Fiber optic sensing and deep learning early warning system for vehicle violations on rural roads

Abstract

The present invention relates to a rural highway vehicle violation optical fiber sensing and deep learning early warning system, including an optical fiber sensing module, a demodulation module, a preprocessing module, an identification module and a host computer display module: the optical fiber sensing module is used to sense the vibration of the vehicle during driving on the highway, and collect optical signals containing vibration information; the demodulation module is used to convert the optical signal collected by the optical fiber sensing module into an electrical signal and demodulate a phase signal that changes with time; the preprocessing module is used to window the collected phase signal, and then obtain the frequency domain information of the phase signal through Fourier transformation to obtain the vibration signal. The present invention captures dangerous driving behaviors such as speeding, illegal parking, and illegal lane changes in real time without destroying the scenery along the highway, and also provides an accurate basis for road maintenance. Secondly, the cloud data processing platform can also be used to statistically analyze the distribution of traffic density, vehicle speed and other information.

Description

Technical Field

[0001] The present invention relates to the field of weak grating sensing, signal processing and deep learning application technology, and more specifically, to a rural road vehicle violation optical fiber sensing and deep learning early warning system and method. Background Art

[0002] Most rural areas are located in remote areas with complex natural and geographical conditions and backward road management. During holidays, the increase in the number of tourists in scenic spots often causes problems such as illegal parking, speeding, and illegal lane changes on some popular scenic spots, which poses a safety hazard. Traditional road management of rural roads mainly relies on manpower, traditional monitoring equipment, and a combination of the two. Manual road management is currently only implemented in a very small number of areas, because it is difficult to provide timely feedback on problems on a section of road, and the cost is relatively high.

[0003] Traditional monitoring technologies such as video cameras and speed radars are limited by technical conditions and have certain monitoring blind spots. Video information is also affected by the weather and cannot be monitored around the clock. At the same time, traditional monitoring installation methods will also destroy the viewing experience of the scenery along the highway. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a rural road vehicle violation optical fiber sensing and deep learning early warning system, which can work in harsh environments, has high measurement sensitivity and accuracy, is easy to network, and the convolutional neural network method has a high recognition rate and strong robustness.

[0005] The technical solution adopted by the present invention to solve the technical problem is: constructing a rural road vehicle violation optical fiber sensing and deep learning early warning system, including an optical fiber sensing module, a demodulation module, a preprocessing module, an identification module and a host computer display module:

[0006] The optical fiber sensing module is used to sense the vibration of the vehicle during driving on the road and collect the optical signal containing the vibration information;

[0007] The demodulation module is used to convert the optical signal collected by the optical fiber sensing module into an electrical signal and demodulate a phase signal that varies with time;

[0008] The preprocessing module is used to window the collected phase signal, then obtain the frequency domain information of the phase signal through Fourier transform, and enhance the signal-to-noise ratio of the frequency amplitude through spectral density calculation to obtain the vibration signal;

[0009] The recognition module loads a model file for vehicle violation recognition for training, uses a one-dimensional convolutional neural network to recognize road vehicle violations, and determines whether there are vehicle violations in the collected sensor data;

[0010] The upper computer display module is used to display the identification result of whether the vehicle violates the rules.

[0011] According to the above scheme, the optical fiber sensing module is an identical weak grating array, and the optical fiber sensing module includes a first optical fiber distributed vibration sensing unit, a second optical fiber distributed vibration sensing unit and a third optical fiber distributed vibration sensing unit.

[0012] According to the above scheme, the optical fiber distributed vibration sensing units are respectively arranged longitudinally and crosswise on the rural road; the first optical fiber distributed vibration sensing unit and the third optical fiber distributed vibration sensing unit are arranged on the right side of the rural road and are used to monitor vehicle speeding and illegal parking; the second optical fiber distributed vibration sensing unit is arranged in the middle of the rural road and is used to monitor vehicle overtaking and lane changing.

[0013] According to the above scheme, the first optical fiber distributed vibration sensing unit is composed of weak grating S1 and weak grating S2; the second optical fiber distributed vibration sensing unit is composed of weak grating S2, weak grating S3 and weak grating S4; the third optical fiber distributed vibration sensing unit is composed of weak grating S4 and weak grating S5.

[0014] According to the above scheme, the demodulation module adopts the phase-sensitive optical time domain reflectometer φ-OTDR technology and 3×3 coupler system design.

[0015] According to the above scheme, the phase-sensitive optical time domain reflectometer φ-OTDR technology uses a photodetector to detect the backscattered Rayleigh light generated by the pulse light at various locations along the optical fiber.

