Wavelet packet fusion-based optical fiber event identification method and system

Abstract

The embodiment of the invention provides an optical fiber event identification method and system based on wavelet packet fusion, and belongs to the technical field of distributed optical fiber sensing. The method comprises the following steps: acquiring optical fiber vibration original data output by a distributed optical fiber sensing link, and preprocessing the optical fiber vibration original data to generate a preprocessed data sequence; event candidate segments are determined, and a corresponding sub-band coefficient set is generated; extracting sub-band energy features based on the sub-band coefficient set and executing multi-scale fusion, executing time-frequency coding on the sub-band coefficient set, and constructing event feature representation based on a multi-scale fusion result and a time-frequency coding result; and inputting the event feature representation into an optical fiber event classification model, and outputting a corresponding optical fiber event identification result. According to the scheme, optical fiber vibration characteristics are represented through multi-scale wavelet packet energy fusion and time-frequency coding in a combined mode, and stable identification and accurate classification of optical fiber events are achieved.

Description

Technical Field

[0001] This invention relates to the field of distributed optical fiber sensing technology, and more specifically to an optical fiber event recognition method and system based on wavelet packet fusion. Background Technology

[0002] Distributed fiber optic sensing technology utilizes optical fibers as a continuous sensing medium to acquire information on external vibrations, strain, or acoustic disturbances along the fiber's length. It is widely used in scenarios such as perimeter protection of oil and gas pipelines, safety monitoring of rail transit, intrusion detection of critical facilities, and structural condition sensing in engineering projects. By demodulating and analyzing the fiber optic echo signals, the location of events and their disturbance characteristics can be perceived, thus providing technical support for safety monitoring and operation management.

[0003] In practical applications, fiber optic vibration signals typically exhibit non-stationarity, strong transientity, and complex background interference. Different types of events often overlap in the time and frequency domains; for example, pedestrians walking, vehicles passing by, mechanical construction, and natural environmental disturbances share similar vibration characteristics in amplitude, duration, and spectral distribution. This makes it difficult to accurately distinguish event types using methods that rely solely on single time-domain features or simple frequency-domain analysis, easily leading to misjudgments or omissions.

[0004] In existing technologies, common fiber optic event identification methods often employ features such as short-time energy, zero-crossing rate, and power spectrum, or combine them with traditional wavelet transform for multi-scale analysis. However, traditional wavelet transform typically only decomposes low-frequency components layer by layer, gradually discarding high-frequency information during the decomposition process, resulting in limited ability to characterize multi-frequency band features in complex vibration events. Furthermore, the lack of an effective fusion mechanism between features at different scales makes it difficult to comprehensively reflect the overall time-frequency structure of the event, thus limiting the discriminative power of classification models.

[0005] Furthermore, in complex application scenarios, fiber optic signals are susceptible to environmental noise, system drift, and distance attenuation. If the event feature construction process lacks a clear data flow and a unified representation method, the uncertainty of event recognition will be further increased. Therefore, how to fully mine the time-frequency information of fiber optic vibration signals across multiple scales and frequency bands, and construct stable and effective event feature representations, while ensuring a clear processing flow and interpretable features, is a problem that current distributed fiber optic event recognition technology urgently needs to solve. Summary of the Invention

[0006] The purpose of this invention is to provide a fiber optic event recognition method and system based on wavelet packet fusion, so as to at least solve the problems of insufficient utilization of multi-scale time-frequency information and inconsistent event feature representation in existing fiber optic event recognition methods.

[0007] To achieve the above objectives, a first aspect of the present invention provides a fiber optic event recognition method based on wavelet packet fusion. The method includes: acquiring raw fiber vibration data output from a distributed fiber optic sensing link, and performing preprocessing on the raw fiber vibration data to generate a preprocessed data sequence; determining event candidate segments based on the preprocessed data sequence, and performing wavelet packet decomposition on the event candidate segments to generate corresponding subband coefficient sets; extracting subband energy features based on the subband coefficient sets and performing multi-scale fusion, while simultaneously performing time-frequency coding on the subband coefficient sets; constructing an event feature representation based on the multi-scale fusion result and the time-frequency coding result; inputting the event feature representation into a fiber optic event classification model, and outputting the corresponding fiber optic event recognition result.

[0008] Optionally, the raw fiber vibration data output from the distributed fiber optic sensing link is acquired, and preprocessing is performed on the raw fiber vibration data to generate a preprocessed data sequence. This includes: performing time alignment processing on the raw fiber vibration data based on the sampling clock information of the distributed fiber optic sensing link to eliminate timing deviations between different sampling channels and generating a time-aligned data sequence; performing normalization processing on the data amplitude corresponding to each sampling channel based on the time-aligned data sequence to map the vibration amplitude under different channels to a unified amplitude scale and generating an amplitude-normalized data sequence; and performing bandpass filtering processing according to preset passband parameters based on the amplitude-normalized data sequence to generate a frequency band-constrained data sequence, thereby determining the frequency band-constrained data sequence as the preprocessed data sequence.

