A distributed optical fiber highway litter detection method and system

Abstract

The application provides a distributed optical fiber highway litter detection method and system, and belongs to the field of road traffic monitoring and sensing. The method comprises the following steps: based on the distributed optical fiber acoustic sensing system, collecting vibration signals at each spatial position point along the sensing optical fiber and carrying out pretreatment; performing multi-scale time-frequency analysis on the pretreated vibration signals, introducing adjacent spatial position joint modeling, and constructing a three-dimensional feature representation; inputting the three-dimensional feature representation into a trained deep learning model for preliminary identification to obtain an identification probability; and based on a preset spatiotemporal consistency rule, performing constraint judgment on the identification probability to determine whether a litter event occurs; for the determined litter event, performing road mileage mapping according to the corresponding optical fiber spatial position to realize positioning and output warning information. The application realizes all-weather, high-precision and low-false-alarm automatic safety monitoring, and improves the active safety guarantee capability of the highway.

Description

Technical Field

[0001] This invention belongs to the field of road traffic monitoring and sensing technology, and in particular relates to a distributed fiber optic highway spill detection method and system. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Current methods for detecting debris on highways mainly include video surveillance, millimeter-wave radar, and manual patrols. Video surveillance is susceptible to changes in lighting, rain, fog, and obstructions, and its reliability is insufficient at night and in adverse environments. Radar equipment is costly to deploy and has limited effectiveness in detecting low-reflectivity or non-metallic debris. Manual patrols rely on human experience, have slow response times, and cannot meet real-time requirements. Therefore, there is an urgent need for a debris detection technology that can achieve all-weather, wide-area, and real-time monitoring.

[0004] Distributed fiber acoustic sensing (DAS) technology transforms roadside or underground communication optical cables into continuously distributed "hearing" sensors, enabling real-time detection and location of weak vibration and sound wave signals along fiber optic cables over distances of several kilometers or even tens of kilometers. DAS technology possesses unique advantages such as resistance to electromagnetic interference, tolerance to harsh environments, long monitoring distances, and the ability to be reused with existing communication infrastructure, theoretically making it highly suitable for large-scale road safety monitoring. However, directly applying DAS technology to highway debris detection still faces significant challenges: highway environmental vibration signals are extremely complex, with background signals such as periodic vibrations from normal vehicle traffic, environmental wind noise, distant construction interference, and equipment noise strongly mixed with the transient, non-stationary, and low-repetition-rate weak impact signals generated at the moment debris lands. Debris signals are fleeting in the time domain, dispersed in the frequency domain, and have a limited spatial range. How to accurately, robustly, and with low false alarm rates identify genuine debris events from massive, noisy, and highly interfering DAS data streams is a core technological bottleneck that urgently needs to be overcome in this field. Summary of the Invention

[0005] To overcome the shortcomings of the prior art, this invention provides a distributed fiber optic highway spill detection method and system. By defining spills as a type of road vibration event with transient impact, spatial localization, and non-periodicity, the invention utilizes multi-scale time-frequency mapping to characterize its sudden energy characteristics and combines the consistency constraints of vibrations in adjacent spatial locations to establish a spill recognition model that integrates deep learning and rule-based decision-making. This effectively distinguishes spill events from normal vehicle traffic vibrations and background noise.

[0006] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions: The first aspect of this invention provides a method for detecting debris spilled on distributed optical fiber highways; A method for detecting debris spilled on a distributed fiber optic highway includes: Based on the deployed distributed fiber optic acoustic sensing system, vibration signals at various spatial locations along the sensing fiber are collected in real time, and the collected vibration signals are preprocessed. Multi-scale time-frequency analysis was performed on the preprocessed vibration signal, and joint modeling of adjacent spatial locations was introduced to construct a three-dimensional feature representation that includes time, frequency and spatial dimensions. The three-dimensional feature representation is input into a trained deep learning model for preliminary recognition to obtain the recognition probability; and the recognition probability is constrained and judged based on a preset spatiotemporal consistency rule to determine whether a spill event has occurred. For any incidents involving spilled materials, road mileage mapping is performed based on the corresponding fiber optic spatial location to locate the issue and output early warning information.

[0007] As a further technical solution, the distributed fiber optic acoustic sensing system specifically uses a narrow linewidth laser as a light source. The continuous wave light from the light source is modulated by an acousto-optic modulator and then amplified by a first fiber optic amplifier to compensate for the insertion loss of the acousto-optic modulator. The modulated light enters the sensing fiber through the optical circulator, and the Rayleigh scattered light is amplified by the second fiber amplifier and detected by the photodetector. The electrical signal collected by the photodetector is transmitted to the computer for analysis and processing.

