Low-altitude and ground unmanned device monitoring method and system based on optical fiber sensing

Abstract

The present application relates to the technical field of optical fiber sensing, in particular to a low-altitude and ground unmanned device monitoring method and system based on optical fiber sensing, the method comprising the following steps: deploying an anti-interference fading DAS system in a monitoring area to obtain signal dynamic components and a composite signal in the monitoring area; converting the signal dynamic components into a dynamic feature matrix; using a noise suppression model to suppress noise of the dynamic feature matrix to obtain a clean dynamic feature map; using the clean dynamic feature map to locate a vibration source, and then combining a vibration signal generation model to obtain a vibration signal spectrum map; using a convolutional neural network to construct a vibration source classification and identification model, and then using the vibration signal spectrum map and the vibration source classification and identification model to classify and identify the vibration source. The present application improves signal stability through the anti-interference fading DAS system, combines multi-strategy noise suppression, CycleGAN signal extraction and convolutional neural network classification, and can realize high-precision monitoring and identification of low-altitude and ground unmanned devices.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of optical fiber sensing technology, in particular to a low-altitude and ground unmanned device monitoring method and system based on optical fiber sensing. BACKGROUND

[0002] With the rapid development of low-altitude economy and unmanned device technology, unmanned devices such as low-altitude unmanned aerial vehicles and unmanned vehicles are increasingly widely used in fields such as logistics, security, and agriculture, but they also pose a potential threat to public safety. Therefore, achieving high-precision, all-weather monitoring of low-altitude and ground unmanned devices has become an important requirement for current technological development. Distributed acoustic sensing (DAS) systems can achieve real-time positioning and signal acquisition of vibration events by detecting backscattered Rayleigh light signals in optical fibers, providing a basic technical support for unmanned device monitoring. However, existing monitoring methods based on optical fiber sensing still face technical bottlenecks such as poor signal quality, strong noise interference, and low vibration source classification accuracy in complex environments, making it difficult to meet the high-reliability monitoring requirements.

[0003] In the prior art, the unmanned device monitoring method based on the DAS system usually adopts the following process: obtaining vibration signals through the DAS system, using wavelet transform, empirical mode decomposition, and other algorithms for noise reduction processing of the signals, extracting time domain or frequency domain features, and finally realizing vibration source classification based on a machine learning model. However, such methods have significant defects: first, traditional DAS systems are susceptible to Rayleigh scattering random interference fading, which can easily form a detection blind area in the optical fiber, resulting in loss of vibration signals or positioning deviation; second, noise reduction algorithms such as wavelet transform rely on fixed basis functions and have limited ability to suppress non-stationary noise, and residual noise can interfere with subsequent feature extraction; third, the feature expression dimension of one-dimensional vibration signals is single, making it difficult to fully capture the spatial distribution and pattern differences of vibration sources, leading to confusion of similar vibration sources by the classification model; fourth, existing methods often use global signal processing when classifying and positioning vibration signals, which can cause computational redundancy, and environmental noise and non-target vibration source signals can reduce classification accuracy. SUMMARY

[0004] In view of the defects in the prior art, the present application provides a low-altitude and ground unmanned device monitoring method and system based on optical fiber sensing.

[0005] In order to achieve the above-mentioned purpose, in a first aspect, the present application provides a low-altitude and ground unmanned device monitoring method based on optical fiber sensing, the method comprising the following steps: deploying an anti-interference fading DAS system in a monitoring area; using the anti-interference fading DAS system to obtain signal dynamic components and a composite signal in the monitoring area; converting the signal dynamic components into a dynamic feature matrix; using a noise suppression model to suppress noise of the dynamic feature matrix to obtain a clean dynamic feature map; using the clean dynamic feature map to locate a vibration source, and then combining a vibration signal generation model to obtain a vibration signal spectrum map; using a convolutional neural network to construct a vibration source classification and identification model, and then using the vibration signal spectrum map and the vibration source classification and identification model to classify and identify the vibration source. The present application improves signal stability through the anti-interference fading DAS system, combines multi-strategy noise suppression, CycleGAN signal extraction and convolutional neural network classification, and can realize high-precision monitoring and identification of low-altitude and ground unmanned devices.

[0006] Optionally, the anti-interference fading DAS system is deployed in the monitoring area, comprising the following steps:

[0007] determining a monitoring area;

[0008] planning a buried path of a sensing optical fiber in the monitoring area;

[0009] burying the sensing optical fiber according to the buried path, and installing a matching DAS system device.

