Distributed optical fiber sensing data processing and monitoring method based on conditional GAN network

Abstract

The invention discloses a distributed optical fiber sensing data processing and monitoring method based on a conditional GAN network. The data processing method comprises the following steps: acquiring original DAS data of a target traffic area within a first preset time; the original DAS data is a vibration signal of the object obtained by differentiating the intensity of reverse Rayleigh scattering light before and after the object passes through the optical fiber laying area; performing different data processing on the original DAS data in the first preset time to obtain a plurality of data images; inputting the data images into the trained generator to obtain a mask image; the mask image represents the motion trail of each object of each category in the target traffic area within the first preset time; the trained generator is obtained by training the conditional GAN network by adopting a plurality of groups of training images and corresponding real mask images, and each group of training images are a plurality of data images. According to the invention, the denoising process can be simplified, the signal-to-noise ratio of signals can be improved, and traffic monitoring can be carried out accurately and effectively.

Description

Technical Field

[0001] The present invention belongs to the technical field of distributed optical fiber sensing, and in particular relates to a distributed optical fiber sensing data processing and monitoring method based on a conditional GAN ​​network. Background Art

[0002] Traffic monitoring technology based on distributed acoustic sensing (DAS) has attracted more and more attention. Compared with other traditional traffic monitoring technologies, it has the advantages of obtaining continuous distributed information, long detection distance, high spatial resolution, low deployment cost, high anti-interference ability, and privacy protection. It can meet the needs of full-time and space traffic monitoring of smart highways. Smart transportation and smart infrastructure management are indispensable links in the process of national development and construction. Among the traffic monitoring and management methods, surveillance cameras are the most intuitive monitoring means, but the camera deployment cost is high, there are privacy issues, the video data processing volume is large, there are huge blind spots, and it is also greatly affected by meteorological factors such as light and rain and snow. Although mobile phone data is also used to monitor traffic flow, this method also has privacy issues like the method of using surveillance cameras. Most people are resistant and wary of their mobile data being monitored.

[0003] Distributed fiber acoustic sensing (DAS) is a sensing technology that emits a pulsed laser beam with a wavelength of 1550nm and uses the Rayleigh scattering principle in optical fibers to detect sound and vibration signals in the environment. Figure 1 As shown in the figure, the optical fiber is laid under the road. When a vehicle passes by, the vibration signal generated by the vehicle will affect the local optical fiber refractive index, thereby affecting the intensity of the reverse Rayleigh scattered light. By differentiating the reverse Rayleigh scattered light intensity before and after the vehicle passes, the vibration signal can be demodulated. Figure 1 This is a schematic diagram of the DAS traffic detection system. Specifically, Figure 1 As shown, optical fiber 1 is laid under road 2. The ultra-narrow pulse laser transmitted in the optical fiber causes reverse Rayleigh scattering. The intensity of the reverse Rayleigh scattering is related to the local refractive index of the optical fiber. When vehicles and pedestrians pass by, the ground vibration causes the local refractive index of the optical fiber to change, thereby changing the intensity of the reverse Rayleigh scattering light. By differentiating the reverse Rayleigh scattering light 4 before the vibration at a certain position (i.e., a certain virtual channel) and the reverse Rayleigh scattering light 5 after the vibration, the vibration information 6 occurring at the position can be demodulated. In the DAS system, optical fibers that are tens of kilometers long are divided into many virtual channels. The channel length may be 2 meters or 1 meter, which needs to be set manually. Figure 1 It shows the status of one of the virtual channels.

[0004] Traffic data analysis using DAS is an emerging technology that is expected to revolutionize the way traffic networks are monitored and managed. High-performance DAS systems require highly coherent radiation sources, high-speed optical modulators, high-performance optical amplifiers and optical detectors, and several kilometers to tens of kilometers of special optical fibers. Such systems are quite expensive, and it does not make sense to deploy such systems for traffic monitoring. Therefore, general DAS systems are deployed on existing idle communication optical fibers. Such systems have seismic noise, factory noise, river noise, and various man-made noises around the optical fiber, so they have extremely low signal-to-noise ratios, for example, Figure 2 As shown in the figure, when hundreds of meters of virtual channels are stitched together, it can be seen that there are many vehicle tracks7, but there is also a lot of noise8. In addition, when the idle optical fiber is laid, some parts are not well coupled with the road, which will cause signal loss. Studies have shown that such signals are difficult to analyze and use for deep learning.

