Construction method and device of rotation Doppler signal classification model, equipment and medium

Abstract

The invention provides a construction method and device of a rotation Doppler signal classification model, equipment and a medium. A training signal for training a rotation Doppler effect is collected; the training rotation Doppler effect is generated when the training light beam irradiates the surface of a training rotation object; the training light beam and the training rotating object are respectively a light beam and a rotating object for generating a training signal; the training signal is used for model training; acquiring a relative pose relationship between a propagation axis of a training light beam and a rotating axis of a training rotating object during acquisition of the training signal, and determining a training type of the training signal according to the relationship; the training type is the type of a signal for model training; designing a target mixed kernel function, and replacing a kernel function in the support vector machine with the kernel function to obtain a target machine learning model; and training a target machine learning model by using the training signal and the training type to obtain a target rotation Doppler signal classification model, and improving the classification capability of the support vector machine for complex data through a target mixed kernel function.

Description

Technical Field

[0001] The present application relates to the field of signal technology, and in particular to a method for constructing a rotational Doppler signal classification model, a device for constructing a rotational Doppler signal classification model, an electronic device, and a computer-readable medium. Background Art

[0002] In related technologies, a vortex beam can be applied to a rotating macroscopic object. This rotation produces an optical rotational Doppler effect, where the frequency of light reflected or scattered by the object changes. By detecting this frequency change, the object's rotational speed can be determined. Therefore, the rotational Doppler effect can be used to measure an object's rotational speed.

[0003] When using the rotational Doppler effect to measure an object's rotational speed, the signal distribution characteristics of the rotational Doppler effect are closely related to the extraction of the object's rotational speed information. Therefore, accurate classification of the rotational Doppler effect signal is necessary to further accurately extract the rotational speed information. However, in related art, object detection systems based on the rotational Doppler effect suffer from difficulties in accurately identifying and classifying signal types, making it difficult to accurately obtain the object's rotational speed information. Summary of the Invention

[0004] The embodiments of the present application provide a method, device, electronic device, and computer-readable storage medium for constructing a rotational Doppler signal classification model to solve the problem that, in an object detection system based on the rotational Doppler effect, the signal type is difficult to accurately identify and classify, resulting in difficulty in accurately obtaining the object's rotational speed information.

[0005] The present application discloses a method for constructing a rotational Doppler signal classification model, comprising:

[0006] Collecting a training signal of a preset training rotational Doppler effect; the training rotational Doppler effect is the Doppler effect generated when a preset training light beam is irradiated on the surface of a preset training rotating object; the training light beam and the training rotating object are respectively used to generate the training signal; the training signal is a signal used for model training;

[0007] Obtaining the relative positional relationship between the propagation axis of the training light beam and the rotation axis of the training rotating object when collecting the training signal, and determining the training type of the training signal based on the relative positional relationship; the training type of the training signal is the type of signal used for model training;

[0008] Designing a target hybrid kernel function, and using the target hybrid kernel function to replace the kernel function in a preset support vector machine to obtain a target machine learning model;

[0009] The target machine learning model is trained using the training signal and the training type of the training signal to obtain a target rotation Doppler signal classification model.

[0010] Optionally, determining the training type of the training signal according to the relative posture relationship includes:

[0011] The training type of the training signal is determined according to the lateral offset and / or tilt angle between the propagation axis and the rotation axis.

[0012] Optionally, the design target hybrid kernel function includes:

[0013] Obtain at least one kernel function to be processed;

[0014] Determining the target weight coefficient of the kernel function to be processed by using a preset genetic algorithm;

[0015] The kernel function to be processed is processed based on the target weight coefficient to obtain the target hybrid kernel function.

[0016] Optionally, the determining the target weight coefficient of the kernel function to be processed by using a preset genetic algorithm includes:

[0017] Determining at least one initial weight coefficient of the kernel function to be processed, and determining a fitness value corresponding to the initial weight coefficient;

[0018] Based on the fitness value, determining a weight coefficient to be processed from the initial weight coefficients;

[0019] Performing crossover and / or mutation processing on the weight coefficient to be processed to generate an updated weight coefficient;

[0020] The updated weight coefficient is used as the initial weight coefficient, and the step of determining the fitness value corresponding to the initial weight coefficient is repeatedly performed until the fitness value corresponding to the initial weight coefficient meets the preset fitness condition, and / or the number of repeated executions meets the preset number condition, thereby obtaining the target weight coefficient.

[0021] Optionally, determining the fitness value corresponding to the initial weight coefficient includes:

[0022] Processing the kernel function to be processed based on the initial weight coefficient to obtain an initial hybrid kernel function;

[0023] Replacing the kernel function in the support vector machine with the initial hybrid kernel function to obtain an initial machine learning model;

[0024] Training the initial machine learning model using the training signal and the training type of the training signal to obtain an initial rotational Doppler signal classification model;

[0025] Based on the initial rotational Doppler signal classification model, a fitness value corresponding to the initial weight coefficient is determined.

[0026] Optionally, the kernel function to be processed includes a Gaussian kernel function and / or a polynomial kernel function.

[0027] Optionally, the method comprises:

[0028] Collecting signals with preset rotational Doppler effect;

[0029] The rotational Doppler effect signal is input into the target rotational Doppler signal classification model to obtain the type of the rotational Doppler effect signal.

[0030] The present application also discloses a device for constructing a rotational Doppler signal classification model, comprising:

[0031] A training signal acquisition module is configured to acquire a training signal for a preset training rotational Doppler effect; the training rotational Doppler effect is the Doppler effect generated when a preset training light beam is irradiated on the surface of a preset training rotating object; the training light beam and the training rotating object are the light beam and the rotating object, respectively, used to generate the training signal; the training signal is a signal used for model training;

[0032] a training type determination module, configured to obtain a relative positional relationship between the propagation axis of the training light beam and the rotation axis of the training rotating object when the training signal is collected, and determine a training type of the training signal based on the relative positional relationship; the training type of the training signal is a type of signal used for model training;

[0033] A function design module is used to design a target hybrid kernel function and use the target hybrid kernel function to replace the kernel function in the preset support vector machine to obtain a target machine learning model;

[0034] The model acquisition module is used to train the target machine learning model using the training signal and the training type of the training signal to obtain a target rotation Doppler signal classification model.

[0035] Optionally, the training type determination module includes:

[0036] The training type determination submodule is configured to determine the training type of the training signal according to the lateral offset and / or tilt angle between the propagation axis and the rotation axis.

[0037] Optionally, the function design module includes:

[0038] A kernel function to be processed acquisition submodule, used to acquire at least one kernel function to be processed;

[0039] A target weight coefficient determination submodule is used to determine the target weight coefficient of the kernel function to be processed using a preset genetic algorithm;

[0040] The target mixed kernel function obtaining submodule is used to process the kernel function to be processed based on the target weight coefficient to obtain the target mixed kernel function.

