A fiber optic audio early warning system, method, and application based on road segment attribute adaptation.

Abstract

This invention, entitled "A Method, System, and Application of Fiber Optic Audio Reconstruction Early Warning Based on Adaptive Road Segment Attributes," belongs to the field of distributed fiber optic sensing and intelligent security technology. The technical problem to be solved is to provide a fiber optic audio reconstruction early warning system and method with low false alarm rate based on adaptive road segment attributes. Key technical points: The fiber optic audio reconstruction early warning system includes a distributed fiber optic sensing unit, a road segment attribute configuration storage module, an adaptive pattern recognition parameter calculation module, a fiber optic sound reconstruction module, a scene-specific audio recognition model library, an audio type-hazard judgment engine, and an alarm execution unit.

Description

Technical Field

[0001] Distributed Vibration Sensing (DVS) technology is an advanced sensing method that uses optical fibers as sensors to detect changes in the phase or intensity of optical signals caused by external disturbances, enabling real-time sensing of vibration events over a range of several kilometers to tens of kilometers along a pipeline. Due to its advantages such as resistance to electromagnetic interference, intrinsic safety, long-distance coverage, and high spatial resolution, it has been widely applied in critical infrastructure safety fields such as oil and gas pipeline safety monitoring, power cable protection against external damage, railway intrusion early warning, border perimeter protection, and urban underground pipeline network protection.

[0002] In the aforementioned application scenarios, the core objective of the system is to promptly and accurately identify abnormal events that may threaten facility safety (such as excavation, drilling, climbing, vehicle running over, etc.) and issue early warnings, while effectively suppressing non-threatening disturbances caused by the natural environment or daily activities (such as wind, rain, animal movement, distant traffic noise, etc.). Therefore, how to accurately extract effective event features from massive, weak, and mixed vibration signals and determine their hazard becomes the key to the application of DVS technology. Background Technology

[0003] In recent years, with the development of signal processing and artificial intelligence technologies, researchers have begun to attempt acoustic reconstruction of vibration signals acquired by DVS systems. This involves using specific algorithms to restore the vibration waveform propagating along optical fibers into an audio signal that is approximately audible to the human ear. This breakthrough enables the system not only to sense whether there is vibration, but also to further analyze what kind of sound it is, thereby introducing audio semantic understanding capabilities and significantly improving the precision of event recognition.

[0004] Based on this, existing mainstream early warning systems typically employ a two-tier processing architecture: Front-end pattern recognition: Perform time-frequency analysis on the original vibration signal (such as short-time Fourier transform, wavelet transform, etc.), extract features such as energy, frequency, and abrupt change coefficient, and set thresholds for preliminary screening to filter out obvious noise; Backend classification and judgment: For suspected valid event fragments, use machine learning models (such as support vector machine SVM, random forest or deep neural network) to classify them and determine whether they are intrusion or destructive behaviors.

[0005] However, such systems generally assume that the entire fiber optic link is in a homogeneous environment, using uniform parameter configurations and universal identification models, ignoring the significant differences in signal propagation characteristics and noise spectrum distribution caused by differences in geological conditions, burial methods, and surrounding environment in different sections during actual deployment.

[0006] Patent document CN113988170A (publication date January 28, 2022) discloses a machine learning algorithm for event classification in a fiber optic early warning system. This algorithm aims to address the problem of inaccurate signal labeling caused by the numerous, long, and wide-ranging pipelines, making it impossible to immediately reach the site for tracking of hazardous events. The key technical aspects include a feature extraction module, a classifier structure module, and a training database module. The feature extraction module includes segmented energy proportion units, energy units, signal continuity units, time stamp units, sensor serial number units, and fiber optic cable burial depth units. The classifier structure module includes a preprocessing classifier, a main classifier, a false alarm prevention classifier, and a false negative prevention classifier. The training database module includes a signal definition database, a daily event database, a correction database one, and a correction database two, achieving the effect of quickly and accurately identifying fault events and filtering information.

[0007] Patent document CN113627470A (published on November 9, 2021) discloses a method for classifying unknown events in a fiber optic early warning system based on zero-order learning. The method includes: Step 1, detecting Rayleigh scattering light intensity signals obtained when a vibration event occurs using the fiber optic early warning system; Step 2, creating vibration event data samples from the Rayleigh scattering light intensity signals; Step 3, inputting the vibration event data samples into a classification network obtained through zero-order learning, and combining the data with an attribute space to determine the category and obtain the classification result. This invention enables fiber optic sensing systems to identify vibration event types that are not pre-trained, improving the applicability of the classification method.

[0008] Patent document CN104269006A (published on January 7, 2015) discloses a fiber optic early warning system and pattern recognition method, relating to the field of pipeline monitoring. The system involves an electrical signal that is amplified, filtered, and converted from analog to digital by a signal acquisition and host computer module, followed by digital signal processing and analysis. Simultaneously, continuous light generated from a Raman light source is split into two beams by a 2x2 splitter, which are then fed into a first wavelength division multiplexer and a second wavelength division multiplexer, respectively, and injected into the sensing fiber from the forward and reverse directions. Through Raman scattering, the light pulses generated by the laser light source are distributedly amplified, ensuring signal strength along the sensing fiber. Finally, the signals obtained from multiple pulse processes are rearranged in the signal acquisition and host computer module to obtain a two-dimensional signal relating to space and time for subsequent use. This invention can effectively identify and locate events such as human walking, manual excavation, and vehicle passage, effectively reducing the false alarm rate of the early warning system.