[0005] The traditional method of DAS system traffic monitoring requires experienced signal processing engineers to perform complex preprocessing on the original signal to improve the signal-to-noise ratio. Signal processing engineers will adjust many parameters to improve the signal-to-noise ratio at each step of the signal preprocessing process. For example, in the normalization stage, it is necessary to choose linear normalization or nonlinear normalization, adjust the frequency band of the signal during bandpass filtering, select the size of the sliding window during moving average and moving difference, and select which wavelet to use for decomposition, how many layers to decompose, and how many thresholds to choose when performing wavelet transform. These parameters seriously affect the signal-to-noise ratio after preprocessing, and if they are not adjusted properly, some signals will be lost.

[0006] Each DAS system has different sampling frequencies and spatial resolutions, so each deployed DAS system requires signal processing engineers to perform signal preprocessing. Even after long-term adjustments, the signal-to-noise ratio of the DAS system may not be greatly improved. Therefore, the preprocessed signal is processed through image processing, and then the vehicle trajectory is extracted using the Hough line extraction algorithm, which will introduce many erroneous trajectories. In addition, the traditional DAS vehicle monitoring algorithm treats vehicle trajectories as straight lines.

[0007] In other words, the traditional signal preprocessing method is not only cumbersome but also requires a certain amount of experience to adjust various parameters. In addition, each step processes the one-dimensional information of each virtual channel of the DAS, so the temporal and spatial correlation of the signal is lost, and the signal-to-noise ratio of the final preprocessed signal is still very low. In addition, the traditional DAS vehicle monitoring algorithm treats vehicle trajectories as straight lines, which does not conform to the actual situation and cannot achieve effective monitoring. Summary of the invention

[0008] In order to solve the above problems existing in the prior art, the present invention provides a distributed optical fiber sensing data processing and monitoring method based on a conditional GAN ​​network.

[0009] The technical problem to be solved by the present invention is achieved through the following technical solutions:

[0010] The present invention provides a distributed optical fiber sensing data processing method based on a conditional GAN ​​network, comprising:

[0011] Acquire the original DAS data of the target traffic area within a first preset time; the original DAS data is a vibration signal of the object obtained by differentiating the intensity of the reverse Rayleigh scattered light before and after the object passes through the optical fiber laying area;

[0012] Performing different data processing on the original DAS data within the first preset time to obtain a plurality of data images;

[0013] The multiple data images are simultaneously input into the trained generator to obtain a mask image; the mask image represents the motion trajectory of each object of each category in the target traffic area within the first preset time; wherein the trained generator is obtained by training the conditional GAN ​​network using multiple groups of training images and multiple real mask images that correspond one-to-one to the multiple groups of training images, and each group of training images is the multiple data images of the original DAS data within the second preset time.

[0014] The present invention also provides a distributed optical fiber sensing vehicle pedestrian monitoring method based on a conditional GAN ​​network, comprising:

[0015] Obtaining a mask image to be detected generated by the above-mentioned distributed optical fiber sensing data processing method based on the conditional GAN ​​network;

[0016] Input the mask image to be detected into the trained target detection network to obtain the detection result of the mask image to be detected; the detection result includes: the subcategory to which each motion trajectory belongs, the number of motion trajectories belonging to each subcategory, and the size and position of the annotation box of each motion trajectory; wherein the trained target detection network is obtained by training an initial target detection network using multiple mask images for training; each mask image for training has an annotation, and the annotation is used to indicate the subcategory to which each motion trajectory in the mask image belongs and the position of the rectangular area occupied by each motion trajectory in the mask image;

[0017] According to the size of the annotation box of each motion trajectory, the motion speed of the motion trajectory is calculated.

[0018] Compared with the prior art, the present invention has the following beneficial effects:

[0019] 1) Traditional trajectory detection algorithms need to go through normalization, bandpass filtering, moving difference moving average, wavelet denoising, image thresholding, image morphological processing, line detection, trajectory selection and other processes, and each step has a large number of parameters that need to be adjusted by professionals. The data processing method described in the present invention only needs simple data processing, and can generate the motion trajectory of each object of each category in the target traffic area based on the data after simple data processing and the trained generator, and only needs to collect and train the data set of the conditional GAN ​​network, which improves the ability of signal denoising and makes the signal processing process very simple.