[0016] According to the above scheme, the pulse light is a highly coherent laser pulse signal obtained by modulating a narrow linewidth laser through an acousto-optic modulator (AOM).

[0017] According to the above scheme, after the pulse light is amplified by an erbium-doped fiber amplifier (EDFA), it enters the optical fiber sensing module through the first circulator and is reflected back as a pulse light sequence, wherein the spacing between adjacent weak gratings is the same.

[0018] According to the above scheme, the reflected pulse light sequence passes through the first circulator, the second circulator and then reaches the unbalanced Michelson interferometer.

[0019] According to the above scheme, the three output ports of the 3×3 coupler respectively output three optical signals with a phase difference of 2π / 3.

[0020] According to the above scheme, two optical signals among the three optical signals output by the 3×3 coupler meet and interfere at the 3×3 coupler after being reflected by two unequally spaced delay optical fibers and a Faraday rotator.

[0021] According to the above scheme, the three optical signals output by the 3×3 coupler are converted into electrical signals using three photodetectors (PD1, PD2, PD3) and collected by a data acquisition card (DAQ).

[0022] According to the above scheme, the preprocessing module is used to extract frequency domain information from the phase signal demodulated by the demodulation module based on Fourier transform; the preprocessing module is used to enhance the signal-to-noise ratio of the frequency amplitude based on spectral density calculation to obtain a vibration signal.

[0023] According to the above scheme, the recognition module includes loading and recognizing a one-dimensional convolutional neural network model. The one-dimensional convolutional neural network specifically includes an input layer, a pooling layer, a convolutional layer, a second pooling layer, a second convolutional layer, a third pooling layer, a Flatten layer, a fully connected layer, a Dropout layer, a second fully connected layer, a second Dropout layer, a third fully connected layer, and a classification layer.

[0024] According to the above scheme, the one-dimensional convolutional neural network inputs the collected optical fiber sensor module signal, and before the collected optical fiber sensor module signal enters the neural network, the collected optical fiber sensor module signal is first randomly shuffled; the classification result is output, and the classification result is optimized using softmax regression and Adam optimizer.

[0025] According to the above scheme, the loss function of the one-dimensional convolutional neural network adopts a binary cross entropy function, which is specifically:

[0026]

[0027] Among them, y is the true label, is the predicted label.

[0028] According to the above scheme, the host computer module is used to display the identification result of whether the vehicle is in violation of the rules.

[0029] The rural road vehicle violation optical fiber sensing and deep learning early warning system implementing the present invention has the following features:

[0030] Beneficial effects:

[0031] 1. The present invention uses a 3×3 coupler The vibration demodulation system with advanced technology uses the intensity information of scattered light to quickly locate disturbance events. The system has high sensitivity and good versatility. It can be easily connected with other optical devices and has high compatibility, solving the problems of inaccurate recognition or low recognition speed in existing detection methods.

[0032] 2. The present invention uses a fully identical weak grating array as a sensor, which is low in cost, easy to install, corrosion-resistant, and highly resistant to electromagnetic interference, and can adapt to a variety of harsh environments;

[0033] 3. The present invention uses Fourier transform and spectral density to perform calculations, extract and enhance the frequency domain features of the vibration signal, improve the sensitivity of the effective features of the vibration signal, and make it easier for the neural network to identify the vibration signal features;

[0034] 4. The present invention utilizes a convolutional neural network to identify vehicle violations, which can solve the problems of inaccurate recognition or low recognition speed in existing detection methods, and can quickly and accurately detect vehicle violations, meet the technical requirements of monitoring, and have a relatively low implementation difficulty. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which:

[0036] Figure 1 A schematic diagram of a rural road vehicle violation optical fiber sensing and deep learning early warning system proposed in an embodiment of the present invention;

[0037] Figure 2 A schematic diagram of the arrangement of optical fiber sensing modules on a rural road according to an embodiment of the present invention;

[0038] Figure 3 The embodiment of the present invention is based on a 3×3 coupler Schematic diagram of demodulation system structure design;

[0039] Figure 4 A schematic diagram of performing Fourier transform and spectral density calculation on a phase signal demodulated by a demodulation module according to an embodiment of the present invention;

[0040] Figure 5 A schematic diagram of sparse connection and weight sharing of a convolutional neural network proposed in an embodiment of the present invention;

[0041] Figure 6 A schematic diagram of a one-dimensional convolutional neural network model proposed in an embodiment of the present invention. DETAILED DESCRIPTION

[0042] In order to have a clearer understanding of the technical features, purposes and effects of the present invention, specific embodiments of the present invention are now described in detail with reference to the accompanying drawings.