[0009] Optionally, based on the amplitude-normalized data sequence, bandpass filtering is performed according to preset passband parameters to generate a frequency-constrained data sequence, including: determining a target frequency range based on the preset passband parameters, and constructing a corresponding bandpass filter according to the target frequency range; inputting the amplitude-normalized data sequence into the bandpass filter, performing suppression processing on non-target frequency components in the amplitude-normalized data sequence to generate a filtered data sequence; performing boundary processing on the filtered data sequence, and determining the filtered data sequence after boundary processing as the frequency-constrained data sequence.

[0010] Optionally, determining event candidate segments based on the preprocessed data sequence includes: performing time segmentation processing on the preprocessed data sequence according to a preset time windowing rule to generate multiple time window data segments; calculating the corresponding short-time energy parameter and zero-crossing rate parameter for each time window data segment to generate an energy feature sequence and a zero-crossing rate feature sequence that correspond one-to-one with each time window data segment; performing joint determination on each time window data segment according to a preset determination rule based on the energy feature sequence and the zero-crossing rate feature sequence to filter out target time window data segments that meet the event determination conditions; and performing merging processing on adjacent and consecutively satisfying target time window data segments to form event candidate segments with a continuous time range.

[0011] Optionally, wavelet packet decomposition is performed on the candidate event segments to generate a corresponding set of sub-band coefficients. This includes: constructing wavelet packet decomposition rules for the candidate event segments based on a preset number of wavelet packet decomposition layers and a preset wavelet packet basis function; performing layer-by-layer decomposition on the candidate event segments according to the wavelet packet decomposition rules, so as to simultaneously decompose low-frequency and high-frequency components at each decomposition layer to generate multiple sub-band components; extracting the corresponding wavelet packet coefficients for each sub-band component, and associating sub-band index information with the wavelet packet coefficients to form a set of sub-band coefficients.

[0012] Optionally, the process involves extracting sub-band energy features based on the sub-band coefficient set and performing multi-scale fusion, while simultaneously performing time-frequency coding on the sub-band coefficient set. An event feature representation is then constructed based on the multi-scale fusion result and the time-frequency coding result. This includes: calculating the energy value corresponding to each sub-band based on the sub-band coefficient set to generate a sub-band energy feature set; performing fusion processing on the energy features of different sub-bands according to a preset multi-scale fusion rule based on the sub-band energy feature set to generate a multi-scale fusion result; constructing a time-frequency band matrix based on the sub-band coefficient set according to a preset time framing rule and sub-band index relationship, and performing coding processing on the time-frequency band matrix to generate a time-frequency coding result; and combining the multi-scale fusion result with the time-frequency coding result to form an event feature representation.

[0013] Optionally, inputting the event feature representation into the fiber optic event classification model and outputting the corresponding fiber optic event recognition result includes: inputting the event feature representation into a preset fiber optic event classification model to obtain a category determination result for the event feature representation; determining the corresponding event category label based on the category determination result and generating a fiber optic event recognition result; associating the fiber optic event recognition result with the event candidate segment and outputting it to the user terminal.

[0014] A second aspect of the present invention provides an optical fiber event recognition system based on wavelet packet fusion. The system includes: an acquisition unit for acquiring raw optical fiber vibration data output from a distributed optical fiber sensing link, and performing preprocessing on the raw optical fiber vibration data to generate a preprocessed data sequence; a processing unit for determining event candidate segments based on the preprocessed data sequence, and performing wavelet packet decomposition processing on the event candidate segments to generate a corresponding subband coefficient set; a fusion unit for extracting subband energy features based on the subband coefficient set and performing multi-scale fusion, while performing time-frequency coding on the subband coefficient set, and constructing an event feature representation based on the multi-scale fusion result and the time-frequency coding result; and an output unit for inputting the event feature representation into an optical fiber event classification model and outputting the corresponding optical fiber event recognition result.

[0015] A second aspect of the present invention provides an electronic device, comprising: one or more processors; and a storage device having stored one or more programs thereon, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the fiber optic event recognition method based on wavelet packet fusion as described above.

[0016] On the other hand, the present invention provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the aforementioned fiber optic event recognition method based on wavelet packet fusion.

