A method and system for identifying underground structure boundaries based on distributed optical fiber sensing

Abstract

This invention provides a method and system for identifying underground structure boundaries based on distributed optical fiber sensing, relating to the field of underground structure detection and boundary identification technology. The invention first deploys a sensor array in the area to be measured and establishes a correspondence between channels and spatial coordinates, acquiring an unexcited baseline; through low-energy probing to estimate the medium and noise, the excitation frequency band, energy, and repetition rate are optimized, emitted by an adaptive seismic source; reflection and transmission responses are synchronously acquired under a unified time reference; wavelet denoising, temperature and dispersion compensation, and time alignment are applied to the data to extract amplitude, phase, and frequency features; gradients are calculated on the feature field and non-maximum suppression is performed; a stable boundary point is obtained by employing an adaptive double threshold based on local statistics combined with temporal continuity and spatial connectivity constraints; a two-dimensional profile is obtained by spline fitting, and a three-dimensional model is reconstructed under multi-profile consistency constraints, outputting the profile diagram and the three-dimensional model, achieving high-precision, real-time, and visualized detection of small-sized boundaries.

Description

Technical Field

[0001] This invention relates to the field of underground structure detection and boundary identification technology, and in particular to an underground structure boundary identification method and system based on distributed optical fiber sensing. Background Technology

[0002] Boundary identification and condition assessment of underground structures (such as tunnels, utility tunnels, foundations, and underground retaining structures) have long relied on a combination of methods, including ground-penetrating radar, resistivity measurement, seismic / vibration surveys, and a limited number of borehole verifications. While these technologies have gradually established a relatively complete engineering process, they still face some common challenges: insufficient spatial continuity, resolution and noise immunity limited by operating conditions, significant disturbance to existing structures, and difficulty in achieving continuous online monitoring.

[0003] Distributed fiber optic sensing technology, with its advantages of long-distance continuous sampling, suitability for embedded deployment in engineering projects, and long-term operation, provides a stable sensing platform and abundant data resources for non-destructive and sustainable identification of underground structure boundaries. Engineering applications are gradually shifting from "single detection + offline analysis" to "continuous monitoring + real-time identification + 3D visualization." Constructing an end-to-end, closed-loop identification process, encompassing the front-end deployment and excitation control of distributed fiber optic sensing arrays, back-end data spatiotemporal synchronization, compensation for environmental factors (such as noise and temperature), statistical analysis and boundary extraction of prior information, as well as multi-view surface reconstruction and reliability verification, has become an important direction for technological development. This direction emphasizes system robustness and reusability in complex media and multi-condition scenarios, and places higher demands on the standardization and automation of methods.

[0004] Current solutions mostly optimize only one aspect of sensing, signal processing, or geometric reconstruction, lacking deep integration of the entire process: "adaptive excitation – quality control and compensation – statistical threshold adaptive boundary detection – connectivity constraint – spline or surface reconstruction – consistency verification." This results in insufficient boundary stability and interpretability under conditions of high noise levels, non-uniform media, or variable operating conditions; consistency constraints across multiple profiles in three dimensions and time are also not fully established, leaving significant room for improvement in accuracy and reliability in engineering applications. Summary of the Invention

[0005] To overcome the shortcomings of existing technologies, the purpose of this invention is to provide a method and system for identifying underground structure boundaries based on distributed optical fiber sensing. This method can achieve high-precision dynamic detection of small-sized boundaries without damaging the underground structure. By combining adaptive thresholds, time and space constraints, and three-dimensional consistency modeling, it not only significantly improves detection resolution and real-time performance, but also makes the boundary identification process intelligent, visualized, and engineered.

[0006] To achieve the above objectives, the present invention provides the following solution:

[0007] A method for identifying underground structure boundaries based on distributed optical fiber sensing, comprising:

[0008] A distributed fiber optic sensor array is deployed in the area to be tested to establish the correspondence between the channel and the spatial coordinates, and the baseline response without external excitation is collected as reference data for compensation and comparison.

[0009] First, low-energy trial sampling is performed to estimate the transmission characteristics and noise spectrum of the medium in the target frequency band. Then, the center frequency, bandwidth, energy and repetition rate of the excitation are optimized within the equipment power and safety boundary. The excitation is then issued by the adaptive source controller.

[0010] Under a unified time reference, a distributed fiber optic sensing array and a real-time signal processor synchronously acquire the seismic wave reflection and transmission responses caused by the excitation, and obtain synchronous reflection response data with channel numbers and timestamps.

[0011] Based on the synchronous reflection response data, wavelet denoising, temperature compensation based on the baseline response, dispersion compensation and time alignment are performed to extract amplitude, phase and frequency characteristics and construct a spatiotemporal-channel feature representation.

[0012] The gradient is calculated on the feature representation and non-maximum suppression is performed. A dual-threshold edge detection method, in which the high and low thresholds are adaptively determined by local statistics, is used to extract candidate boundary regions and candidate boundary points.

[0013] Temporal continuity and spatial connectivity constraints are applied to the candidate boundary points to obtain a stable set of boundary points; wherein the candidate boundary regions are subsets to be refined by threshold filtering on the feature representation, and maintain a one-to-one correspondence with channel-space coordinates;

[0014] The stable boundary point set is spline fitted to obtain a two-dimensional profile, and a three-dimensional boundary model is reconstructed under multi-profile consistency constraints, outputting profile views and three-dimensional model files that are compatible with engineering interpretation.

[0015] Preferably, a distributed fiber optic sensor array is deployed in the area to be tested to establish the correspondence between channels and spatial coordinates, and baseline responses without external excitation are collected as reference data for compensation and comparison, including:

[0016] Determine the spatial boundary of the area to be measured, establish a three-dimensional coordinate system and layout path consistent with the engineering benchmark, and determine the set of benchmark points used for positioning;

[0017] A distributed optical fiber sensor array is laid along the deployment path to complete the fixing and turning radius control. Identifiable markers are set at each reference point to record the correspondence between optical fiber mileage and channel number.

[0018] The distributed optical fiber sensor array is connected to the distributed optical fiber demodulator to complete the link connectivity and loss check, establish a channel-space coordinate mapping table based on the reference point coordinates and mileage data, and calibrate the propagation delay of each channel under a unified time reference.

[0019] Static responses are continuously acquired under conditions of external excitation being off and environmental stability. The mean and variance of amplitude for each channel are calculated to generate baseline responses and statistics, which are then archived by timestamp as reference data for compensation and threshold determination.