[0043] like Figure 1-6 As shown, an optical fiber sensing and deep learning early warning system for vehicle violations on rural roads proposed in an embodiment of the present invention specifically includes the following modules: an optical fiber sensing module, a demodulation module, a preprocessing module, an identification module, and a host computer display module. The optical fiber sensing module is used to sense the vibration of vehicles traveling on the road and collect optical signals containing vibration information.

[0044] According to the long-distance detection of rural roads, the fiber optic sensor module uses the same weak grating array, as follows:

[0045] Among them, the fully identical weak grating array multiplexes multiple FBG sensors on the same optical fiber, has high capacity, long measurement distance, is not affected by electromagnetic interference, and is corrosion-resistant, which can adapt to the harsh environment of outdoor roads.

[0046] Specifically, the optical fiber sensing module is composed as follows: Figure 2 As shown, the optical fiber sensing module includes a first optical fiber distributed vibration sensing unit, a second optical fiber distributed vibration sensing unit and a third optical fiber distributed vibration sensing unit. The optical fiber distributed vibration sensing units are respectively arranged on the rural road in a longitudinally cross-arranged manner; the first optical fiber distributed vibration sensor is arranged on the right side of the rural road and is used to monitor vehicle speeding and illegal parking, the first optical fiber distributed vibration sensing unit and the third optical fiber distributed vibration sensing unit are arranged on the right side of the rural road and are used to monitor vehicle speeding and illegal parking; the second optical fiber distributed vibration sensing unit is arranged in the middle of the rural road and is used to monitor vehicle overtaking and lane changing. The first optical fiber distributed vibration sensing unit is composed of a weak grating S1 and a weak grating S2; the second optical fiber distributed vibration sensing unit is composed of a weak grating S2, a weak grating S3 and a weak grating S4; the third optical fiber distributed vibration sensing unit is composed of a weak grating S4 and a weak grating S5. The spacing of the weak gratings S1 to S5 on the optical fiber is the same.

[0047] Furthermore, when the vehicle is driving normally, the signal characteristics collected by the optical fiber sensing module are that the time difference between the first optical fiber distributed vibration sensing unit and the third optical fiber distributed vibration sensing unit detecting the vibration is within a normal range; when a vehicle is speeding, the signal characteristics collected by the optical fiber sensing module are that the time difference between the first optical fiber distributed vibration sensing unit and the third optical fiber distributed vibration sensing unit detecting the vibration is shorter than that during normal driving; when a vehicle is illegally parked, the signal characteristics collected by the optical fiber sensing module are that the time difference between the first optical fiber distributed vibration sensing unit and the third optical fiber distributed vibration sensing unit detecting the vibration is longer than that during normal driving; when a vehicle overtakes and changes lanes during driving, the signal characteristics collected by the optical fiber sensing module are that the second optical fiber distributed vibration sensing unit detects the vehicle driving signal.

[0048] The demodulation module converts the optical signal collected by the optical fiber sensing module into an electrical signal and demodulates the phase signal that changes with time. The details are as follows:

[0049] like Figure 3As shown, in the demodulation module, the light emitted by the narrow linewidth laser is modulated into a light pulse sequence with a fixed pulse width by a semiconductor amplifier (SOA). Then, the pulse light is amplified by an erbium-doped fiber amplifier (EDFA) and enters the fiber optic sensing module through the first circulator. The pulse light sequence reflected back from the fiber optic sensing module passes through the second circulator and the second circulator to the unbalanced Michelson interferometer. The three output ports of the 3×3 coupler respectively output three signals with a phase difference of 2π / 3, of which two signals meet and interfere at the 3×3 coupler after being reflected by two non-equidistant delay fibers and a Faraday rotator. Since the length difference between the delay fibers is the distance of the sensing area, the delay between the two interfering signals is equal to the time difference between the detection light reaching the two weak gratings in the sensing area, that is, each interference pulse is obtained by the coherence of the reflected light of the two weak gratings in the corresponding sensing area. Finally, three photodetectors (PD1, PD2, and PD3) are used to convert the three output optical signals of the 3×3 coupler into electrical signals, and the phase signals that vary with time are demodulated and collected by a data acquisition card (DAQ).

[0050] In the above demodulation process, the phase signal that changes with time is demodulated, which refers to the phase signal between the weak gratings S1 and S2, S2 and S3, S3 and S4, and S4 and S5. The sampling frequency of the data acquisition card is set to 10KHZ.