[0017] Through the above technical solution, this invention preprocesses the raw fiber vibration data output from the distributed fiber optic sensing link to form a unified preprocessed data sequence, ensuring that subsequent event analysis is based on data with consistent timing and controlled amplitude. On this basis, candidate event segments are identified first, avoiding complex analysis of data without events and ensuring the targetedness and stability of the feature extraction process. Furthermore, by performing wavelet packet decomposition on the candidate event segments, combined with multi-scale fusion of sub-band energy features and time-frequency coding, an event feature representation is constructed, enabling a complete expression of the time-frequency information of the fiber optic vibration signal across multiple frequency bands and scales, thus forming a clearly structured and consistent event feature input. Event classification based on this event feature representation effectively reduces the instability in event judgment caused by insufficient or inconsistent feature representation, achieving reliable output of fiber optic event recognition results.

[0018] Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of the steps of a fiber optic event recognition method based on wavelet packet fusion provided by one embodiment of the present invention; Figure 2 This is a system architecture diagram of an optical fiber event recognition system based on wavelet packet fusion provided by one embodiment of the present invention; Figure 3 This is an internal structural diagram of a computer device provided in one embodiment of the present invention. Detailed Implementation

[0020] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0021] like Figure 1 As shown, embodiments of the present invention provide a fiber optic event recognition method based on wavelet packet fusion, the method comprising: Step S10: Obtain the raw fiber vibration data output by the distributed fiber optic sensing link, and perform preprocessing on the raw fiber vibration data to generate a preprocessed data sequence.

[0022] Specifically, based on the sampling clock information of the distributed optical fiber sensing link, time alignment processing is performed on the raw optical fiber vibration data to eliminate timing deviations between different sampling channels and generate a time-aligned data sequence. Based on the time-aligned data sequence, normalization processing is performed on the data amplitude corresponding to each sampling channel to map the vibration amplitude under different channels to a unified amplitude scale and generate an amplitude-normalized data sequence. Based on the amplitude-normalized data sequence, bandpass filtering processing is performed according to preset passband parameters to generate a frequency band-constrained data sequence, which is then used as the preprocessed data sequence.

[0023] Furthermore, based on the amplitude-normalized data sequence, bandpass filtering is performed according to preset passband parameters to generate a frequency-constrained data sequence, including: determining a target frequency range based on the preset passband parameters, and constructing a corresponding bandpass filter according to the target frequency range; inputting the amplitude-normalized data sequence into the bandpass filter, performing suppression processing on non-target frequency components in the amplitude-normalized data sequence to generate a filtered data sequence; performing boundary processing on the filtered data sequence, and determining the filtered data sequence after boundary processing as the frequency-constrained data sequence.

[0024] In this embodiment of the invention, the fiber optic event recognition method based on wavelet packet fusion first includes preprocessing the raw fiber vibration data output from the distributed fiber optic sensing link to generate the preprocessed data sequence required for subsequent event analysis. The distributed fiber optic sensing link can be a distributed vibration sensing link constructed based on Rayleigh scattering, phase-sensitive OTDR, or other equivalent mechanisms. This link forms multiple spatial sampling channels along the length of the fiber, with different sampling channels corresponding to different spatial positions on the fiber.

[0025] Because distributed fiber optic sensing systems typically employ multi-channel parallel sampling or time-division multiplexing sampling during data acquisition, slight deviations may exist in the sampling start time and sampling trigger phase between different sampling channels. Directly analyzing the raw fiber vibration data can easily introduce timing errors during cross-channel or cross-spatial location comparisons, thus affecting the stability of subsequent event determination. Therefore, in this embodiment, time alignment processing is first performed on the raw fiber vibration data based on the sampling clock information of the distributed fiber optic sensing link.

[0026] Specifically, let the first The discrete vibration signal acquired from each sampling channel is represented as follows: ,in This represents the sampling point index, and the system sampling period is... Based on the channel clock offset recorded by the system. The vibration signals from each sampling channel are resampled or interpolated to align the data from different channels on a unified reference time axis. The time-aligned signal can be represented as:

[0027] in, Indicates the time-aligned first... Vibration data from each channel. For the ideal interpolation kernel function, The interpolation truncation length is used to control computational complexity while ensuring time alignment accuracy. Through the above time alignment processing, vibration data from different sampling channels have a consistent sampling benchmark in the time dimension, thus forming a time-aligned data sequence.

[0028] After time alignment, to eliminate amplitude scale differences caused by factors such as sensor sensitivity, optical path loss, and demodulation gain in different sampling channels, amplitude normalization is further performed on the time-aligned data sequence. This step maps the vibration amplitudes from different sampling channels to a unified amplitude scale, thereby improving the comparability between multi-channel data.