[0009] However, the existing technology still has the following major technical defects: (1) The parameter configuration is rigid and lacks environmental adaptability. The unified parameters cannot take into account the perception needs of various road sections; (2) The audio recognition model is generalized and lacks semantic understanding ability, and cannot accurately distinguish based on the road section context; (3) The physical attributes of the road section are disconnected from the intelligent recognition logic, and no structured association mechanism has been established between "physical environment attributes" and "signal processing strategy + recognition model"; (4) It relies on a large amount of manual parameter tuning, resulting in high deployment and maintenance costs and poor system stability. Summary of the Invention

[0010] The purpose of this invention is to provide: A fiber optic audio restoration early warning system based on road segment attribute adaptation, and related technologies, to solve the technical problems of high false alarm rate in existing fiber optic early warning systems, or a combination thereof.

[0011] Terminology Explanation: Unless otherwise defined, all technical terms used herein have the same meanings as commonly understood by one of ordinary skill in the art to which this subject matter pertains. Unless otherwise stated, all patents, patent inventions, and disclosures cited throughout this document are incorporated herein by reference in their entirety. Where multiple definitions exist for terms herein, the definitions provided in this chapter shall prevail.

[0012] It should be understood that the above brief description and the following detailed description are exemplary and for illustrative purposes only, and do not limit the subject matter of the invention in any way. In this invention, the singular is used in conjunction with the plural unless otherwise specifically stated. It should also be noted that, unless otherwise stated, the use of “or” or “or” means “and / or”. Furthermore, the use of the term “comprising” and other forms such as “including,” “containing,” and “contains” are not limiting.

[0013] The definition of the standard term can be found in the reference "Fully Distributed Fiber Optic Sensing Technology" by Zhang Xuping, published by Science Press in 2013.

[0014] Unless specifically defined herein, the use of all commercially available products herein employs standard techniques. For example, it may be carried out using the manufacturer's instructions for use with the kit, or in accordance with methods known in the art or the description of this invention. The techniques and methods described herein can generally be implemented according to conventional methods well known in the art, based on the descriptions in the various summary and more specific documents cited and discussed in this specification.

[0015] The “range” disclosed in this document takes the form of a lower limit and an upper limit. It can be one or more lower limits and one or more upper limits, respectively. A given range is defined by selecting a lower limit and an upper limit. The selected lower and upper limits define the boundaries of the particular range. All ranges that can be defined in this way are inclusive and composable; that is, any lower limit can be combined with any upper limit to form a range. For example, if ranges of 60-120 and 80-110 are listed for a specific parameter, it is also expected that ranges of 60-110 and 80-120 are also expected. Furthermore, if the minimum range values ​​are listed as 1 and 2, and if the maximum ranges are listed as 3, 4, and 5, then the following ranges are all expected: 1-2, 1-4, 1-5, 2-3, 2-4, and 2-5.

[0016] In this invention, unless otherwise specified, the numerical range "ab" represents a shortened representation of any combination of real numbers from a to b, where a and b are both real numbers. For example, the numerical range "0-5" means that all real numbers between "0-5" have been listed in this document, and "0-5" is simply a shortened representation of these numerical combinations.

[0017] Distributed Vibration Sensing (DVS): A sensing technology that uses optical fiber as a distributed sensing medium to detect changes in the phase or intensity of optical signals caused by external disturbances at various points along a line, thereby achieving real-time sensing of vibration events along the entire line. This invention employs the Φ-OTDR (phase-sensitive optical time-domain reflectometry) principle to implement the DVS function.

[0018] Φ-OTDR (Phase-sensitive Optical Time Domain Reflectometry): A fiber optic sensing technology that emits highly coherent laser pulses into an optical fiber and detects the phase change of the backscattered Rayleigh light signal to sense vibration events along the fiber, thus obtaining the precise spatial location of the vibration event.

[0019] Road Section Attribute Label: In this invention, structured description information is pre-configured for each physical segment of the fiber optic link, including physical and environmental characteristic parameters such as soil type, soil moisture content, burial depth, and type of adjacent infrastructure (such as manholes, culverts, farmland, and roads), which are used to drive subsequent adaptive parameter selection and model invocation.

[0020] Sensitivity Threshold (Ts): A parameter used to determine whether the amplitude of the vibration signal exceeds the background noise level of the current road section. Different Ts values ​​are used for different road sections due to different noise bases.

[0021] Transient Coefficient (Tc): Defined as the ratio of the standard deviation of the signal within adjacent time windows, used to quantify the suddenness of vibration events; the higher the transient coefficient, the more instantaneous the impact characteristics of the vibration event. Natural disturbances such as raindrops usually have a lower transient coefficient.

[0022] Transient Ratio (Tr): The ratio of the mutation coefficient to the local mean of the signal, used to suppress the confusion between persistent disturbances (such as wind noise, continuous vibration of agricultural machinery) and real impact events.

[0023] Acoustic Reconstruction: The signal processing procedure that restores vibration signals acquired by fiber optic sensing to an approximate audible audio waveform using an inverse filtering algorithm.

[0024] Channel Transfer Function (H(f)): describes the frequency domain characteristics of an acoustic vibration signal from the sound source through the "soil-optical fiber" propagation path to the sensor receiver, reflecting the attenuation and phase shift of each frequency component under different media; this invention performs experimental calibration for different soil types.