[0010] Optionally, the anti-interference fading DAS system is deployed in the monitoring area, comprising the following steps:

[0011] modulating continuous light generated by a laser into quasi-orthogonal chirp pulses and injecting the quasi-orthogonal chirp pulses into a sensing optical fiber, and using an optical detector to obtain beat frequency signals of backscattered Rayleigh light and local light;

[0012] extracting high and low frequency band signals of the quasi-orthogonal chirp pulses through matched filtering and non-matched filtering, and then suppressing interference fading of the beat frequency signals through rotation vector and synthesis to obtain a composite signal;

[0013] performing dynamic detection on the composite signal to obtain signal dynamic components, the dynamic detection being time difference;

[0014] performing digital signal processing on the signal dynamic components to obtain a composite signal containing vibration signals.

[0015] The method modulates laser pulses into quasi-orthogonal chirp signals, separates high and low frequency signals by matched filtering and non-matched filtering, effectively suppresses interference fading, compensates for random phase changes through vector synthesis technology, solves the signal loss problem caused by fading in traditional DAS, significantly improves the integrity of the signal, and improves the accuracy of the composite signal, which is conducive to improving the accuracy of monitoring low-altitude and ground unmanned devices.

[0016] Optionally, the converting the signal dynamic component into a dynamic feature matrix comprises the following steps:

[0017] Determining the row index and column index of the dynamic feature matrix;

[0018] According to the row index and the column index, the amplitude of the signal dynamic component is used to construct the dynamic feature matrix.

[0019] The method converts one-dimensional signals into a three-dimensional feature matrix of time-space-intensity, breaks through the limitation of single feature dimension of traditional one-dimensional signals, provides more rich spatial pattern information for subsequent noise suppression and vibration source classification, and improves the accuracy of monitoring low-altitude and ground unmanned devices.

[0020] Optionally, the noise suppression model is used to suppress noise of the dynamic feature matrix to obtain a clean dynamic feature map, comprising the following steps:

[0021] Constructing a noise suppression model of the dynamic feature matrix;

[0022] Inputting the dynamic feature matrix into the noise suppression model to obtain a clean dynamic feature map.

[0023] Optionally, the noise suppression model performs the following steps when running:

[0024] According to the dynamic feature matrix, a multi-strategy noise suppression scheme is used to generate an original solution set of the noise suppression model;

[0025] According to each solution in the original solution set, a variation individual is generated through a difference vector disturbance, and then a test vector is generated through a crossover operation;

[0026] Selecting a high fitness individual in the variation individual and the test vector as an excellent solution, and taking a set of all excellent solutions as a solution set for the next iteration;

[0027] When the maximum number of iterations is reached or the fitness value converges, output the current optimal solution, and reconstruct the optimal solution into the clean dynamic feature map.

[0028] The method combines multiple noise reduction schemes to generate an initial solution set, and then iteratively optimizes the solution set through mutation, crossover and selection operations to dynamically search for the optimal noise suppression strategy, solve the residual noise problem of traditional fixed basis function noise reduction, significantly improve the signal purity, and adapt to complex noise scenes.

[0029] Optionally, the step of generating the original solution set of the noise suppression model using the multi-strategy noise suppression scheme according to the dynamic feature matrix comprises the following steps:

[0030] The dynamic feature matrix is grayed, and multiple noise suppression schemes are used to reduce noise of the dynamic feature matrix, and then a set of denoising image sets is obtained;

[0031] Each denoising image in the denoising image set is flattened into a one-dimensional vector, and then an original solution set of the noise suppression model is obtained.

[0032] Optionally, the step of locating the vibration source using the clean dynamic feature map and then generating a vibration signal spectrum graph by combining the vibration signal generation model comprises the following steps:

[0033] The clean dynamic feature map is used to locate the vibration source;

[0034] According to the positioning result of the vibration source, the composite signal is intercepted to obtain a composite signal segment;

[0035] A vibration signal generation model is constructed using CycleGAN, and then a vibration signal spectrum graph is obtained according to the composite signal segment.

[0036] The method first locates the vibration source position, and then intercepts the composite signal at the corresponding position to obtain a composite signal segment, which not only can alleviate the calculation redundancy caused by global signal processing when classifying the vibration source, but also can reduce the classification error caused by high-frequency noise; the vibration signal generation model constructed using CycleGAN is suitable for various types of noise and signals, and can solve the problem that traditional methods such as wavelet transform cannot effectively process non-stationary noise in vibration signals.

[0037] Optionally, the step of using a convolutional neural network to construct a vibration source classification and identification model, and then using the vibration signal spectrum graph and the vibration source classification and identification model to classify and identify the vibration source comprises the following steps:

[0038] A vibration signal spectrum data set is constructed;

[0039] The vibration signal spectrum data set is used to train and verify the convolutional neural network to obtain a vibration source classification and identification model;

[0040] The vibration signal spectrum graph is input into the vibration source classification and identification model to realize classification and identification of the vibration source.