[0008] As a further technical solution, the collected vibration signals are preprocessed, including: An adaptive wavelet denoising algorithm is used to denoise the collected vibration signal; The collected vibration signals are separated using a blind source separation algorithm. The periodic vibration interference generated by normally driving vehicles is then suppressed by a periodic interference suppression method based on adaptive comb filtering on the signal after blind source separation.

[0009] As a further technical solution, multi-scale time-frequency analysis is performed on the preprocessed vibration signal, and joint modeling of adjacent spatial locations is introduced to construct a three-dimensional feature representation that includes time, frequency, and spatial dimensions, including: The vibration signal at each spatial location point is processed by time framing, dividing the vibration signal at each spatial location point into signal frames of a specific length according to time. Perform synchronous compressed wavelet time-frequency transform on the signal of each time frame to obtain the time-frequency energy distribution; For the current target spatial location, the effective spatial neighborhood range for joint modeling is adaptively determined based on the spatiotemporal correlation of its vibration energy with that of neighboring locations. By combining the time-frequency energy distributions of all locations within the effective spatial neighborhood, a three-dimensional feature representation including time, frequency, and spatial dimensions is obtained.

[0010] As a further technical solution, the adaptive determination of the effective spatial neighborhood range for joint modeling based on the spatiotemporal correlation of vibration energy with adjacent locations includes: Calculate the spatial correlation coefficient between the vibration energy sequences of the target location and each of its adjacent locations; A correlation threshold is set, and starting from the target location point, the process extends point by point along the optical fiber to both sides, incorporating adjacent location points whose spatial correlation coefficient is not lower than the correlation threshold into the effective spatial neighborhood, until a point with a correlation coefficient lower than the threshold is encountered.

[0011] As a further technical solution, the three-dimensional feature representation is input into a trained deep learning model for preliminary recognition to obtain a recognition probability; and the recognition probability is constrained and judged based on a preset spatiotemporal consistency rule to determine whether a spill event has occurred, including: The constructed 3D feature representation is input into the trained deep learning model for preliminary identification, and the output is the identification probability of the corresponding spill event. ; Introduce spatiotemporal consistency rules and set thresholds for consecutive time frames. Threshold of adjacent spatial locations and judgment threshold When continuous Within a time frame, and adjacent The recognition probability corresponding to each spatial location point satisfies If the condition is met, an event of spillage is determined to have occurred; otherwise, an event of spillage is determined not to have occurred.

[0012] A second aspect of the present invention provides a distributed fiber optic highway spill detection system.

[0013] A distributed fiber optic highway spill detection system includes: The data acquisition and preprocessing module is configured to: acquire vibration signals at various spatial locations along the sensing fiber in real time based on the deployed distributed fiber acoustic sensing system, and preprocess the acquired vibration signals. The three-dimensional feature construction module is configured to perform multi-scale time-frequency analysis on the preprocessed vibration signal, introduce joint modeling of adjacent spatial locations, and construct a three-dimensional feature representation that includes time, frequency, and spatial dimensions. The spill event judgment module is configured to: input the three-dimensional feature representation into a trained deep learning model for preliminary recognition and obtain the recognition probability; and perform a constraint judgment on the recognition probability based on a preset spatiotemporal consistency rule to determine whether a spill event has occurred. The location and early warning module is configured to: for a determined spill event, perform road mileage mapping based on the corresponding fiber optic spatial location, realize location and output early warning information.

[0014] A third aspect of the present invention provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps of a distributed fiber optic highway spill detection method as described in the first aspect of the present invention.

[0015] A fourth aspect of the present invention provides an electronic device, including a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of a distributed fiber optic highway spill detection method as described in the first aspect of the present invention.

[0016] The fifth aspect of the present invention provides a computer program product, including a computer program / instructions, which, when executed by a processor, implement the steps of the distributed fiber optic highway spill detection method described in the first aspect of the present invention.

[0017] The above one or more technical solutions have the following beneficial effects: (1) This invention employs distributed fiber acoustic sensing (DAS) technology, using road vibration signals as the core detection basis, thus eliminating the environmental and material dependence of traditional detection methods. By constructing a time-frequency-space three-dimensional feature representation system, multi-scale time-frequency analysis is used to characterize the sudden energy characteristics of spill impact vibration. Combined with the consistency constraint of vibration at adjacent spatial locations, the transient impact, spatial localization, and non-periodic nature of spill events are fully captured. The core advantage of long-distance continuous spatial perception of the DAS system is fully utilized, effectively distinguishing spill events from interference signals such as normal vehicle traffic vibration and environmental noise, significantly improving the stability and robustness of event recognition.