[0020] Preferably, low-energy trial sampling is first performed to estimate the transmission characteristics and noise spectrum of the medium in the target frequency band, and the center frequency, bandwidth, energy, and repetition rate of the excitation are optimized within the equipment power and safety boundaries. The excitation is then issued by an adaptive source controller, including:

[0021] Under a unified time reference, an adaptive source controller is configured to transmit low-energy linear sweep frequency signals or short pulse trains, record start and end frequencies, pulse width, energy and repetition rate parameters, and a distributed fiber optic demodulator provides a common trigger to achieve excitation-acquisition synchronization, thereby obtaining experimental sampling synchronous reflection response data with channel number and timestamp.

[0022] Based on the channel-space coordinate mapping and the synchronous reflection response data of the trial sampling, the amplitude-frequency response and group delay in the target frequency band are calculated, the noise power spectral density and frequency-resolved signal-to-noise ratio are estimated, and the medium transmission characteristics and noise statistics of the test area are formed for excitation design.

[0023] With the goal of improving the target frequency band identification sensitivity and signal-to-noise ratio, under the constraints of the upper limit of energy, the upper limit of peak power and the upper limit of repetition rate determined by the equipment specifications and on-site safety regulations, the optimized parameter set of center frequency, bandwidth, energy and repetition rate is obtained, and the corresponding adaptive excitation configuration is generated.

[0024] The optimized parameters are sent to the adaptive source controller, which transmits the adaptive excitation. The distributed fiber optic demodulator performs trigger verification and first-cycle alignment check. After confirming that the excitation-acquisition time synchronization meets the preset error threshold, synchronous reflection response data is acquired.

[0025] Preferably, based on the synchronous reflection response data, wavelet denoising, temperature compensation based on the baseline response, dispersion compensation and time alignment are performed to extract amplitude, phase and frequency representations, and construct a spatiotemporal-channel feature representation, including:

[0026] For the synchronous reflection response data, wavelet decomposition is performed by channel and thresholds are determined based on noise statistics to complete denoising;

[0027] The signal-to-noise ratio of the channel segment and the trigger alignment residual are used to form a quality label, and low-quality segments are eliminated or downweighted.

[0028] Based on the baseline response statistics, a channel temperature drift model and gain correction coefficient are established. Temperature compensation and gain normalization are then applied to the denoised channel signals to obtain a compensated set of channel signals.

[0029] By combining the channel propagation delay parameter with the timestamp written during the acquisition process with the common trigger, the compensated channel signal set is subjected to group delay correction and inter-channel time alignment, and a channel signal set with consistent timing is output.

[0030] Amplitude, phase, and frequency representations are extracted from a time-consistent channel signal set and bound to channel-space coordinate mapping, organized into a time-space-channel feature representation according to time, space, and channel dimensions.

[0031] Preferably, gradients are calculated on the feature representation and non-maximum suppression is performed. A dual-threshold edge detection method, where high and low thresholds are adaptively determined by local statistics, is employed to extract candidate boundary regions and candidate boundary points, including:

[0032] The gradient magnitude is calculated along the spatial dimension on the feature representation. With gradient direction And maintain a one-to-one correspondence with the channel-space coordinate mapping to obtain a gradient field organized by time slice and channel index;

[0033] Calculate the local mean using a sliding window on the gradient field. with standard deviation in accordance with and Determine the high threshold With low threshold This forms a threshold pair for dual-threshold discrimination; where This is a preset constant;

[0034] Non-maximum suppression is performed on the gradient magnitude along the gradient direction to obtain candidate edge pixels;

[0035] by Mark strong edge seeds, to to The interval is marked as a weak edge, and the weak edge is connected to the strong edge according to the connectivity rule. The connected weak edges are retained and the isolated weak edges are removed, and a stable set of boundary points is output.

[0036] Connectivity aggregation is performed based on a stable set of boundary points to generate candidate boundary regions. Each candidate boundary region is then bound to its corresponding time slice, channel number, and spatial coordinates. The set of boundary points is used as the candidate boundary points.

[0037] Preferably, temporal continuity and spatial connectivity constraints are applied to the candidate boundary points to obtain a stable set of boundary points; wherein the candidate boundary regions are subsets to be refined through threshold filtering on the feature representation, and maintain a one-to-one correspondence with channel-space coordinates, including:

[0038] Within adjacent time slices, with a preset spatial search radius, the closest corresponding point with a directional difference not exceeding a preset limit is found around each candidate boundary point of the previous time slice to establish a time trajectory. The time trajectories are then filtered out according to the preset minimum trajectory length and maximum average speed, retaining only the trajectories that simultaneously meet both conditions.

[0039] Within each time slice, candidate boundary points are aggregated into connected components according to eight neighborhoods, connected components with an area smaller than a preset threshold are deleted, and a closing operation is performed to fill holes with an area not exceeding the preset threshold, thus obtaining a set of spatial connected components.

[0040] With time trajectory and spatial connectivity as dual constraints, only candidate boundary points that belong to both time trajectory and spatial connectivity are retained. The direction difference and curvature change of the retained points in adjacent time slices are limited to preset upper limits, and a stable set of boundary points is output.

[0041] The stable set of boundary points is aggregated into connected components to generate candidate boundary regions, and each candidate boundary region is bound to its corresponding time slice, channel number and spatial coordinates.

[0042] Preferably, a two-dimensional profile is obtained by spline fitting of the stable boundary point set, and a three-dimensional boundary model is reconstructed under multi-profile consistency constraints, outputting profile views and three-dimensional model files that are compatible with engineering interpretation, including:

[0043] The stable boundary point set is sorted according to channel number and time slice, and combined with channel-space coordinate mapping, the boundary point set of each survey line direction is grouped to obtain the profile dataset divided by survey line.

[0044] Spline curve fitting is performed on the boundary points in each profile dataset, and a smoothing factor is set during the fitting process to ensure that the obtained curve has overall continuity while preserving local details, thus obtaining a two-dimensional profile contour.

[0045] Multiple two-dimensional profile contours are interpolated and fused in the global coordinate system. Based on the geometric consistency and physical rationality constraints between adjacent profiles, a three-dimensional boundary model covering the area to be measured is generated.

[0046] The reconstructed 3D boundary model and profiles are output in a standard data format to generate cross-sectional views and 3D model files that can be used for engineering interpretation.

[0047] A system for identifying the boundary of underground structures based on distributed optical fiber sensing, comprising:

[0048] The sensor array deployment and baseline response acquisition unit is used to deploy a distributed fiber optic sensor array in the area to be tested, establish the correspondence between the channel and the spatial coordinates, and acquire the baseline response without external excitation as reference data for compensation and comparison.

[0049] The adaptive excitation optimization and emission unit is used to first perform low-energy trial sampling, estimate the transmission characteristics and noise spectrum of the medium in the target area in the target frequency band, and optimize the center frequency, bandwidth, energy and repetition rate of the excitation within the equipment power and safety boundary, and the excitation is emitted by the adaptive source controller.