[0041] The method utilizes the local mode extraction capability of the CNN for the time-frequency features, fully captures the frequency components and time dynamics of different vibration sources, solves the problem of easy confusion of traditional one-dimensional feature classification, and improves the accuracy of vibration source classification.

[0042] In a second aspect, the present application provides a low-altitude and ground unmanned device monitoring system based on optical fiber sensing, which comprises a data acquisition device, a data output device, a processor and a storage, the storage comprises a computer readable storage medium, the computer readable storage medium stores a computer program, the computer program comprises program instructions, and the program instructions enable the processor to realize the low-altitude and ground unmanned device monitoring method based on optical fiber sensing provided by the present application when executed by the processor. The system can improve the practicability of the method and facilitate the popularization of the method. BRIEF DESCRIPTION OF DRAWINGS

[0043] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiments will be briefly introduced as follows. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0044] Figure 1 The flowchart of the low-altitude and ground unmanned device monitoring method based on optical fiber sensing of the embodiments of the present application is shown.

[0045] Figure 2 The local diagram of the slotting condition in the monitoring area of the embodiments of the present application is shown.

[0046] Figure 3 The local diagram of the trench backfilling condition in the monitoring area of the embodiments of the present application is shown.

[0047] Figure 4 The partial clean dynamic feature map obtained based on the noise suppression model of the embodiments of the present application is shown.

[0048] Figure 5 The dynamic feature map after noise reduction using wavelet transform of the embodiments of the present application is shown.

[0049] Figure 6 The framework diagram of the low-altitude and ground unmanned device monitoring system based on optical fiber sensing of the embodiments of the present application is shown. DETAILED DESCRIPTION

[0050] Specific embodiments of the present invention will now be described in detail. It should be noted that the embodiments described herein are for illustrative purposes only and are not intended to limit the invention. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the invention. However, it will be apparent to those skilled in the art that these specific details are not necessary to practice the invention. In other instances, well-known circuits, software, or methods have not been specifically described to avoid obscuring the invention.

[0051] Throughout this specification, references to "an embodiment," "an embodiment," "an example," or "an example" mean that a particular feature, structure, or characteristic described in connection with that embodiment or example is included in at least one embodiment of the invention. Therefore, the phrases "in an embodiment," "in an embodiment," "an example," or "an example" appearing in various places throughout the specification do not necessarily refer to the same embodiment or example. Furthermore, specific features, structures, or characteristics can be combined in one or more embodiments or examples in any suitable combination and / or sub-combination. Moreover, those skilled in the art will understand that the illustrations provided herein are for illustrative purposes and are not necessarily drawn to scale.

[0052] It should be noted in advance that, in one alternative embodiment, except for independent descriptions, the same symbols or letters appearing in all formulas have the same meaning and value.

[0053] In one optional embodiment, please refer to Figure 1 This invention provides a method for monitoring low-altitude and ground-based unmanned equipment based on fiber optic sensing, the method comprising the following steps:

[0054] S1. Deploy an anti-interference fading DAS system in the monitoring area.

[0055] Specifically, step S1 includes the following steps:

[0056] S11. Determine the monitoring area.

[0057] Specifically, in this embodiment, the selected monitoring area is a temporarily designated rectangular monitoring experimental field, covering an area of ​​approximately 0.5 square kilometers, with a 10cm thick layer of cement laid within the experimental field.

[0058] S12. Plan the burial path of the sensing optical fiber in the monitoring area.

[0059] Specifically, in this embodiment, the activity range of the drones and unmanned vehicles is first determined to delineate the monitoring boundary. Then, key monitoring areas, such as the start and stop locations of the drones or unmanned vehicles within the monitoring area, are identified. For key monitoring areas, the density of fiber optic cable deployment needs to be increased, while the density can be appropriately reduced in ordinary areas to control costs.

[0060] S13, bury the sensing optical fiber according to the burying path, and install the matched DAS system equipment.

[0061] Specifically, in the embodiment, a groove is excavated in the monitoring area according to the planned burying path, the groove depth is 2 cm, the groove width is 2 cm, and the total length of the groove is 1.4 km. After the optical cable is placed in the groove, the groove is backfilled and compacted with cement of the same material as the road surface to establish good coupling with the road surface. The local groove and the local groove after backfilling are shown in Figs. 2 and 3, respectively. Figure 2 Figure 3

[0062] After the burying of the sensing optical fiber is completed, the installation of the DAS equipment, including the narrow-linewidth laser, the optical detector, the coupler, the erbium-doped fiber amplifier, the filter, the circulator, and the data acquisition card, etc., needs to be completed in the ground monitoring center.

[0063] S2, acquiring the signal dynamic component and the composite signal in the monitoring area by using the anti-interference fading DAS system.

[0064] Specifically, the step S2 includes the following steps.

