DAS signal noise reduction method, device, medium, and product

Abstract

The application provides a DAS signal denoising method, device, medium and product, relates to the technical field of signal processing, and the method comprises the following steps: when a target vibration signal meets an abnormal feature, denoising strategies are executed in turn according to priorities. The execution process comprises the following steps: reference noise data is used to denoise the signal, the denoised signal is input into a neural network classifier, and the maximum classification probability value output by the neural network classifier is used as a denoising score. If the denoising score obtained by a certain strategy exceeds a preset threshold value, subsequent strategies are immediately stopped, and the denoised signal and the classification result corresponding to the strategy are output as the final result. If all the strategies are executed and none of the denoising scores exceeds the threshold value, the signal and the classification result corresponding to the strategy with the highest denoising score are selected as the final result. The scheme solves the technical problem that the denoising effect of a monitoring signal is poor when a spatial reference noise denoising scheme is used for a DAS system that encounters large-scale and continuous environmental noise interference.

Description

Technical Field

[0001] This application relates to the field of signal processing technology, specifically to a DAS signal noise reduction method, device, medium, and product. Background Technology

[0002] Distributed fiber optic acoustic sensing (DAS) technology enables real-time monitoring of intrusion events such as excavation and vehicle movement along long-distance pipelines. To improve alarm accuracy, modern DAS systems typically employ a cascaded architecture of "signal processing + artificial intelligence." This involves first using front-end digital signal processing algorithms to remove environmental background noise and enhance effective signal characteristics, then inputting the processed signal into a neural network for classification and recognition. Due to the complex environment along the fiber optic route, the signal often contains non-stationary noise such as wind, rain, and traffic noise; therefore, efficient noise reduction is crucial for system performance.

[0003] To address non-stationary environmental noise, existing technologies utilize the multi-channel characteristics of DAS systems, employing a spatial reference-based noise reduction scheme. This scheme works by selecting a channel spatially adjacent to the target channel as a reference channel. Assuming that the common-mode noise of adjacent channels is similar, the signals acquired from neighboring channels are treated as background noise references. Subsequently, the system uses spectral subtraction or adaptive filtering algorithms to cancel out the environmental noise in the target channel using this spatial reference signal, thereby separating the target vibration signal for use by subsequent identification modules.

[0004] However, the aforementioned spatial proximity reference-based schemes have limitations when facing large-scale, continuous interference scenarios. Especially in long-distance pipeline monitoring, when encountering widespread, strong noise interference such as thunderstorms, floods, or continuous traffic flow on parallel roads, the spatial proximity reference-based denoising algorithm will fail due to the high signal-to-noise ratio of the reference signal itself. In such cases, if the system still solely uses the nearest channel as the reference for denoising calculations, the reference signal itself contains extremely strong interference, which can easily lead to the algorithm incorrectly canceling the target signal, resulting in severe "over-subtraction" or "under-subtraction." This causes the final output signal to be severely distorted, failing to meet the requirements for subsequent recognition accuracy. Summary of the Invention

[0005] This application provides a DAS signal noise reduction method, device, medium, and product to solve the technical problem that the existing DAS system has poor noise reduction effect when using a spatial reference noise reduction scheme to reduce the noise of the monitoring signal when encountering large-scale continuous environmental noise interference.

[0006] Firstly, this application provides a DAS signal denoising method, including: When the target vibration signal of the target pipeline segment meets the preset abnormal characteristics, several noise reduction strategies are executed in sequence according to the priority of strategy execution, until all the noise reduction strategies are executed or the noise reduction score output by any of the noise reduction strategies is greater than the preset score threshold. The noise reduction strategy includes: Select target noise reference data corresponding to the current target noise reduction strategy from the reference noise database corresponding to the target pipeline segment. The target noise reduction strategy can be any one of the noise reduction strategies. Based on the target noise reference data and the target noise reduction strategy, perform noise reduction processing on the target vibration signal to obtain the target noise-reduced vibration signal. Input the target noise-reduced vibration signal into a neural network classifier to obtain several classification results, where each classification result corresponds to a classification probability value. Take the classification probability value with the largest probability value as the target noise reduction score, and take the classification result corresponding to the target noise reduction score as the first final classification result. When any of the noise reduction scores is greater than the preset score threshold, the corresponding noise-reduced vibration signal is taken as the final vibration signal, and the final vibration signal and the first final classification result corresponding to the final vibration signal are output. When all the noise reduction strategies have been executed, the noise-reduced vibration signal with the highest noise reduction score is determined as the final vibration signal, and the final vibration signal and the first final classification result corresponding to the final vibration signal are output.