[0041] Optionally, the target weight coefficient determination submodule includes:

[0042] a fitness value determining unit, configured to determine at least one initial weight coefficient of the kernel function to be processed, and determine a fitness value corresponding to the initial weight coefficient;

[0043] a to-be-processed weight coefficient determining unit, configured to determine a to-be-processed weight coefficient from the initial weight coefficients based on the fitness value;

[0044] An updated weight coefficient generating unit, configured to perform crossover and / or mutation processing on the weight coefficient to be processed to generate an updated weight coefficient;

[0045] A repeated execution unit is used to use the updated weight coefficient as the initial weight coefficient, and repeatedly execute the step of determining the fitness value corresponding to the initial weight coefficient until the fitness value corresponding to the initial weight coefficient meets the preset fitness condition, and / or the number of repeated executions meets the preset number condition, thereby obtaining the target weight coefficient.

[0046] Optionally, the fitness value determining unit includes:

[0047] An initial hybrid kernel function obtaining subunit is used to process the kernel function to be processed based on the initial weight coefficient to obtain an initial hybrid kernel function;

[0048] An initial machine learning model obtaining subunit is used to replace the kernel function in the support vector machine with the initial hybrid kernel function to obtain an initial machine learning model;

[0049] an initial rotation Doppler signal classification model obtaining subunit, configured to train the initial machine learning model using the training signal and the training type of the training signal to obtain an initial rotation Doppler signal classification model;

[0050] The fitness value determining subunit is configured to determine the fitness value corresponding to the initial weight coefficient based on the initial rotating Doppler signal classification model.

[0051] Optionally, the kernel function to be processed includes a Gaussian kernel function and / or a polynomial kernel function.

[0052] Optionally, the device comprises:

[0053] A signal acquisition module, used for acquiring a signal of a preset rotational Doppler effect;

[0054] The type obtaining module is used to input the rotational Doppler effect signal into the target rotational Doppler signal classification model to obtain the type of the rotational Doppler effect signal.

[0055] The embodiment of the present application further discloses an electronic device, comprising a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus;

[0056] The memory is used to store computer programs;

[0057] The processor is used to implement the method described in the embodiment of the present application when executing the program stored in the memory.

[0058] The embodiments of the present application also disclose one or more computer-readable media having instructions stored thereon, which, when executed by one or more processors, enable the processors to perform the method described in the embodiments of the present application.

[0059] The embodiments of the present application include the following advantages:

[0060] In an embodiment of the present application, a training signal of a preset training rotational Doppler effect is collected; the training rotational Doppler effect is the Doppler effect generated when a preset training light beam is irradiated on the surface of a preset training rotating object; the training light beam and the training rotating object are the light beam and the rotating object used to generate the training signal, respectively; the training signal is a signal used for model training. When acquiring the training signal, the relative posture relationship between the propagation axis of the training light beam and the rotation axis of the training rotating object is obtained, and the training type of the training signal is determined based on the relative posture relationship; the training type of the training signal is the type of signal used for model training. A target hybrid kernel function is designed, and the target hybrid kernel function is used to replace the kernel function in the preset support vector machine to obtain a target machine learning model; the target machine learning model is trained using the training signal and the training type of the training signal to obtain a target rotational Doppler signal classification model. In an embodiment of the present application, a target hybrid kernel function is designed for the complex signal characteristics of the rotational Doppler signal, which effectively improves the support vector machine's ability to classify complex data. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1This is a flowchart of the steps of a method for constructing a rotational Doppler signal classification model provided in an embodiment of the present application;

[0062] Figure 2 This is a single-peak rotational Doppler frequency shift signal diagram provided in an embodiment of the present application;

[0063] Figure 3 This is a signal detection result diagram provided in an embodiment of the present application when the purity of the beam mode is not high;

[0064] Figure 4 This is a rotational Doppler frequency shift signal diagram provided in an embodiment of the present application under the condition that the vortex beam is not aligned with the rotation axis;

[0065] Figure 5 An embodiment of the present application provides a graph of equally spaced frequency signals under conditions of tilted light beam illumination;

[0066] Figure 6 This is a time domain and frequency domain signal diagram provided in an embodiment of the present application under the condition that the light beam is completely not aligned with the rotation axis;

[0067] Figure 7 Schematic diagram of the classification of the support vector machine model provided in the embodiment of the present application;

[0068] Figure 8 This is a signal classification diagram provided in an embodiment of the present application;

[0069] Figure 9 is a classification flow chart of a rotational Doppler signal provided in an embodiment of the present application;

[0070] Figure 10 This is a structural block diagram of a device for constructing a rotational Doppler signal classification model provided in an embodiment of the present application;

[0071] Figure 11 is a block diagram of an electronic device provided in an embodiment of the present application;

[0072] Figure 12 It is a schematic diagram of a computer-readable medium provided in an embodiment of the present application. DETAILED DESCRIPTION

[0073] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0074] To facilitate understanding of the technical solutions and technical effects of the embodiments of the present application, the relevant technologies of the present application are briefly described below.

[0075] In the related art, based on the superposition state vortex beam, the rotation speed of the macroscopic object can be detected and measured. Specifically, the superposition state vortex beam can be applied to the rotating macroscopic object. Under the rotation of the macroscopic object, an optical rotational Doppler effect will be generated, that is, the frequency of the light reflected or scattered by the macroscopic object will change; by detecting the frequency change of the reflected or scattered light, the rotation speed information of the object can be determined. Therefore, the rotational Doppler effect can be used to measure the rotation speed of the object. In addition, the use of supercontinuum white light can also realize the observation of the rotational Doppler effect of the object. The rotational Doppler effect can also be used to detect the composite motion speed of the object, as well as the acceleration, axis position, rotation direction and other motion information of the object.

[0076] The basic principle of measuring an object's rotational speed based on the rotational Doppler effect is that it is essentially a Doppler effect, arising from the relative motion between the wave source and the object. While the classic linear Doppler effect is caused by the object's motion along the wave source's propagation direction, the rotational Doppler effect stems from the object's relative rotational motion within a cross-section perpendicular to the wave source's propagation direction. Therefore, the Doppler effect manifests itself primarily in the relative motion between the object and the wave source, causing the frequency of the beam received by the object to increase or decrease. This indicates that to detect the rotational Doppler effect, the acquired signal must be converted to the frequency domain.

[0077] When a laser beam is irradiated onto a moving object, the frequency shift Δf of the laser beam due to the movement of the object is calculated as follows:

[0078] Δf=f0νcosα / c

[0079] Among them, f0 represents the frequency of the laser light source, v represents the speed of the object, c represents the speed of light, and α is the angle between the direction of the object's speed and the direction of wave source propagation.