[0065] S21, modulating the continuous light generated by the laser into quasi-orthogonal chirp pulses and injecting the quasi-orthogonal chirp pulses into the sensing optical fiber, and acquiring the beat frequency signal of the backscattered Rayleigh light and the local light by using the optical detector.

[0066] Specifically, in the embodiment, the continuous light generated by the narrow-linewidth laser is divided into first light and second light by the coupler. The first light is modulated into quasi-orthogonal chirp light pulses including positive chirp pulses and negative chirp pulses by the Mach-Zehnder modulator, and then the quasi-orthogonal chirp light pulses are input into the erbium-doped fiber amplifier to improve the optical power, and after the first-order sideband is filtered out by the optical filter, the quasi-orthogonal chirp light pulses enter the sensing optical fiber through the circulator. The second light is used as the local light to beat with the backscattered Rayleigh light output by the circulator and input into the optical detector. Then, the optical detector performs photoelectric conversion on the received optical signal, and after sampling by the data acquisition card, the beat frequency signal is obtained.

[0067] In the embodiment, the chirp rate of the quasi-orthogonal chirp light pulses is 20 MHz / μs, the pulse width is 5 μs, and the sampling rate is 500 MS / s. In other alternative embodiments, the pulse width can be further shortened to improve the multi-target detection capability.

[0068] S22, extracting the high-frequency band signal and the low-frequency band signal of the quasi-orthogonal chirp pulses through matched filtering and non-matched filtering, and then suppressing the interference fading of the beat frequency signal through rotation vector and synthesis to obtain a composite signal.

[0069] ​​Specifically, in the embodiment, when the interference fading of the beat signal is inhibited by rotating the vector and synthesizing, the real-time adjustment of the synthetic phase is calibrated by adopting the closed-loop phase feedback. This step is a prior art means, and thus will not be described in more detail.

[0070] S23, dynamically detecting the synthetic signal to obtain a signal dynamic component, wherein the dynamic detection is time difference.

[0071] Specifically, in the embodiment, the dynamic detection of the synthetic signal can be expressed by using the following relationship:

[0072]

[0073] wherein, is the signal dynamic component, is the synthetic signal, is the delay of the synthetic signal, .

[0074] S24, performing digital signal processing on the signal dynamic component to obtain a composite signal containing the vibration signal.

[0075] Specifically, in the embodiment, the composite signal satisfies the following relationship:

[0076]

[0077] wherein, is the composite signal, and j is an imaginary unit, is the Hilbert transform.

[0078] In the embodiment, the laser pulse is modulated into a quasi-orthogonal chirp signal, the high and low frequency band signals are separated by combining the matched filtering and the unmatched filtering, the interference fading is effectively inhibited , the phase random variation is compensated by using the vector synthesis technology, the signal loss problem caused by the fading in the traditional DAS is solved, the integrity of the signal is significantly improved, and thus the accuracy of the composite signal is improved, which is beneficial to improving the accuracy of monitoring the low-altitude and ground unmanned equipment.

[0079] S3, converting the signal dynamic component into a dynamic feature matrix.

[0080] Specifically, the step S3 includes the following steps.

[0081] S31, determining the row index and the column index of the dynamic feature matrix.

[0082] Specifically, in the embodiment, the spatial position in the optical fiber is taken as the row index of the dynamic feature matrix, and the time is taken as the column index of the dynamic feature matrix.

[0083] S32, constructing the dynamic feature matrix using amplitudes of signal dynamic components according to the row index and the column index.

[0084] Specifically, in the embodiment, the data acquisition card records the accurate emission timestamp each time the laser pulse triggers. According to the reception timestamp of the backscattered Rayleigh light, the position of the signal in the optical fiber is calculated by the formula of the propagation speed of the light pulse in the optical fiber:

[0085]

[0086] wherein z is the position of the signal in the optical fiber, c is the speed of light, is the group refractive index of the sensing optical fiber, , is the reception timestamp.

[0087] Further, by emitting a sequence of laser pulses with a repetition frequency of 500 Hz and correlating the position of the signal in the optical fiber with the signal amplitude, a time-amplitude map covering all spatial positions can be constructed, and the amplitudes at different positions at different times can be obtained.

[0088] The embodiment converts one-dimensional signals into a three-dimensional feature matrix of time-space-intensity, breaks through the limitation of single feature dimension of traditional one-dimensional signals, provides more abundant spatial pattern information for subsequent noise suppression and vibration source classification, and further improves the accuracy of monitoring low-altitude and ground unmanned devices.

[0089] S4, using a noise suppression model to suppress noise of the dynamic feature matrix to obtain a clean dynamic feature map.

[0090] S41, constructing a noise suppression model of the dynamic feature matrix.