[0029] In one implementation, amplitude normalization can be performed as follows. For the first... Each sampling channel, with its time-aligned signal The mean and standard deviation are calculated within a preset statistical window, and amplitude mapping is performed accordingly. The normalized vibration data is represented as follows:

[0030] in, This represents the normalized vibration data. Indicates the first The average value of each sampling channel within a preset time window is used to characterize the DC bias level of that channel. The standard deviation represents the statistical fluctuation range of the vibration amplitude of that channel. Through the above amplitude normalization process, the vibration signals of different sampling channels are mapped to a unified amplitude scale, thereby generating an amplitude-normalized data sequence.

[0031] After obtaining the amplitude-normalized data sequence, bandpass filtering is further performed according to preset passband parameters to suppress non-target frequency components and retain effective vibration components related to fiber optic events, generating a frequency band-constrained data sequence. The preset passband parameters can be set according to the type of event of interest in a specific fiber optic vibration monitoring scenario, such as determining the target frequency range based on the dominant frequency distribution range of common intrusion events, mechanical vibrations, or environmental disturbances.

[0032] Specifically, the target frequency range is determined based on preset passband parameters. And construct a corresponding bandpass filter based on the target frequency range. In one implementation, a finite impulse response bandpass filter can be used to filter the amplitude-normalized data sequence. The filtering process is expressed as follows:

[0033] in, This represents the data sequence obtained after filtering. This represents the impulse response coefficient of the bandpass filter. The impulse response coefficient is determined by the target frequency range and the system sampling frequency, allowing frequency components within the target frequency range to pass through while effectively suppressing non-target frequency components.

[0034] After bandpass filtering, boundary processing is performed on the filtered data sequence to avoid introducing boundary distortion at the beginning and end of the data segment. Boundary processing can employ methods such as mirror continuation, periodic continuation, or zero-padding to ensure the continuity and stability of the filtered signal in the boundary region. The filtered data after boundary processing is ultimately determined as a band-constrained data sequence and serves as the output of the preprocessed data sequence.

[0035] Through the above-mentioned time alignment, amplitude normalization, and frequency band constraint processing, the generated preprocessed data sequence meets the unified constraints in terms of time reference, amplitude scale, and spectral range, providing a stable and controllable data input basis for subsequent event candidate segment extraction and wavelet packet decomposition processing.

[0036] Step S20: Based on the preprocessed data sequence, determine the event candidate segments, and perform wavelet packet decomposition processing on the event candidate segments to generate the corresponding sub-band coefficient set.

[0037] Specifically, determining event candidate segments based on the preprocessed data sequence includes: performing time segmentation processing on the preprocessed data sequence according to a preset time windowing rule to generate multiple time window data segments; calculating the corresponding short-time energy parameter and zero-crossing rate parameter for each time window data segment to generate an energy feature sequence and a zero-crossing rate feature sequence corresponding one-to-one with each time window data segment; performing joint judgment on each time window data segment according to a preset judgment rule based on the energy feature sequence and the zero-crossing rate feature sequence to filter out target time window data segments that meet the event judgment conditions; and performing merging processing on adjacent and consecutive target time window data segments that meet the event judgment conditions to form event candidate segments with a continuous time range.

[0038] Furthermore, wavelet packet decomposition is performed on the candidate event segments to generate a corresponding set of sub-band coefficients. This includes: constructing wavelet packet decomposition rules for the candidate event segments based on a preset number of wavelet packet decomposition layers and a preset wavelet packet basis function; performing layer-by-layer decomposition on the candidate event segments according to the wavelet packet decomposition rules, so as to simultaneously decompose low-frequency and high-frequency components at each decomposition layer to generate multiple sub-band components; extracting the corresponding wavelet packet coefficients for each sub-band component, and associating sub-band index information with the wavelet packet coefficients to form a set of sub-band coefficients.

[0039] In this embodiment of the invention, after generating the preprocessed data sequence, candidate event segments are further determined based on the preprocessed data sequence, and wavelet packet decomposition is performed on the candidate event segments to generate corresponding sub-band coefficient sets. This step is used to locate time intervals that may contain fiber optic events in continuous vibration data, and on this basis, multi-band expansion analysis is performed on event-related signals, thereby providing a basic data structure for subsequent feature construction.

[0040] In this embodiment, the determination of event candidate segments is based on time windowing analysis. Specifically, the preprocessed data sequence is segmented according to a preset time windowing rule, dividing the continuous vibration data into multiple independent time window data segments. The time windowing rule can be determined by both the window length and the window shift, where the window length limits the number of sampling points included in each time window, and the window shift controls the degree of overlap between adjacent time windows. Through the above time segmentation process, the preprocessed data sequence is mapped into a set of time window data segments arranged in chronological order.

