Target rotating speed extraction method and system based on neural network

Abstract

The invention relates to the technical field of neural network training, and particularly discloses a target rotating speed extraction method and system based on a neural network, and the method comprises the steps: S1, building a rotating Doppler detection simulation program according to an actual detection condition, and taking a set target rotating speed as a label to generate a training data set; the training data set at least comprises a sampling rate, a detection beam topological charge and a time domain signal; s2, building a neural network framework based on a multilayer feedforward principle, and training the neural network framework according to the training data set to obtain a trained network model; the neural network framework comprises m layers of neural networks, and each layer of neural network comprises n neurons; m and n are positive integers; s3, preparing a vortex light beam and irradiating the vortex light beam on the surface of the rotating target; s4, detecting the rotating target to obtain acquisition data with the same specification as the training data; and S5, inputting the acquired data into the network model for prediction calculation to obtain a target rotating speed.

Description

Technical Field

[0001] The present invention relates to the technical field of neural network training, and in particular to a method and system for extracting a target rotation speed based on a neural network. Background Art

[0002] Rotational motion is ubiquitous in nature, from cosmic bodies to electrons outside the nucleus. The precise measurement of target rotational motion parameters has always been a research hotspot. For cooperative targets, rotational speed measurement can be performed by marking and visual measurement; however, for non-cooperative targets, such as unstable debris in space and mid-course missiles, rotational speed measurement is always difficult to solve.

[0003] Momentum is a term in physics, which means the product of the mass and velocity of an object, and refers to the effect of a moving object. Momentum is a vector whose direction is the same as the direction of velocity. Since the wave-particle duality of light was established, people have discovered that photons also carry momentum (mc), but this momentum generally refers to linear momentum. In 1992, the existence of orbital angular momentum (OAM) in electromagnetic waves was confirmed for the first time. This discovery shows that in addition to momentum along the propagation direction of the light beam, photons can also carry momentum rotating around the propagation axis of the light beam. The rotational Doppler effect (RDE) is a new type of physical effect discovered when people studied orbital angular momentum.

[0004] The rotational Doppler effect is essentially still a Doppler effect, and this effect is also caused by the relative motion between the wave source and the object. The only difference is that the classical linear Doppler effect is caused by the movement of the object along the propagation direction of the wave source, while the rotational Doppler effect is caused by the relative rotational motion of the object in the section perpendicular to the propagation direction of the wave source. The early discovery of the classical Doppler effect came from people's observation of moving sound sources. Later, it was discovered that the Doppler effect exists in various types of beams. Especially since the invention of lasers in the 1960s, the Doppler effect of light waves has gradually been taken seriously, and various types of laser Doppler meters have been developed. After a beam of ordinary laser irradiates a rotating target, the frequency shift Δf produced by the frequency of the beam received by the rotating target compared to the wave source can be expressed as,

[0005] Δf=f 0 vcosθ / c (1)

[0006] Among them, f 0 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.

[0007] Obviously, when the relative motion angle between the object's motion direction and the wave source's propagation direction is 90°, the linear Doppler shift generated is zero. When a classical plane wave is vertically irradiated on the surface of a rotating object, since each tiny scattering point is perpendicular to the plane wave's propagation direction, the generated Doppler shift is zero, so the classical linear Doppler effect is difficult to measure the object's rotation speed. The rotational Doppler effect is different. When the wave source is vertically coaxially irradiated on a rotating target, the object's rotation vector is exactly parallel to the orbital angular momentum vector. Therefore, there is a natural coupling between the orbital angular momentum of the vortex beam and the rotating object.

[0008] Since the vortex light carries the spiral phase factor The direction of the beam energy flow is no longer along the beam propagation direction, but has an angle with the propagation direction. The size of this angle is β = lλ / 2πr. Since the movement direction of each tiny scattering point in the target rotation plane is exactly perpendicular to the beam propagation direction, the angle between the energy flow direction of the final beam and the direction of the scattering point movement speed has the following relationship:

[0009] α=π / 2-β (2)

[0010] Substituting equation (2) into equation (1) and using the approximation that sinθ≈θ when θ is very small, we can obtain the expression for the rotational Doppler shift:

[0011] Δf=lΩ / 2π (3)

[0012] In actual measurement, there are many ways to measure the rotational Doppler frequency shift Δf. Since the beam frequency is as high as 1014Hz, it is difficult to measure directly. Therefore, the common frequency shift measurement method is to use the beat frequency detection method of the beam to couple the signal light and the reference light to achieve the detection of the frequency shift. According to the actual frequency shift measurement value Δf, combined with the beam topological charge information l, the rotation speed of the rotating target can be measured.