[0080] According to the frequency shift calculation formula, when the relative motion angle between the object's motion direction and the wave source propagation direction is 90°, the linear Doppler frequency shift generated by the laser is zero. Vortex light is different because the rotational motion itself has a rotation vector, and the rotation vector is perpendicular to the rotation plane. Therefore, when the light beam is perpendicular to the rotation plane, the rotation vector is actually parallel to the beam propagation axis, and the rotational Doppler effect will continue to exist. For vortex light, its definition contains a spiral phase factor The Poynting Vector of the beam is no longer along the propagation direction of the beam, but has an angle with the propagation direction. The size of this angle is in, Refers to the topological charge information of the beam.

[0081] Since the movement direction of each tiny scattering point in the rotating plane of the object is exactly perpendicular to the propagation direction of the light beam, the relationship between the angle between the energy flow direction of the final light beam and the direction of the speed of the scattering point is:

[0082] α=π / 2-β

[0083] According to the frequency shift calculation formula and the relationship between the angle between the energy flux direction of the final light beam and the velocity direction of the scattering point, and because sinα≈α when α is very small, the general expression of the rotational Doppler effect can be obtained:

[0084]

[0085] In actual measurement, there are many ways to measure the frequency shift Δf of the rotational Doppler effect. 14 Hz, it is difficult to measure directly, so the common frequency shift measurement method is to use the beat frequency detection method of the light beam to couple the signal light and the reference light to beat the frequency, so as to realize the detection of the frequency shift. According to the actual frequency shift measurement value Δf, combined with the beam topological charge information The speed of rotating objects can be measured.

[0086] However, the general expression for the rotational Doppler effect is valid only if the propagation axis of the beam is parallel to the object's rotational axis. When the object's rotational axis is not coaxial with the beam's propagation axis, the frequency shift due to the rotational Doppler effect can be calculated using the following formula:

[0087]

[0088] These two formulas represent the situation when there is a lateral offset between the beam propagation axis and the object's rotational axis, respectively, and when there is a tilt angle between the two. These two formulas indicate that when there is an offset between the beam propagation axis and the object's rotational axis, the rotational Doppler signal is no longer simply a function of rotational speed and beam topological charge. Instead, it becomes a parameter closely related to the object's rotational speed, the beam topological charge, and the illumination pose (lateral offset d, tilt angle θ). Further analysis reveals that the object's rotational speed and beam topological charge only affect the magnitude of the rotational Doppler effect frequency shift signal, while the relative pose between the beam propagation axis and the object's rotational axis affects the signal's distribution. Because the signal distribution is closely related to the extraction of object rotational speed information, accurate extraction of rotational speed information can only be achieved by accurately classifying the detection signal based on the relative pose between the beam propagation axis and the object's rotational axis.

[0089] In the related art, the method for extracting the rotational speed information of an object is only for the rotational speed signal under a specific posture, and cannot handle situations under any variety of postures. In any variety of postures, the signal classification of the rotational Doppler effect mainly relies on manual spectrum analysis. However, manual spectrum analysis has at least the following problems: first, the spectrum conversion process is complicated, requiring Fourier transform and manual interpretation, which is inefficient; second, it is only applicable to ideal conditions where the light beam and the rotation axis are strictly aligned, and lacks the ability to classify complex posture scenes such as offset and tilt. Therefore, in the related art, there is a problem that under the object detection system based on the rotational Doppler effect, the signal type is difficult to accurately identify and classify, resulting in the difficulty in accurately obtaining the rotational speed information of the object.

[0090] Reference Figure 1 , shows a flowchart of the steps of a method for constructing a rotational Doppler signal classification model provided in an embodiment of the present application, which may specifically include the following steps:

[0091] Step 101: Acquire a training signal of a preset training rotational Doppler effect; the training rotational Doppler effect is the Doppler effect generated when a preset training light beam is irradiated on the surface of a preset training rotating object; the training light beam and the training rotating object are the light beam and the rotating object, respectively, used to generate the training signal; the training signal is a signal used for model training;

[0092] In an embodiment of the present application, a training beam is irradiated onto the surface of a training rotating object. As the training rotating object rotates, a training rotational Doppler effect is generated, resulting in a change in the frequency of light reflected or scattered by the training rotating object. By detecting this change in the frequency of the reflected or scattered light, the rotational speed of the training rotating object can be determined. The training beam and the training rotating object are, respectively, the beam used to generate the training signal, and the training signal is the signal used for model training.

[0093] In the embodiment of the present application, a training signal for training the rotational Doppler effect may be collected, where the training signal is a frequency variation signal of reflected or scattered light.

[0094] Step 102: Obtain the relative position relationship between the propagation axis of the training light beam and the rotation axis of the training rotating object when the training signal is collected, and determine the training type of the training signal based on the relative position relationship; the training type of the training signal is the type of signal used for model training;

[0095] In the present embodiment, the signal classification of the rotational Doppler effect is based on the relative positional relationship between the propagation axis of the light beam and the rotational axis of the object. Therefore, the relative positional relationship between the propagation axis of the training light beam and the rotational axis of the training rotating object can be obtained during training signal acquisition, and the training type of the training signal can be determined based on this relative positional relationship. The training type of the training signal is the type of signal used for model training.

[0096] It should be noted that the training signals and training signal types in this application can be based on the signal classification basis of the rotational Doppler effect, using a vortex optical rotational Doppler signal simulation detection program to obtain a data set with classification labels. The data set includes the training signal, and the classification label is the training type of the training signal.

[0097] In some embodiments of the present application, determining the training type of the training signal according to the relative posture relationship includes:

[0098] The training type of the training signal is determined according to the lateral offset and / or tilt angle between the propagation axis and the rotation axis.

[0099] In an embodiment of the present application, the signal classification of the rotational Doppler effect is based on the relative posture relationship between the beam propagation axis and the object rotation axis. The relative posture relationship can be divided into four categories according to the size of the lateral offset and / or tilt angle between the beam propagation axis and the object rotation axis, and the corresponding signal categories are also divided into four types.

[0100] Therefore, the training type of the training signal can be determined based on the lateral offset and / or tilt angle between the propagation axis of the training light beam and the rotation axis of the training rotating object.

[0101] In the embodiment of the present application, the signal classification principle obtained by rotational Doppler detection is as follows:

[0102] Rotational Doppler detection is an active detection method. When the detection beam strikes the surface of a rotating object, the primary physical parameter that changes is frequency. Therefore, spectrum analysis methods are often used in rotational Doppler signal processing and extraction. Depending on the detection conditions, the corresponding spectrum detection signal will exhibit different characteristics. Common signal forms include single peak signals, equally spaced spread spectrum signals, and random spread signals.

[0103] Ideal detection conditions are when the detection beam has only a single mode component, the detection beam illuminates the rotating object vertically and coaxially, and the object's rotational speed is constant. In this case, the generated rotational Doppler frequency shift signal is a stable single peak signal. Figure 2 , shows a single-peak rotational Doppler frequency shift signal diagram provided in an embodiment of the present application. Figure 2The time domain signal and frequency domain rotational Doppler shift signal are shown under the conditions of the detection beam topological charge m = ±15 and the object rotation speed ω = 40rps (Revolutions Per Second). The characteristics of this type of signal are that the time domain signal has obvious periodic changes and the frequency domain signal is a single peak signal. By reading this frequency value, the object rotation speed can be accurately calculated.