[0041] For each time window data segment, the corresponding short-time energy parameter and zero-crossing rate parameter are calculated to characterize the intensity and structural change features of the vibration signal within that time window. Let the... Vibration data within a time window are represented as follows: ,in This is the index of the sampling points within the window, and the time window length is... The short-time energy parameter can be expressed as:

[0042] in, Indicates the first The short-time energy parameters corresponding to each time window This represents a window function used to reduce the impact of abrupt changes at the boundaries of a time window. This represents the vibration amplitude within a time window. Short-time energy parameters are used to reflect the overall energy level of the vibration signal within that time window.

[0043] Simultaneously, the zero-crossing rate parameter is calculated for data segments within the same time window to reflect the frequency variation trend of the vibration signal within that time window. The zero-crossing rate parameter can be expressed as:

[0044] in, Indicates the first The zero-crossing rate parameter corresponding to each time window This is the sign function, used to determine whether the sign changes between adjacent sampling points. The zero-crossing rate parameter reflects the high-frequency components and structural complexity of the vibration signal within the time window.

[0045] By calculating short-time energy parameters and zero-crossing rate parameters for each time window data segment, energy feature sequences and zero-crossing rate feature sequences corresponding one-to-one with each time window data segment are generated. Subsequently, based on the energy feature sequences and zero-crossing rate feature sequences, a joint judgment is performed on each time window data segment according to preset judgment rules. The preset judgment rules may include energy threshold conditions, zero-crossing rate change conditions, or a combination of both, to distinguish between background vibrations and time window data segments that may contain event characteristics.

[0046] During the joint determination process, time window data segments that meet the event determination criteria are selected as target time window data segments. Further, adjacent and consecutive target time window data segments that meet the event determination criteria on the timeline are merged, combining multiple consecutive time windows into a single event candidate segment with a continuous time range. This merging process avoids splitting the same physical event into multiple discrete segments, thus ensuring the integrity of the event candidate segment in the time dimension.

[0047] After obtaining candidate event segments, wavelet packet decomposition is performed on each segment to generate a corresponding set of sub-band coefficients. In this embodiment, the wavelet packet decomposition rules are determined by a preset number of wavelet packet decomposition levels and a preset number of wavelet packet basis functions. The wavelet packet basis functions can be selected based on the characteristics of the vibration signal to achieve a balance between the time and frequency domains, and the number of wavelet packet decomposition levels is used to control the fineness of the frequency band division.

[0048] According to the constructed wavelet packet decomposition rules, the event candidate segments are decomposed layer by layer. In each decomposition layer, the event candidate segments are further decomposed in both low-frequency and high-frequency component directions, thus forming multi-layer sub-band components covering the entire frequency range. Unlike traditional wavelet decomposition, which only decomposes low-frequency components, wavelet packet decomposition can completely characterize the spectral structure of vibration signals across multiple scales and frequency bands.

[0049] After completing the layer-by-layer decomposition, wavelet packet coefficients are extracted for each sub-band component, and sub-band index information is associated with each set of wavelet packet coefficients. The sub-band index information identifies the hierarchical and frequency band positions of the wavelet packet coefficients in the decomposition tree, thus establishing the correspondence between the wavelet packet coefficients and specific frequency intervals. Finally, the wavelet packet coefficients with sub-band index information form a sub-band coefficient set, which serves as the input data for subsequent sub-band energy feature extraction and time-frequency coding processing.

[0050] Step S30: Extract sub-band energy features based on the sub-band coefficient set and perform multi-scale fusion. At the same time, perform time-frequency coding on the sub-band coefficient set and construct an event feature representation based on the multi-scale fusion result and the time-frequency coding result.

[0051] Specifically, based on the sub-band coefficient set, the energy value corresponding to each sub-band is calculated to generate a sub-band energy feature set; based on the sub-band energy feature set, the energy features of different sub-bands are fused according to a preset multi-scale fusion rule to generate a multi-scale fusion result; based on the sub-band coefficient set, a time-frequency band matrix is ​​constructed according to a preset time framing rule and sub-band index relationship, and the time-frequency band matrix is ​​encoded to generate a time-frequency coding result; the multi-scale fusion result is combined with the time-frequency coding result to form an event feature representation.

[0052] In this embodiment of the invention, after obtaining the sub-band coefficient set, sub-band energy features are further extracted based on the sub-band coefficient set, and multi-scale fusion processing is performed on this basis. Simultaneously, time-frequency coding is performed on the same sub-band coefficient set. Finally, an event feature representation is constructed based on the multi-scale fusion result and the time-frequency coding result. This process is used to organize and integrate vibration features from candidate event segments from different scales and different representation forms to form stable and discriminative feature inputs.