[0091] Specifically, in the embodiment, the noise suppression model performs the following steps when running:

[0092] 1. Generating an original solution set of the noise suppression model according to the dynamic feature matrix using a multi-strategy noise suppression scheme.

[0093] Step 1 specifically includes the following steps:

[0094] (1) Grayscale the dynamic feature matrix, and use multiple noise suppression schemes to perform noise reduction processing on the dynamic feature matrix, and then obtain a set of noise reduction images.

[0095] Each value in the dynamic feature matrix is ​​mapped to a grayscale range, thus converting the dynamic feature matrix to grayscale and obtaining a dynamic feature grayscale image. Then, wavelet transform, empirical mode decomposition, and Gaussian filtering are used to denoise the dynamic feature grayscale image, resulting in a set of denoised images. This process is a current technique and will not be described in detail here.

[0096] (2) Flatten each denoised image in the denoised image set into a one-dimensional vector to obtain the original solution set of the noise suppression model.

[0097] First, each denoised image in the denoised image set is flattened into a one-dimensional vector in row order, resulting in a basic solution set. Then, for any two one-dimensional vectors in the basic solution set, their values ​​at the same position are randomly swapped, resulting in a new one-dimensional vector, which is then added to the basic solution set. This process is repeated until the number of one-dimensional vectors in the basic solution set reaches 30. Finally, the final basic solution set is used as the original solution set for the differential evolution algorithm, where each one-dimensional vector is treated as an individual.

[0098] 2. Based on each solution in the original solution set, generate mutated individuals through difference vector perturbation, and then generate experimental vectors through crossover operations.

[0099] First, regarding the first The i-th target individual in generation This requires randomly selecting three different individuals to generate mutation vectors. Meanwhile, considering the problems of traditional differential evolution algorithms, such as insufficient balance between exploration and exploitation in mutation operations, slow convergence speed due to fixed scaling and crossover factors, susceptibility to local optima, and difficulty in achieving a balance between optimization efficiency and accuracy, this embodiment makes the following improvements to the traditional differential evolution algorithm:

[0100] Considering the significant impact of basis vectors on the convergence speed of the differential evolution algorithm, a mutation coefficient is introduced into the mutation operation of the traditional differential evolution algorithm. The coefficient of variation can measure the number of iterations up to the maximum number of iterations. Multiplying the coefficient of variation with the basis vector can improve the effectiveness of the algorithm's local search while maintaining the algorithm's global search capability, thereby improving the algorithm's convergence speed and reducing the possibility of the algorithm getting stuck in local optima.

[0101] The scaling factor and crossover factor of the traditional differential evolution algorithm are improved. In the early stage of iteration, the scaling factor is larger and the crossover factor is smaller, which enhances the global search capability of the algorithm. As the number of iterations increases, the scaling factor gradually decreases and the crossover factor gradually increases, thereby increasing the local search capability of the algorithm while preserving the excellent information in the population, and improving the accuracy and convergence speed of the algorithm.

[0102] In the improved differential evolution algorithm, the scaling factor, crossover factor, and mutation vector satisfy the following relationships respectively:

[0103]

[0104]

[0105]

[0106] in, For the first The mutation vector of the i-th individual in the generation. The maximum number of iterations, For the first The r1th individual in generation, where F is the scaling factor. For the first The r-th individual in generation r, For the first The r-th individual in the generation. When using the original solution set. .

[0107] Then, the improved crossover factor is used to perform a binary crossover between the variant vector and the target individual to generate the experimental vector.

[0108] 3. Select the high-fitness individuals from the mutated individuals and experimental vectors as excellent solutions, and use the set of all excellent solutions as the solution set for the next iteration.

[0109] The fitness function used in this embodiment is:

[0110]

[0111] in, The fitness value is denoted by n, where n is the number of values ​​in the one-dimensional vector. and For the total change regularization term And authenticity items The balance weight, This represents the value at the (j+1)th position in the mutated individual or the experimental vector. This represents the value at the j-th position in the mutated individual or the experimental vector. For variant individuals or experimental vectors, It is the first individual in the original solution set. and Take values ​​of 0.55 and 0.45 respectively.

[0112] 4. When the maximum number of iterations or the fitness value converges, output the current optimal solution and reconstruct the optimal solution into the clean dynamic feature map.

[0113] In this embodiment, the current optimal solution, i.e., the optimal individual, is output when the maximum number of iterations is reached or the fitness value converges, otherwise, step 2 is returned to enter the next iteration, but the solution set used in step 2 of the next iteration is the solution set obtained in step 3 of the last iteration. Wherein, the maximum number of iterations is set to 200, and the fitness value convergence condition is that the last 5 generations satisfy .