[0104] In addition to ideal detection conditions, deviations are more common in actual detection. The first thing that is prone to deviation is the mode purity of the light beam. The main methods for preparing vortex beams for detection include spatial light modulators, vortex slides, and arrays. Due to the limitations of phase modulation methods and the accuracy of optical components, the generated vortex beams cannot achieve 100% mode purity, and there will always be a certain amount of stray light. In actual applications, in order to ensure the simplicity of the beam preparation method and the requirements of higher optical power, there will also be a phenomenon of sacrificing part of the beam mode purity to achieve the best detection effect. In this detection situation where the mode purity is not high, there will usually be certain noise around the frequency domain signal. In the case of poor mode purity, it may even submerge the main peak signal, which brings certain challenges to signal interpretation.

[0105] Reference Figure 3 , shows a signal detection result diagram when the purity of a beam mode provided in an embodiment of the present application is not high. Figure 3 It can be seen that the detection time domain signal has large fluctuations and the frequency domain signal has multiple peaks, but within a certain range, the main peak frequency signal is still the largest component.

[0106] Another area where deviations can occur in actual detection is when it's difficult to ensure the beam is perfectly aligned with the rotational axis and the rotating object. This alignment requires that the beam propagation axis completely coincide with the object's rotational axis. Otherwise, from the perspective of the microscatterer analysis model, the speeds of each small scatterer within the light field will no longer be uniform, resulting in varying rotational Doppler shifts. This results in a series of broadened spectrum signals around the main peak of the frequency domain signal. When the beam misalignment is minimal, the frequency domain signal broadening is relatively small. When the beam deflects significantly, the Doppler frequency domain signal becomes a series of broadened spectra starting from zero, making the ideal Doppler shift signal difficult to observe.

[0107] Reference Figure 4 , shows a rotational Doppler frequency shift signal diagram provided in an embodiment of the present application under the condition that the vortex light beam is not aligned with the rotation axis. Figure 4 (a)-(b) in the figure represent the time domain and frequency domain signal diagrams when the lateral offset distance is less than the vortex beam radius, that is, d<r, that is, when the object's rotation center is still within the coverage range of the vortex beam. Figure 4(c)-(d) in the figure represent the time domain and frequency domain signal diagrams when the lateral offset distance is greater than the beam radius, that is, d>r, that is, when the object's rotation center has left the coverage range of the vortex beam. Figure 4 In the two cases shown, it is difficult to identify the signal by the main peak, and the signal is broadened into a series of wider spectra, which is a common signal feature in actual detection.

[0108] In addition to lateral misalignment, there is also the case of tilted irradiation when the beam is not aligned with the axis of rotation. Under the condition of tilted beam irradiation, due to factors such as uneven scattering point speed, the signal will also be broadened, and the peak signal will be difficult to identify. Figure 5 , shows a graph of equally spaced frequency signals under an oblique beam illumination condition, as provided in an embodiment of this application. Unlike the aforementioned signal characteristics, when the beam is obliquely illuminated at the object's rotational center, the Doppler frequency shift signal broadens at equal intervals. Theoretically, this frequency shift interval is an integer multiple of the object's rotational speed.

[0109] When the light beam is completely misaligned with the rotation axis and illuminates the rotating object, that is, there is a lateral offset and an oblique incident angle between the propagation axis of the vortex beam and the rotation axis of the object, the detection signal has the most complex characteristics. Figure 6 , showing the time and frequency domain signal diagrams for an embodiment of the present application, where the light beam is completely misaligned with the axis of rotation. From a phase perspective, the surface of the object is non-uniform, which further distorts the phase of the scattered light after interacting with the detection beam, making the frequency domain signal irregular. In this case, the scattered echo signal also stretches from zero to a maximum value, with no main peak signal, making it impossible to determine the object's rotational speed based on the highest peak value.

[0110] Step 103: design a target hybrid kernel function, and use the target hybrid kernel function to replace the kernel function in the preset support vector machine to obtain a target machine learning model;

[0111] In the embodiments of the present application, the SVM (Support Vector Machine) model is a machine learning model. The kernel function is an important component of the SVM model, which can map the input data from the original space to the high-dimensional feature space without explicitly calculating the coordinates after the mapping. Through the kernel function, the inner product can be calculated directly in the high-dimensional space, which can effectively avoid the dimensionality curse. In related technologies, the SVM model usually only uses a single kernel function.

[0112] In an embodiment of the present application, a target hybrid kernel function is designed to target the complex signal characteristics of the rotating Doppler signal. The target kernel function can be used to replace the single kernel function in the support vector machine to obtain a target machine learning model.

[0113] In some embodiments of the present application, the design target hybrid kernel function includes:

[0114] Obtain at least one kernel function to be processed;

[0115] Determining the target weight coefficient of the kernel function to be processed by using a preset genetic algorithm;

[0116] The kernel function to be processed is processed based on the target weight coefficient to obtain the target hybrid kernel function.

[0117] In an embodiment of the present application, when designing a target hybrid kernel function, at least one kernel function to be processed is first obtained. A genetic algorithm is then used to determine a target weight coefficient for the kernel function to be processed. The kernel function to be processed is then processed based on the target weight coefficient to obtain the target hybrid kernel function. Specifically, the kernel function to be processed and the target weight coefficient of the kernel function to be processed are used to perform a weighted combination of the at least one kernel function to obtain the target hybrid kernel function.

[0118] In some embodiments of the present application, the kernel function to be processed includes a Gaussian kernel function and / or a polynomial kernel function.

[0119] In the embodiment of the present application, the kernel function to be processed is a single kernel function, which may include a Gaussian kernel function and / or a polynomial kernel function.

[0120] In some embodiments of the present application, determining the target weight coefficient of the kernel function to be processed by using a preset genetic algorithm includes:

[0121] Determining at least one initial weight coefficient of the kernel function to be processed, and determining a fitness value corresponding to the initial weight coefficient;

[0122] Based on the fitness value, determining a weight coefficient to be processed from the initial weight coefficients;

[0123] Performing crossover and / or mutation processing on the weight coefficient to be processed to generate an updated weight coefficient;

[0124] The updated weight coefficient is used as the initial weight coefficient, and the step of determining the fitness value corresponding to the initial weight coefficient is repeatedly performed until the fitness value corresponding to the initial weight coefficient meets the preset fitness condition, and / or the number of repeated executions meets the preset number condition, thereby obtaining the target weight coefficient.