[0018] (2) To address the differences in the propagation range of impact vibration from spilled materials under varying masses, drop heights, and road surface conditions, this invention proposes an adaptive spatial scale dynamic selection mechanism based on spatial correlation and energy attenuation characteristics. This mechanism automatically adjusts the spatial neighborhood modeling range according to the propagation characteristics of the actual vibration signal, ensuring the integrity of the core feature information of the spilled material while effectively suppressing the introduction of irrelevant spatial noise. Compared with fixed spatial scale modeling, this design significantly improves the system's anti-interference capability in complex traffic environments and enhances the model's generalization performance for different scenarios and types of spilled material events.

[0019] (3) This invention employs a two-level decision mechanism combining intelligent models and engineering rules. Based on the automatic extraction of high-dimensional features and output of discrimination probabilities by a three-dimensional convolutional neural network, it introduces spatiotemporal consistency rule constraints. By jointly verifying the decision results of continuous time frames and adjacent spatial locations, it effectively filters false alarms caused by occasional noise, single-point abnormal vibrations, and other factors, solving the problem of insufficient stability of pure deep learning models in complex engineering scenarios. This fusion mechanism balances the high recognition accuracy of intelligent algorithms with the high reliability of engineering systems, meeting the practical application requirements for the long-term stable operation of highways.

[0020] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0021] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0022] Figure 1 This is a flowchart of the method in the first embodiment.

[0023] Figure 2 This is a schematic diagram of the internal architecture of the DAS system in the first embodiment.

[0024] Figure 3 This is a system structure diagram of the second embodiment. Detailed Implementation

[0025] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0026] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations of the present invention.

[0027] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0028] Example 1 This embodiment discloses a distributed fiber optic method for detecting spilled materials on highways. It collects road vibration signals through a distributed fiber optic sensing system, performs synchronous compressed time-frequency analysis after signal preprocessing, adaptively determines the effective neighborhood range based on spatial correlation, constructs a three-dimensional feature representation, and uses a deep learning model combined with spatiotemporal consistency rules for fusion judgment to identify spilled material events, ultimately achieving precise event location and real-time early warning. This invention enables all-weather, high-precision, and low-false-alarm automatic detection of spilled materials, effectively improving highway safety monitoring capabilities.

[0029] Specifically, such as Figure 1 As shown, a method for detecting debris spilled on a distributed fiber optic highway includes: Step S1: Based on the deployed distributed fiber optic acoustic sensing system, the vibration signals at various spatial locations along the sensing fiber are collected in real time, and the collected vibration signals are preprocessed.

[0030] like Figure 2 As shown, the distributed fiber-optic acoustic sensing system uses a narrow-linewidth laser as the core light source to generate highly coherent, low-noise continuous-wave light. This continuous-wave light first enters an acousto-optic modulator and is modulated into a series of optical pulses with tunable repetition frequency. The modulation process introduces insertion loss; therefore, the modulated pulsed light is then power-up boosted by a first-stage erbium-doped fiber amplifier to compensate for this loss. The amplified modulated optical pulses are then injected into the sensing fiber optic cable deployed along the highway through an optical circulator.

[0031] As the optical pulse propagates forward in the sensing fiber, its Rayleigh scattered light undergoes phase modulation due to the minute strain caused by external vibrations. This backscattered Rayleigh light, carrying vibration information, returns along the original path and re-enters the signal receiving branch through the optical circulator. Because the scattered signal is extremely weak, a second-stage erbium-doped fiber amplifier is used to amplify it. The amplified optical signal is then converted into an electrical signal by a high-speed photodetector. Finally, this electrical signal is captured by a high-speed data acquisition card and transmitted to a host computer for subsequent real-time analysis and processing.

[0032] In this embodiment, the fiber core of an idle communication optical cable buried under the central green median of a highway is selected as the optical fiber to be tested in this scenario experiment.

[0033] Specifically, the DAS equipment can be installed in the communication room of a toll station within the selected highway section. The fiber optic interface of the equipment can be coupled to the starting interface of the field communication fiber optic cable through a fiber optic coupler, and data can be collected.

[0034] During data acquisition, the system continuously collects vibration data from various spatial locations along the sensing fiber according to preset parameters. The acquisition parameters are set as follows: repetition frequency 2000Hz, pulse width 100ns, and 512 spatial locations, covering a 10km long highway monitoring section. This achieves distributed vibration sensing at 20m intervals, ensuring no monitoring blind spots. The acquisition process is continuous, capturing signals corresponding to various vibration events such as impacts from spilled materials, vehicle traffic, and environmental disturbances in real time.