[0050] The seismic response synchronous acquisition unit is used to synchronously acquire the seismic wave reflection and transmission responses caused by the excitation under a unified time reference by a distributed fiber optic sensor array and a real-time signal processor, and obtain synchronous reflection response data with channel numbers and timestamps.

[0051] The response preprocessing and feature construction unit is used to perform wavelet denoising, temperature compensation based on the baseline response, dispersion compensation and time alignment based on the synchronous reflection response data, extract amplitude, phase and frequency characteristics, and construct a spatiotemporal-channel feature representation.

[0052] The candidate boundary detection unit is used to calculate the gradient on the feature representation and perform non-maximum suppression. It adopts dual-threshold edge detection, which adaptively determines the high and low thresholds by local statistics, to extract candidate boundary regions and candidate boundary points.

[0053] A stable boundary point set extraction unit is used to apply temporal continuity and spatial connectivity constraints to the candidate boundary points to obtain a stable boundary point set; wherein the candidate boundary region is a subset to be refined by threshold filtering on the feature representation, and maintains a one-to-one correspondence with the channel-space coordinates;

[0054] The boundary model fitting and 3D reconstruction unit is used to perform spline fitting on the stable boundary point set to obtain a two-dimensional profile contour, and reconstruct a three-dimensional boundary model under multi-profile consistency constraints, outputting profile views and 3D model files that are compatible with engineering interpretation.

[0055] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0056] This invention constructs a boundary detection and fitting method based on feature representation, introducing adaptive dual-threshold filtering, temporal continuity and spatial connectivity constraints, and multi-profile consistency reconstruction in the data processing stage, significantly improving boundary extraction accuracy. Compared to the limitation of traditional ground-penetrating radar or resistivity methods, which typically have a spatial resolution greater than 0.5 meters, this invention can identify and locate small-sized boundaries, meeting the high-precision requirements in underground structure detection, thus solving the problem of traditional methods' inability to capture subtle boundary changes.

[0057] While traditional borehole exploration methods can directly obtain stratigraphic information, they can damage the integrity of underground structures, increasing safety and environmental risks. This invention employs a non-contact signal acquisition and intelligent processing workflow. Through steps such as feature representation, boundary point set constraints, and profile contour reconstruction, complete boundary information can be obtained without physically damaging the strata or structure. This significantly improves the safety and environmental friendliness of the exploration, achieving truly non-destructive boundary detection.

[0058] This invention introduces a time continuity mechanism into the boundary point constraint and contour fitting process, enabling the boundary identification results to be dynamically updated with time-series data. This feature overcomes the limitation of traditional methods that can only provide static results, realizing real-time or near-real-time monitoring of underground structure boundaries, and is particularly suitable for deformation observation and risk warning during the operation of underground engineering projects.

[0059] This invention integrates intelligent algorithms such as adaptive threshold determination, non-maximum suppression, spatial connectivity constraints, and multi-profile consistency modeling, enabling a high level of automation and intelligence in the boundary identification process, from raw signal processing to 3D model output. Compared to traditional methods relying on human experience for interpretation, this invention reduces interference from subjective human factors, improves the stability and objectivity of identification results, and thus enhances the scientific rigor and decision support capabilities of underground exploration.

[0060] By generating two-dimensional profile contours and three-dimensional boundary models through spline fitting, this invention not only provides intuitive visualization results but also interfaces with engineering design software to output standardized profile diagrams and three-dimensional model files. This result can directly serve underground engineering planning, risk analysis, and construction decisions, significantly enhancing the application value of detection data and achieving an integrated closed loop from data acquisition to engineering interpretation. Attached Figure Description

[0061] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0062] Figure 1 A flowchart of the method provided in an embodiment of the present invention;

[0063] Figure 2 This is a schematic diagram of the system structure provided in an embodiment of the present invention. Detailed Implementation

[0064] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0065] The purpose of this invention is to provide a method and system for identifying underground structure boundaries based on distributed optical fiber sensing, which achieves non-destructive, high-precision, and dynamic identification of underground structure boundaries.

[0066] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0067] Figure 1 The method flowchart provided in the embodiments of the present invention is as follows: Figure 1 As shown, this invention provides a method for identifying underground structure boundaries based on distributed optical fiber sensing, comprising:

[0068] Step 100: Deploy a distributed fiber optic sensor array in the area to be tested, establish the correspondence between the channel and the spatial coordinates, and collect the baseline response without external excitation as reference data for compensation and comparison.

[0069] Step 200: First, perform low-energy trial sampling to estimate the transmission characteristics and noise spectrum of the medium in the target frequency band, and optimize the center frequency, bandwidth, energy and repetition rate of the excitation within the equipment power and safety boundary, and then issue the excitation by the adaptive source controller.

[0070] Step 300: Under a unified time reference, the distributed fiber optic sensing array and the real-time signal processor synchronously acquire the seismic wave reflection and transmission responses caused by the excitation to obtain synchronous reflection response data with channel numbers and timestamps.

[0071] Step 400: Based on the synchronous reflection response data, perform wavelet denoising, temperature compensation based on baseline response, dispersion compensation and time alignment, extract amplitude, phase and frequency representations, and construct a spatiotemporal-channel feature representation;

[0072] Step 500: Calculate the gradient on the feature representation and perform non-maximum suppression. Use dual-threshold edge detection, where the high and low thresholds are adaptively determined by local statistics, to extract candidate boundary regions and candidate boundary points.

[0073] Step 600: Apply temporal continuity and spatial connectivity constraints to the candidate boundary points to obtain a stable set of boundary points; wherein the candidate boundary regions are subsets to be refined by threshold filtering in feature representation, and maintain a one-to-one correspondence with channel-space coordinates;

[0074] Step 700: Perform spline fitting on the stable boundary point set to obtain the two-dimensional profile contour, and reconstruct the three-dimensional boundary model under the multi-profile consistency constraint, outputting the profile view and three-dimensional model file that are compatible with engineering interpretation.

[0075] Specifically, step 100 of this embodiment involves deploying a distributed fiber optic sensor array within the test area to establish a correspondence between channels and spatial coordinates. This includes: first, determining the boundary of the test area; establishing a three-dimensional coordinate system based on engineering construction benchmarks; planning the fiber optic deployment path; setting control radii at path bends to avoid excessive bending; and installing identifiable markers at several known reference points. Subsequently, the distributed fiber optic sensor array is laid along the path, and the correspondence between fiber optic mileage and channel numbers is recorded. After the fiber optic cables are fixed, this embodiment connects them to a distributed fiber optic demodulator and performs link connectivity and loss checks to confirm normal fiber optic communication. By combining the spatial coordinates of the reference points with the fiber optic mileage data, this embodiment establishes a one-to-one mapping table between channels and spatial coordinates and calibrates the propagation delay of each channel under a unified time reference.