[0007] Optionally, the reference noise database includes a spatial reference data pool, a temporal reference data pool, and a statistical reference data pool; The spatial reference data pool is used to store the first vibration signals of several first pipeline segments within a first preset range before and after the target pipeline segment at the current moment. The time reference data pool is used to store the second vibration signal of the target pipeline segment within a first preset time period before the current time, and the noise of the second vibration signal is less than a preset noise threshold. The statistical reference data pool is used to store the background noise spectrum feature data. The background noise spectrum feature data is calculated based on the third vibration signal collected within a second preset time period before the current moment, based on a number of second pipeline segments within a second preset range before and after the target pipeline segment. The second preset time period is longer than the first preset time period, and the second preset range is larger than the first preset range.

[0008] Optionally, the noise reduction strategy specifically includes: When the target noise reduction strategy is the first noise reduction strategy, the first vibration signal with the least noise is selected as the target noise reference data, and the target vibration signal is noise-reduced using the minimum mean square error noise reduction algorithm based on the target noise reference data. When the target noise reduction strategy is the second noise reduction strategy, the second vibration signal that is closest to the current time of the target vibration signal is selected as the target noise reference data, and the target vibration signal is noise-reduced by the minimum mean square error noise reduction algorithm according to the target noise reference data. When the target noise reduction strategy is the third noise reduction strategy, the background noise spectrum feature data is selected as the target noise reference data, and the target vibration signal is noise-reduced using the target minimum mean square error noise reduction algorithm based on the target noise reference data. When the target noise reduction strategy is the fourth noise reduction strategy, the second vibration signal closest to the current time of the vibration signal is selected as the target noise reference data, and the gain parameter of the minimum mean square error noise reduction algorithm is adjusted according to the preset strategy, and the target vibration signal is denoised according to the target noise reference data. Optionally, the method further includes: Within a preset statistical period, the denoised vibration signal of each of the denoising strategies is determined as the cumulative number of times the final vibration signal is obtained. Based on the cumulative number of each noise reduction strategy in descending order, the noise reduction strategies are reordered, and the sorting result is used as the updated strategy execution priority order.

[0009] Optionally, the method further includes: When all the noise reduction strategies have been executed, the noise-reduced vibration signal and noise reduction score output by each noise reduction strategy are obtained. Based on the noise reduction scores, calculate the fusion weight value corresponding to each of the noise-reduced vibration signals; Based on the fusion weight values, the denoised vibration signals are weighted and superimposed to obtain the fused denoised signal. The fused and denoised signal is input into the neural network classifier to obtain several fused classification results, wherein each fused classification result corresponds to a fused classification probability value, and the fused classification result with the highest fused classification probability value is taken as the second final classification result. Output the fused and denoised signal and the second final classification result.

[0010] Optionally, the method further includes: The final vibration signal with the classification result belonging to the preset normal background noise category and the fused noise-reduced signal with the fusion classification result belonging to the preset normal background noise category are stored in the time reference data pool.

[0011] Optionally, the method further includes: Calculate the energy variance of the target vibration signal; When the energy variance is greater than or equal to the preset noise threshold, the target vibration signal is determined to meet the preset abnormal characteristics.

[0012] In a second aspect, embodiments of this application provide a DAS signal noise reduction device, which includes: one or more processors and a memory; the memory is coupled to the one or more processors, and the memory is used to store computer program code, which includes computer instructions, and the one or more processors call the computer instructions to cause the DAS signal noise reduction device to perform the method described in the first aspect and any possible implementation thereof.

[0013] Thirdly, embodiments of this application provide a computer program product containing instructions that, when the computer program product is run on a DAS signal noise reduction device, cause the DAS signal noise reduction device to perform the method described in the first aspect and any possible implementation thereof.

[0014] Fourthly, embodiments of this application provide a computer-readable storage medium including instructions that, when executed on a DAS signal noise reduction device, cause the DAS signal noise reduction device to perform the method described in the first aspect and any possible implementation thereof.