[0035] Furthermore, the acquired vibration signals undergo preprocessing. Firstly, the DAS system, based on the principle of phase-sensitive optical time-domain reflectometry, outputs a signal that can be expressed as:

[0036] in, The light intensity signal is at position z and time t; The incident light intensity; Visibility coefficient; The phase change is caused by external vibration; This represents the system noise phase.

[0037] An adaptive wavelet noise reduction algorithm is used to process the collected vibration signals.

[0038] in, These are the wavelet coefficients at the j-th scale and k-th position after thresholding, i.e., the coefficients after noise reduction. This represents the original wavelet coefficients at the j-th scale and k-th position obtained after wavelet transform of the original vibration signal; This is a scale-dependent threshold.

[0039] Subsequently, a blind source separation algorithm was used to separate the mixed vibration signals:

[0040] in, The observed signal vector; It is a mixed matrix; The source signal vector; This is the noise vector.

[0041] The source signal is estimated using independent component analysis (ICA), where the source signal includes both useful vibration and noise signals, and noise separation is performed on the source signal.

[0042] Where W is the separation matrix, which is solved by maximizing the non-Gaussianity.

[0043] After blind source separation, to further suppress periodic vibration interference generated by normally driving vehicles, this embodiment employs a periodic interference suppression method based on adaptive comb filtering. First, the vibration signal at each spatial location point after blind source separation is processed in frames, and the short-time autocorrelation function of each frame is calculated. R ( τ )=E[ x ( t ) x ( t + τ By detecting the peak position of the autocorrelation function at a non-zero time delay, the fundamental frequency of vibration in the frame signal is estimated. f 0. Subsequently, an adaptive comb filter is designed based on the estimated fundamental frequency. Its frequency response produces notch filtering at the fundamental frequency and its integer multiples. The transfer function can be expressed as:

[0044] in, The sampling period corresponding to the fundamental frequency. Sampling rate, K For the harmonic order to be considered, The notch depth coefficients are adaptively updated. This is achieved by adaptively adjusting the filter output signal's periodicity metrics (such as the peak value of the autocorrelation function). This allows for the dynamic filtering of periodic interference that changes over time. The filtered signal is the effective vibration signal after removing periodic vibrations, which is used for subsequent 3D feature construction.

[0045] Step S2 involves performing multi-scale time-frequency analysis on the preprocessed vibration signal, introducing joint modeling of adjacent spatial locations, and constructing a three-dimensional feature representation that includes time, frequency, and spatial dimensions.

[0046] After vibration signal preprocessing, projectile event features are constructed from vibration signals obtained at various spatial locations along the sensing fiber. Let the preprocessed vibration signal be represented as:

[0047] Where z represents the spatial location point index corresponding to the optical fiber, t represents the time sampling point, Z is the total number of location points in the monitoring section, and T is the number of time sampling points in a single acquisition.

[0048] Considering the sudden and short-duration nature of spill events, the vibration signal is first processed by time framing. The vibration signal at each spatial location is divided into signal frames of length L seconds, with an overlap ratio set between adjacent frames. This is to prevent the impact vibration of the projectiles from being truncated by the frame boundary.

[0049] The signal of the nth frame can be represented as:

[0050] Where n is the frame number.

[0051] In response to the non-stationary, transient impact, and rapidly changing frequency characteristics of vibration signals caused by highway spills, this embodiment employs synchronous compressed continuous wavelet transform for high-resolution time-frequency analysis of the signals.

[0052] First, perform a continuous wavelet transform (CWT) on the time frame signal, the expression of which is:

[0053] in, This is a scale parameter used to control the analysis frequency resolution; For time shift parameters; For the mother wavelet function; Represents the complex conjugate operation; These represent the wavelet coefficients of the signal in the scale-time plane.

[0054] To overcome the problems of energy dispersion and inaccurate frequency localization in traditional wavelet transform, a synchronous compression operation is introduced, which calculates the instantaneous frequency of the wavelet coefficients:

[0055] in, Instantaneous frequency, This indicates the operation of taking the imaginary part.

[0056] Based on the above instantaneous frequency estimation results, the energy of the original wavelet coefficients is remapped from the scale axis to the frequency axis to construct a synchronous compressed time-frequency energy spectrum:

[0057] in, This represents the frequency variable after redistribution; This is the frequency tolerance parameter; Indicates spatial location ,time ,frequency Synchronous compression of time-frequency energy at the location.

[0058] This synchronous compression spectrum can significantly improve frequency concentration while maintaining time resolution, and is particularly suitable for characterizing transient high-energy vibrations generated by projectile impacts.