[0025] Scene-Specific Audio Recognition Model: A fine-grained audio classification neural network model that collects training samples and trains independently for specific road scenes (such as farmland, urban roads, and areas near manholes). It can output a probability distribution covering the typical sound categories of the scene and has a stronger scene discrimination ability compared to general models.

[0026] Harm determination rule engine: Based on the event type output by the audio recognition model and the current road segment scene category, the logic processing module makes structured decisions according to the preset "event type × scene → hazard" mapping rule table, realizing the context awareness capability of "same sound, different scenes, differentiated judgment".

[0027] Mel spectrogram: A time-frequency representation obtained by mapping the short-time Fourier transform of an audio signal onto a Mel frequency scale. It is closer to the characteristics of human auditory perception and is widely used as an input feature for audio recognition models.

[0028] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides: a fiber optic audio restoration early warning method based on road segment attribute adaptation, comprising the following steps: S1. Collect vibration signals from the fiber optic link; S2. When an abnormal vibration signal V(t) is detected at location x of the optical link, query and obtain the road segment attributes of location x; S3. Based on the road segment attributes obtained in step S2, select the threshold parameter of the corresponding road segment attribute, and use the threshold parameter to filter the vibration signal V(t) to obtain the vibration signal V(t) that has passed the filter. S4. Perform acoustic inversion on the vibration signal V(t) selected in step S3 to generate the restored audio. S5. Based on the road segment attributes obtained in step S2, select the appropriate fine-grained audio classification model to perform fine-grained classification on the restored audio obtained in step S4, and obtain the event type of the restored audio. S6. Combining the road segment attributes obtained in step S2 and the event type obtained in step S5, determine whether the event occurring at location x is harmful. If the result of step S7 and step S6 is determined to be harmful, an alarm is triggered and the event log is retained.

[0029] Further, the acquisition in step S1 is as follows: the vibration signal of the entire length of the optical fiber is acquired in real time using distributed optical fiber sensing to generate a time-space two-dimensional vibration data matrix V(t,x).

[0030] Furthermore, the distributed optical fiber sensing employs phase-sensitive optical time-domain reflectometry (Φ-OTDR) to emit probe light pulses into the sensing optical fiber and receives backscattered Rayleigh light signals. By demodulating the phase change of the backscattered Rayleigh light signals, a time-space two-dimensional vibration data matrix V(t,x) is generated, where t is time in seconds and x is the position of the optical fiber in meters.

[0031] Furthermore, the road segment attributes mentioned in step S2 include physical and environmental attribute information corresponding to each optical fiber link segment, wherein the physical and environmental attribute information includes spatial location characteristics, medium physical characteristics, environmental condition characteristics, and scene classification characteristics. Furthermore, the spatial location features include the starting and ending positions of the optical fiber; the medium physical features include soft soil, concrete, and metal; the environmental condition features include humidity, open space, and enclosed road surface; and the scene classification features include farmland, road, and manhole.

[0032] Furthermore, the threshold parameters mentioned in step S3 include the sensitivity threshold Ts, the mutation coefficient threshold Tc, and the mutation coefficient ratio threshold Tr.

[0033] Furthermore, the threshold parameter for the corresponding road segment attribute is: When the scene classification of the road segment attribute is farmland, Ts=0.3, Tc=1.8, Tr=3.5; When the scene classification of the road segment attribute is "road", Ts=0.5, Tc=2.5, and Tr=2.0; When the scenario classification of the road segment attribute is manhole, Ts=0.4, Tc=2.2, and Tr=3.0.

[0034] Further, the screening in step S3 is as follows: when the vibration signal V(t) simultaneously meets the conditions of three threshold parameters, it is recorded as passing the screening.

[0035] Furthermore, the acoustic inversion described in step S4 is as follows: Signal restoration is performed by combining inverse filtering with spectrum mapping algorithm, and the collected vibration signal v(t) is equivalent to the output signal obtained after the sound source signal is filtered through the soil-fiber transmission channel; Based on the channel transfer function H(f) corresponding to different soil types, the sound source signal is restored using the channel transfer function to obtain the restored audio.

[0036] Furthermore, the formula for the reduction is: ; in, To reconstruct the audio, F represents the Fourier transform, and f represents the frequency.

[0037] The channel transfer function H(f) corresponding to different soil types is as follows: When the road segment is classified as farmland, H(f) = e^(-(f / 200)²). Farmland soil is loose, resulting in significant attenuation of high-frequency signals, with effective event energy primarily concentrated in the 0-200Hz low-frequency band. This channel conduction function effectively suppresses high-frequency interference above 200Hz, matching the vibration propagation characteristics of farmland soil. When the scene classification of the road segment attribute is road, H(f)=1 / (1+(f / 500)²). The road medium is hard, and vibration can be transmitted to higher frequency bands. Setting the cutoff frequency to 500Hz can preserve the low- and medium-frequency characteristic signals of vehicle traffic and construction operations, while suppressing high-frequency noise interference. When the road segment is classified as a manhole, H(f)=(f / 100)² / (1+(f / 100)²). Low-frequency interference such as water sloshing is easily generated in the manhole. Using a 100Hz high-pass transfer function can highlight the effective mid-to-high frequency signals such as human prying and hard object knocking, and filter out low-frequency interference.