[0020] 2) Traditional signal denoising algorithms such as wavelet denoising, Wiener filtering, short-time Fourier transform denoising, etc. all make certain assumptions about noise, such as noise is random, noise is high frequency, etc., but there are various rich signals in the DAS system, such as factories, rivers, concerts, earthquake waves, etc. These signals do not meet people's definition of noise, so traditional signal denoising algorithms cannot completely eliminate these signals. The data processing method described in the present invention regards these signals as background, so the signal-to-noise ratio can be greatly improved.

[0021] 3) The traditional signal processing steps focus on denoising one-dimensional signals (denoising a certain virtual channel signal). However, the present invention uses the loss function of the conditional GAN ​​network to focus on the characteristic of image structural information, and can learn the spatiotemporal characteristics of the motion trajectory. Moreover, after learning the spatiotemporal characteristics of the motion trajectory, it can also complete the trajectory of some very faint trajectories.

[0022] 4) Traditional traffic detection algorithms use straight line detection algorithms. When the vehicle speed is low, accelerating, decelerating, or stopping, the trajectory is not a straight line, and pedestrian trajectories cannot be detected using straight line detection algorithms. However, the present invention uses a target detection network for traffic detection, which can detect and classify any form of trajectory, and can also calculate the speed of each motion trajectory, achieving accurate and effective monitoring.

[0023] The present invention will be further described in detail below with reference to the accompanying drawings and specific implementation methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a schematic diagram of the principle of a DAS traffic detection system provided by an embodiment of the present invention;

[0025] Figure 2 It is a schematic diagram of signals of a virtual channel spliced ​​together over hundreds of meters provided by an embodiment of the present invention;

[0026] Figure 3is a two-dimensional space-time diagram of a stack of virtual channel signals after thresholded traditional signal processing provided by an embodiment of the present invention;

[0027] Figure 4 This is an exemplary embodiment of the present invention. Figure 3 The image shown is a detection result diagram obtained after trajectory detection using a traditional straight line detection algorithm;

[0028] Figure 5 It is a flow chart of a distributed optical fiber sensing data processing method based on a conditional GAN ​​network provided by an embodiment of the present invention;

[0029] Figure 6 is a schematic diagram of a template for making an exemplary mask image provided by an embodiment of the present invention;

[0030] Figure 7 It is a first principle schematic diagram of an exemplary conditional GAN ​​network provided by an embodiment of the present invention;

[0031] Figure 8 is a second principle schematic diagram of an exemplary conditional GAN ​​network provided by an embodiment of the present invention;

[0032] Fig. 9 is a third principle schematic diagram of an exemplary conditional GAN ​​network provided by an embodiment of the present invention;

[0033] Fig.10 is a first schematic diagram of a mask image generated by an exemplary trained generator provided in an embodiment of the present invention;

[0034] Fig.11 is a second schematic diagram of a mask image generated by an exemplary trained generator provided in an embodiment of the present invention;

[0035] Fig.12 is a third schematic diagram of a mask image generated by an exemplary trained generator provided in an embodiment of the present invention;

[0036] Fig.13 is a fourth schematic diagram of a mask image generated by an exemplary trained generator provided in an embodiment of the present invention;

[0037] Fig.14 It is a flow chart of a distributed optical fiber sensing vehicle pedestrian monitoring method based on a conditional GAN ​​network provided by an embodiment of the present invention;

[0038] Fig.15 is an exemplary result diagram obtained by performing target detection on a mask image containing a motion track of a vehicle and calculating the vehicle speed according to the target detection result, provided by an embodiment of the present invention;

[0039] Fig.16 This is an exemplary result diagram provided by an embodiment of the present invention, which is obtained by performing target detection on a mask image containing a motion trajectory of a pedestrian, and calculating the pedestrian speed based on the target detection result. DETAILED DESCRIPTION

[0040] The present invention is further described in detail below with reference to specific embodiments, but the embodiments of the present invention are not limited thereto.

[0041] like Figure 3 As shown in Figure 2, after data preprocessing using traditional methods, we can see that the signal-to-noise ratio has been greatly improved, but there is still a lot of noise, such as Figure 3 As shown in 8. Using the traditional straight line detection algorithm to detect vehicle trajectory Figure 4 As shown in Figure 2, it can be seen that due to the influence of noise, the detected trajectory has a large number of erroneous vehicle trajectories, such as Figure 4 As shown in 15, the vehicle cannot be accurately detected.