[0059] Considering the spatial propagation characteristics of the impact vibration of the spilled material along the road structure and fiber optic path, the target spatial location point... Centered on, select the areas before and after it. Construct a spatial neighborhood set by placing points in adjacent spaces:

[0060] The synchronous compressed time-frequency energy spectra corresponding to each location point within the spatial neighborhood are combined to form a time-frequency-space three-dimensional feature tensor:

[0061] in, This indicates the spatial neighborhood scale, used to describe the spatial diffusion range of impact vibrations from projectiles along the optical fiber.

[0062] Because different ejected material events vary in mass, drop height, and road surface conditions, the propagation distance and spatial attenuation characteristics of the resulting impact vibrations along the optical fiber also differ. To avoid insufficient feature information or noise introduction problems caused by using a fixed spatial neighborhood scale, this embodiment introduces an adaptive spatial scale selection mechanism based on spatial correlation and energy attenuation characteristics. Specifically: Synchronous compression of time-frequency energy spectrum Based on this, we first perform energy integration over the frequency dimension to obtain the space-time energy distribution function:

[0063] in, and These represent the upper and lower limits of the analysis frequency band, respectively. Indicates time Location, spatial position The total vibrational energy.

[0064] With target location point Using the center as the reference point, calculate its relationship with neighboring points. The spatial correlation coefficient, defined as the correlation on the energy sequence, is:

[0065] in, This represents the time mean of the energy sequence at position z; Used to characterize the spatial consistency of impact vibrations from projectiles.

[0066] Based on the spatial correlation coefficient, the valid criteria for determining spatial neighborhood are defined as follows:

[0067] in, This is the correlation threshold used to distinguish between the effective impact propagation location and the background noise location.

[0068] On both sides of the target location, expand point by point in both the forward and reverse directions until the above conditions are no longer met, thus finally determining the spatial neighborhood scale:

[0069] This yields an adaptive spatial neighborhood set for the current spill event:

[0070] Based on the determined adaptive spatial scale Construct a three-dimensional feature representation that includes time, frequency, and spatial dimensions:

[0071] This feature tensor can adaptively adjust the spatial modeling range according to the impact intensity and propagation characteristics of different projectiles, thereby effectively suppressing the introduction of irrelevant noise while ensuring the integrity of key information.

[0072] Step S3: Input the three-dimensional feature representation into the trained deep learning model for preliminary recognition to obtain the recognition probability; and make a constraint decision on the recognition probability based on the preset spatiotemporal consistency rule to determine whether a spill event has occurred.

[0073] A three-dimensional feature representation incorporating time, frequency, and spatial dimensions is constructed as input and fed into a three-dimensional convolutional neural network for feature learning and event discrimination. The three-dimensional convolutional neural network consists of several three-dimensional convolutional layers, nonlinear activation layers, and an output layer. Layer convolution operation is represented as:

[0074] in, This is the input feature map for the previous layer; For the first Layer 3D convolution kernel parameters; For bias terms; It is a non-linear activation function.

[0075] In addition, to further enhance the model's ability to perceive key features of spill events, a multi-dimensional attention mechanism module is introduced into the three-dimensional convolutional neural network. By adaptively focusing on key information in the channel, spatial and temporal dimensions, the feature expression capability is enhanced, significantly improving detection accuracy and robustness.

[0076] Let the input feature tensor be Where T is the number of time frames, F is the frequency dimension, and S is the number of spatial location points. The 3D attention mechanism module first performs global average pooling and global max pooling on the input features to generate two different context descriptors:

[0077]

[0078] in, To utilize the context descriptor obtained through global average pooling; To utilize the context descriptor obtained through global max pooling; This is a global average pooling operation; This is a global max pooling operation.

[0079] The pooling result is then fed into a shared 3D convolutional layer (or fully connected layer) to generate the attention weight tensor:

[0080] in For the Sigmoid activation function, These are the normalized attention weights. Finally, the original feature map is multiplied element-wise by the attention weights to obtain the enhanced feature representation:

[0081] in, For the enhanced feature representation; This indicates element-wise multiplication.

[0082] Through layer-by-layer feature extraction, the network outputs the probability of discriminating against the spilled object event corresponding to the input feature tensor. .

[0083] Obtaining the enhanced feature representation Subsequently, the deep learning model further extracts high-level features and outputs discrimination probabilities through a three-dimensional convolutional neural network. The specific process is as follows: Enhanced features As input, it passes through multiple 3D convolutional layers sequentially. Each 3D convolutional operation extracts local time-frequency-space joint features, and introduces non-linear expressive power through non-linear activation functions (such as ReLU), gradually abstracting higher-level event features. Let the th... l The convolution operation of a layer is:

[0084] in, The output of the previous layer (the first layer is) ), It is a three-dimensional convolution kernel. For bias terms, σ This is the activation function.