[0053] In this embodiment, the extraction of subband energy features is accomplished using the subband coefficient set as input. Let the first... Layer, First The wavelet packet coefficient sequence corresponding to each sub-band is represented as follows: ,in This is the coefficient index within the sub-band. For each sub-band, the corresponding sub-band energy value is calculated to characterize the energy distribution of the vibration signal within the frequency range of that sub-band. The sub-band energy value can be calculated as follows:

[0054] in, Indicates the first Layer, First The energy value of each child. This represents the wavelet packet coefficients of the corresponding sub-band. This indicates the number of wavelet packet coefficients contained in the sub-band. By calculating the energy value for each sub-band separately, a sub-band energy feature set consisting of multiple sub-band energy values ​​is generated.

[0055] After obtaining the sub-band energy feature set, the energy features of different sub-bands are fused according to a preset multi-scale fusion rule to generate a multi-scale fusion result. The multi-scale fusion rule can be set based on the position of the sub-bands in the decomposition hierarchy. For example, the sub-band energies at different decomposition levels can be grouped according to the hierarchy, and energy convergence operations can be performed within each scale. In this way, while maintaining the frequency band distinguishability, the impact of single sub-band energy fluctuations on the overall feature stability can be reduced.

[0056] In one implementation, the multi-scale fusion result can be expressed as:

[0057] in, Indicates the first The fusion energy characteristics corresponding to each scale Indicates being divided into the first A set of sub-band indices for each scale. This represents the subband energy weighting coefficient, used to reflect the importance of different subbands at this scale. Through the above fusion process, the set of subband energy features is mapped to a set of multi-scale energy features, forming a multi-scale fusion result.

[0058] While performing subband energy feature extraction and multi-scale fusion, time-frequency coding results are further constructed based on the subband coefficient set. Specifically, time-framing processing is performed on the wavelet packet coefficients corresponding to the event candidate segments according to the preset time framing rules, and the subband coefficients in each time frame are organized into a time-frequency band matrix in combination with the subband index relationship. The row direction of the time-frequency band matrix is ​​used to represent the subband index, and the column direction is used to represent the time frame index, thus preserving information in both the time dimension and the frequency band dimension in a unified data structure.

[0059] After obtaining the time-frequency band matrix, encoding processing is performed on the matrix to generate time-frequency encoded results. Encoding processing may include statistical mapping, symbolic mapping, or interval quantization mapping of each element in the time-frequency band matrix, used to convert wavelet packet coefficients in continuous numerical form into an encoded representation more suitable for classification model input. In one implementation, the encoded time-frequency features can be represented as:

[0060] in, Indicates the first The number of belts in the first Encoding results on each time frame Indicates the first The set of coefficient indices corresponding to each time frame This represents the wavelet packet coefficients of the corresponding sub-band. This represents the coding mapping function, used to convert energy statistics results into coded values ​​of a predetermined form. Through the above processing, time-frequency coding results reflecting the time-varying characteristics of subbands are generated.

[0061] After generating the multi-scale fusion results and time-frequency coding results, the two are combined to form an event feature representation that describes the overall characteristics of the event candidate segments. The combination method can include feature splicing, hierarchical organization, or other equivalent methods, ensuring that multi-scale energy information and time-frequency structure information are preserved in the same feature representation. This event feature representation serves as input to the subsequent fiber optic event classification model, supporting the determination of different types of fiber optic events.

[0062] Step S40: Input the event feature representation into the fiber optic event classification model and output the corresponding fiber optic event recognition result.

[0063] Specifically, the event feature representation is input into a preset fiber optic event classification model to obtain a category determination result for the event feature representation; based on the category determination result, the corresponding event category label is determined to generate a fiber optic event recognition result; the fiber optic event recognition result is associated with the event candidate segment and output to the user terminal.

[0064] In this embodiment of the invention, after forming the event feature representation, the event feature representation is input into the fiber optic event classification model, and the corresponding fiber optic event identification result is output. The fiber optic event classification model is used to determine the event category based on the event feature representation. Its model structure can be pre-built and trained according to actual application requirements, and the input format of the model is consistent with the aforementioned event feature representation.

[0065] In practice, the event feature representation is fed into the fiber optic event classification model as an input vector. Through the model's internal feature mapping and determination mechanism, a category determination result is obtained for the event feature representation. The category determination result can be output as a category number, category probability distribution, or other equivalent form to characterize the attribution of the event feature representation under each preset event category.