[0042] In view of the problem that the traditional method of preprocessing data is cumbersome and the signal-to-noise ratio of the final preprocessed signal is still very low, the present invention provides a distributed optical fiber sensing data processing method based on a conditional GAN ​​network, which can simplify the signal processing process and improve the ability to denoise the signal. In view of the problem that the traditional DAS vehicle monitoring algorithm treats vehicle trajectories as straight lines, which does not conform to the actual situation and cannot achieve effective monitoring, the present invention provides a distributed optical fiber sensing vehicle and pedestrian monitoring method based on a conditional GAN ​​network, which can detect and classify any form of trajectory, and can also calculate the speed of each motion trajectory, achieving accurate and effective monitoring.

[0043] Figure 5 is a flow chart of a distributed optical fiber sensing data processing method based on a conditional GAN ​​network provided by an embodiment of the present invention, such as Figure 5 As shown, the method includes:

[0044] S101, obtaining original DAS data of a target traffic area within a first preset time; the original DAS data is a vibration signal of the object obtained by differentiating the intensity of reverse Rayleigh scattered light before and after the object passes through the optical fiber laying area.

[0045] Here, the original DAS data may be the original DAS data of a preset number of virtual channels, and the preset number may be set according to actual needs, for example, 170 virtual channels; the first preset time may also be set according to actual needs, for example, 60 seconds (S). Based on this, illustratively, the original DAS data within the first preset time of the target traffic area is the original DAS data within 60S of the 170 virtual channels of the target traffic area.

[0046] Here, the target traffic area may be a target road, a target parking lot, a target airport, etc., and the present invention is not limited to this.

[0047] S102: Perform different data processing on the original DAS data within the first preset time to obtain a plurality of data images.

[0048] S103, inputting multiple data images into the trained generator at the same time to obtain a mask image; the mask image represents the motion trajectory of each object of each category in the target traffic area within the first preset time; wherein the trained generator uses multiple groups of training images and multiple real mask images that correspond one to one to the multiple groups of training images to train the conditional GAN ​​network, and each group of training images is a plurality of data images of the original DAS data within the second preset time.

[0049] Here, the second preset time may be the same as or different from the first preset time; similarly, the number of virtual channels of the original DAS data within the second preset time may be the same as or different from the number of virtual channels of the original DAS data within the first preset time. In addition, the original DAS data within the second preset time may come from other traffic areas.

[0050] Here, each motion track in the mask image is colored, and different colors represent different categories, and the width of the motion tracks belonging to different subcategories in each category is different. The categories include: pedestrians and vehicles; the subcategories of pedestrian category include single person and multi-person, and the subcategories of vehicle category include large vehicle, medium vehicle and small vehicle.

[0051] In the present invention, S102 is implemented by the following steps:

[0052] S1021 . Perform nonlinear log normalization processing on the original DAS data within a first preset time to obtain a first data image.

[0053] S1022: Perform a bandpass filtering process within a first preset frequency band on the original DAS data within a first preset time to obtain a second data image.

[0054] Exemplarily, the first preset frequency band ranges from 0.1 to 2 Hz, because the sound waves caused by vehicle vibration are clearest in the 0.1 to 2 Hz frequency band.

[0055] S1023: Perform bandpass filtering within a second preset frequency band on the original DAS data within the first preset time to obtain a third data image.

[0056] Exemplarily, the second preset frequency band ranges from 1 to 5 Hz, because the vibration sound waves caused by pedestrians walking are clearest in the 1 to 5 Hz frequency band.

[0057] S1024, performing moving average and moving difference processing on the original DAS data within the first preset time to obtain a fourth data image.

[0058] Here, for the original DAS data of a certain period of time, when the four data images of the original DAS data of this period of time are obtained, the human eye can distinguish the approximate vehicle and pedestrian trajectories by combining these four images, but these trajectories often have a very low signal-to-noise ratio, and it is difficult to extract vehicle trajectories using traditional straight line monitoring algorithms. Therefore, by combining these four data images, a trajectory mask image corresponding to the four data images of the original DAS data of this period of time can be manually produced. Exemplarily, in the mask image, the present invention distinguishes the motion trajectories into large vehicles, medium-sized vehicles, small vehicles, and multi-person walking trajectories and single-person walking trajectories according to the amplitude, and pedestrians and vehicles are distinguished by different colors. Figure 6 is an exemplary template for making a mask map, such as Figure 6 As shown, the original trajectory is divided into several types according to the color and width of the motion trajectory, including large vehicles, medium-sized vehicles, small vehicles, multiple people walking, and single people walking. In addition, vehicles are represented by white and pedestrians are represented by red.