[0051] like Figure 4 As shown, the preprocessing module is used to perform Fourier transform and spectral density calculation on the phase signal demodulated by the demodulation module to obtain a vibration signal.

[0052] Specifically, after obtaining the phase signal varying with time demodulated by the demodulation module, the phase signal is preprocessed by windowing, and the intercepted phase signal is converted into a single-sided frequency domain signal by Fourier transform, and the number of sampling points is halved after the transformation.

[0053] Furthermore, after obtaining the frequency domain signal, the spectral density calculation is used to enhance the signal-to-noise ratio of the frequency amplitude. The spectral density calculation is to improve the effective characteristics of the vibration signal and enhance the signal-to-noise ratio by performing autocorrelation calculation on the frequency domain signal.

[0054] From the above, it can be seen that after Fourier transform and spectral density calculation of the phase signal, a vibration signal with more significant effective features and stronger signal-to-noise ratio is obtained.

[0055] In the embodiment of the present invention, after the vibration signal is obtained as described above, it is arranged into a one-dimensional sequence according to the order of the optical fiber distributed vibration sensing units and sent to the recognition module for training and recognition.

[0056] The recognition module loads the model file for vehicle violation recognition for training, uses a one-dimensional convolutional neural network to recognize road vehicle violations, and determines whether there are vehicle violations in the collected sensor data, specifically including the following:

[0057] The vibration signal sample set corresponding to vehicle speeding, illegal parking, overtaking and lane changing is used to train a one-dimensional convolutional neural network to obtain a model file for vehicle violation recognition. Then, the real-time collected vibration signal is used to train the model file for vehicle violation recognition to output the vehicle violation classification result.

[0058] The one-dimensional convolutional neural network mentioned above is characterized by sparse connections (local perception) and weight sharing. Figure 5 The figure shows sparse connections and weight sharing. Sparse connections and weight sharing can reduce the parameters to be trained and reduce computational complexity.

[0059] Specifically, the one-dimensional convolutional neural network is composed as follows: Figure 6 As shown, it specifically includes input layer, pooling layer, convolution layer, second pooling layer, second convolution layer, third pooling layer, Flatten layer, fully connected layer, Dropout layer, second fully connected layer, second Dropout layer, third fully connected layer, classification layer, where S1 is the input layer, P1~P3 are pooling layers, C1~C2 are convolution layers, FC1~FC3 are fully connected layers, D1~D2 are Dropout layers, F is the Flatten layer, and O is the output layer.

[0060] Furthermore, the input layer is used to input the vibration signals corresponding to vehicle speeding, illegal parking, overtaking and lane changing calculated by the preprocessing module; the pooling layer is used for feature dimensionality reduction, which is equivalent to dimensionality reduction in the spatial range, so that the model can extract a wider range of features, reducing the input size of the next layer, thereby reducing the amount of calculation and the number of parameters; the convolution layer is used to extract different features of the input, each convolution layer is composed of several convolution units, and the parameters of each convolution unit are optimized through the back propagation algorithm; the Flatten layer is used to "flatten" the input, that is, to make the multi-dimensional input one-dimensional, which is often used in the transition from the convolution layer to the fully connected layer; the fully connected layer acts as a classifier, mapping the learned features to the sample label space; the Dropout layer is used in the training process of the neural network, and temporarily discards some neural network units from the network according to a certain probability.

[0061] Specifically, before the vibration signal calculated by the input preprocessing module enters the input layer, it is first randomly shuffled in order to reduce the randomness of the data, which is beneficial to the generalization ability of the model; the Dropout layer ratio is selected as 0.5, which eliminates the weakening of the joint adaptability between neuron nodes and enhances the generalization ability.

[0062] The convolution layer uses the convolution operation, replacing the full connection in the feedforward network with a local connection, and sharing weights between different connections. When a convolution kernel is applied to an image, the convolution kernel is like an observation window during detection, which gradually slides from the upper left corner of the image to the lower right corner. Each position of its sliding corresponds to an output node, which is only connected to the input node in the window (each point on the image corresponds to an input node), and the weights of the connections between different output nodes and corresponding input nodes are the same. The convolution operation can naturally retain the spatial information of the image. It only acts locally, so the spatial structure of the output node corresponds to the spatial structure of the input node.

[0063] The output layer is the classifier, which uses softmax regression, and the optimization algorithm in the softmax regression method uses a binary cross entropy function. The vibration signal can be outputted after the model is identified to determine the vehicle's violation status.