[0038] Furthermore, the fine-grained audio classification model mentioned in step S5 is a lightweight convolutional neural network (such as MobileNetV2), Transformer, or a hybrid model, with the audio Mel spectrogram as the input and the event category probability distribution as the output.

[0039] Furthermore, the fine-grained audio classification model includes a farmland-specific model, a road-specific model, and a hand well enhancement model; The training data for the farmland-specific model includes at least one of the following: tractor, harvester, drilling and piling, wind sound, bird song, and human voice. The training data for the road-specific model includes at least one of the following: excavator, vehicle rolling, braking, construction drilling, and pedestrian footsteps. The training data for the hand well enhancement model includes at least one of the following: raindrops hitting the metal cover, water sloshing, and human prying.

[0040] Furthermore, the event log described in step S7 includes at least one of the following: the time, location, audio clip, and event category of the event.

[0041] Based on further solutions to the technical problems of the present invention, or simultaneous solutions to multiple technical problems, the preferred solution in the technical solution provided in the first aspect of the present invention includes: The first preferred solution involves setting threshold parameters for different road segment attributes to better align the initial screening with local environmental characteristics and effectively suppress non-threatening vibrations. This technical solution further reduces the false alarm rate while already lowering the overall false alarm rate.

[0042] The second preferred solution is to set up dedicated models for different scenario classifications to achieve finer granularity and more accurate judgment. This technical solution further reduces the false alarm rate on the basis of reducing the false alarm rate.

[0043] Secondly, the present invention provides: a fiber optic audio restoration early warning system based on road segment attribute adaptation, comprising: a distributed fiber optic sensing unit, a road segment attribute configuration storage module, an adaptive pattern recognition parameter calculation module, a fiber optic sound restoration module, a scene-specific audio recognition model library, an audio type-hazard judgment engine, and an alarm execution unit. The fiber optic sound restoration module, the scene-specific audio recognition model library, the audio type-hazard judgment engine, and the alarm execution unit are connected in sequence; the road segment attribute configuration storage module is connected to the adaptive pattern recognition parameter calculation module and the scene-specific audio recognition model library.

[0044] Furthermore, the distributed optical fiber sensing unit is used to collect vibration signals along the entire length of the optical fiber in real time.

[0045] Furthermore, the road segment attribute configuration storage module is used to store the road segment attributes corresponding to each fiber optic link, providing a basis for the selection of the adaptive pattern recognition parameter calculation module and the scene-specific audio recognition model library.

[0046] Furthermore, the adaptive pattern recognition parameter calculation module is used to store threshold parameters for different road segment attributes and to filter the vibration signal at the event location x.

[0047] Furthermore, the filtering is as follows: matching the threshold parameter of the corresponding road segment attribute according to the road segment attribute of the event location x, and filtering the vibration signal V(t) at the location x according to the threshold parameter.

[0048] Furthermore, the threshold parameters include a sensitivity threshold Ts, a mutation coefficient threshold Tc, and a mutation coefficient ratio threshold Tr.

[0049] Furthermore, the fiber optic sound restoration module is used to perform acoustic inversion on the screened vibration signal V(t) to generate restored audio.

[0050] Furthermore, the scene-specific audio recognition model library is used to store fine-grained audio classification models trained independently for specific scenes and to identify the event types of the reconstructed audio.

[0051] Furthermore, the identification method is as follows: Select the appropriate fine-grained audio classification model based on the road segment attributes of the location x where the event occurred; The fine-grained audio classification model is used to perform fine-grained classification on the reconstructed audio to obtain the event type of the reconstructed audio.

[0052] Furthermore, the audio type-hazard assessment engine is used to determine whether an event is harmful.

[0053] Furthermore, the method for making the judgment is as follows: combining the road segment attributes and event type of location x, it is determined whether the event occurring at location x is harmful.

[0054] Furthermore, the alarm execution unit is used to receive alarm signals and execute alarm actions.

[0055] Furthermore, the alarm execution unit is a multi-level alarm.

[0056] Furthermore, the alarm execution unit outputs digital signals to an audible and visual alarm, an SMS push module, a video linkage interface, or a SCADA system, and records the event time, location, audio clips, and event category.

[0057] Furthermore, the connection is a data bus or communication interface connection.

[0058] Thirdly, the present invention provides: a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the fiber optic audio restoration warning method.

[0059] Fourthly, the present invention provides: a computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the fiber optic audio restoration warning method.

[0060] The beneficial effects of this invention are as follows: 1. Achieve parameter-level self-adaptation: Dynamically configure the optimal pattern recognition parameters (including sensitivity, mutation coefficient, mutation coefficient ratio, etc.) according to the road segment attributes corresponding to each optical fiber segment, so that the initial screening and filtering process is accurately matched with the local physical environment.

[0061] 2. Achieve model-level personalization: Bind exclusive fine-grained audio recognition models (such as farmland model, road model, and manhole enhancement model) to different road segment attributes, so that the semantic classification of the restored audio fully considers the scene context and effectively distinguishes between "normal interference" and "real threat".

[0062] 3. Construct a closed-loop collaborative mechanism of "attribute-parameter-model": By pre-setting the mapping relationship between road segment attributes and processing strategies, the entire process from signal acquisition, initial screening, sound reconstruction to event discrimination can be adaptive, maintaining high accuracy in heterogeneous environments without manual intervention.