[0059] In the present invention, the trained generator is trained using the following method:

[0060] S10. Acquire a plurality of pieces of original DAS data within a second preset time.

[0061] S11. Determine multiple data images and one real mask image for each piece of original DAS data within the second preset time, and obtain multiple sets of training images corresponding one-to-one to the multiple pieces of original DAS data within the second preset time, and multiple real mask images corresponding one-to-one to the multiple sets of training images.

[0062] Here, by adopting the principle of S102 above, each piece of original DAS data within the second preset time is processed to obtain multiple data images of the original DAS data within the second preset time. The obtained multiple data images can be used as a group of training images, and a corresponding mask image can be produced based on the multiple data images as a real mask image of this group of training images.

[0063] Exemplarily, 1000 pieces of original DAS data within the second preset time may be obtained, thereby generating 1000 sets of training images accordingly, and producing 1000 real mask images accordingly, so as to train the conditional GAN ​​network.

[0064] S12. Use multiple groups of training images, multiple real mask images corresponding to the multiple groups of training images, and preset noise to iteratively train the conditional GAN ​​network composed of a generator and a discriminator to obtain a trained generator and a trained discriminator; wherein each group of training images and the preset noise are inputs to the generator, the real mask image, the preset noise and each group of training images are one input to the discriminator, and the mask image generated by the generator, the preset noise and each group of training images are another input to the discriminator.

[0065] Here, the preset noise may be Gaussian noise or other noise.

[0066] Here, the following formula (1.1) shows the loss function of the conditional GAN ​​network, where G represents the generator, D represents the discriminator, E represents the mathematical expectation, x is the real data, y represents the condition, and z represents the noise. data (x) means x follows the probability distribution of real data, z~p z (z) indicates that z obeys the probability distribution of noise. In general, the purpose of this loss function is to learn a mapping from the probability distribution of noise to the probability distribution of real data under the condition of y.

[0067] From the perspective of the discriminator D, we hope to maximize the loss function, so the optimization target during training is as shown in the following formula (1.2). From the perspective of the generator G, we hope to minimize the loss function, so the optimization target during training is as shown in the following formula (1.3). In this way, the generator and the discriminator will learn against each other, the discriminator's judgment ability will become stronger and stronger, and the generator will generate fake data that looks more and more like the real data x according to the instructions of condition y.

[0068]

[0069] According to the principle of the conditional GAN ​​network, the present invention trains a conditional GAN ​​network so that its generator starts directly from the raw data after simple data processing and directly maps it into a mask image with only signals and background.

[0070] Each piece of raw DAS data within the second preset time is subjected to log normalization, 0.1-2HZ bandpass filtering, 1-5HZ bandpass filtering, and moving average moving difference to become four data images 11 (i.e., a set of training images 11). Masks are made based on the trajectory information on this set of training images. Fig.14 , a set of training images 11 is used as the condition y in the conditional GAN ​​network, and the mask made based on a set of training images 11 Fig.14 is the real data x. When training the discriminator, the generator generates a fake image D(z|y) (i.e., 13 in the figure) according to y and the preset noise z (i.e., 12 in the figure) and sends it to the discriminator. At this time, it is hoped that the discriminator will judge it as false; and when the real data x and condition y are sent to the discriminator, it is hoped that the discriminator will judge it as true. When training the generator, the generator generates a fake image D(z|y) according to the condition y and noise z and sends it to the discriminator. It is hoped that the discriminator will judge it as true. In this way, the trained generator can obtain the corresponding mask map according to each set of training images by alternating training. At this time, the trajectory will be very clear, and the generated mask map has a very high signal-to-noise ratio. Different from the L2 loss function, the loss function of the conditional GAN ​​network pays great attention to the structural information in the image rather than strict control of each pixel. Therefore, the network can make good use of the two-dimensional spatiotemporal information in the original signal. It has a certain completion function for the trajectories of pedestrians and vehicles. In addition, the mask map contains only the signal and the background, so the signal-to-noise ratio is greatly improved. Figure 7-Figure 9 This is a schematic diagram of the principle of the conditional GAN ​​network. Figure 7-Figure 9 As shown, 11 and 12 are fed into generator 9, which generates a false mask. Fig.13 , 13 and 11 are sent to the discriminator 10 together. When 13 reaches the level of being indistinguishable from the real thing, the discriminator 10 judges 13 to be true; when 13 is not real enough, the discriminator 10 judges 13 to be false. When 14 and condition 11 are sent to the discriminator 10, the discriminator 10 judges it to be true. The present invention improves the signal preprocessing process by using the conditional GAN ​​network, improves the ability to denoise the signal, and makes the signal processing process very simple.