[0125] In an embodiment of the present application, based on a genetic algorithm, the steps for determining the target weight coefficient of the kernel function to be processed are as follows: First, randomly generate at least one initial weight coefficient of the kernel function to be processed. Multiple sets of initial weight coefficients are generated for the kernel function to be processed, and the sum of each set of initial weight coefficients is 1. Fitness evaluation is performed on each set of initial weight coefficients to determine the fitness value corresponding to each set of initial weight coefficients. Then, based on the fitness value corresponding to each set of initial weight coefficients, the weight coefficient to be processed is determined from the multiple sets of initial weight coefficients. Specifically, the initial weight coefficients corresponding to the fitness values of the top K% with the highest fitness values can be used as the weight coefficient to be processed.

[0126] Next, the weight coefficients to be processed can be subjected to crossover and / or mutation processing to generate updated weight coefficients. Crossover processing of the weight coefficients to be processed refers to taking the weight coefficients to be processed as parent individuals, randomly selecting two parent individuals to perform crossover operations such as single-point crossover, multi-point crossover, or uniform crossover, and generating offspring individuals, which are the updated weight coefficients. Mutation processing of the weight coefficients to be processed refers to perturbing the weight coefficients to be processed with a certain probability to form a mutation and obtain an updated weight coefficient. Perturbing the weight coefficients to be processed with a certain probability can refer to changing a certain value of the weight coefficients to be processed with a certain probability or adding random noise to the weight coefficients to be processed.

[0127] In an embodiment of the present application, the updated weight coefficient may be used as the initial weight coefficient, and the step of determining the fitness value corresponding to the initial weight coefficient may be repeated until the fitness value corresponding to the initial weight coefficient satisfies a preset fitness condition and / or the number of repetitions satisfies a preset number condition, thereby obtaining a target weight coefficient. The preset fitness condition refers to the stability of the fitness value corresponding to the initial weight coefficient.

[0128] In some embodiments of the present application, determining the fitness value corresponding to the initial weight coefficient includes:

[0129] Processing the kernel function to be processed based on the initial weight coefficient to obtain an initial hybrid kernel function;

[0130] Replacing the kernel function in the support vector machine with the initial hybrid kernel function to obtain an initial machine learning model;

[0131] Training the initial machine learning model using the training signal and the training type of the training signal to obtain an initial rotational Doppler signal classification model;

[0132] Based on the initial rotational Doppler signal classification model, a fitness value corresponding to the initial weight coefficient is determined.

[0133] In an embodiment of the present application, a fitness evaluation is performed on each set of initial weight coefficients, and the steps for determining the fitness value corresponding to each set of initial weight coefficients are as follows: processing the kernel function to be processed based on the initial weight coefficients to obtain an initial mixed kernel function. That is, using the kernel function to be processed and the initial weight coefficients of the kernel function to be processed, at least one kernel function to be processed can be weightedly combined to obtain an initial mixed kernel function. The kernel function in the support vector machine is replaced by the initial mixed kernel function to obtain an initial machine learning model. Then, the initial machine learning model is trained using the training signal and the training type of the training signal to obtain an initial rotation Doppler signal classification model. Based on the initial rotation Doppler signal classification model, the fitness value corresponding to the initial weight coefficient can be determined. Specifically, the initial rotation Doppler signal classification model can be subjected to a performance evaluation of signal classification, and the evaluation result is the fitness value corresponding to the initial weight coefficient.

[0134] In a specific example, there are two kernel functions to be processed, namely Gaussian kernel function and polynomial kernel function. The Gaussian kernel (RBF, Radial Basis Function) expression is:

[0135]

[0136] Here, σ represents the kernel width parameter, which controls the decay rate of the Gaussian kernel function. The Gaussian kernel function excels at capturing local nonlinear characteristics of data and is suitable for classifying signals with complex distributions in high-dimensional space.

[0137] The expression of the polynomial kernel function is:

[0138] K poly (x,y)=(γ·x·y+c) d

[0139] Here, γ is the scaling factor that controls the kernel function's scale; c is a constant term that adjusts the kernel function's offset; and d is the polynomial degree, which determines the kernel function's complexity. Polynomial kernel functions can model global characteristics of data and are particularly well-suited for fitting regularities in frequency-domain broadened signals.

[0140] By performing a weighted combination of the Gaussian kernel function and the polynomial kernel function, the target hybrid kernel function K can be achieved. hybrid Build:

[0141] K hybird (x,y)=α·K RBF (x,y)+β·K poly (x,y)

[0142] Here, α and β represent weight coefficients, satisfying α+β=1.

[0143] The steps of using genetic algorithm to obtain the target weight coefficients of Gaussian kernel function and polynomial kernel function are as follows: ① Initialize the population, randomly generate N groups of initial weight coefficients (α i ,β i ), satisfying α i +β i =1. ②Fitness evaluation: Based on each set of initial weight coefficients, the corresponding initial machine learning model is trained to obtain the initial rotational Doppler signal classification model, and the cross-validation accuracy (fitness value) of the initial rotational Doppler signal classification model is calculated. ③Selection operation: The top k% individuals with the highest fitness are retained as the weight coefficients to be processed. ④Crossover and mutation: Randomly select two weight coefficients to be processed as parent individuals, generate offspring individuals, and obtain updated weight coefficients to form a crossover; or perturb the weights with a certain probability to form a mutation, and obtain updated weight coefficients to ensure the diversity of the population; ⑤Iterative convergence: Use the updated weight coefficient as the initial weight coefficient and repeat the above steps ②-④ until the fitness value of the initial weight coefficient is stable or the maximum number of iterations is reached.

[0144] In the embodiment of the present application, the target hybrid kernel function support vector machine learning framework principle is:

[0145] A support vector machine (SVM) is a binary classification model whose basic model is a linear classifier defined in feature space with the largest margin. Simply put, a SVM uses a line or surface in feature space to separate a training dataset into two categories. The principle behind this is to maximize the margin, which means that the distance from the point closest to the separating line or surface in feature space to this line or surface is maximized. Finding the hyperplane with the largest geometric margin in the training dataset means classifying the training data with sufficient confidence. This means that the hyperplane not only separates positive and negative instances, but also separates even the most challenging instances with sufficient confidence. This gives it excellent classification and prediction capabilities for unknown new instances.

[0146] Reference Figure 7 , which shows a classification diagram of the support vector machine model provided in an embodiment of the present application. Figure 7 Several simple classification diagrams of the support vector machine model are shown, where (a) represents linear classification, (b) represents nonlinear classification, and (c) represents hyperplane classification.

[0147] In practical applications, simple linear and nonlinear classification applications are rare, and most of them are hyperplane classification. A hyperplane refers to a subspace of dimension n-1 in an n-dimensional linear space, which can divide the linear space into two non-intersecting parts. For example, in a two-dimensional space, a straight line is one-dimensional, and it divides the plane into two pieces; in a three-dimensional space, a plane is two-dimensional, and it divides the space into two pieces. In the sample space, the dividing hyperplane can be obtained by the linear equation ω T x+b=0, where ω=(ω1,ω2,...,ω d ) is the normal vector, determining the direction of the hyperplane; b is the displacement term, determining the distance between the hyperplane and the origin. A partitioning hyperplane is determined by ω and the displacement b. The support vector machine prediction model uses support vectors to determine the partitioning hyperplane, thereby achieving classification and prediction.