[0013] The above is the basic principle of measuring the target speed based on the rotational Doppler effect. It can be seen that the time domain signal obtained by detection contains the target speed information, which can be extracted through spectrum conversion, but the operation steps are relatively cumbersome. There is an urgent need for a method to achieve different spectrum conversions to directly obtain the target speed from the time domain signal, which can greatly improve the efficiency and automation of target speed signal extraction. Summary of the invention

[0014] In view of the above problems, the purpose of the present invention is to provide a target speed extraction method and system based on a neural network, which can realize efficient and high-precision measurement of the speed of space targets such as engine rotors, drone propellers, and unstable rotating debris in space.

[0015] The present invention provides a target speed extraction method based on a neural network, comprising:

[0016] Step S1, establishing a rotational Doppler detection simulation program according to actual detection conditions, and generating a training data set by setting a target rotation speed as a label; the training data set at least includes: a sampling rate, a detection beam topological charge number, and a time domain signal;

[0017] Step S2, building a neural network framework based on the multi-layer feedforward principle, training the neural network framework according to the training data set to obtain a trained network model; the neural network framework includes m layers of neural networks, each layer of neural network includes n neurons; m, n are positive integers;

[0018] Step S3, preparing a vortex beam and irradiating it onto a rotating target surface;

[0019] Step S4, detecting the rotating target to obtain collected data with the same specifications as the training data;

[0020] Step S5, inputting the collected data into the network model for prediction calculation to obtain the target speed.

[0021] In a possible implementation, establishing a rotational Doppler detection simulation program according to actual detection conditions includes:

[0022] Calculate the surface amplitude distribution of the object transmitted to the surface of the object according to the transfer matrix method;

[0023] Multiplying the vortex beam amplitude distribution by the object surface amplitude to obtain a scattered light field;

[0024] The scattered light field is collected by a lens, and the intensity of the scattered light is calculated by a convergence algorithm;

[0025] The target rotation speed and the set sampling rate are associated to obtain a time domain signal according to the intensity of the scattered light at a certain moment sampled at an arbitrary sampling rate.

[0026] In a possible implementation, S3 includes:

[0027] The vortex beam is prepared by using a spatial light modulator method, a vortex slide method or an array method, and the vortex beam is irradiated onto a rotating target surface.

[0028] In a possible implementation, the S4 includes:

[0029] The echo signal on the surface of the rotating target is collected by using a photoelectric detector and a data acquisition card to obtain collected data with the same specifications as the training data.

[0030] In a possible implementation, the S5 includes:

[0031] The target speed is calculated according to the following formula:

[0032]

[0033] Among them, (x 1 ,x 2 ,...x i ) represents the time domain signal of the external input, (w 1 ,w 2 ,...w i ) represents the weight coefficient of external input, (v 1 ,v 2 ,...v i ) represents the output weight coefficient, β and λ represent the bias, ω represents the target speed, (q 1 ,q 2 ,...q i ) represents the parameter value of each neuron in the hidden layer.

[0034] The present invention also provides a target speed extraction system based on a neural network, referring to any of the above target speed extraction methods, including:

[0035] A data set generation module is used to establish a rotational Doppler detection simulation program according to actual detection conditions, and to generate a training data set by setting a target rotation speed as a label; the training data set at least includes: a sampling rate, a detection beam topological charge number, and a time domain signal;

[0036] Building a network model module, which is used to build a neural network framework based on the multi-layer feedforward principle, and train the neural network framework according to the training data set to obtain a trained network model; the neural network framework includes m layers of neural networks, each layer of neural network includes n neurons; m and n are positive integers;

[0037] A vortex beam preparation module is used to prepare a vortex beam and irradiate it on a rotating target surface;

[0038] A detection module, used to detect the rotating target and obtain collected data with the same specifications as the training data;

[0039] The prediction module is used to input the collected data into the network model for prediction calculation to obtain the target speed.