[0076] After completing the fiber optic deployment and establishing the channel mapping table, this embodiment continuously acquires the static response signals of the distributed fiber optic sensor array under conditions of external excitation being turned off and the environment remaining stable. After demodulation processing, the mean and variance of the amplitude of each channel are calculated to generate a baseline response and corresponding statistics, used to characterize the inherent response features under conditions of no external interference. This embodiment archives and saves the baseline response and statistics according to timestamps, serving as reference data for subsequent signal compensation and threshold setting, thereby ensuring that during dynamic detection, the response caused by actual boundary deformation can be accurately distinguished from errors caused by environmental noise or system drift.

[0077] In this specification, the "channel-spatial coordinate mapping table" refers to a parameter table established by mapping fiber optic mileage to actual three-dimensional coordinates. This table is used to map the signals from each channel acquired by the distributed fiber optic sensor array to their specific locations within the area under test. The function of this mapping table is to provide the basis for signal spatial positioning, enabling subsequent detection results to not only reflect amplitude and phase changes but also to pinpoint specific spatial points, thus achieving high-precision identification and dynamic monitoring of underground structure boundaries.

[0078] Furthermore, this embodiment deploys a distributed fiber optic sensor array within the area to be measured to acquire signals from the underground structure boundaries and perform temperature compensation. The sensor array preferably employs a Brillouin scattering type distributed fiber optic sensor with a spatial resolution ranging from 0.5 to 2 meters, capable of covering various types of underground structures under different deployment patterns. Specifically, when the object to be measured is a tunnel, shaft, or circular foundation, a ring layout pattern can be adopted, forming multiple fiber optic loops with radii of 12.5 meters, 25 meters, and 50 meters to achieve 360-degree full-coverage monitoring. When the object to be measured is a pipeline or linear structure, a grid layout pattern can be adopted, forming a sensor grid within a 100-meter × 60-meter area with node spacing of 2 meters horizontally and 2 meters vertically, resulting in a total of 1581 sensor units, thereby achieving high-density, high-precision positioning capabilities.

[0079] In the signal excitation stage, this embodiment employs an adaptive source controller to generate seismic wave excitation. The source controller has an adjustable frequency range of 10 to 1000 Hz and an impact force output of no less than 50 kN. It can automatically optimize excitation parameters, including center frequency, bandwidth, energy, and repetition rate, under the constraints of equipment power and on-site safety. After the excitation signal acts on the underground structure, it forms reflected and transmitted waves. The wave signals are acquired by a distributed fiber optic sensor array and converted into response data with timestamps and channel numbers by a distributed fiber optic demodulator. This embodiment further utilizes the baseline response for temperature compensation and dispersion correction to ensure the accuracy and stability of signal processing.

[0080] In the signal processing and boundary recognition stages, this embodiment inputs the acquired response data into a real-time signal processor for hardware-accelerated preprocessing. The real-time signal processor employs a field-programmable gate array (FPGA) architecture, supporting 128 channels of parallel processing and a sampling rate of one million times per second, enabling rapid wavelet denoising, amplitude, phase, and frequency feature extraction. Subsequently, the boundary reconstruction server invokes an integrated improved Canny operator and B-spline fitting algorithm. First, gradient calculation, non-maximum suppression, and double-threshold edge detection are performed on the feature representation to extract a set of candidate boundary points. Then, stable boundary points are selected through temporal continuity and spatial connectivity constraints. Finally, B-spline fitting is used to generate a two-dimensional profile contour, and a three-dimensional boundary model is reconstructed under the condition of multi-profile consistency. This model can be directly output as the profile views and three-dimensional model files required for engineering interpretation, achieving non-destructive, high-precision, and dynamic recognition of underground structure boundaries.

[0081] In step 200 of this embodiment, low-energy trial sampling is first performed to estimate the transmission characteristics and noise level of the medium in the test area within the target frequency band. Specifically, this includes configuring an adaptive source controller to emit a low-energy linear sweep signal or short pulse train under a unified time reference, setting parameters such as start and end frequencies, pulse width, excitation energy, and repetition rate, and recording these parameters completely. To ensure time synchronization of the sampling, this embodiment utilizes a distributed fiber optic demodulator to provide a common trigger signal, ensuring strict consistency between the excitation and acquisition processes. After acquisition and demodulation by the fiber optic array, trial sampling synchronous reflection response data with channel numbers and timestamps are obtained.

[0082] After acquiring the trial sampling response data, this embodiment performs amplitude-frequency analysis and group delay calculation on the signal based on the established channel-space coordinate mapping relationship to obtain the amplitude-frequency response characteristics within the target frequency band. Simultaneously, a power spectral density estimation method is used to obtain noise statistics, and the signal-to-noise ratio under frequency resolution conditions is further calculated. Through the above analysis, this embodiment can form complete dielectric transport characteristics and noise statistical parameters of the test area, providing an accurate basis for the optimized design of the excitation signal.

[0083] Based on an understanding of the medium transmission characteristics and noise statistics, this embodiment aims to improve the target frequency band identification sensitivity and signal-to-noise ratio. Combining equipment specifications and on-site safety regulations, it determines upper limits for energy, peak power, and repetition rate. Under these constraints, an optimization calculation method is used to obtain the optimal parameter combination of center frequency, bandwidth, energy, and repetition rate, generating a corresponding adaptive excitation configuration. This configuration is sent to the adaptive source controller for execution, and the source emits the optimized adaptive excitation. The distributed fiber optic demodulator performs trigger verification and first-cycle alignment checks on the excitation process to ensure that the time synchronization error between excitation and acquisition does not exceed a preset threshold. After verification, formal synchronous reflection response data acquisition can begin.

[0084] In this specification, "channel-space coordinate mapping" refers to a parameter table that establishes a correspondence between each acquisition channel in the distributed fiber optic sensing array and its actual three-dimensional coordinate point within the area to be measured. Its function is to achieve signal spatial positioning, enabling the acquired response data to be associated with a specific spatial location, providing a spatial reference for subsequent amplitude-frequency characteristic calculation, noise estimation, and boundary reconstruction.