[0148] According to the above definition, the most important thing in the support vector machine model is to train and find the dividing hyperplane. Since the dividing hyperplane is not certain, it is necessary to compare the distance from the same point to different hyperplanes. A two-dimensional space point p = (x p ,y p ) to the straight line Ax+By+C=0 is:

[0149]

[0150] After expanding to n-dimensional space, point x=(x1,x2,x3,...,x n ) to the hyperplane ω T The distance formula for x+b=0 is:

[0151]

[0152] in,

[0153] When considering finding the dividing hyperplane with the greatest separation between two data sets, it's intuitive to start from the point closest to the two data sets. This is because only points at the extreme edges of the two spaces are likely to be closest to the dividing hyperplane, while other points play no role in determining the final position of the hyperplane. Therefore, these edge points can be considered to support the establishment of the hyperplane. In mathematics, sequences of points are also called vectors. For example, a two-dimensional point (x, y) can be considered a two-dimensional vector. Similarly, an n-dimensional point is an n-dimensional vector. Therefore, these useful edge points are called "support vectors."

[0154] The optimization goal of the support vector machine model is to find the farthest distance between each sample point and the hyperplane, that is, to find the maximum margin hyperplane. According to the point x=(x1,x2,x3,...,x n ) to the hyperplane ω TThe distance formula of x+b=0 can be used to obtain the mathematical expression of the support vector machine regression prediction model:

[0155]

[0156]

[0157] Among them, ω, b are the parameters to be determined for the model; x i is the sample eigenvalue; y i is the sample label value; C is the penalty coefficient; ξ i is the error term; ε is the function f(x i ) and y i deviation.

[0158] Step 104: Use the training signal and the training type of the training signal to train the target machine learning model to obtain a target rotation Doppler signal classification model.

[0159] In the embodiment of the present application, the training signal and the training type of the training signal are used as a training set, and the target machine learning model is trained using the training set to obtain a target rotation Doppler signal classification model. The training set is a preprocessed time domain signal data set.

[0160] In some embodiments of the present application, the method includes:

[0161] Collecting signals with preset rotational Doppler effect;

[0162] The rotational Doppler effect signal is input into the target rotational Doppler signal classification model to obtain the type of the rotational Doppler effect signal.

[0163] In the embodiment of the present application, a vortex beam is prepared and irradiated onto the surface of a rotating object to generate a rotational Doppler effect. Then, a photodetector and a data acquisition card are used to collect time domain signal data with the same specifications as the training data.

[0164] In an embodiment of the present application, the collected time domain signal data is input into the target rotational Doppler signal classification model for prediction and classification to obtain a signal category label. Based on the category label information, the signal can be accurately classified to obtain the type of rotational Doppler effect signal without the need for spectrum conversion or drawing judgment.

[0165] Reference Figure 8 , shows a signal classification schematic diagram provided in an embodiment of the present application. Figure 8 The graph shows the detection signal classification results obtained under different postures when the detection light topological charge number is ±12 and the sampling rate is 10000 Hz. The accuracy of the signal classification result is 100%.

[0166] Reference Figure 9 , shows a classification flow chart of a rotational Doppler signal provided in an embodiment of the present application. First, a training data set is prepared. Based on the principle of rotational Doppler signal feature classification under different postures, a data set with classification labels is generated using a vortex light rotational Doppler signal simulation program. Then, the model is constructed and trained. A hybrid kernel SVM machine learning model is constructed, specifically using a weighted combination of a Gaussian kernel and a polynomial kernel, and the kernel weights are optimized by a genetic algorithm to obtain a target machine learning model. The target machine learning model is trained using the training data set to obtain a trained grid model, which is the target rotational Doppler signal classification model.

[0167] Next, actual detection data is generated. A vortex beam is prepared and irradiated onto the surface of a rotating object. A photodetector and data acquisition card are then used to acquire the signal and obtain the actual data. By combining the sampling rate and the beam's topological charge information, actual detection signal data with the same specifications as the training dataset can be obtained. The rotating object can also be called a rotating target.

[0168] Finally, signal category prediction: The collected data is input into the trained network model. After algorithm processing, the signal category label is obtained. Based on the category label information, the signal can be accurately classified.

[0169] In an embodiment of the present application, a training signal of a preset training rotational Doppler effect is collected; the training rotational Doppler effect is the Doppler effect generated when a preset training light beam is irradiated on the surface of a preset training rotating object; the training light beam and the training rotating object are the light beam and the rotating object used to generate the training signal, respectively; the training signal is a signal used for model training. When acquiring the training signal, the relative posture relationship between the propagation axis of the training light beam and the rotation axis of the training rotating object is obtained, and the training type of the training signal is determined based on the relative posture relationship; the training type of the training signal is the type of signal used for model training. A target hybrid kernel function is designed, and the target hybrid kernel function is used to replace the kernel function in the preset support vector machine to obtain a target machine learning model; the target machine learning model is trained using the training signal and the training type of the training signal to obtain a target rotational Doppler signal classification model. In an embodiment of the present application, a target hybrid kernel function is designed for the complex signal characteristics of the rotational Doppler signal, which effectively improves the support vector machine's ability to classify complex data.

[0170] In an embodiment of the present application, the target weight coefficients of the Gaussian kernel function and the polynomial kernel function are determined by a genetic algorithm. Based on these target weight coefficients, the Gaussian kernel function and the polynomial kernel function are weighted and combined to obtain a target hybrid kernel function. This target hybrid kernel function can achieve the fusion of local and global features and has dynamic adaptability. By replacing the kernel function in the support vector machine with the target hybrid kernel function, a target machine learning model is obtained. The target machine learning model is trained to obtain a target rotational Doppler signal classification model. Using this classification model, accurate classification of rotational Doppler time domain signals can be achieved without performing a Fourier spectrum transform. The target rotational Doppler signal classification model can directly classify time domain detection data, thereby further achieving higher-precision rotational speed detection and extraction. This technology is novel, simple to operate, easy to implement, and highly adaptable under practical conditions. It has high credibility in identifying rotational Doppler signal types and can achieve frequency signal recognition in different postures, opening up a new path for the accurate extraction and identification of different types of rotational speed signals based on intelligent algorithms. This technology is suitable for various unmanned environments. Due to the use of intelligent algorithms, it can autonomously classify signals without the need for further manual identification and processing, greatly improving the degree of automation and the efficiency of target speed extraction.