[0040] The target speed extraction method and system based on neural network provided by the present invention firstly trains the neural network model using the vortex light speed detection simulation data set to obtain a trained network model; secondly, detects the rotating target using the vortex light detection device, and uses the photoelectric detector and the data acquisition card to collect the signal; finally, the sampled time domain signal is input into the trained network model for prediction calculation, thereby obtaining the target speed. The method is easy to operate and can achieve efficient and high-precision measurement of the spatial target speed without complex spectrum conversion. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 A first flow chart of a target speed extraction method provided by an embodiment of the present invention;

[0042] Figure 2 A second flow chart of the target speed extraction method provided by an embodiment of the present invention;

[0043] Figure 3 A schematic diagram of the structure of a multi-layer feedforward neural network provided by an embodiment of the present invention;

[0044] FIG4( a ) is a schematic diagram of the light field intensity distribution of the detection beam provided in an embodiment of the present invention;

[0045] FIG4( b ) is a schematic diagram of the intensity distribution of scattered light irradiated on the surface of an object provided by an embodiment of the present invention;

[0046] FIG4( c ) is a detection schematic diagram provided by an embodiment of the present invention;

[0047] FIG5( a) is a schematic diagram of a time domain signal of a rotational Doppler detection simulation provided by an embodiment of the present invention;

[0048] FIG5(b) is a schematic diagram of a frequency domain signal after Fourier transform of FIG5(a) provided by an embodiment of the present invention;

[0049] Figure 6 A schematic diagram of prediction results provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0050] The following detailed description of the embodiments of the present invention is further described in detail in conjunction with the accompanying drawings and examples. The detailed description of the following embodiments and the accompanying drawings are used to exemplarily illustrate the principles of the present invention, but cannot be used to limit the scope of the present invention, that is, the present invention is not limited to the preferred embodiments described, and the scope of the present invention is defined by the claims.

[0051] In the description of the present invention, it should be noted that, unless otherwise specified, “plurality” means two or more than two; the terms “first”, “second”, etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance; for ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to the specific circumstances.

[0052] The accurate measurement of the rotation speed of space targets is of great significance for improving the target acquisition capability in space control and the ability to identify true and false targets in air defense and anti-missile. Studies have shown that vortex beams have the potential to detect target rotation speed, but due to the complexity and variability of rotation speed signals, high-precision and automated signal extraction has become a problem.

[0053] Figure 1 A first flow chart of a target speed extraction method provided by an embodiment of the present invention is shown. Figure 2 A second flow chart of the target speed extraction method provided by an embodiment of the present invention, combined with Figure 1 and Figure 2 The present invention provides a target speed extraction method based on a neural network, comprising:

[0054] Step S1, establishing a rotational Doppler detection simulation program according to actual detection conditions, and generating a training data set by setting a target rotation speed as a label;

[0055] The training data set includes at least: sampling rate, detection beam topological charge and time domain signal.

[0056] The target speed is set as the label of each set of training data, which is also the object to be predicted in the end. The time domain signal actually detected has the same data format and dimension as the training data set, thus ensuring the accuracy of the training model prediction.

[0057] In one possible implementation, a rotational Doppler detection simulation program is established according to actual detection conditions, including: calculating the surface amplitude distribution of the object transmitted to the surface of the object according to the transmission matrix method; multiplying the vortex light beam amplitude distribution with the object surface amplitude to obtain the scattered light field; collecting the scattered light field through a lens, and calculating the intensity of the scattered light through a convergence algorithm; associating the target rotation speed and the set sampling rate to obtain a time domain signal according to the intensity of the scattered light at a certain moment sampled at an arbitrary sampling rate.

[0058] Specifically, in order to obtain a large number of accurately labeled training data sets, the present invention has built a full-process simulation detection system based on superposition state vortex beams. Referring to the radiation protection material coated on the surface of the satellite, the rotating target in the system is set as a flat circular plate-shaped object, and its surface is a tin foil reflective material. The numerical simulation system can be divided into five steps, among which the superposition state vortex beam can be directly calculated and generated according to the standard Laguerre-Gaussian beam expression, and when lateral displacement or tilt and other conditions are required, the corresponding offset can be directly added to the expression parameters.