[0085] In one specific embodiment, step 300 is implemented as follows: A distributed fiber optic sensor array is deployed in the target monitoring area. This array uses optical fiber as a continuous sensing medium, enabling real-time sensing of changes in external stress or vibration at different points along the fiber. When an external excitation source (such as a controlled seismic source or natural seismic waves) acts on the strata, it generates reflected and transmitted waves in the underground medium. The array transmits the received seismic wave response signals from different locations to the fiber optic port in the form of optical phase or intensity changes. A real-time signal processor synchronizes with the array under a unified time reference, which can be provided by a high-precision clock or GPS timing module to ensure complete data alignment between different sensing channels. The real-time signal processor demodulates and digitizes the optical signals returned from the optical fiber, generating reflection response data with channel numbers and precise timestamps, thereby obtaining a high-density spatial sampling signal covering the monitoring area. This method ensures accurate acquisition of seismic wave reflection and transmission responses within a unified time frame, avoiding phase deviation and time drift problems caused by asynchrony between channels.

[0086] In this specification, a distributed fiber optic sensing array refers to a sensing system that utilizes optical fibers as long-distance continuous sensing units. Through physical effects such as Rayleigh scattering, Brillouin scattering, or Raman scattering, it converts changes in external stress, temperature, or vibration into phase or intensity changes in optical signals, achieving continuous or quasi-continuous spatial sampling along the fiber's length. In this invention, this array is primarily used to acquire seismic wave reflection and transmission response information in large-scale underground spaces. Its function is to achieve high-density detection over a wide area using only one or multiple optical fibers, without the need for point-by-point deployment of traditional seismic detectors, thereby significantly improving data acquisition efficiency and spatial resolution.

[0087] In a preferred embodiment, for the synchronous reflection response data obtained in step 300, step 400 of this embodiment first performs wavelet denoising processing on a channel-by-channel basis. Specifically, a suitable wavelet basis function is selected and the signal of each channel is decomposed into multiple scales. Based on the noise statistics obtained from the aforementioned baseline response, a threshold is adaptively determined and the wavelet coefficients are shrunk, ultimately reconstructing a denoised signal sequence. Simultaneously, this embodiment calculates the signal-to-noise ratio and trigger alignment residual of each channel segment and generates a quality label. For segments with low signal-to-noise ratios or large trigger residuals, they are either eliminated or given reduced weight in subsequent processing steps to ensure overall signal quality.

[0088] After signal quality labeling is completed, this embodiment uses the baseline response and its statistics acquired without external excitation to establish a channel temperature drift model and gain correction coefficients. By performing temperature compensation and gain normalization on the denoised channel signals, response deviations caused by ambient temperature fluctuations and system drift can be eliminated, thus obtaining a compensated channel signal set. Based on this, combined with channel propagation delay parameters and the common trigger timestamp written by the distributed fiber optic demodulator during acquisition, group delay correction and inter-channel time alignment are performed on the compensated channel signal set to obtain a time-consistent channel signal set.

[0089] After time alignment, this embodiment extracts three types of characterization parameters—amplitude, phase, and frequency—from the time-consistent channel signal set. The extracted characterization parameters are bound to the previously established channel-space coordinate mapping, ensuring that each signal feature can be precisely located to its specific spatial position within the test area. Subsequently, these feature data are organized according to the three dimensions of time, space, and channel, forming a spatiotemporal-channel feature representation. This feature representation provides complete data support for subsequent boundary detection, boundary point constraints, and 3D model reconstruction.

[0090] In this specification, the spatiotemporal-channel feature representation refers to a multidimensional data structure formed by binding signal feature parameters (amplitude, phase, frequency) with timestamps, spatial coordinates, and acquisition channel numbers. This representation can simultaneously characterize the temporal evolution, spatial distribution, and channel-level correspondence of the signal, thereby achieving comprehensive modeling of the reflection characteristics of underground structures. Its function is to provide a unified data interface and high-precision input for subsequent gradient calculation, boundary detection, and 3D model fitting.

[0091] In a preferred embodiment, step 500 is as follows: First, based on the spatiotemporal-channel feature representation obtained in step 400, the gradient magnitude and gradient direction are calculated along the spatial dimension. During calculation, a one-to-one correspondence with the channel-spatial coordinate mapping is maintained, ensuring that each gradient result accurately corresponds to a specific channel number and spatial location point. The calculation results are organized using time slices and channel indices to form a gradient field containing gradient magnitude and direction information, providing a foundation for subsequent edge detection.

[0092] In the gradient field, this embodiment calculates the local mean and standard deviation using a sliding window method, and adaptively determines the high and low thresholds based on these results, forming a threshold pair for dual-threshold discrimination. Subsequently, non-maximum suppression is performed on the gradient magnitude along the gradient direction to eliminate non-edge points, retaining only candidate edge pixels that may be boundaries. This process ensures the precision of boundary detection and avoids false edges caused by noise or local fluctuations.

[0093] After extracting candidate edge pixels, this embodiment marks points exceeding a high threshold as strong edges and points within the high-to-low threshold range as weak edges. Using connectivity rules, weak edges are connected to adjacent strong edges, retaining weak edges connected to strong edges and discarding isolated weak edges. A stable set of boundary points is ultimately output. Further, based on this stable set of boundary points, connected component aggregation is performed to form candidate boundary regions. Each candidate region is bound to its corresponding time slice, channel number, and spatial coordinates, while the boundary point set is retained as candidate boundary points, providing accurate input for subsequent connectivity constraints and 3D model reconstruction.

[0094] In a specific embodiment of the present invention, to ensure the stability of candidate boundary points in the dynamic sequence, step 600 first establishes a correspondence between points between adjacent time slices. Specifically, taking each candidate boundary point in the previous time slice as a reference, a fixed spatial search radius is set in the current time slice. Within this radius, the candidate point whose direction is closest to the boundary point is searched. When the direction difference does not exceed a preset limit, the point is considered its corresponding point, and the two points are connected to form a time trajectory. Further, the formed time trajectory is subjected to dual screening based on the minimum trajectory length and the maximum average speed. Only trajectories with a duration greater than the minimum trajectory length and an average speed lower than the maximum speed limit are retained, thereby eliminating isolated points and points with abnormally rapid drift, ensuring temporal continuity.

[0095] In another embodiment, to further ensure the spatial stability of boundary points, eight-neighbor aggregation is performed on candidate boundary points in each time slice to form several connected components. Connected components with an area smaller than a preset threshold are directly deleted to remove noise point sets. Simultaneously, a morphological closing operation is performed within the connected components to fill holes with areas not exceeding the threshold, resulting in a set of structurally complete spatial connected components. Through this processing step, candidate boundary points are limited to continuously distributed regions, avoiding isolated points or overly scattered distributions, thus improving the stability and identifiability of the boundary regions.

