction finding device based on a neural network, the device being used to implement the direction finding method based on a neural network as claimed in any one of claims 1 to 7, characterized in that: The device comprises: A data configuration module is used to configure the direction finding parameters of the direction finder, wherein the direction finding parameters include: operating frequency band, frequency resolution, direction finding accuracy, number of antenna arrays and antenna array; A first construction module is used to construct a first training set based on the working frequency band, frequency resolution, direction finding accuracy, number of antenna arrays and antenna formation of the direction finder; A second construction module is used to add preset noise to the first training set to construct a second training set; A third construction module is used to perform a range test on the direction finder based on the direction finding parameters, collect measured data of the range test of the direction finder, and construct a third training set based on the measured data; The network training module is used to perform three rounds of training on the neural network model using the first training set, the second training set and the third training set in sequence. In the first two rounds of training, when the output error of the neural network model reaches the threshold set in each round, the next round of training is performed. In the third round of training, when the output error of the neural network model reaches the threshold set in the third round, a trained neural network model is obtained.
9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the neural network-based direction finding method according to any one of claims 1 to 7 is implemented.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the neural network-based direction finding method described in any one of claims 1 to 7 is implemented.