[0085] After multiple 3D convolutional layers, a global pooling layer (such as global average pooling or global max pooling) or a flattening operation is usually applied to compress the high-dimensional feature map into a fixed-length feature vector, aggregating discriminative information in the time-frequency-space dimensions.

[0086] The aggregated feature vector is input to a fully connected layer, and finally passes through an output node with a sigmoid activation function (for binary classification tasks) or a softmax layer (for multi-class classification tasks) to generate the discrimination probability of the projectile event. , representing the confidence level that the current input feature belongs to the spill event.

[0087] Guided by the attention mechanism, the model can focus more on the local abrupt changes in the time-frequency-spatial domain of the spill event, effectively suppressing interference from periodic vehicle vibrations, environmental noise, etc., and significantly improving recognition accuracy and robustness.

[0088] Finally, to improve the stability of the decision results and reduce the false alarm rate caused by occasional noise, a spatiotemporal consistency rule constraint is introduced based on the output of the deep learning model, by setting a threshold for consecutive time frames. Threshold of adjacent spatial locations and judgment threshold When continuous Within a time frame, and adjacent The recognition probability corresponding to each spatial location point satisfies If the condition is met, an event of spillage is determined to have occurred; otherwise, an event of spillage is determined not to have occurred.

[0089] After a deep learning model and spatiotemporal consistency rules jointly determine that a spill event has occurred, the system immediately triggers a pre-set high-definition camera and LiDAR to synchronously collect data on the event area. Specifically, based on the fiber optic spatial location z and time τ corresponding to the event, the system determines the road segment and lane where the event occurs through a pre-calibrated fiber-road mileage-sensor viewpoint mapping relationship. The system then controls the camera to focus on the area and continuously capture multiple frames of images, while the LiDAR scans to acquire 3D point cloud data around the event point.

[0090] Based on the collected multi-source data, features for weight estimation are first extracted from the DAS vibration signal, and four key features are calculated: peak impact energy, dominant frequency distribution, energy decay rate, and spatial propagation distance.

[0091] Among them, the peak impact energy for:

[0092] Main frequency distribution for:

[0093] in, The time-frequency energy is obtained by synchronously compressing wavelet transform, and Ω represents the spatiotemporal domain of the event's influence.

[0094] Energy decay rate The energy decay rate at the event center is obtained by linear fitting of the logarithmic curve of energy decay over time, reflecting the rate of attenuation of the impact signal. The spatial propagation distance L is defined as the distance from the event center along the fiber direction where the energy decays to... The number of spatial points covered at that time.

[0095] The above features are combined into a feature vector. The input is fed into a pre-trained weight regression model, which outputs an estimated weight. .

[0096] Secondly, images captured by a camera are preprocessed and then input into a pre-trained convolutional neural network for debris type recognition. The network output layer is a softmax layer, corresponding to categories including stones, metal parts, plastic boxes, tires, wood, and scattered debris. The category with the highest confidence level is selected as the recognition type. .

[0097] Next, the lidar point cloud data is processed: first, ground points and background noise are filtered out, and then the point cloud clusters corresponding to the spilled material are segmented using the DBSCAN algorithm. The convex hull volume of the point cloud cluster is calculated. .

[0098] The attributes output from the three sensors mentioned above are fused at the decision level to form a unified attribute description vector. Color information can be obtained through image recognition. Combining attribute vectors with real-time environmental information, XGBoost is used to assess the risk level. The risk level is divided into three levels: (1) Low risk: weighing less than 10kg and having a volume greater than 0.1m³, or located in the emergency lane; (2) Medium risk: The weight is between 10 and 30 kg and the volume is medium, or it is located in a section of road with poor visibility such as a bend or tunnel entrance; (3) High risk: weighing more than 30kg, or of the type being sharp metal parts or stones, or located in the fast lane with low visibility.

[0099] Step S4: For the determined spill event, perform road mileage mapping based on the corresponding fiber optic spatial location to achieve positioning and output early warning information.

[0100] When a valid ejection event is determined to exist, it is based on the corresponding fiber optic spatial location. With time of occurrence By combining the mapping relationship between fiber optic cable length and highway mileage, the system can accurately locate the spilled material within the road space. The location result, along with information such as the event time and confidence level, is sent to the monitoring and early warning platform, triggering corresponding alarm prompts and handling procedures.

[0101] Step S4: In the process of positioning the projectile, in order to further improve the positioning accuracy, this invention proposes a fine positioning method based on subspatial resolution energy centroid estimation and multi-spatial point signal arrival time difference analysis.