[0096] Furthermore, in another embodiment of the present invention, the temporal trajectory and spatial connectivity are combined as dual constraints for the final screening of candidate boundary points. Only candidate boundary points that belong to both the temporal trajectory set and the spatial connectivity set are retained, thereby ensuring that the candidate points are continuous in time and coherent in space. In addition, to avoid excessive offset, the direction difference and curvature change of the retained points in adjacent time slices are limited to preset upper limits; if they exceed these limits, they are directly discarded. The final output stable boundary point set can maintain morphological consistency and boundary continuity across multiple time slices, making it suitable as a reliable input for subsequent fine-tuning and region generation.

[0097] In this specification, "candidate boundary region" refers to the subset to be refined after thresholding the feature representation results. This region maintains a one-to-one correspondence with channel-spatial coordinates, meaning that each candidate point can be mapped to the original channel number, time slice, and spatial coordinate position. The purpose of the candidate boundary region is to limit the scope of subsequent filtering and constraint processing, thereby reducing redundant computation while ensuring that the processed point set all have the potential to possess boundary features.

[0098] Optionally, step 700 in this embodiment includes:

[0099] The stable boundary point set is sorted according to channel number and time slice, and the boundary points in each survey line direction are grouped based on the channel-space coordinate mapping relationship to form a two-dimensional profile dataset divided by survey line. Each profile dataset contains stable boundary points selected in continuous time slices along that survey line direction, ensuring that the profile data corresponds one-to-one with the channel position in space and can be used for subsequent curve fitting operations.

[0100] For each boundary point in the profile dataset, cubic or higher-order spline curves are fitted, and a smoothing factor is set during the fitting process to balance the preservation of local details with the continuity of the overall curve. By adjusting the smoothing factor, the influence of noise and outliers on the continuity of the curve can be eliminated while preserving the local fluctuation characteristics of the boundary points, thus obtaining an accurate and smooth two-dimensional profile contour.

[0101] The two-dimensional profile contours are spatially interpolated and fused in a global coordinate system. Based on the geometric consistency constraints and physical rationality constraints between adjacent profiles, a three-dimensional boundary model covering the area to be measured is generated. Finally, the three-dimensional boundary model and the two-dimensional profile contours are output in a standardized data format to form profile drawings and three-dimensional model files that can be directly used for engineering interpretation, realizing the visualization and engineering application of boundary data.

[0102] As an optional implementation, this embodiment first deploys distributed fiber optic sensors in a ring or grid pattern in the area to be measured. Preferably, the spatial spacing between adjacent sampling points is set to 0.5–2 meters to improve spatial resolution and cover key monitoring areas. After establishing a one-to-one mapping between channels and spatial coordinates, baseline responses are acquired without external excitation for subsequent compensation and comparison. In the excitation stage, an adaptive seismic source generates a 10–1000 Hz sweep frequency signal, and the reflected waveforms are synchronously acquired by the distributed fiber optic sensor array under a unified time reference, completing the time alignment of excitation and acquisition and data archiving.

[0103] Synchronous reflection data is denoised by channel using wavelet denoising, preferably employing the Daubechies wavelet basis with 4–6 levels of decomposition. After denoising and time alignment, three-dimensional features are extracted, including amplitude A(t), phase Φ(t), and frequency f(t), while maintaining their binding relationship with channel-space coordinates. Based on this, an improved Canny operator is used for boundary identification.

[0104] (1) Gradient calculation: Calculated on the feature field F(x, y) in the spatial dimension. Where * denotes discrete convolution, , For the first-order difference operator in space, , These are the gradient components in the x and y directions, respectively; the gradient magnitude and direction can be further defined as follows:

[0105] ;

[0106] (2) Adaptive dual threshold: Calculate the local mean μ and standard deviation σ for G within a sliding window, and set... ,by Mark strong edges, with [ , Weak edges are marked, and combined with nonmaximum suppression and connectivity rules, connected weak edges are attached to strong edges to form a candidate boundary point set. Parameter explanation: F(x, y) is a feature scalar field organized along the spatial dimension (composed of A(t), Φ(t), f(t) or a combination thereof); , is the first-order spatial difference operator; μ and σ are the mean and standard deviation of the gradient magnitudes within the local window, respectively; , For adaptive high and low thresholds; A(t) is the instantaneous amplitude, Φ(t) is the instantaneous phase, and f(t) is the instantaneous frequency.

[0107] Subsequently, in this embodiment, B-spline curve fitting is first performed within each survey line profile to obtain a two-dimensional boundary contour. Then, a three-dimensional contour model is reconstructed under multi-profile consistency constraints, and engineering results are output. The B-spline curve can be represented as:

[0108]

[0109] in, For parameters The lower curve point, As control points, For the number of times B-spline basis functions, The parameter range defined for the node vectors. The final output includes: a 2D cross-sectional view with dimensions and a 3D model file in OBJ format for integration with engineering interpretation and design software. Parameter explanation: Coordinates of control points that control the shape of the curve; These are the basis functions determined by the node vectors; The degree of the spline; These are curve parameters.

[0110] Optionally, in urban stormwater pipe scenarios, the sensors are deployed in a 12×20 rectangular grid with a node spacing of 1 meter; the excitation uses a 50J pulse with a frequency sweep of 10–1000Hz, and the reflected waveforms are simultaneously acquired for processing and interpretation. Experimental results show that, based on this deployment and excitation parameter configuration, the structural depth estimation error is 1.3%, and the identification rate of cavities inside and around the pipe reaches 100%.

[0111] In the subway tunnel scenario, the sensor is deployed in a ring array with 48 nodes in each ring to cover the circumferential information of the tunnel. Combined with the signal processing and boundary recognition process of this invention, it can stably identify void targets with a size of 0.3m×0.3m and effectively detect cracks with a width of not less than 2mm, which can be used for rapid location and quantitative assessment of tunnel lining defects.

[0112] In the boundary identification stage, a convolutional neural network can be introduced on the basis of the improved Canny operator to assist in the discrimination, thereby improving the reliability and robustness of the boundary under complex backgrounds. In the excitation control stage, a distributed rhythm control strategy can be used to time and space the seismic source, thereby improving the coverage uniformity and effective action time of the sweep frequency energy in the monitoring area.

[0113] The electronic device of this invention uses an FPGA+ARM combined chip as the main control processor, integrating a quad-core ARM and programmable logic resources to accelerate tasks such as signal processing, edge detection, and model building in parallel. The hardware-accelerated Canny edge detection process is implemented on the programmable logic side, including module-level pipelines such as "gaussian filtering—gradient calculation—non-max suppression—double threshold processing." The supporting storage system includes an 8GB DDR4 high-speed cache for real-time sensor data caching and intermediate calculation results, and a 1TB SSD for storing algorithm models and historical data. The communication interface supports high-speed links such as Gigabit Ethernet, USB 3.0, and PCIe to meet the high-bandwidth data transmission requirements between the device and the host computer and sensor array.