[0059] The first step is that before interacting with the rotating object surface, the detection beam will be transmitted for a distance in free space as in actual detection. The light field distribution transmitted to the object surface can be calculated based on the ABCD transfer matrix method.

[0060] The second step is to simulate the rotating target. The amplitude distribution of the rotating scattering surface is considered in the rotating Doppler detection simulation program, but its phase distribution is not considered. In fact, the photodetector can only directly reflect the amplitude of the scattered light. There are two ways to simulate the scattering characteristics of the target surface. One is to generate a random scattered plate field by digital means, which can be used to simulate the surface scattering characteristics of general targets. The other is to record the scattering characteristics of the object surface by flash photography in a dark room. Because vortex light detection is also an active detection method, the scattering characteristics of the target surface can be simulated by taking pictures with a camera to record the intensity information of the return light on the object surface.

[0061] The third step is to match the target speed with the system set sampling rate. Since the target rotation and data sampling are two continuous dynamic processes, in order to accurately simulate the actual measurement process, the rotation rate must match the sampling rate. Since the target rotation and data sampling are two continuous dynamic processes, in order to accurately simulate the actual measurement process, the rotation rate must match the sampling rate. Set the system sampling rate to Fs and the target speed frequency to ω, then the angle rotated between each two frames should be θ = 360ω / Fs.

[0062] The fourth step is to collect scattered light. After the objects interact, the echo light beam needs to be collected through a lens. Similar to the transmission of the light beam in free space, the ABCD matrix is ​​also used in the light beam collection process to simulate the process of being converged by the lens.

[0063] The fifth step is the spectrum conversion analysis of the sampling signal. Similar to the actual detection, the direct sampling signal is the time domain amplitude information of the scattered light. In addition to the time domain amplitude information, since the frequency domain signal characteristics are also related to the parameters such as the topological charge of the detection beam and the sampling rate, the corresponding parameters should be added to the sampling data set.

[0064] In one example, when the beam is aligned with the center of the rotation axis, the simulation flow chart of using the superposition state vortex beam with a topological charge of ±15 for detection is as follows: Figure 4(a) to Figure 4(c) As shown. Figure 4(a) is a schematic diagram of the light field intensity distribution of the detection beam provided in an embodiment of the present invention, Figure 4(b) is a schematic diagram of the intensity distribution of the scattered light irradiated on the surface of an object provided in an embodiment of the present invention, and Figure 4(c) is a detection schematic diagram provided in an embodiment of the present invention. When calculating the intensity of the scattered light, the scattered light field is obtained by directly multiplying the vortex beam amplitude distribution with the object surface amplitude, and then the scattered light field is converged by a lens convergence algorithm, which completely simulates the scattered light collection process in the experiment; finally, according to the light intensity calculation formula, the intensity of the scattered light can be obtained. By associating the rotational speed and time information, it is possible to sample the intensity of the scattered light at a certain moment at any sampling rate. After setting a sampling time, the detection time domain signal can be obtained.

[0065] Figure 5(a) shows the time domain signal obtained by sampling at a sampling rate of 10000Hz during the experiment. It can be seen that the light intensity fluctuates around 0 after normalization. The frequency domain signal can be obtained after Fourier transform of the time domain signal, as shown in Figure 5(b). It can be seen that there is a clear peak signal in the frequency domain, and its frequency value is about 1780Hz. By substituting it into the Doppler effect calculation formula, the target speed can be calculated to be 59.33rps, which is basically consistent with the experimental setting value of 59.5rps, proving the effectiveness of the simulation program.

[0066] Step S2, building a neural network framework based on the multi-layer feedforward principle, and training the neural network framework according to the training data set to obtain a trained network model;

[0067] The neural network framework includes m layers of neural networks, each layer of neural network includes n neurons; m and n are positive integers. The number of layers of the neural network framework and the number of neurons in each layer can actually be set according to the detection data to achieve the best network prediction results.

[0068] A multi-layer feedforward neural network is a neural network trained according to the error propagation algorithm, commonly known as the BP neural network, which is the most widely used neural network in machine learning. This neural network adds several layers of neurons between the input layer and the output layer. These neurons become hidden units. They have no direct connection with the outside world, but their state changes can affect the relationship between the input and output layers. Each layer can have several nodes, and the connection status of the nodes between layers is reflected by weights, such as Figure 2 shown.