[0114] Since each individual in the original solution set in the differential evolution algorithm is obtained by flattening the corresponding denoised image in the denoised image set according to the row, after obtaining the optimal individual, the optimal solution can be reconstructed into a gray image, i.e., a clean dynamic feature map.

[0115] S42, input the dynamic feature matrix into the noise suppression model, and further obtain a clean dynamic feature map.

[0116] This embodiment generates an initial solution set in combination with multiple denoising schemes, and then iteratively optimizes the solution set through mutation, crossover and selection operations, dynamically searches for the optimal noise suppression strategy, solves the residual noise problem of traditional fixed basis function denoising, significantly improves the signal purity, and adapts to complex noise scenes.

[0117] S5, using the clean dynamic feature map to locate the vibration source, and further combining the vibration signal generation model to obtain a vibration signal spectrum.

[0118] Wherein, step S5 specifically includes the following steps:

[0119] S51, using the clean dynamic feature map to locate the vibration source.

[0120] Specifically, in this embodiment, the hovering experiment of the unmanned aerial vehicle is carried out at 890m of the sensing optical fiber, the hovering height is 7m, and part of the clean dynamic feature map obtained by using the noise suppression model is as shown in Figure 4 From Figure 4 , it is not difficult to see that the vibration caused by the hovering of the unmanned aerial vehicle causes the intensity of the back Rayleigh scattering light of the sensing optical fiber at 890m to change periodically, and under the concrete cover layer, the high frequency component is attenuated, so only the main stripe of the fundamental frequency is observed, that is, the bright narrowband straight line at 890m.

[0121] Further, in order to verify the superiority of the present scheme, wavelet transform is used for comparison with the noise suppression model. The denoised dynamic feature map obtained by using the conventional wavelet transform is as shown in Figure 5 It is not difficult to see that Figure 5 there is also a bright straight line at 890m, but compared with Figure 4 , Figure 5There are still a large number of short-range artifact stripes with random directions in other positions, which are typical residual noise patterns caused by the fixed basis function failing to fit the non-stationary noise. In some cases, these short-range artifact stripes can be misjudged as vibration sources, and further cause classification errors of the vibration sources. As can be seen, the noise suppression model provided in this embodiment can better locate the vibration source position, laying a foundation for subsequent vibration source classification.

[0122] S52, according to the positioning result of the vibration source, the composite signal is intercepted to obtain a composite signal segment.

[0123] Specifically, in this embodiment, after the vibration source is located, the signal segment at the corresponding position in the composite signal can be intercepted according to the time stamp to obtain the composite signal segment.

[0124] S53, using CycleGAN to construct a vibration signal generation model, and then obtaining a vibration signal spectrum according to the composite signal segment.

[0125] Specifically, in this embodiment, first, a plurality of unmanned vehicle or unmanned aerial vehicle running experiments are performed in the monitoring area. The noise-containing signal caused by the unmanned aerial vehicle or unmanned vehicle running is collected by the sensing optical fiber, and the vibration signal of the ground directly below the unmanned vehicle or unmanned aerial vehicle is measured by using a laser Doppler vibration instrument to obtain a clean signal. The noise-containing signal and the clean signal need to be time-synchronized. Then, the vibration signal is intercepted in the clean signal to obtain a clean vibration signal, and the noise-containing signal is intercepted according to the time stamp of the clean vibration signal to obtain the corresponding noise-containing vibration signal. When intercepting the clean vibration signal, the clean vibration signal should have the same time span as the composite signal segment on the basis of not damaging the integrity of the vibration signal. Next, the obtained clean vibration signal and noise-containing vibration signal are subjected to short-time Fourier transform to obtain a clean vibration signal spectrum and a noise-containing vibration signal spectrum, and they are uniformly scaled to 256x256 pixels and used to construct a data set for training and verifying CycleGAN (cyclegan). Finally, the obtained data set is divided into a training set and a verification set according to a ratio of 7:3, which is used to complete the training and verification of CycleGAN, so that CycleGAN can generate a corresponding clean vibration signal spectrum according to the input noise-containing vibration signal spectrum. Finally, the generator of CycleGAN is used as a vibration signal generation model.

[0126] In this embodiment, the generator of CycleGAN adopts a U-Net network structure, the discriminator uses a 70x70 PatchGAN, and the loss function of CycleGAN includes a generative adversarial loss and a cycle consistency loss. These are prior art means, and therefore will not be described in detail here.

[0127] After obtaining the vibration signal generation model, the short-time Fourier transform is performed on the composite signal segment to obtain a composite signal spectrum diagram, and the composite signal spectrum diagram is input into the vibration signal generation model to obtain a vibration signal spectrum diagram.