[0071] For example, Figure 10-13 Illustration of mask images generated for a trained generator. Figure 10-13 Figure A shows the 2D spatiotemporal image of the stacked original DAS signals (0.1-2HZ filtering), and Figure B shows the mask image generated by the trained generator. It can be seen that the generated mask image only contains the signal and background, with a very high signal-to-noise ratio. The intermittent vehicle and pedestrian trajectories in the original signal have been completed in the generated mask image. The spatial resolution has been greatly improved. The noise introduced by various factors has also been completely removed, and the denoising speed using the generator is much faster than the traditional method.

[0072] Compared with the prior art, the distributed optical fiber sensing data processing method based on the conditional GAN ​​network provided by the present invention has the following beneficial effects:

[0073] 1) Traditional trajectory detection algorithms need to go through normalization, bandpass filtering, moving difference moving average, wavelet denoising, image thresholding, image morphological processing, line detection, trajectory selection and other processes, and each step has a large number of parameters that need to be adjusted by professionals. The data processing method described in the present invention only needs simple data processing, and can generate the motion trajectory of each object of each category in the target traffic area based on the data after simple data processing and the trained generator, and only needs to collect and train the data set of the conditional GAN ​​network, which improves the ability of signal denoising and makes the signal processing process very simple.

[0074] 2) Traditional signal denoising algorithms such as wavelet denoising, Wiener filtering, short-time Fourier transform denoising, etc. all make certain assumptions about noise, such as noise is random, noise is high frequency, etc., but there are various rich signals in the DAS system, such as factories, rivers, concerts, earthquake waves, etc. These signals do not meet people's definition of noise, so traditional signal denoising algorithms cannot completely eliminate these signals. The data processing method described in the present invention regards these signals as background, so the signal-to-noise ratio can be greatly improved.

[0075] 3) The traditional signal processing steps focus on denoising one-dimensional signals (denoising a certain virtual channel signal). However, the present invention uses the loss function of the conditional GAN ​​network to focus on the characteristic of image structural information, and can learn the spatiotemporal characteristics of the motion trajectory. Moreover, after learning the spatiotemporal characteristics of the motion trajectory, it can also complete the trajectory of some very faint trajectories.

[0076] The present invention also provides a distributed optical fiber sensing vehicle pedestrian monitoring method based on a conditional GAN ​​network, such as Fig.14 As shown, the method includes:

[0077] S201. Obtain a mask image to be detected that is generated by a distributed optical fiber sensing data processing method based on a conditional GAN ​​network.

[0078] S202. Input the mask image to be detected into the trained target detection network to obtain the detection result of the mask image to be detected; the detection result includes: the subcategory to which each motion trajectory belongs, the number of motion trajectories belonging to each subcategory, and the size and position of the annotation box of each motion trajectory; wherein the trained target detection network is obtained by training the initial target detection network using multiple mask images for training; each mask image for training carries an annotation, and the annotation is used to indicate the subcategory to which each motion trajectory in the mask image belongs and the position of the rectangular area occupied by each motion trajectory in the mask image.

[0079] Here, when the category to which the subcategory belongs is pedestrian, the subcategory includes a single person and multiple persons; when the category to which the subcategory belongs is vehicle, the subcategory includes a large vehicle, a medium vehicle, and a small vehicle.

[0080] S203: Calculate the movement speed of each movement trajectory according to the size of the marked box of the movement trajectory.