[0069] In a multi-layer feedforward neural network, after the data is input from the input end, it is directed along the network, multiplied by the corresponding weights and then added. The result is then used as input to calculate in the activation function, and the calculated result is passed as input to the next node, and the calculation is continued until the final result is obtained. Through the perceptrons of each layer, the output is calculated layer by layer. This process is forward propagation. Next, the output result is compared with the expected result, and the error generated by the comparison is back-propagated through the network. This is essentially a negative feedback process. Through multiple iterations, the weights between the nodes on the network are continuously adjusted and updated. The weight adjustment uses the gradient descent method to finally obtain the optimal extraction model. This process is called error back propagation. Therefore, the training purpose of this type of neural network is to find the input data x and the true result y.

[0070] Find an optimal set of neural network parameters so that the network calculates the estimated value It can best match the true value y.

[0071] In the present invention, the input layer parameters are mainly time domain signals, but the characteristics of the time domain signals are closely related to the topological charge of the detection beam and the sampling rate, so the above two parameters are also synchronously included in the training data set. The output value of the neural network is the target speed. According to the above neural network structure, the following target speed is obtained:

[0072]

[0073] Among them, (x 1 ,x 2 ,...x i ) represents the time domain signal of the external input, (w 1 ,w 2 ,...w i ) represents the weight coefficient of external input, (v 1 ,v 2 ,...v i ) represents the output weight coefficient, β and λ represent the bias, ω represents the target speed, (q 1 ,q 2 ,...q i ) represents the parameter value of each neuron in the hidden layer.

[0074] According to the basic principle of multi-layer feedforward network, the training data set generated in the simulation system is delivered to the extraction model for training, and the change of the pre-set parameters is calculated from back to front, and then the parameter values ​​are updated, and the output is recalculated until the input result is consistent with the expected result (the error is less than a certain set value σ). When repeated training still cannot achieve the expected result, the prediction results can be observed by adjusting the number of neural network layers and the number of neurons in each layer to ultimately achieve the training purpose.

[0075] Step S3, preparing a vortex beam and irradiating it onto a rotating target surface;

[0076] In a possible implementation, a vortex beam is prepared by using a spatial light modulator method, a vortex slide method, or an array method, and the vortex beam is irradiated onto a rotating target surface.

[0077] Step S4, detecting the rotating target to obtain collected data with the same specifications as the training data;

[0078] In a possible implementation, a photoelectric detector and a data acquisition card are used to collect echo signals from the surface of a rotating target to obtain collected data with the same specifications as the training data.

[0079] Step S5, inputting the collected data into the network model for prediction calculation to obtain the target speed.

[0080] The present invention utilizes a trained neural extraction model to directly obtain an accurate target rotation speed value without the need to extract signals through complex spectrum conversion and then calculate the target rotation speed.

[0081] As attached Figure 6 As shown, by the method of the present invention, when the detection light topological charge number is ±20 and the sampling rate is 10000 Hz, the comparison result of the network predicted value and the true value of the detection rotation speed obtained has an average prediction accuracy of 93.7%.

[0082] The present invention has the following advantages:

[0083] (1) This method is novel in technology, simple in operation, easy to implement, and has strong adaptability under practical conditions.

[0084] (2) This method has high reliability in detecting the rotation speed of rotating targets. According to the principle of the method, it can be seen that this method is a training data policy system built based on actual detection conditions. The resulting extraction model is highly targeted and has high stability and detection accuracy under specific conditions.

[0085] (3) This device is still applicable to extreme conditions such as high-speed rotation. Using light waves as the detection medium, it has a long transmission distance, fast measurement speed, is not affected by factors such as temperature and environment, and responds quickly, which has broad application prospects.

[0086] The present invention also provides a target speed extraction system based on a neural network, referring to any of the above target speed extraction methods, including:

[0087] A data set generation module is used to establish a rotational Doppler detection simulation program according to actual detection conditions, and to generate a training data set by setting a target rotation speed as a label; the training data set at least includes: sampling rate, detection beam topological charge number and time domain signal;

[0088] Build a network model module, which is used to build a neural network framework based on the multi-layer feedforward principle, train the neural network framework according to the training data set, and obtain a trained network model; the neural network framework contains m layers of neural networks, and each layer of neural network includes n neurons; m and n are positive integers;

[0089] A vortex beam preparation module is used to prepare a vortex beam and irradiate it on a rotating target surface;

[0090] A detection module is used to detect the rotating target and obtain the collected data with the same specifications as the training data;

[0091] The prediction module is used to input the collected data into the network model for prediction calculation to obtain the target speed.