[0066] After obtaining the category determination result, the corresponding event category label is determined according to the preset category mapping relationship, and this event category label is recorded as part of the fiber optic event identification result. The event category label is used to clearly identify the event type corresponding to the event candidate segment, so that the identification result can correspond to the specific fiber optic vibration event.

[0067] Furthermore, the generated fiber optic event identification results are associated with the aforementioned event candidate segments, ensuring that each event candidate segment corresponds to a unique event category label within its time frame. This association method clearly identifies the start and end points of each event and its corresponding event type on the timeline. Finally, the associated fiber optic event identification results are output to the user terminal. The output format may include information such as event occurrence time, duration, and event category, allowing users to perform subsequent analysis or processing.

[0068] like Figure 2As shown, this invention provides an optical fiber event recognition system based on wavelet packet fusion. The system includes: a data acquisition unit for acquiring raw optical fiber vibration data output from a distributed optical fiber sensing link, and performing preprocessing on the raw optical fiber vibration data to generate a preprocessed data sequence; a processing unit for determining event candidate segments based on the preprocessed data sequence, and performing wavelet packet decomposition processing on the event candidate segments to generate a corresponding subband coefficient set; a fusion unit for extracting subband energy features based on the subband coefficient set and performing multi-scale fusion, while performing time-frequency coding on the subband coefficient set, and constructing an event feature representation based on the multi-scale fusion result and the time-frequency coding result; and an output unit for inputting the event feature representation into an optical fiber event classification model and outputting the corresponding optical fiber event recognition result.

[0069] A third aspect of the present invention provides an electronic device, comprising: one or more processors; and a storage device having stored one or more programs thereon, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the above-described fiber optic event recognition method based on wavelet packet fusion.

[0070] The present invention also provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the aforementioned fiber optic event recognition method based on wavelet packet fusion.

[0071] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 3 As shown, the computer device includes a processor A01, a network interface A02, memory (not shown), and a database (not shown) connected via a system bus. The processor A01 provides computing and control capabilities. The memory includes internal memory A03 and a non-volatile storage medium A04. The non-volatile storage medium A04 stores an operating system B01, a computer program B02, and a database (not shown). The internal memory A03 provides an environment for the operation of the operating system B01 and the computer program B02 stored in the non-volatile storage medium A04. The network interface A02 is used for communication with external terminals via a network connection. When the computer program B02 is executed by the processor A01, it implements a fiber optic event recognition method based on wavelet packet fusion.

[0072] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a microcontroller, chip, or processor to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0073] The optional embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the embodiments of the present invention are not limited to the specific details described above. Within the scope of the technical concept of the embodiments of the present invention, various simple modifications can be made to the technical solutions of the embodiments of the present invention, and these simple modifications all fall within the protection scope of the embodiments of the present invention. It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, the embodiments of the present invention will not further describe the various possible combinations.

[0074] Furthermore, various different embodiments of the present invention can be combined in any way, as long as they do not violate the spirit of the embodiments of the present invention, they should also be regarded as the content disclosed by the embodiments of the present invention.

Claims

1. A fiber optic event recognition method based on wavelet packet fusion, characterized in that, The method includes: The raw fiber vibration data output from the distributed fiber optic sensing link is acquired, and the raw fiber vibration data is preprocessed to generate a preprocessed data sequence. Based on the preprocessed data sequence, candidate event segments are determined, and wavelet packet decomposition is performed on the candidate event segments to generate corresponding sub-band coefficient sets. Subband energy features are extracted based on the subband coefficient set and multi-scale fusion is performed. At the same time, time-frequency coding is performed on the subband coefficient set, and event feature representation is constructed based on the multi-scale fusion result and the time-frequency coding result. The event feature representation is input into the fiber optic event classification model, and the corresponding fiber optic event recognition result is output.

2. The fiber optic event recognition method based on wavelet packet fusion according to claim 1, characterized in that, The raw fiber vibration data is preprocessed to generate a preprocessed data sequence, including: Based on the sampling clock information of the distributed optical fiber sensing link, time alignment processing is performed on the raw optical fiber vibration data to eliminate timing deviations between different sampling channels and generate a time-aligned data sequence. Based on the time-aligned data sequence, the data amplitude corresponding to each sampling channel is normalized to map the vibration amplitude under different sampling channels to a unified amplitude scale, thereby generating an amplitude-normalized data sequence. Based on the amplitude-normalized data sequence, bandpass filtering is performed according to preset passband parameters to generate a frequency band-constrained data sequence, which is then used as the preprocessed data sequence.