[0015] In summary, one or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: 1. By adopting the above technical solution, a dynamic strategy optimization mechanism based on classification confidence feedback was established. The system is no longer limited to a single, fixed processing flow, but instead attempts multiple paths for the target vibration signal based on the priority order of strategy execution and different target noise reference data. Crucially, this method directly uses the classification probability value output by the neural network classifier to quantify and generate a target denoising score, and compares this score with a preset score threshold in real time, using this as an evaluation standard for the denoising effect. This logic prioritizes backend recognition capability as the screening criterion for signal processing, ensuring that only when the target vibration signal retains features sufficient for high-confidence recognition after denoising is it output as the final vibration signal. This achieves deep coupling between the denoising algorithm and the recognition model, ensuring that the system can automatically adapt to the signal processing path most conducive to machine recognition in changing environments.

[0016] 2. By adopting the above technical solution, the limitations of traditional noise reduction methods that rely solely on spatial neighborhood are overcome, and a multi-dimensional reference architecture comprising a spatial reference data pool, a temporal reference data pool, and a statistical reference data pool is constructed. By storing the first vibration signal within a first preset range, the second vibration signal within a first preset duration, and the background noise spectrum feature data calculated based on a larger spatiotemporal range, the system can flexibly retrieve silent data from historical periods or long-term statistical features as alternative references when spatial references fail (e.g., high noise across the entire domain). This multi-dimensional reference source reserve ensures that the noise reduction algorithm can always obtain effective noise reference samples under various extreme or complex environmental noise interferences, avoiding noise reduction failures caused by a lack of reference data or reference data contamination, and guaranteeing the system's environmental adaptability in complex all-weather scenarios.

[0017] 3. By adopting the above technical solutions, the system clarifies the specific combination methods of reference data of different dimensions and denoising algorithms, realizing a refined classification of denoising methods. During the execution of the first to fourth denoising strategies, the system selects the first vibration signal with the least noise, the second vibration signal with the closest proximity, or the background noise spectrum feature data as the target noise reference data, and applies or adjusts the minimum mean square error denoising algorithm and its gain parameters accordingly. This step-by-step strategy design allows the system to prioritize the use of the most correlated spatial information, smoothly transitioning to the use of temporal correlation or statistical regularity when spatial information is unavailable. By matching corresponding algorithm processing logic to reference data of different properties, the system can remove non-stationary noise to the greatest extent while retaining the effective characteristics of the vibration signal, improving the robustness of the algorithm in dynamically changing environments.

[0018] 4. By adopting the above technical solution, a weighted fusion mechanism based on multi-strategy results is introduced to address challenging scenarios where a single denoising strategy fails to achieve the desired effect. After all strategies have been executed, the system does not simply select one; instead, it calculates the fusion weight value using the denoising scores of each strategy and performs weighted superposition processing on each denoised vibration signal accordingly. This processing method comprehensively utilizes the advantages of denoising strategies from different dimensions such as space, time, and statistics, weakening residual noise through superposition and complementarity to generate a higher-quality fused denoised signal. This signal is then input into a neural network classifier to obtain a second final classification result, enabling further mining of signal features under extremely low signal-to-noise ratio conditions. This effectively improves the system's ability to detect covert intrusion events under weak signals or strong interference, reducing the risk of missed detections. Attached Figure Description

[0019] Figure 1 This is a schematic flowchart of a DAS signal noise reduction method in an embodiment of this application; Figure 2This is another flowchart illustrating the DAS signal noise reduction method in the embodiments of this application; Figure 3 This is a schematic diagram of the physical device structure of a DAS signal noise reduction device in the embodiments of this application. Detailed Implementation

[0020] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0021] In the description of the embodiments of this application, the words "for example" or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design that is described as "for example" or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design options. Rather, the use of the words "for example" or "for instance" is intended to present the relevant concepts in a specific manner.

[0022] In the description of the embodiments of this application, the term "multiple" means two or more. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the indicated technical features. Thus, a feature defined with "first," "second," or "third" may explicitly or implicitly include one or more of that feature. The terms "comprising," "including," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized. It should be noted that all data collection in this scheme is conducted after obtaining user consent.