[0114] Corresponding to the above methods, such as Figure 2 As shown, this embodiment also provides an underground structure boundary identification system based on distributed optical fiber sensing, including:

[0115] The sensor array deployment and baseline response acquisition unit is used to deploy a distributed fiber optic sensor array in the area to be tested, establish the correspondence between the channel and the spatial coordinates, and acquire the baseline response without external excitation as reference data for compensation and comparison.

[0116] The adaptive excitation optimization and emission unit is used to first perform low-energy trial sampling, estimate the transmission characteristics and noise spectrum of the medium in the target area in the target frequency band, and optimize the center frequency, bandwidth, energy and repetition rate of the excitation within the equipment power and safety boundary, and the excitation is emitted by the adaptive source controller.

[0117] The seismic response synchronous acquisition unit is used to synchronously acquire the seismic wave reflection and transmission responses caused by the excitation under a unified time reference by a distributed fiber optic sensor array and a real-time signal processor, and obtain synchronous reflection response data with channel numbers and timestamps.

[0118] The response preprocessing and feature construction unit is used to perform wavelet denoising, temperature compensation based on the baseline response, dispersion compensation and time alignment based on the synchronous reflection response data, extract amplitude, phase and frequency characteristics, and construct a spatiotemporal-channel feature representation.

[0119] The candidate boundary detection unit is used to calculate the gradient on the feature representation and perform non-maximum suppression. It adopts dual-threshold edge detection, which adaptively determines the high and low thresholds by local statistics, to extract candidate boundary regions and candidate boundary points.

[0120] A stable boundary point set extraction unit is used to apply temporal continuity and spatial connectivity constraints to the candidate boundary points to obtain a stable boundary point set; wherein the candidate boundary region is a subset to be refined by threshold filtering on the feature representation, and maintains a one-to-one correspondence with the channel-space coordinates;

[0121] The boundary model fitting and 3D reconstruction unit is used to perform spline fitting on the stable boundary point set to obtain a two-dimensional profile contour, and reconstruct a three-dimensional boundary model under multi-profile consistency constraints, outputting profile views and 3D model files that are compatible with engineering interpretation.

[0122] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0123] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for identifying the boundary of underground structures based on distributed optical fiber sensing, characterized in that, include: A distributed fiber optic sensor array is deployed in the area to be tested to establish the correspondence between the channel and the spatial coordinates, and the baseline response without external excitation is collected as reference data for compensation and comparison. First, low-energy trial sampling is performed to estimate the transmission characteristics and noise spectrum of the medium in the target frequency band. Then, the center frequency, bandwidth, energy and repetition rate of the excitation are optimized within the equipment power and safety boundary. The excitation is then issued by the adaptive source controller. Under a unified time reference, a distributed fiber optic sensing array and a real-time signal processor synchronously acquire the seismic wave reflection and transmission responses caused by the excitation, and obtain synchronous reflection response data with channel numbers and timestamps. Based on the synchronous reflection response data, wavelet denoising, temperature compensation based on the baseline response, dispersion compensation and time alignment are performed to extract amplitude, phase and frequency characteristics and construct a spatiotemporal-channel feature representation. The gradient is calculated on the feature representation and non-maximum suppression is performed. A dual-threshold edge detection method, in which the high and low thresholds are adaptively determined by local statistics, is used to extract candidate boundary regions and candidate boundary points. Temporal continuity and spatial connectivity constraints are applied to the candidate boundary points to obtain a stable set of boundary points; wherein the candidate boundary regions are subsets to be refined by threshold filtering on the feature representation, and maintain a one-to-one correspondence with channel-space coordinates; The stable boundary point set is spline fitted to obtain a two-dimensional profile, and a three-dimensional boundary model is reconstructed under multi-profile consistency constraints, outputting profile views and three-dimensional model files that are compatible with engineering interpretation.

2. The method for identifying underground structure boundaries based on distributed optical fiber sensing according to claim 1, characterized in that, A distributed fiber optic sensor array is deployed in the area to be tested to establish the correspondence between channels and spatial coordinates. Baseline responses without external excitation are collected as reference data for compensation and comparison, including: Determine the spatial boundary of the area to be measured, establish a three-dimensional coordinate system and layout path consistent with the engineering benchmark, and determine the set of benchmark points used for positioning; A distributed optical fiber sensor array is laid along the deployment path to complete the fixing and turning radius control. Identifiable markers are set at each reference point to record the correspondence between optical fiber mileage and channel number. Connect the distributed optical fiber sensor array to the distributed optical fiber demodulator to complete the link connectivity and loss check, establish a channel-space coordinate mapping table based on the reference point coordinates and mileage data, and calibrate the propagation delay of each channel under a unified time reference. Static responses are continuously acquired under conditions of external excitation being off and environmental stability. The mean and variance of amplitude for each channel are calculated to generate baseline responses and statistics, which are then archived by timestamp as reference data for compensation and threshold determination.

3. The method for identifying underground structure boundaries based on distributed optical fiber sensing according to claim 1, characterized in that, First, low-energy trial sampling is performed to estimate the transmission characteristics and noise spectrum of the medium in the target frequency band. Then, within the equipment power and safety boundaries, the center frequency, bandwidth, energy, and repetition rate of the excitation are optimized. The excitation is then generated by an adaptive source controller, including: Under a unified time reference, an adaptive source controller is configured to transmit low-energy linear sweep frequency signals or short pulse trains, record start and end frequencies, pulse width, energy and repetition rate parameters, and a distributed fiber optic demodulator provides a common trigger to achieve excitation-acquisition synchronization, thereby obtaining experimental sampling synchronous reflection response data with channel number and timestamp. Based on the channel-space coordinate mapping and the synchronous reflection response data of the trial sampling, the amplitude-frequency response and group delay in the target frequency band are calculated, the noise power spectral density and frequency-resolved signal-to-noise ratio are estimated, and the medium transmission characteristics and noise statistics of the test area are formed for excitation design. With the goal of improving the target frequency band identification sensitivity and signal-to-noise ratio, under the constraints of the upper limit of energy, the upper limit of peak power and the upper limit of repetition rate determined by the equipment specifications and on-site safety regulations, the optimized parameter set of center frequency, bandwidth, energy and repetition rate is obtained, and the corresponding adaptive excitation configuration is generated. The optimized parameters are sent to the adaptive source controller, which transmits the adaptive excitation. The distributed fiber optic demodulator performs trigger verification and first-cycle alignment check. After confirming that the excitation-acquisition time synchronization meets the preset error threshold, synchronous reflection response data is acquired.