[0128] The embodiment first locates the vibration source position, and then obtains the composite signal segment by intercepting the composite signal at the corresponding position, which not only can alleviate the calculation redundancy caused by the global signal processing in the prior art when classifying the vibration source to a certain extent, but also can reduce the classification errors caused by high-frequency noise; the vibration signal generation model constructed by using the CycleGAN is suitable for various types of noise and signals, and can solve the problem that traditional methods such as wavelet transform cannot effectively process non-stationary noise in the vibration signal.

[0129] S6, using a convolutional neural network to construct a vibration source classification and identification model, and then using the vibration signal spectrum diagram and the vibration source classification and identification model to classify and identify the vibration source.

[0130] Specifically, in the embodiment, the clean vibration signal spectrum diagram obtained in step S53 is labeled to distinguish the source of the vibration signal, and then a vibration signal spectrum dataset is constructed.

[0131] S61, constructing a vibration signal spectrum dataset.

[0132] Specifically, in the embodiment, the clean vibration signal spectrum diagram obtained in step S53 is labeled to distinguish the source of the vibration signal, and then a vibration signal spectrum dataset is constructed.

[0133] S62, using the vibration signal spectrum dataset to train and verify the convolutional neural network to obtain a vibration source classification and identification model.

[0134] Specifically, in the embodiment, the vibration signal dataset is divided into a training set and a verification set according to a ratio of 7:3, which is used to complete the training and verification of the convolutional neural network to obtain the vibration source classification and identification model.

[0135] S63, inputting the vibration signal spectrum diagram into the vibration source classification and identification model to realize classification and identification of the vibration source.

[0136] The embodiment uses the local pattern extraction capability of CNN for time-frequency features to fully capture the frequency components and time dynamics of different vibration sources, solves the problem that traditional one-dimensional feature classification is easy to confuse, and improves the accuracy of vibration source classification.

[0137] It should be noted that in some cases, the actions described in the specification can be performed in different orders and still achieve the desired results, and in the embodiment, the order of the steps given is only to make the embodiment look clearer and more understandable, and is not a limitation.

[0138] In an alternative embodiment, see Figure 6, in order to improve the practicability of the method, facilitate the popularization of the method, the application also provides a low altitude and ground unmanned device monitoring system based on optical fiber sensing, the low altitude and ground unmanned device monitoring system based on optical fiber sensing comprises: data acquisition equipment 1, data output equipment 2, processor 3 and storage 4, the storage 4 includes a computer readable storage medium, the computer readable storage medium stores computer programs, the computer programs include program instructions, the program instructions are executed by the processor 3, and the processor 3 realizes the contents of steps S1 to S6.

[0139] In summary, the present application has at least the following beneficial effects:

[0140] 1, the method is used for starting and stopping the unmanned aerial vehicle or unmanned vehicle through the optical fiber, so that the unmanned aerial vehicle or unmanned vehicle still has sensing ability in the scene of electromagnetic interference.

[0141] 2, the method is used for modulating laser pulse into quasi-orthogonal chirp signal, combining matched filtering and non-matched filtering to separate high and low frequency signals, effectively suppressing Interference fading; through the vector synthesis technology, the signal loss problem caused by the traditional DAS due to fading is solved, the integrity of the signal is improved, and the accuracy of the composite signal is improved, which is beneficial to improve the accuracy of low altitude and ground unmanned device monitoring.

[0142] 3, the method is used for converting one-dimensional signal into time-space-intensity three-dimensional feature matrix, breaking through the limitation of single feature dimension of traditional one-dimensional signal, and combining various noise reduction schemes and improving differential evolution algorithm to realize accurate positioning of vibration source.

[0143] 4, the method is used for positioning the vibration source position, then intercepting the corresponding position of the composite signal to obtain the composite signal segment, which can not only alleviate the calculation redundancy caused by global signal processing in the prior art when classifying vibration sources, but also reduce the classification errors caused by high frequency noise

[0144] 5, the method uses CycleGAN to construct a vibration signal generation model, which is suitable for various types of noise and signals, and can solve the problem that traditional methods such as wavelet transform cannot effectively process non-stationary noise in vibration signals.

[0145] 6, the method uses CNN to extract the local mode of time-frequency features, fully captures the frequency components and time dynamics of different vibration sources, solves the problem that traditional one-dimensional feature classification is easy to confuse, and improves the accuracy of vibration source classification.

[0146] 7, a system suitable for the method is provided, which can improve the practicability of the method and facilitate the popularization of the method.

[0147] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit the present application. Although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions recorded in the above embodiments can be modified, or some or all of the technical features can be replaced by equivalent replacements. These modifications or replacements do not change the essence of the corresponding technical solutions, which should be covered in the scope of the claims and the specification of the present application.