[0064] Furthermore, the loss function of the one-dimensional convolutional neural network adopts the binary cross entropy function, which is specifically:

[0065]

[0066] Among them, y is the true label, is the predicted label.

[0067] As mentioned above, the vibration signal passes through the pooling layer, the convolution layer, the second pooling layer, the second convolution layer, the third pooling layer, the Flatten layer, the fully connected layer, the Dropout layer, the second fully connected layer, the second Dropout layer, the third fully connected layer, and the classification layer outputs the vehicle violation status.

[0068] The host computer display module is used to display the recognition result of whether the vehicle is in violation of the rules, and the result predicted by the neural network is displayed on the host computer module.

[0069] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation modes, which are merely illustrative rather than restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims, all of which are within the protection of the present invention.

Claims

1. A rural road vehicle violation optical fiber sensing and deep learning early warning system, characterized in that: It includes optical fiber sensing module, demodulation module, pre-processing module, identification module and host computer display module: The optical fiber sensing module is used to sense the vibration of the vehicle during driving on the road and collect the optical signal containing the vibration information; The demodulation module is used to convert the optical signal collected by the optical fiber sensing module into an electrical signal and demodulate a phase signal that varies with time; The preprocessing module is used to window the collected phase signal, then obtain the frequency domain information of the phase signal through Fourier transform, and enhance the signal-to-noise ratio of the frequency amplitude through spectral density calculation to obtain the vibration signal; The recognition module loads a model file for vehicle violation recognition for training, uses a one-dimensional convolutional neural network to recognize road vehicle violations, and determines whether there are vehicle violations in the collected sensor data; The host computer display module is used to display the identification result of whether the vehicle violates the rules; The optical fiber sensing module is an identical weak grating array, and the optical fiber sensing module includes a first optical fiber distributed vibration sensing unit, a second optical fiber distributed vibration sensing unit and a third optical fiber distributed vibration sensing unit; The optical fiber distributed vibration sensing units are respectively arranged longitudinally and crosswise on the rural road; the first optical fiber distributed vibration sensing unit and the third optical fiber distributed vibration sensing unit are arranged on the right side of the rural road and are used to monitor vehicle speeding and illegal parking; the second optical fiber distributed vibration sensing unit is arranged in the middle of the rural road and is used to monitor vehicle overtaking and lane changing.

2. The rural road vehicle violation optical fiber sensing and deep learning early warning system according to claim 1 is characterized in that: The first optical fiber distributed vibration sensing unit is composed of a weak grating S1 and a weak grating S2; the second optical fiber distributed vibration sensing unit is composed of a weak grating S2, a weak grating S3 and a weak grating S4; the third optical fiber distributed vibration sensing unit is composed of a weak grating S4 and a weak grating S5.

3. The rural road vehicle violation optical fiber sensing and deep learning early warning system according to claim 1 is characterized in that: The demodulation module adopts the phase-sensitive optical time domain reflectometer φ-OTDR technology.

4. The rural road vehicle violation optical fiber sensing and deep learning early warning system according to claim 1 is characterized in that: The preprocessing module is used to extract frequency domain information from the phase signal demodulated by the demodulation module based on Fourier transform; the preprocessing module is used to enhance the signal-to-noise ratio of the frequency amplitude based on spectral density calculation to obtain a vibration signal.

5. The rural road vehicle violation optical fiber sensing and deep learning early warning system according to claim 1 is characterized in that: The recognition module includes loading and recognizing a one-dimensional convolutional neural network model. The one-dimensional convolutional neural network specifically includes an input layer, a pooling layer, a convolutional layer, a second pooling layer, a second convolutional layer, a third pooling layer, a Flatten layer, a fully connected layer, a Dropout layer, a second fully connected layer, a second Dropout layer, a third fully connected layer, and a classification layer.

6. The rural road vehicle violation optical fiber sensing and deep learning early warning system according to claim 5 is characterized in that: The one-dimensional convolutional neural network inputs the vibration signal calculated by the preprocessing module, and before the vibration signal calculated by the preprocessing module is input into the neural network, the vibration signal calculated by the preprocessing module is randomly shuffled; The classification result is output, and the classification result is optimized using softmax regression and Adam optimizer.

7. The rural road vehicle violation optical fiber sensing and deep learning early warning system according to claim 6 is characterized in that: The loss function of the one-dimensional convolutional neural network adopts a binary cross entropy function, which is specifically: in, is the true label, is the predicted label.

Images