[0063] 4. Significantly reduce false alarm rate and improve system robustness and engineering applicability: In complex and ever-changing actual deployment scenarios, it can effectively suppress typical interferences such as wind, rain, animals, and agricultural machinery, and reliably detect harmful events such as excavation, drilling, and climbing, truly achieving the intelligent early warning goal of "low false alarm, zero false alarm, and strong adaptability".

[0064] 5. In summary, this invention overcomes the inherent limitations of the traditional "one-size-fits-all" approach in fiber optic early warning systems. For the first time, it deeply integrates geographic / physical context awareness into the entire process of fiber optic sound early warning, providing key technical support for the large-scale application of distributed fiber optic sensing in the field of high-reliability security monitoring. Attached Figure Description

[0065] Figure 1 This is a block diagram of a fiber optic audio early warning system based on road segment attribute adaptation. Detailed Implementation

[0066] The following non-limiting embodiments are intended to enable those skilled in the art to gain a more comprehensive understanding of the present invention, but do not limit the invention in any way. The following content is merely an exemplary description of the scope of protection claimed by the present invention, and those skilled in the art can make various changes and modifications to the present invention based on the disclosed content, and such changes should also fall within the scope of protection claimed by the present invention.

[0067] The present invention will be further described below by way of specific embodiments. Unless otherwise specified, all instruments, devices, equipment, reagents, products, etc., used in the embodiments of the present invention are obtained through conventional commercial means.

[0068] Example 1 A fiber optic audio reconstruction early warning method based on road segment attribute adaptation includes the following steps: S1. Acquire vibration signals of the fiber optic link: Specifically, use distributed fiber optic sensing to collect vibration signals of the entire length of the fiber in real time, and generate a time-space two-dimensional vibration data matrix V(t,x), where t is time in seconds and x is the position of the fiber in meters.

[0069] S2. When an abnormal vibration signal V(t) is detected at location x of the optical link, query and obtain the segment attributes of location x: The road segment attributes include physical and environmental attribute information corresponding to each fiber optic link segment, wherein the physical and environmental attribute information includes spatial location characteristics, medium physical characteristics, environmental condition characteristics, and scene classification characteristics.

[0070] S3. Based on the road segment attributes obtained in step S2, select the threshold parameters (Ts, Tc, Tr) of the corresponding road segment attributes, and use the threshold parameters to filter the vibration signal V(t). When the vibration signal V(t) simultaneously meets the conditions of the three threshold parameters, it is recorded as passing the filter.

[0071] S4. Perform acoustic inversion on the vibration signal V(t) selected in step S3 to generate the restored audio.

[0072] S5. Based on the road segment attributes obtained in step S2, select the appropriate fine-grained audio classification model to perform fine-grained classification on the restored audio obtained in step S4, and obtain the event type and confidence level of the restored audio. The fine-grained audio classification model is a lightweight convolutional neural network, Transformer, or hybrid model, and the input is the audio Mel-spectrum.

[0073] S6. Combining the road segment attributes obtained in step S2 and the event type obtained in step S5, determine whether the event occurring at location x is harmful.

[0074] If the result of step S7 and step S6 is determined to be harmful, an alarm is triggered, and an event log is retained (e.g., the time, location, audio clip, and event category of the event).

[0075] In some preferred embodiments, the distributed optical fiber sensing in step S1 involves emitting probe light pulses into the sensing optical fiber using phase-sensitive optical time-domain reflectometry (Φ-OTDR) and receiving backscattered Rayleigh light signals. By demodulating the phase change of the backscattered Rayleigh light signals, a time-space two-dimensional vibration data matrix V(t,x) is generated, where t is time in seconds and x is the position of the optical fiber in meters.

[0076] In some preferred embodiments, the spatial location features in step S2 include the starting and ending positions of the optical fiber; the medium physical features include soft soil, concrete, and metal; the environmental condition features include humidity, open space, and enclosed road surface; and the scene classification features include farmland, road, and manhole.

[0077] For example, taking a section of optical fiber that crosses farmland, roads, and hand wells as an example, the format of its segment attributes can be shown in Table 1 below.

[0078] Table 1

[0079] In some preferred embodiments, the filtering in step S3 is as follows: based on the scene classification in the road segment attributes obtained in step S2, the corresponding sensitivity threshold Ts, mutation coefficient threshold Tc, and mutation coefficient ratio threshold Tr are selected to filter the vibration signal V(t).

[0080] For example, the filtering can be: (1) Calculate the average amplitude A of the vibration signal: Take the average amplitude of the vibration signal V(t) within the current time window; If A≥Ts, the sensitivity threshold requirement is met; otherwise, it is directly judged as interference. (2) Calculate the mutation coefficient Tc_calc: Tc_calc = current signal standard deviation / previous window signal standard deviation; If Tc_calc≥Tc, it is determined that the mutation coefficient threshold requirement is met; otherwise, it is directly judged as interference. (3) Calculate the mutation coefficient ratio Tr_calc: Tr_calc = Tc_calc / local mean of the current window signal; If Tr_calc≥Tr, then the mutation coefficient ratio threshold requirement is met; otherwise, it is directly judged as interference.