[0102] First, establish a precise correspondence between optical fiber spatial sampling points and actual road mileage. Record the road mileage L(z) corresponding to each sampling point z by measuring along the road or combining it with GIS data, and establish an interpolation function L=f(z). Since optical fibers may bend with the road or have redundancy, this mapping is usually nonlinear and can be described using piecewise linear interpolation or spline interpolation.

[0103] When the DAS system is in space When a littering incident is detected nearby, extract the following: Centered N neighboring spatial points ( ,…, At the moment the event occurred τ Nearby vibration energy amplitude Because the impact vibration of projectiles exhibits a decaying spatial distribution, its energy peak location does not necessarily fall precisely at the discrete sampling points. Therefore, the energy centroid method is used to estimate the true impact location with subspatial resolution.

[0104] in, For the estimated continuous fiber location index, For spatial points The formula calculates the vibration energy amplitude at a given time. By using an energy-weighted average, the position of the projectile is located at the energy center of gravity, achieving an accuracy that surpasses the spatial sampling interval of a DAS system.

[0105] Further calculation of the spatial standard deviation of the energy distribution To verify the reliability of the estimation results:

[0106] like A smaller value indicates concentrated energy and high reliability of the location; conversely, if the energy is dispersed, it may correspond to multiple projectiles or interference.

[0107] In obtaining continuous fiber positions Then, it can be done through the mapping function. Converted to actual road mileage Simultaneously, by combining lane information based on time, a precise positioning result is output.

[0108] Example 2 This embodiment discloses a distributed fiber optic highway spill detection system; like Figure 3 As shown, a distributed fiber optic highway spill detection system includes: The data acquisition and preprocessing module is configured to: acquire vibration signals at various spatial locations along the sensing fiber in real time based on the deployed distributed fiber acoustic sensing system, and preprocess the acquired vibration signals. The three-dimensional feature construction module is configured to perform multi-scale time-frequency analysis on the preprocessed vibration signal, introduce joint modeling of adjacent spatial locations, and construct a three-dimensional feature representation that includes time, frequency, and spatial dimensions. The spill event judgment module is configured to: input the three-dimensional feature representation into a trained deep learning model for preliminary recognition and obtain the recognition probability; and perform a constraint judgment on the recognition probability based on a preset spatiotemporal consistency rule to determine whether a spill event has occurred. The location and early warning module is configured to: for a determined spill event, perform road mileage mapping based on the corresponding fiber optic spatial location, realize location and output early warning information.

[0109] Example 3 The purpose of this embodiment is to provide a computer-readable storage medium.

[0110] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a distributed fiber optic highway spill detection method as described in Example 1.

[0111] Example 4 The purpose of this embodiment is to provide an electronic device.

[0112] An electronic device includes a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the distributed fiber optic highway spill detection method described in Embodiment 1.

[0113] Example 5 Embodiment 5 of the present invention provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps in the distributed fiber optic highway spill detection method described in Embodiment 1.

[0114] The steps and methods involved in the apparatuses of Embodiments 2, 3, 4, and 5 above correspond to those in Embodiment 1. For specific implementation details, please refer to the relevant description section of Embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood as including any medium capable of storing, encoding, or carrying an instruction set for execution by a processor and enabling the processor to perform any of the methods in this invention.

[0115] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computer devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computer device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. The present invention is not limited to any particular combination of hardware and software.

[0116] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for detecting debris spilled on a distributed optical fiber highway, characterized in that, include: Based on the deployed distributed fiber optic acoustic sensing system, vibration signals at various spatial locations along the sensing fiber are collected in real time, and the collected vibration signals are preprocessed. Multi-scale time-frequency analysis is performed on the preprocessed vibration signal. Joint modeling of adjacent spatial locations is introduced to construct a three-dimensional feature representation that includes time, frequency and spatial dimensions. Specifically, the vibration signal at each spatial location point is processed by time framing, and the vibration signal at each spatial location point is divided into signal frames of a specific length according to time. Perform synchronous compressed wavelet time-frequency transform on the signal of each time frame to obtain the time-frequency energy distribution; For the current target spatial location point, based on the spatiotemporal correlation of its vibration energy with that of adjacent locations, the effective spatial neighborhood range for joint modeling is adaptively determined. Specifically, the spatial correlation coefficient between the target location point and the vibration energy sequences of each adjacent location point is calculated. A correlation threshold is set, and starting from the target location point, the model is expanded point by point along the optical fiber to both sides. Adjacent locations with spatial correlation coefficients not lower than the correlation threshold are included in the effective spatial neighborhood until a point with a correlation coefficient lower than the threshold is encountered. The time-frequency energy distributions of all locations within the effective spatial neighborhood are combined to obtain a three-dimensional feature representation that includes time, frequency, and spatial dimensions; The three-dimensional feature representation is input into a trained deep learning model for preliminary recognition to obtain the recognition probability; and the recognition probability is constrained and judged based on a preset spatiotemporal consistency rule to determine whether a spill event has occurred. For any incidents involving spilled materials, road mileage mapping is performed based on the corresponding fiber optic spatial location to locate the issue and output early warning information.