[0171] In the embodiment of the present application, a support vector machine signal classification model of a target hybrid kernel function is designed based on the principle of machine learning, specifically combining the Gaussian kernel (capturing nonlinear characteristics) and the polynomial kernel (fitting the frequency domain broadening law), and determining the weights of the Gaussian kernel function and the polynomial kernel function through a genetic algorithm to improve the classification robustness of the model for complex posture signals. In actual operation, the time domain signal obtained by the acquisition is directly used as the input parameter of the target rotation Doppler signal classification model to accurately identify the detection signal category, thereby laying a good foundation for further accurate extraction of the rotation speed. The operation is simple and the execution is efficient. Compared with the traditional manual identification and classification method after spectrum conversion, it is more efficient and convenient, has strong adaptability, solves the classification problem of multimodal signals, and has broad application prospects in harsh scenarios such as target speed measurement occasions with high automation requirements, such as spacecraft attitude detection, high-speed rotating machinery fault diagnosis, etc.

[0172] It should be noted that for the method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the embodiments of the present application are not limited by the order of the actions described, because according to the embodiments of the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily required by the embodiments of the present application.

[0173] Reference Figure 10 , shows a structural block diagram of a device for constructing a rotational Doppler signal classification model provided in an embodiment of the present application, which may specifically include the following modules:

[0174] The training signal acquisition module 1001 is configured to acquire a training signal for a preset training rotational Doppler effect; the training rotational Doppler effect is the Doppler effect generated when a preset training light beam is irradiated on the surface of a preset training rotating object; the training light beam and the training rotating object are the light beam and the rotating object, respectively, used to generate the training signal; the training signal is a signal used for model training;

[0175] A training type determination module 1002 is configured to obtain a relative positional relationship between the propagation axis of the training beam and the rotation axis of the training rotating object when the training signal is collected, and determine a training type of the training signal based on the relative positional relationship; the training type of the training signal is the type of signal used for model training;

[0176] Function design module 1003, used to design a target hybrid kernel function, and use the target hybrid kernel function to replace the kernel function in the preset support vector machine to obtain a target machine learning model;

[0177] The model acquisition module 1004 is used to train the target machine learning model using the training signal and the training type of the training signal to obtain a target rotation Doppler signal classification model.

[0178] In an optional embodiment of the present application, the training type determination module includes:

[0179] The training type determination submodule is configured to determine the training type of the training signal according to the lateral offset and / or tilt angle between the propagation axis and the rotation axis.

[0180] In an optional embodiment of the present application, the function design module includes:

[0181] A kernel function to be processed acquisition submodule, used to acquire at least one kernel function to be processed;

[0182] A target weight coefficient determination submodule is used to determine the target weight coefficient of the kernel function to be processed using a preset genetic algorithm;

[0183] The target mixed kernel function obtaining submodule is used to process the kernel function to be processed based on the target weight coefficient to obtain the target mixed kernel function.

[0184] In an optional embodiment of the present application, the target weight coefficient determination submodule includes:

[0185] a fitness value determining unit, configured to determine at least one initial weight coefficient of the kernel function to be processed, and determine a fitness value corresponding to the initial weight coefficient;

[0186] a to-be-processed weight coefficient determining unit, configured to determine a to-be-processed weight coefficient from the initial weight coefficients based on the fitness value;

[0187] An updated weight coefficient generating unit, configured to perform crossover and / or mutation processing on the weight coefficient to be processed to generate an updated weight coefficient;

[0188] A repeated execution unit is used to use the updated weight coefficient as the initial weight coefficient, and repeatedly execute the step of determining the fitness value corresponding to the initial weight coefficient until the fitness value corresponding to the initial weight coefficient meets the preset fitness condition, and / or the number of repeated executions meets the preset number condition, thereby obtaining the target weight coefficient.

[0189] In an optional embodiment of the present application, the fitness value determining unit includes:

[0190] An initial hybrid kernel function obtaining subunit is used to process the kernel function to be processed based on the initial weight coefficient to obtain an initial hybrid kernel function;

[0191] An initial machine learning model obtaining subunit is used to replace the kernel function in the support vector machine with the initial hybrid kernel function to obtain an initial machine learning model;

[0192] an initial rotation Doppler signal classification model obtaining subunit, configured to train the initial machine learning model using the training signal and the training type of the training signal to obtain an initial rotation Doppler signal classification model;

[0193] The fitness value determining subunit is configured to determine the fitness value corresponding to the initial weight coefficient based on the initial rotating Doppler signal classification model.

[0194] In an optional embodiment of the present application, the kernel function to be processed includes a Gaussian kernel function and / or a polynomial kernel function.

[0195] In an optional embodiment of the present application, the device includes:

[0196] A signal acquisition module, used for acquiring a signal of a preset rotational Doppler effect;

[0197] The type obtaining module is used to input the rotational Doppler effect signal into the target rotational Doppler signal classification model to obtain the type of the rotational Doppler effect signal.

[0198] As for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0199] In addition, the present invention also provides an electronic device, such as Figure 11 As shown, it includes a processor 1101, a communication interface 1102, a memory 1103 and a communication bus 1104, wherein the processor 1101, the communication interface 1102, and the memory 1103 communicate with each other through the communication bus 1104.

[0200] Memory 1103, used for storing computer programs;

[0201] The processor 1101 is configured to execute the program stored in the memory 1103 by performing the following steps:

[0202] Collecting a training signal of a preset training rotational Doppler effect; the training rotational Doppler effect is the Doppler effect generated when a preset training light beam is irradiated on the surface of a preset training rotating object; the training light beam and the training rotating object are respectively used to generate the training signal; the training signal is a signal used for model training;

[0203] Obtaining the relative positional relationship between the propagation axis of the training light beam and the rotation axis of the training rotating object when collecting the training signal, and determining the training type of the training signal based on the relative positional relationship; the training type of the training signal is the type of signal used for model training;

[0204] Designing a target hybrid kernel function, and using the target hybrid kernel function to replace the kernel function in a preset support vector machine to obtain a target machine learning model;

[0205] The target machine learning model is trained using the training signal and the training type of the training signal to obtain a target rotation Doppler signal classification model.

[0206] In an optional embodiment of the present application, determining the training type of the training signal according to the relative posture relationship includes:

[0207] The training type of the training signal is determined according to the lateral offset and / or tilt angle between the propagation axis and the rotation axis.

[0208] In an optional embodiment of the present application, the design target hybrid kernel function includes:

[0209] Obtain at least one kernel function to be processed;

[0210] Determining the target weight coefficient of the kernel function to be processed by using a preset genetic algorithm;

[0211] The kernel function to be processed is processed based on the target weight coefficient to obtain the target hybrid kernel function.

[0212] In an optional embodiment of the present application, determining the target weight coefficient of the kernel function to be processed by using a preset genetic algorithm includes:

[0213] Determining at least one initial weight coefficient of the kernel function to be processed, and determining a fitness value corresponding to the initial weight coefficient;

[0214] Based on the fitness value, determining a weight coefficient to be processed from the initial weight coefficients;

[0215] Performing crossover and / or mutation processing on the weight coefficient to be processed to generate an updated weight coefficient;

[0216] The updated weight coefficient is used as the initial weight coefficient, and the step of determining the fitness value corresponding to the initial weight coefficient is repeatedly performed until the fitness value corresponding to the initial weight coefficient meets the preset fitness condition, and / or the number of repeated executions meets the preset number condition, thereby obtaining the target weight coefficient.