[0081] In some preferred embodiments, the threshold parameter for the corresponding road segment attribute in step S3 is: When the scene classification of the road segment attribute is farmland, Ts=0.3, Tc=1.8, Tr=3.5; When the scene classification of the road segment attribute is "road", Ts=0.5, Tc=2.5, and Tr=2.0; When the scenario classification of the road segment attribute is manhole, Ts=0.4, Tc=2.2, and Tr=3.0.

[0082] In some preferred embodiments, the acoustic inversion in step S4 is as follows: using inverse filtering combined with a spectrum mapping algorithm, the vibration signal v(t) is regarded as the output signal of the sound source after being filtered by the "soil-fiber" channel; Using a pre-calibrated channel transfer function H(f) (calibrated separately for different soil types), the audio can be reconstructed using the following formula: ; Where F represents the Fourier transform, and f is the frequency. To reproduce the audio, a standard audio format (such as 16kHz PCM) is used for subsequent recognition.

[0083] Preferably, the channel transfer function H(f) is calibrated for different soil types, specifically as follows: When the road segment is classified as farmland, H(f) = e^(-(f / 200)²). Farmland soil is loose, resulting in significant attenuation of high-frequency signals, with effective event energy primarily concentrated in the 0-200Hz low-frequency band. This channel conduction function effectively suppresses high-frequency interference above 200Hz, matching the vibration propagation characteristics of farmland soil. When the scene classification of the road segment attribute is road, H(f)=1 / (1+(f / 500)²). The road medium is hard, and vibration can be transmitted to higher frequency bands. Setting the cutoff frequency to 500Hz can preserve the low- and medium-frequency characteristic signals of vehicle traffic and construction operations, while suppressing high-frequency noise interference. When the road segment is classified as a manhole, H(f)=(f / 100)² / (1+(f / 100)²). Low-frequency interference such as water sloshing is easily generated in the manhole. Using a 100Hz high-pass transfer function can highlight the effective mid-to-high frequency signals such as human prying and hard object knocking, and filter out low-frequency interference.

[0084] In some preferred embodiments, the fine-grained audio classification model in step S5 is a fine-grained audio classification model trained for a specific scene category.

[0085] The fine-grained audio classification model includes a farmland-specific model, a road-specific model, and a hand well enhancement model.

[0086] The training data for the farmland-specific model includes tractors, harvesters, drilling and piling, wind sounds, birdsong, and human voices. The training data for the road-specific model includes excavator data, vehicle traffic data, braking data, construction drilling data, and pedestrian footsteps. The training data for the enhanced manhole model includes raindrops hitting the metal cover, water sloshing, and human prying.

[0087] In some preferred embodiments, the method for determining whether an event occurring at location x is harmful in step S6 is exemplified as follows: If scene category = "hand well" and event type = "raindrop", the output is harmless; If Scene Category = "Hand Well" and Event Type = "Human Tampering", output harmful; If scene category = "farmland" and event type = "tractor", the output is harmless; If Scene Category = "Farmland" and Event Type = "Drilling and Piling", output harmful. If scene category = "road" and event type = "brake", output is harmless; If Scene Category = "Road" and Event Type = "Excavator", output is harmful.

[0088] Example 2 The fiber optic audio early warning system based on road segment attribute adaptation includes: a distributed fiber optic sensing unit, a road segment attribute configuration storage module, an adaptive pattern recognition parameter calculation module, a fiber optic sound restoration module, a scene-specific audio recognition model library, an audio type-hazard judgment engine, and an alarm execution unit. The fiber optic sound restoration module, the scene-specific audio recognition model library, the audio type-hazard judgment engine, and the alarm execution unit are connected in sequence; the road segment attribute configuration storage module is connected to the adaptive pattern recognition parameter calculation module and the scene-specific audio recognition model library, such as... Figure 1 As shown.

[0089] Specifically, (1) Distributed fiber optic sensing unit Function: To acquire vibration signals along the entire length of the optical fiber in real time; Implementation method: Φ-OTDR (phase-sensitive optical time-domain reflectometry) is used to transmit probe light pulses into the sensing fiber and receive backscattered Rayleigh light signals; by demodulating the phase change of the backscattered Rayleigh light signals, a time-space two-dimensional vibration data matrix V(t,x) is generated, where t is time in seconds and x is the fiber position in meters. Output: The original vibration waveform sequence is output in segments according to a preset spatial resolution (e.g., 1 meter / point).

[0090] (2) Road segment attribute configuration storage module Function: Stores the segment attributes corresponding to each fiber optic link, providing a basis for the selection of the adaptive pattern recognition parameter calculation module and the scene-specific audio recognition model library; Implementation method: It is composed of non-volatile memory (such as Flash or EEPROM); Configuration method: During the system deployment phase, engineers input the configuration data via host computer software or through automatic import via the GIS system.

[0091] Output: Road segment attribute table, with an example format as shown in Table 1 above.

[0092] (3) Adaptive pattern recognition parameter calculation module Function: Used to store threshold parameters for different road segment attributes, and to filter vibration signals at the location x where the event occurs based on the threshold parameters for different road segment attributes.

[0093] (4) Fiber optic sound restoration module Function: Used to perform acoustic inversion on the screened vibration signal V(t) to generate the restored audio; (5) Scene-specific audio recognition model library Function: Used to store fine-grained audio classification models trained independently for specific scenarios and to identify the event types that reproduce audio. Implementation method: Implemented by embedded AI acceleration chips (such as NPU) or server-side model repository.