2. The method for detecting spilled material on a distributed optical fiber highway as described in claim 1, characterized in that, The distributed fiber optic acoustic sensing system specifically uses a narrow linewidth laser as the light source. The continuous wave light from the light source is modulated by an acousto-optic modulator and then amplified by a first fiber amplifier to compensate for the insertion loss of the acousto-optic modulator. The modulated light enters the sensing fiber through the optical circulator, and the Rayleigh scattered light is amplified by the second fiber amplifier and detected by the photodetector. The electrical signal collected by the photodetector is transmitted to the computer for analysis and processing.

3. The method for detecting spilled material on a distributed optical fiber highway as described in claim 1, characterized in that, The collected vibration signals are preprocessed, including: An adaptive wavelet denoising algorithm is used to denoise the collected vibration signal; The collected vibration signals are separated using a blind source separation algorithm. The periodic vibration interference generated by normally driving vehicles is then suppressed by a periodic interference suppression method based on adaptive comb filtering on the signal after blind source separation.

4. The method for detecting spilled material on a distributed optical fiber highway as described in claim 1, characterized in that, The three-dimensional feature representation is input into a trained deep learning model for preliminary recognition to obtain the recognition probability. The identification probability is constrained and judged based on a preset spatiotemporal consistency rule to determine whether a spill event has occurred, including: The constructed 3D feature representation is input into the trained deep learning model for preliminary identification, and the output is the identification probability of the corresponding spill event. ; Introduce spatiotemporal consistency rules and set thresholds for consecutive time frames. Threshold of adjacent spatial locations and judgment threshold When continuous Within a time frame, and adjacent The recognition probability corresponding to each spatial location point satisfies When this occurs, it is determined that a spill incident has occurred; Conversely, if no such event occurs, it is determined that no spillage incident has taken place.

5. A distributed fiber optic highway spill detection system, characterized in that, include: The data acquisition and preprocessing module is configured to: acquire vibration signals at various spatial locations along the sensing fiber in real time based on the deployed distributed fiber acoustic sensing system, and preprocess the acquired vibration signals. The 3D feature construction module is configured to perform multi-scale time-frequency analysis on the preprocessed vibration signal, introduce joint modeling of adjacent spatial locations, and construct a 3D feature representation that includes time, frequency, and spatial dimensions. Specifically, it performs time-frame processing on the vibration signal at each spatial location point, dividing the vibration signal at each spatial location point into signal frames of a specific length according to time. Perform synchronous compressed wavelet time-frequency transform on the signal of each time frame to obtain the time-frequency energy distribution; For the current target spatial location point, based on the spatiotemporal correlation of its vibration energy with that of adjacent locations, the effective spatial neighborhood range for joint modeling is adaptively determined. Specifically, the spatial correlation coefficient between the target location point and the vibration energy sequences of each adjacent location point is calculated. A correlation threshold is set, and starting from the target location point, the model is expanded point by point along the optical fiber to both sides. Adjacent locations with spatial correlation coefficients not lower than the correlation threshold are included in the effective spatial neighborhood until a point with a correlation coefficient lower than the threshold is encountered. The time-frequency energy distributions of all locations within the effective spatial neighborhood are combined to obtain a three-dimensional feature representation that includes time, frequency, and spatial dimensions; The spill event judgment module is configured to: input the three-dimensional feature representation into a trained deep learning model for preliminary recognition and obtain the recognition probability; and perform a constraint judgment on the recognition probability based on a preset spatiotemporal consistency rule to determine whether a spill event has occurred. The location and early warning module is configured to: for a determined spill event, perform road mileage mapping based on the corresponding fiber optic spatial location, realize location and output early warning information.

6. A computer-readable storage medium having a program stored thereon, characterized in that, When executed by the processor, the program implements the steps of the distributed fiber optic highway spill detection method as described in any one of claims 1-4.

7. An electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the distributed fiber optic highway spill detection method as described in any one of claims 1-6.

8. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps in the distributed fiber optic highway spill detection method as described in any one of claims 1-4.

Images