4. The method for identifying underground structure boundaries based on distributed optical fiber sensing according to claim 1, characterized in that, Based on the synchronous reflection response data, wavelet denoising, temperature compensation based on the baseline response, dispersion compensation, and time alignment are performed. Amplitude, phase, and frequency representations are extracted to construct a spatiotemporal-channel feature representation, including: For the synchronous reflection response data, wavelet decomposition is performed by channel and thresholds are determined based on noise statistics to complete denoising; The signal-to-noise ratio of the channel segment and the trigger alignment residual are used to form a quality label, and low-quality segments are eliminated or downweighted. Based on the baseline response statistics, a channel temperature drift model and gain correction coefficient are established. Temperature compensation and gain normalization are then applied to the denoised channel signals to obtain a compensated set of channel signals. By combining the channel propagation delay parameter with the timestamp written during the acquisition process with the common trigger, the compensated channel signal set is subjected to group delay correction and inter-channel time alignment, and a channel signal set with consistent timing is output. Amplitude, phase, and frequency representations are extracted from a time-consistent channel signal set and bound to channel-space coordinate mapping, organized into a time-space-channel feature representation according to time, space, and channel dimensions.

5. The method for identifying underground structure boundaries based on distributed optical fiber sensing according to claim 1, characterized in that, The gradient is calculated on the feature representation and non-maximum suppression is performed. A dual-threshold edge detection method, where high and low thresholds are adaptively determined by local statistics, is used to extract candidate boundary regions and candidate boundary points, including: The gradient magnitude is calculated along the spatial dimension on the feature representation. With gradient direction And maintain a one-to-one correspondence with the channel-space coordinate mapping to obtain a gradient field organized by time slice and channel index; Calculate the local mean using a sliding window on the gradient field. with standard deviation in accordance with and Determine the high threshold With low threshold This forms a threshold pair for dual-threshold discrimination; where This is a preset constant; Non-maximum suppression is performed on the gradient magnitude along the gradient direction to obtain candidate edge pixels; by Mark strong edge seeds, to to The interval is marked as a weak edge, and the weak edge is connected to the strong edge according to the connectivity rule. The connected weak edges are retained and the isolated weak edges are removed, and a stable set of boundary points is output. Connectivity aggregation is performed based on a stable set of boundary points to generate candidate boundary regions. Each candidate boundary region is then bound to its corresponding time slice, channel number, and spatial coordinates. The set of boundary points is used as the candidate boundary points.

6. The method for identifying underground structure boundaries based on distributed optical fiber sensing according to claim 1, characterized in that, By applying temporal continuity and spatial connectivity constraints to the candidate boundary points, a stable set of boundary points is obtained. The candidate boundary regions are subsets to be refined through threshold filtering on the feature representation, maintaining a one-to-one correspondence with channel-space coordinates, including: Within adjacent time slices, with a preset spatial search radius, the closest corresponding point with a directional difference not exceeding a preset limit is found around each candidate boundary point of the previous time slice to establish a time trajectory. The time trajectories are then filtered out according to the preset minimum trajectory length and maximum average speed, retaining only the trajectories that simultaneously meet both conditions. Within each time slice, candidate boundary points are aggregated into connected components according to eight neighborhoods, connected components with an area smaller than a preset threshold are deleted, and a closing operation is performed to fill holes with an area not exceeding the preset threshold, thus obtaining a set of spatial connected components. With time trajectory and spatial connectivity as dual constraints, only candidate boundary points that belong to both time trajectory and spatial connectivity are retained. The direction difference and curvature change of the retained points in adjacent time slices are limited to preset upper limits, and a stable set of boundary points is output. The stable set of boundary points is aggregated into connected components to generate candidate boundary regions, and each candidate boundary region is bound to its corresponding time slice, channel number and spatial coordinates.

7. The method for identifying underground structure boundaries based on distributed optical fiber sensing according to claim 1, characterized in that, Spline fitting is performed on the stable boundary point set to obtain a two-dimensional profile contour, and a three-dimensional boundary model is reconstructed under multi-profile consistency constraints. The output includes profile views and a three-dimensional model file compatible with engineering interpretation, including: The stable boundary point set is sorted according to channel number and time slice, and combined with channel-space coordinate mapping, the boundary point set of each survey line direction is grouped to obtain the profile dataset divided by survey line. Spline curve fitting is performed on the boundary points in each profile dataset, and a smoothing factor is set during the fitting process to ensure that the obtained curve has overall continuity while preserving local details, thus obtaining a two-dimensional profile contour. Multiple two-dimensional profile contours are interpolated and fused in the global coordinate system. Based on the geometric consistency and physical rationality constraints between adjacent profiles, a three-dimensional boundary model covering the area to be measured is generated. The reconstructed 3D boundary model and profiles are output in a standard data format to generate cross-sectional views and 3D model files that can be used for engineering interpretation.

8. A system for identifying the boundary of underground structures based on distributed optical fiber sensing, characterized in that, include: The sensor array deployment and baseline response acquisition unit is used to deploy a distributed fiber optic sensor array in the area to be tested, establish the correspondence between the channel and the spatial coordinates, and acquire the baseline response without external excitation as reference data for compensation and comparison. The adaptive excitation optimization and emission unit is used to first perform low-energy trial sampling, estimate the transmission characteristics and noise spectrum of the medium in the target area in the target frequency band, and optimize the center frequency, bandwidth, energy and repetition rate of the excitation within the equipment power and safety boundary, and the excitation is emitted by the adaptive source controller. The seismic response synchronous acquisition unit is used to synchronously acquire the seismic wave reflection and transmission responses caused by the excitation under a unified time reference by a distributed fiber optic sensor array and a real-time signal processor, and obtain synchronous reflection response data with channel numbers and timestamps. The response preprocessing and feature construction unit is used to perform wavelet denoising, temperature compensation based on the baseline response, dispersion compensation and time alignment based on the synchronous reflection response data, extract amplitude, phase and frequency characteristics, and construct a spatiotemporal-channel feature representation. The candidate boundary detection unit is used to calculate the gradient on the feature representation and perform non-maximum suppression. It adopts dual-threshold edge detection, which adaptively determines the high and low thresholds by local statistics, to extract candidate boundary regions and candidate boundary points. A stable boundary point set extraction unit is used to apply temporal continuity and spatial connectivity constraints to the candidate boundary points to obtain a stable boundary point set; wherein the candidate boundary region is a subset to be refined by threshold filtering on the feature representation, and maintains a one-to-one correspondence with the channel-space coordinates; The boundary model fitting and 3D reconstruction unit is used to perform spline fitting on the stable boundary point set to obtain a two-dimensional profile contour, and reconstruct a three-dimensional boundary model under multi-profile consistency constraints, outputting profile views and 3D model files that are compatible with engineering interpretation.

Images