[0094] (6) Audio type - hazard assessment engine Function: Combines road segment attributes and event type to determine whether an event is harmful.

[0095] (7) Alarm Execution Unit Function: To trigger an alarm when an event is determined to be harmful; Implementation method: Output digital signals to the sound and light alarm, SMS push module, video linkage interface or SCADA system, and record the event time, location, audio segment and event type label for later traceability.

[0096] Example 3 The difference between this embodiment and Embodiment 2 lies in the method of obtaining road segment attribute information, specifically: Using a GIS geographic information system or digital twin platform, attribute data such as geological layers, soil types, and underground facilities can be automatically matched based on the coordinates of the fiber optic cable laying.

[0097] Example 4 The difference between this embodiment and Embodiment 2 lies in the method of obtaining road segment attribute information, specifically: The road segment attributes are modeled as continuous functions (such as interpolation curves showing how water content changes with location), and the attribute values ​​at the current location are calculated in real time during runtime.

[0098] Example 5 The difference between this embodiment and Embodiment 2 lies in the method of obtaining road segment attribute information, specifically: Historical vibration signals are segmented and clustered using an unsupervised clustering algorithm, automatically generating "behavioral similarity segments" and assigning them implicit attribute labels, without the need for manual intervention.

[0099] Example 6 The difference between this embodiment and embodiment 2 is that the method for obtaining the threshold parameter is as follows: a lightweight neural network is used, with road segment attributes as input, to dynamically output the optimal parameter combination.

[0100] Example 7 The difference between this invention and Embodiment 2 is that the method for obtaining the threshold parameter is as follows: construct a composite scoring index (such as weighted energy + mutation + stability), set only a single threshold, and adjust the weight according to the road segment attributes.

[0101] Example 8 The difference between this embodiment and embodiment 2 is that the method for obtaining the threshold parameter is as follows: an online adaptive mechanism is adopted: the threshold is dynamically updated based on background noise statistics, but the adjustment range is constrained by road segment attributes (such as soft soil segments allowing greater fluctuations).

[0102] Example 9 The difference between this embodiment and Embodiment 2 is that the fine-grained audio classification model is a Mixture of Experts model: it shares a bottom-level feature extraction network, and the top-level network is activated by the corresponding expert sub-network based on road segment attributes.

[0103] Example 10 The difference between this embodiment and Embodiment 2 is that the fine-grained audio classification model is a multi-model soft fusion: multiple scene types of fine-grained audio classification models are run simultaneously, and the classification results are fused by weighting the similarity between the current road segment and each scene classification.

[0104] Example 11 The difference between this embodiment and Embodiment 2 is that the fine-grained audio classification model is dynamically loaded at the edge: multiple lightweight models are pre-stored on the embedded device and loaded from the memory to the main memory as needed during runtime.

[0105] Example 12 The difference between this embodiment and embodiment 2 is that the restoration is: a time-domain deconvolution method, if the channel impulse response h(t) is known, the original audio is restored by deconvolution; a data-driven deep learning model (such as U-Net, WaveNet) directly maps the vibration signal to audio; a conditional generation model: "soil type" or "road segment attribute" is introduced as a conditional input in the AI ​​restoration network to achieve single-model multi-scenario adaptation.

[0106] Example 13 The difference between this embodiment and embodiment 2 is that the method for determining whether an event is harmful is: a machine learning scoring model: input audio type, confidence level and road segment attributes, output hazard probability, and when the hazard probability is greater than 80%, it is judged as harmful.

[0107] Example 14 The difference between this embodiment and Embodiment 2 is that the fiber optic audio restoration warning system is a centralized single-cable system.

[0108] Example 15 The difference between this embodiment and Embodiment 2 is that the fiber optic audio restoration early warning system is a distributed edge computing architecture: the initial screening at the front end is completed at the edge node, and the back-end identification is executed at the central server; multi-core fiber or multi-cable fusion: different fiber cores are used to sense multi-dimensional physical quantities to assist in attribute judgment; multi-source sensor fusion: when the audio confidence is low, external devices such as video and weather stations are linked for cross-verification.

[0109] Verification of technical effectiveness and / or analysis of technical problem solving The present invention achieves the following through the above technical solution: Parameter adaptation: Optimal initial screening parameters are used for different road sections; Model personalization: Dedicated audio models accurately identify scene-specific events; Context-aware decision-making: Judging the degree of harm by combining both "where" and "what sound" information; Fully automated process: It can operate stably in complex environments without human intervention.

[0110] Compared with existing technologies, this invention solves the core problem of high false alarm rate in existing technologies.

[0111] Finally, it should be noted that the above content is only used to illustrate the technical solution of the present invention, and is not intended to limit the scope of protection of the present invention. Simple modifications or equivalent substitutions made by those skilled in the art to the technical solution of the present invention do not depart from the essence and scope of the technical solution of the present invention.