3. The fiber optic event recognition method based on wavelet packet fusion according to claim 2, characterized in that, Based on the amplitude-normalized data sequence, bandpass filtering is performed according to preset passband parameters to generate a frequency band-constrained data sequence, including: The target frequency range is determined based on the preset passband parameters, and a corresponding bandpass filter is constructed based on the target frequency range. The amplitude-normalized data sequence is input into the bandpass filter to suppress non-target frequency components in the amplitude-normalized data sequence, thereby generating a filtered data sequence. Boundary processing is performed on the filtered data sequence, and the filtered data sequence after boundary processing is determined as a frequency band constrained data sequence.

4. The fiber optic event recognition method based on wavelet packet fusion according to claim 1, characterized in that, Determining candidate event segments based on the preprocessed data sequence includes: Based on the preprocessed data sequence, time segmentation processing is performed on the preprocessed data sequence according to a preset time window rule to generate multiple time window data segments; For each of the time window data segments, the corresponding short-time energy parameters and zero-crossing rate parameters are calculated to generate energy feature sequences and zero-crossing rate feature sequences that correspond one-to-one with each of the time window data segments; Based on the energy feature sequence and the zero-crossing rate feature sequence, a joint judgment is performed on each time window data segment according to a preset judgment rule to filter out the target time window data segment that meets the event judgment condition; The target time window data segments that are adjacent and consecutively satisfy the event determination conditions are merged to form event candidate segments with a continuous time range.

5. The fiber optic event recognition method based on wavelet packet fusion according to claim 4, characterized in that, Wavelet packet decomposition is performed on the candidate event segments to generate a corresponding set of sub-band coefficients, including: Based on a preset number of wavelet packet decomposition layers and a preset wavelet packet basis function, wavelet packet decomposition rules for the event candidate segments are constructed. According to the wavelet packet decomposition rules, the event candidate segments are decomposed layer by layer to simultaneously decompose low-frequency and high-frequency components at each decomposition layer, generating multi-layer sub-band components. The corresponding wavelet packet coefficients are extracted for each sub-band component, and sub-band index information is associated with the wavelet packet coefficients to form a sub-band coefficient set.

6. The fiber optic event recognition method based on wavelet packet fusion according to claim 1, characterized in that, Subband energy features are extracted from the subband coefficient set and multi-scale fusion is performed. Simultaneously, time-frequency coding is performed on the subband coefficient set. An event feature representation is constructed based on the multi-scale fusion result and the time-frequency coding result, including: Based on the sub-band coefficient set, the energy value corresponding to each sub-band is calculated to generate a sub-band energy feature set; Based on the sub-band energy feature set, the energy features of different sub-bands are fused according to a preset multi-scale fusion rule to generate a multi-scale fusion result. Based on the sub-band coefficient set, a time-frequency band matrix is ​​constructed according to the preset time framing rules and sub-band index relationship, and the time-frequency band matrix is ​​encoded to generate time-frequency coding results; The multi-scale fusion result is combined with the time-frequency coding result to form an event feature representation.

7. The fiber optic event recognition method based on wavelet packet fusion according to claim 1, characterized in that, The event feature representation is input into the fiber optic event classification model, and the corresponding fiber optic event recognition result is output, including: The event feature representation is input into a preset fiber optic event classification model to obtain the category determination result for the event feature representation; Based on the category determination result, the corresponding event category label is determined, and the fiber optic event recognition result is generated. The fiber optic event identification result is associated with the event candidate fragment and output to the user terminal.

8. A fiber optic event recognition system based on wavelet packet fusion, characterized in that, The system is used to perform the fiber optic event recognition method based on wavelet packet fusion as described in any one of claims 1-7, the system comprising: The acquisition unit is used to acquire the raw optical fiber vibration data output by the distributed optical fiber sensing link, and to perform preprocessing on the raw optical fiber vibration data to generate a preprocessed data sequence. The processing unit is used to determine event candidate segments based on the preprocessed data sequence, and to perform wavelet packet decomposition processing on the event candidate segments to generate a corresponding sub-band coefficient set. The fusion unit is used to extract sub-band energy features based on the sub-band coefficient set and perform multi-scale fusion, while performing time-frequency coding on the sub-band coefficient set, and constructing event feature representation based on the multi-scale fusion result and the time-frequency coding result; The output unit is used to input the event feature representation into the optical fiber event classification model and output the corresponding optical fiber event recognition result.

9. An electronic device, characterized in that, include: One or more processors; A storage device having stored one or more programs thereon, which, when executed by the one or more processors, cause the one or more processors to implement the fiber optic event recognition method based on wavelet packet fusion as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the fiber optic event recognition method based on wavelet packet fusion as described in any one of claims 1-7.

Images