[0081] Specifically, the calculation formula for the motion speed of each motion trajectory is as follows:

[0082]

[0083] Among them, V i Indicates the speed of the motion trajectory, H i Indicates the height of the rectangular area occupied by the motion trajectory in the mask image, W i Indicates the width of the rectangular area occupied by the motion trajectory in the mask image, N vc Represents the number of virtual channels of each data image in the multiple data images used to generate the mask image, R s represents the spatial resolution of the DAS system, R T represents the temporal resolution of the DAS system, the DAS system is used to generate the original DAS data, and the multiple data images used to generate the mask image are generated based on the original DAS data generated by the DAS system, N pix Indicates the number of pixels in the length or width of each data image in the multiple data images used to generate the mask image, N samplenum Indicates the preset number of sampling points. For example, R T =0.002s, R s =2m,N samplenum =30000, N vc =176.

[0084] For example, Fig.15 It is the result image obtained by performing target detection on a mask image containing the vehicle's motion trajectory and calculating the vehicle speed based on the target detection result. Fig.15 As shown in the figure, the red frame represents a large car, the yellow frame represents a medium-sized car, and the green frame represents a small car. In addition, the number of frames of each color represents the number of motion tracks belonging to this subcategory, and the corresponding number in each frame is the speed of the white track in the marked frame. Fig.15 Among them, the speeds of the two large cars are 12.46m / s and 9.4m / s respectively, the speed of one medium-sized car is 9.79m / s, and the speeds of the three small cars are 13.33m / s, 13.2m / s, and 13.4m / s respectively.

[0085] For example, Fig.16 It is the result image obtained by performing target detection on a mask image containing the movement trajectory of pedestrians and calculating the pedestrian speed based on the target detection result. Fig.16 As shown in the figure, the purple box represents the walking track of multiple people, the blue box represents the walking track of a single person, and the number of boxes of each color represents the number of motion tracks belonging to this subcategory, and the corresponding number in each box is the speed of the white track in the marked box. Fig.16 In the figure, the speed of one multi-person trajectory is 1.45 m / s, and the speeds of the other two single-person trajectories are 1.4 m / s and 1.49 m / s respectively.

[0086] In the present invention, the trained target detection network is trained using the following method:

[0087] S20. Collecting original DAS data.

[0088] S21. Generate multiple different mask images according to the collected original DAS data.

[0089] S22. In each mask image of the multiple different mask images, mark the subcategory of each motion trajectory and the position of the rectangular area occupied by each motion trajectory in the mask image, to obtain multiple mask images for training.

[0090] S23. Using multiple mask images and loss functions for training, iteratively train the initial target detection network to obtain a trained target detection network.

[0091] Exemplarily, the initial target detection network may be YOLOv8 or other networks, which is not limited in the present invention.

[0092] It should be noted that the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more features. In the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0093] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification.

[0094] In the specification, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude a plurality of components. Certain measures are recorded in different embodiments, but this does not mean that these measures cannot be combined to produce good effects.

[0095] The above contents are further detailed descriptions of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as falling within the protection scope of the present invention.

Claims

1. A distributed optical fiber sensing data processing method based on a conditional GAN ​​network, characterized in that: include: Obtaining raw DAS data within a first preset time of a target traffic area; The original DAS data is a vibration signal of the object obtained by differentiating the intensity of the reverse Rayleigh scattered light before and after the object passes through the optical fiber laying area; Performing different data processing on the original DAS data within the first preset time to obtain a plurality of data images; The multiple data images are simultaneously input into the trained generator to obtain a mask image; the mask image represents the motion trajectory of each object of each category in the target traffic area within the first preset time; wherein the trained generator is obtained by training the conditional GAN ​​network using multiple groups of training images and multiple real mask images that correspond one-to-one to the multiple groups of training images, and each group of training images is the multiple data images of the original DAS data within the second preset time.

2. The distributed optical fiber sensing data processing method based on conditional GAN ​​network according to claim 1 is characterized in that: The performing different data processing on the original DAS data within the first preset time to obtain a plurality of data images includes: Performing nonlinear log normalization processing on the original DAS data within the first preset time to obtain a first data image; Performing a bandpass filtering process within a first preset frequency band on the original DAS data within the first preset time to obtain a second data image; Performing a bandpass filtering process within a second preset frequency band on the original DAS data within the first preset time to obtain a third data image; The original DAS data within the first preset time is processed by moving average and moving difference to obtain a fourth data image.