Claims

1. A fiber optic audio reconstruction early warning method based on road segment attribute adaptation, characterized in that, Includes the following steps: S1. Collect vibration signals from the fiber optic link; S2. When an abnormal vibration signal V(t) is detected at location x of the optical link, query and obtain the road segment attributes of location x; S3. Based on the road segment attributes obtained in step S2, select the threshold parameter of the corresponding road segment attribute, and use the threshold parameter to filter the vibration signal V(t) to obtain the vibration signal V(t) that has passed the filter. S4. Perform acoustic inversion on the vibration signal V(t) selected in step S3 to generate the restored audio. S5. Based on the road segment attributes obtained in step S2, select the appropriate fine-grained audio classification model to perform fine-grained classification on the restored audio obtained in step S4, and obtain the event type of the restored audio. S6. Combining the road segment attributes obtained in step S2 and the event type obtained in step S5, determine whether the event occurring at location x is harmful. S7. If the result of step S6 is deemed harmful, an alarm is triggered and the event log is retained. The road segment attributes mentioned in step S2 include physical and environmental attribute information corresponding to each fiber optic link segment, wherein the physical and environmental attribute information includes spatial location characteristics, medium physical characteristics, environmental condition characteristics, and scene classification characteristics. The threshold parameters mentioned in step S3 include the sensitivity threshold Ts, the mutation coefficient threshold Tc, and the mutation coefficient ratio threshold Tr; The screening in step S3 is as follows: when the vibration signal V(t) simultaneously meets the conditions of three threshold parameters, it is recorded as passing the screening.

2. The fiber optic audio restoration early warning method according to claim 1, characterized in that, The acquisition described in step S1 is as follows: the vibration signal of the entire length of the optical fiber is acquired in real time using distributed optical fiber sensing, and a time-space two-dimensional vibration data matrix V(t,x) is generated, where t is time in seconds and x is the position of the optical fiber in meters.

3. The fiber optic audio restoration early warning method according to claim 1, characterized in that, The spatial location features include the starting and ending positions of the optical fiber; the medium physical features include soft soil, concrete, and metal; the environmental condition features include humidity, open air, and enclosed road surface; and the scene classification features include farmland, roads, and manholes.

4. The fiber optic audio restoration early warning method according to claim 1, characterized in that, The threshold parameter for the corresponding road segment attribute is: When the scene classification of the road segment attribute is farmland, Ts=0.3, Tc=1.8, Tr=3.5; When the scene classification of the road segment attribute is "road", Ts=0.5, Tc=2.5, and Tr=2.0; When the scenario classification of the road segment attribute is manhole, Ts=0.4, Tc=2.2, and Tr=3.

0.

5. The fiber optic audio restoration early warning method according to claim 1, characterized in that, The acoustic inversion described in step S4 is as follows: Signal restoration is performed by combining inverse filtering with spectrum mapping algorithm, and the collected vibration signal v(t) is equivalent to the output signal obtained after the sound source signal is filtered through the soil-fiber transmission channel; Based on the channel transfer function H(f) corresponding to different soil types, the sound source signal is restored using the channel transfer function to obtain the restored audio. The formula for the reduction is: ; in, To reconstruct the audio, F represents the Fourier transform, and f represents the frequency.

6. The fiber optic audio restoration early warning method according to claim 5, characterized in that, The channel transfer function H(f) corresponding to the different soil types is specifically as follows: When the scene classification of the road segment attribute is farmland, H(f) = e^(-(f / 200)²); When the scene classification of the road segment attribute is road, H(f) = 1 / (1 + (f / 500)²); When the scenario classification of the road segment attribute is a hand well, H(f) = (f / 100)² / (1+(f / 100)²).

7. The fiber optic audio restoration early warning method according to claim 1, characterized in that, The fine-grained audio classification model mentioned in step S5 is a lightweight convolutional neural network, Transformer, or hybrid model, and the input is the audio Mel-spectrum. And / or, the fine-grained audio classification model includes a farmland-specific model, a road-specific model, and a hand well enhancement model; The training data for the farmland-specific model includes at least one of the following: tractor, harvester, drilling and piling, wind sound, bird song, and human voice. The training data for the road-specific model includes at least one of the following: excavator, vehicle rolling, braking, construction drilling, and pedestrian footsteps. The training data for the hand well enhancement model includes at least one of the following: raindrops hitting the metal cover, water sloshing, and human prying.

8. A fiber optic audio restoration early warning system corresponding to the fiber optic audio restoration early warning method according to any one of claims 1-7, characterized in that, It includes a distributed fiber optic sensing unit, a road segment attribute configuration storage module, an adaptive pattern recognition parameter calculation module, a fiber optic sound restoration module, a scene-specific audio recognition model library, an audio type-hazard judgment engine, and an alarm execution unit; The fiber optic sound restoration module, the scene-specific audio recognition model library, the audio type-hazard judgment engine, and the alarm execution unit are connected in sequence; the road segment attribute configuration storage module is connected to the adaptive pattern recognition parameter calculation module and the scene-specific audio recognition model library; The distributed optical fiber sensing unit is used to collect vibration signals along the entire length of the optical fiber in real time. The road segment attribute configuration storage module is used to store the road segment attributes corresponding to each fiber optic link. The adaptive pattern recognition parameter calculation module is used to store threshold parameters for different road segment attributes and to filter vibration signals at the event location x. The fiber optic sound restoration module is used to perform acoustic inversion on the screened vibration signal V(t) to generate restored audio. The scene-specific audio recognition model library is used to store fine-grained audio classification models trained independently for specific scenes and to identify and reconstruct the event types of audio. The audio type-hazard assessment engine is used to determine whether an event is harmful. The alarm execution unit is used to receive alarm signals and execute alarm actions.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the fiber optic audio restoration warning method according to any one of claims 1-7.

10. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the fiber optic audio restoration warning method according to any one of claims 1-7.

Images