[0217] In an optional embodiment of the present application, determining the fitness value corresponding to the initial weight coefficient includes:

[0218] Processing the kernel function to be processed based on the initial weight coefficient to obtain an initial hybrid kernel function;

[0219] Replacing the kernel function in the support vector machine with the initial hybrid kernel function to obtain an initial machine learning model;

[0220] Training the initial machine learning model using the training signal and the training type of the training signal to obtain an initial rotational Doppler signal classification model;

[0221] Based on the initial rotational Doppler signal classification model, a fitness value corresponding to the initial weight coefficient is determined.

[0222] In an optional embodiment of the present application, the kernel function to be processed includes a Gaussian kernel function and / or a polynomial kernel function.

[0223] In an optional embodiment of the present application, the method includes:

[0224] Collecting signals with preset rotational Doppler effect;

[0225] The rotational Doppler effect signal is input into the target rotational Doppler signal classification model to obtain the type of the rotational Doppler effect signal.

[0226] The communication bus mentioned in the terminal can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus. This communication bus can be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, only one thick line is used in the figure, but this does not mean that there is only one bus or only one type of bus.

[0227] The communication interface is used for communication between the above terminal and other devices.

[0228] The memory may include random access memory (RAM) or non-volatile memory, such as at least one disk storage. Alternatively, the memory may be at least one storage device located away from the processor.

[0229] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, and discrete hardware components.

[0230] like Figure 12 As shown, in another embodiment provided in the present application, a computer-readable storage medium 1201 is also provided, in which instructions are stored. When the computer-readable storage medium 1201 is run on a computer, the computer executes a method for constructing a rotational Doppler signal classification model described in the above embodiment.

[0231] In another embodiment provided by the present application, a computer program product including instructions is also provided. When the computer program product is run on a computer, the computer executes the method for constructing a rotational Doppler signal classification model described in the above embodiment.

[0232] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When software is used for implementation, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from a website, computer, server or data center to another website, computer, server or data center via a wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) method. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrations. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)).

[0233] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply the existence of any such actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or device comprising the element.

[0234] Each embodiment in this specification is described in a related manner. Similar parts between the various embodiments can be referred to in conjunction with each other. Each embodiment focuses on the differences between the other embodiments. In particular, the system embodiment is generally similar to the method embodiment, so the description is relatively simple. For related parts, refer to the description of the method embodiment.

[0235] The above description is only a preferred embodiment of the present application and is not intended to limit the scope of protection of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application are included in the scope of protection of the present application.

Claims

1. A method for constructing a rotational Doppler signal classification model, characterized in that: include: Collecting a training signal of a preset training rotation Doppler effect; The training rotation Doppler effect is a Doppler effect generated when a preset training light beam is irradiated on the surface of a preset training rotating object; the training light beam and the training rotating object are respectively a light beam and a rotating object used to generate the training signal; The training signal is a signal used for model training; Obtaining the relative positional relationship between the propagation axis of the training light beam and the rotation axis of the training rotating object when collecting the training signal, and determining the training type of the training signal based on the relative positional relationship; the training type of the training signal is the type of signal used for model training; Designing a target hybrid kernel function, and using the target hybrid kernel function to replace the kernel function in a preset support vector machine to obtain a target machine learning model; The target machine learning model is trained using the training signal and the training type of the training signal to obtain a target rotation Doppler signal classification model.

2. The method according to claim 1, characterized in that Determining the training type of the training signal according to the relative posture relationship includes: The training type of the training signal is determined according to the lateral offset and / or tilt angle between the propagation axis and the rotation axis.

3. The method according to claim 1, characterized in that The design target hybrid kernel function includes: Obtain at least one kernel function to be processed; Determining the target weight coefficient of the kernel function to be processed by using a preset genetic algorithm; The kernel function to be processed is processed based on the target weight coefficient to obtain the target hybrid kernel function.

4. The method according to claim 3, characterized in that The method of using a preset genetic algorithm to determine the target weight coefficient of the kernel function to be processed includes: Determining at least one initial weight coefficient of the kernel function to be processed, and determining a fitness value corresponding to the initial weight coefficient; Based on the fitness value, determining a weight coefficient to be processed from the initial weight coefficients; Performing crossover and / or mutation processing on the weight coefficient to be processed to generate an updated weight coefficient; The updated weight coefficient is used as the initial weight coefficient, and the step of determining the fitness value corresponding to the initial weight coefficient is repeatedly performed until the fitness value corresponding to the initial weight coefficient meets the preset fitness condition, and / or the number of repeated executions meets the preset number condition, thereby obtaining the target weight coefficient.

5. The method according to claim 4, characterized in that Determining the fitness value corresponding to the initial weight coefficient includes: Processing the kernel function to be processed based on the initial weight coefficient to obtain an initial hybrid kernel function; Replacing the kernel function in the support vector machine with the initial hybrid kernel function to obtain an initial machine learning model; Training the initial machine learning model using the training signal and the training type of the training signal to obtain an initial rotational Doppler signal classification model; Based on the initial rotational Doppler signal classification model, a fitness value corresponding to the initial weight coefficient is determined.

6. The method according to claim 3, characterized in that The kernel function to be processed includes a Gaussian kernel function and / or a polynomial kernel function.

7. The method according to claim 1, characterized in that The method comprises: Collecting signals with preset rotational Doppler effect; The rotational Doppler effect signal is input into the target rotational Doppler signal classification model to obtain the type of the rotational Doppler effect signal.

8. A device for constructing a rotational Doppler signal classification model, characterized in that: include: A training signal acquisition module is used to acquire a training signal for a preset training rotation Doppler effect; The training rotational Doppler effect is a Doppler effect generated when a preset training light beam is irradiated on the surface of a preset training rotating object; the training light beam and the training rotating object are respectively a light beam and a rotating object used to generate the training signal; the training signal is a signal used for model training; a training type determination module, configured to obtain a relative positional relationship between the propagation axis of the training light beam and the rotation axis of the training rotating object when the training signal is collected, and determine a training type of the training signal based on the relative positional relationship; the training type of the training signal is a type of signal used for model training; A function design module is used to design a target hybrid kernel function and use the target hybrid kernel function to replace the kernel function in the preset support vector machine to obtain a target machine learning model; The model acquisition module is used to train the target machine learning model using the training signal and the training type of the training signal to obtain a target rotation Doppler signal classification model.

9. An electronic device, characterized in that: comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus; The memory is used to store computer programs; The processor is configured to implement the method according to any one of claims 1 to 7 when executing a program stored in the memory.

10. One or more computer-readable media having instructions stored thereon, which, when executed by one or more processors, cause the processors to perform the method of any one of claims 1-7.

Images