3. The distributed optical fiber sensing data processing method based on conditional GAN ​​network according to claim 1 is characterized in that: Each motion track in the mask image has a color, and different colors represent different categories. The widths of motion tracks belonging to different subcategories in each category are different.

4. The distributed optical fiber sensing data processing method based on conditional GAN ​​network according to claim 3 is characterized in that: The categories include: pedestrians and vehicles; subcategories of the pedestrian category include single person and multi-person, and subcategories of the vehicle category include large vehicle, medium vehicle and small vehicle.

5. The distributed optical fiber sensing data processing method based on conditional GAN ​​network according to claim 1 is characterized in that: The trained generator is trained using the following method: Acquire a plurality of pieces of raw DAS data within the second preset time; Determine a plurality of data images and a true mask image for each piece of original DAS data within the second preset time, and obtain a plurality of sets of training images corresponding one-to-one to the plurality of pieces of original DAS data within the second preset time, and a plurality of true mask images corresponding one-to-one to the plurality of sets of training images; The conditional GAN ​​network composed of a generator and a discriminator is iteratively trained using the multiple groups of training images, the multiple real mask images corresponding to the multiple groups of training images, and the preset noise to obtain the trained generator and the trained discriminator; wherein each group of training images and the preset noise are inputs to the generator, the real mask image, the preset noise and each group of training images are one input to the discriminator, and the mask image generated by the generator, the preset noise and each group of training images are another input to the discriminator.

6. The distributed optical fiber sensing data processing method based on conditional GAN ​​network according to claim 2 is characterized in that: The first preset frequency band ranges from 0.1 to 2 Hz, and the second preset frequency band ranges from 1 to 5 Hz.

7. A distributed optical fiber sensing vehicle pedestrian monitoring method based on conditional GAN ​​network, characterized in that: include: Acquire a mask image to be detected generated by the distributed optical fiber sensing data processing method based on the conditional GAN ​​network according to any one of claims 1 to 6 above; Input the mask image to be detected into the trained target detection network to obtain the detection result of the mask image to be detected; the detection result includes: the subcategory to which each motion trajectory belongs, the number of motion trajectories belonging to each subcategory, and the size and position of the annotation box of each motion trajectory; wherein the trained target detection network is obtained by training an initial target detection network using multiple mask images for training; each mask image for training has an annotation, and the annotation is used to indicate the subcategory to which each motion trajectory in the mask image belongs and the position of the rectangular area occupied by each motion trajectory in the mask image; According to the size of the annotation box of each motion trajectory, the motion speed of the motion trajectory is calculated.

8. The distributed optical fiber sensing vehicle pedestrian monitoring method based on conditional GAN ​​network according to claim 7 is characterized in that: The calculation formula for the motion speed of each motion trajectory is as follows: Among them, V i Indicates the speed of the motion trajectory, H i represents the height of the rectangular area occupied by the motion trajectory in the mask image, W i Indicates the width of the rectangular area occupied by the motion trajectory in the mask image, N vc represents the number of virtual channels of each of the multiple data images used to generate the mask image, R s represents the spatial resolution of the DAS system, R T represents the temporal resolution of the DAS system, the DAS system is used to generate raw DAS data, and the plurality of data images used to generate the mask image are generated based on the raw DAS data generated by the DAS system, N pix represents the number of pixels in length or width of each data image in the multiple data images used to generate the mask image, N samplenum Indicates the preset number of sampling points.

9. The distributed optical fiber sensing vehicle pedestrian monitoring method based on conditional GAN ​​network according to claim 7 is characterized in that: When the category to which the subcategory belongs is pedestrian, the subcategory includes a single person and multiple persons; when the category to which the subcategory belongs is vehicle, the subcategory includes a large vehicle, a medium vehicle and a small vehicle.

10. The distributed optical fiber sensing vehicle pedestrian monitoring method based on conditional GAN ​​network according to claim 7 is characterized in that: The trained target detection network is trained using the following method: Collect raw DAS data; Generate multiple different mask images based on the collected original DAS data; In each mask image of the plurality of different mask images, a subcategory of each motion trajectory and a position of a rectangular area occupied by each motion trajectory in the mask image are marked to obtain the plurality of mask images for training; The initial object detection network is iteratively trained using the multiple mask images and loss function for training to obtain the trained object detection network.

Images