Claims

1. A method for monitoring low altitude and ground unmanned devices based on fiber optic sensing, characterized by, The method comprises the following steps: Laying an anti-interference fading DAS system in a monitoring area; Modulating continuous light generated by a laser into quasi-orthogonal chirp pulses and injecting the quasi-orthogonal chirp pulses into a sensing optical fiber, and using an optical detector to obtain beat frequency signals of back Rayleigh scattering light and local light; Extracting high and low frequency band signals of the quasi-orthogonal chirp pulses through matched filtering and non-matched filtering, and then suppressing interference fading of the beat frequency signals through rotation vector and synthesis to obtain a synthesis signal; Performing dynamic detection on the synthesis signal to obtain a signal dynamic component, the dynamic detection being time difference; Performing digital signal processing on the signal dynamic component to obtain a composite signal containing a vibration signal; Converting the signal dynamic component into a dynamic feature matrix; Using a noise suppression model to suppress noise of the dynamic feature matrix to obtain a clean dynamic feature map, the noise suppression model performing the following steps when running: According to the dynamic feature matrix, generating an original solution set of the noise suppression model using a multi-strategy noise suppression scheme; According to each solution in the original solution set, generating a mutated individual through a difference vector disturbance, and then generating a test vector through a crossover operation; Selecting a high fitness individual from the mutated individual and the test vector as an excellent solution, and taking a set of all excellent solutions as a solution set for the next iteration; Outputting a current optimal solution when a maximum number of iterations is reached or a fitness value converges, and reconstructing the optimal solution into the clean dynamic feature map; Using the clean dynamic feature map to locate a vibration source, and then combining a vibration signal generation model to obtain a vibration signal spectrum map; Using a convolutional neural network to construct a vibration source classification and identification model, and then using the vibration signal spectrum map and the vibration source classification and identification model to classify and identify the vibration source. 2.The low altitude and ground unmanned device monitoring method based on fiber sensing according to claim 1, wherein, The method of laying an anti-interference fading DAS system in a monitoring area comprises the following steps: Determining a monitoring area; Planning a burying path of a sensing optical fiber in the monitoring area; Burying the sensing optical fiber according to the burying path, and installing a matching DAS system device. 3.The low altitude and ground unmanned device monitoring method based on fiber sensing according to claim 1, wherein, The method of converting the signal dynamic component into a dynamic feature matrix comprises the following steps: Determining a row index and a column index of the dynamic feature matrix; According to the row index and the column index, constructing the dynamic feature matrix using amplitudes of the signal dynamic component. 4.The low altitude and ground unmanned device monitoring method based on fiber sensing according to claim 1, wherein, The method of using a noise suppression model to suppress noise of the dynamic feature matrix to obtain a clean dynamic feature map comprises the following steps: Constructing a noise suppression model of the dynamic feature matrix; Inputting the dynamic feature matrix into the noise suppression model to obtain the clean dynamic feature map. 5.The low altitude and ground unmanned device monitoring method based on fiber sensing according to claim 1, wherein, The method of generating an original solution set of the noise suppression model using a multi-strategy noise suppression scheme according to the dynamic feature matrix comprises the following steps: Graying the dynamic feature matrix, and performing noise reduction processing on the dynamic feature matrix using a plurality of noise suppression schemes to obtain a set of noise reduction images; Flattening each noise reduction image in the set of noise reduction images into a one-dimensional vector to obtain the original solution set of the noise suppression model. 6.The low altitude and ground unmanned device monitoring method based on fiber sensing according to claim 1, wherein, The method of using the clean dynamic feature map to locate a vibration source, and then combining a vibration signal generation model to obtain a vibration signal spectrum map comprises the following steps: The clean dynamic characteristic map is used to locate the vibration source; According to the positioning result of the vibration source, the composite signal is intercepted to obtain a composite signal segment; A vibration signal generation model is constructed using CycleGAN, and then a vibration signal spectrum is obtained according to the composite signal segment. 7.The low altitude and ground unmanned device monitoring method based on fiber sensing according to claim 1, wherein, The vibration source classification and identification model is constructed using the convolutional neural network, and then the vibration source is classified and identified using the vibration signal spectrum and the vibration source classification and identification model, including the following steps: A vibration signal spectrum data set is constructed; The vibration signal spectrum data set is used to train and verify the convolutional neural network to obtain a vibration source classification and identification model; The vibration signal spectrum is input into the vibration source classification and identification model to realize classification and identification of the vibration source.

8. A low altitude and ground unmanned device monitoring system based on fiber optic sensing, characterized by, The low-altitude and ground unmanned device monitoring system based on optical fiber sensing includes a data acquisition device, a data output device, a processor and a storage, the storage includes a computer readable storage medium, the computer readable storage medium stores a computer program, the computer program includes program instructions, the program instructions are executed by the processor to realize the method for monitoring the low-altitude and ground unmanned device based on optical fiber sensing as claimed in any one of claims 1-7.

Images