[0092] The target speed extraction method and system based on neural network provided by the present invention firstly trains the neural network model using the vortex light speed detection simulation data set to obtain a trained network model; secondly, the rotating target is detected using the vortex light detection device, and the signal is collected using the photoelectric detector and the data acquisition card; finally, the sampled time domain signal is input into the trained network model for prediction calculation, so as to obtain the target speed. For targets such as engine rotors, drone propellers, unstable rotating debris in space, etc., it is difficult to obtain the speed of such targets concisely, efficiently and accurately using existing detection methods; the present invention directly inputs the collected time domain signal into the extraction model through the introduction of machine learning algorithms to obtain the target speed information; the method is simple to operate, more efficient and convenient than the traditional speed extraction method of spectrum conversion calculation, has strong adaptability, and has broad application prospects for speed measurement occasions with high automation requirements.

[0093] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A target speed extraction method based on neural network, characterized in that: include: Step S1, establishing a rotational Doppler detection simulation program according to actual detection conditions, and generating a training data set by setting a target rotation speed as a label; The training data set includes at least: sampling rate, detection beam topological charge and time domain signal; Step S2, building a neural network framework based on the multi-layer feedforward principle, training the neural network framework according to the training data set to obtain a trained network model; the neural network framework includes m layers of neural networks, each layer of neural network includes n neurons; m, n are positive integers; Step S3, preparing a vortex beam and irradiating it onto a rotating target surface; Step S4, detecting the rotating target to obtain collected data with the same specifications as the training data; Step S5, inputting the collected data into the network model for prediction calculation to obtain the target speed.

2. The target speed extraction method according to claim 1, characterized in that: The rotation Doppler detection simulation program established according to the actual detection conditions comprises: Calculate the surface amplitude distribution of the object transmitted to the surface of the object according to the transfer matrix method; Multiplying the vortex beam amplitude distribution by the object surface amplitude to obtain a scattered light field; The scattered light field is collected by a lens, and the intensity of the scattered light is calculated by a convergence algorithm; The target rotation speed and the set sampling rate are associated to obtain a time domain signal according to the intensity of the scattered light at a certain moment sampled at an arbitrary sampling rate.

3. The target speed extraction method according to claim 1, characterized in that: The S3 includes: The vortex beam is prepared by using a spatial light modulator method, a vortex slide method or an array method, and the vortex beam is irradiated onto a rotating target surface.

4. The target speed extraction method according to claim 1, characterized in that: The S4 includes: The echo signal on the surface of the rotating target is collected by using a photoelectric detector and a data acquisition card to obtain collected data with the same specifications as the training data.

5. The target speed extraction method according to claim 1, characterized in that: The S5 includes: The target speed is calculated according to the following formula: Among them, (x1,x2,...x i ) represents the external input time domain signal, (w1,w2,...w i ) represents the weight coefficient of external input, (v1,v2,...v i ) represents the output weight coefficient, β and λ represent the bias, ω represents the target speed, (q1,q2,...q i ) represents the parameter value of each neuron in the hidden layer.

6. A target speed extraction system based on a neural network, referring to the target speed extraction method according to any one of claims 1 to 5, characterized in that: include: A data set generation module is used to establish a rotational Doppler detection simulation program according to actual detection conditions and to generate a training data set by setting a target rotation speed as a label; The training data set includes at least: sampling rate, detection beam topological charge and time domain signal; Building a network model module, which is used to build a neural network framework based on the multi-layer feedforward principle, and train the neural network framework according to the training data set to obtain a trained network model; the neural network framework includes m layers of neural networks, each layer of neural network includes n neurons; m and n are positive integers; A vortex beam preparation module is used to prepare a vortex beam and irradiate it on a rotating target surface; A detection module, used to detect the rotating target and obtain collected data with the same specifications as the training data; The prediction module is used to input the collected data into the network model for prediction calculation to obtain the target speed.

Images