[0023] This application provides a DAS signal noise reduction method, referencing... Figure 1 , Figure 1 This is a flowchart of a DAS signal denoising method provided in an embodiment of this application. The method includes: Step S101: When the target vibration signal of the target pipeline segment meets the preset abnormal characteristics, several noise reduction strategies are executed sequentially according to the priority order of strategy execution until all the noise reduction strategies are executed or the noise reduction score output by any of the noise reduction strategies is greater than the preset score threshold. The target pipeline segment refers to a specific physical length of optical fiber segment located and under real-time monitoring in the distributed optical fiber sensing system, such as a 50-meter optical cable section between chainages K20+100 and K20+150 of an oil pipeline. The target vibration signal refers to raw time-series data containing phase, frequency, or amplitude information acquired from the target pipeline segment, such as a coherent Rayleigh scattering signal with a sampling rate of 2kHz and a duration of 0.5 seconds. Preset anomaly characteristics represent the signal statistical conditions that trigger the noise reduction processing flow, such as a short-term signal energy exceeding the background noise level by 3dB, or a zero-crossing rate within a specific frequency band. The strategy execution priority order refers to a pre-configured list of noise reduction strategy invocation sequences within the system, for example, the list order is defined as executing strategy A first, and then strategy B if the condition is not met. The noise reduction strategy refers to a set of pre-defined algorithmic logic used by the system to optimize signal processing and evaluate its quality, including signal processing algorithms and corresponding quality assessment models. The noise reduction score is a numerical metric used to quantitatively evaluate the signal quality or identifiability after noise reduction strategy processing. For example, a confidence probability value of 0.92 output by a classifier. A preset score threshold is the minimum numerical standard required for the system to deem the processing result valid or reliable, for example, 0.85.

[0024] Specifically, the system monitors the data flow of the target pipeline segment in real time. Once the system calculates and finds that the data characteristics of the collected target vibration signal match the preset abnormal characteristics, it determines that there is a potential intrusion or interference event, and triggers an adaptive processing flow. The system reads the pre-stored strategy execution priority order and strictly calls the noise reduction strategy one by one to process the target vibration signal. When executing each noise reduction strategy, the system processes the target vibration signal according to the algorithm logic defined within the strategy to generate an intermediate signal, and further calculates the quality index of the intermediate signal, thereby outputting a corresponding target noise reduction score. After each strategy execution and obtaining the target noise reduction score, the system immediately compares the score with a preset score threshold. If the current target noise reduction score does not exceed the preset score threshold, the system will automatically call the next noise reduction strategy in the list to continue execution, and so on. This loop will terminate when either of the following conditions is met: first, the current target noise reduction score has exceeded the preset score threshold (a solution that meets the condition has been found); second, all noise reduction strategies in the list have been tried (traversal ends).

[0025] Optional, see reference Figure 2 For implementing the noise reduction strategy, please refer to steps S101A1-S101A4; Step S101A1: Select target noise reference data corresponding to the current target noise reduction strategy from the reference noise database corresponding to the target pipeline segment; The reference noise database refers to a pre-built storage set containing background noise data corresponding to different environmental conditions or operating conditions. The target noise reference data refers to specific auxiliary data samples retrieved from the database to assist the current algorithm's operation, such as historical background vibration data segments adjacent to the current monitoring period, or synchronous environmental data collected by parallel-laid non-sensitive optical cables.

[0026] Specifically, after the system determines to execute a specific target noise reduction strategy, it first needs to obtain the auxiliary input data necessary for the algorithm to run. The system indexes the associated reference noise database based on the target pipeline segment's identifier and performs a search and matching within the database according to the algorithm input requirements of the current target noise reduction strategy (such as the need for time-domain reference, frequency-domain reference, or reference with a specific statistical distribution). The system selects data samples that meet the operational conditions of the current strategy as target noise reference data to provide a benchmark or cancellation source for subsequent signal processing steps.

[0027] Step S101A2: Based on the target noise reference data and the target noise reduction strategy, the target vibration signal is subjected to noise reduction processing to obtain the target noise-reduced vibration signal; Noise reduction refers to the process of correcting, filtering, or separating components of the main signal using reference data based on a specific mathematical model or signal processing algorithm. Examples include performing minimum mean square error (LMS) adaptive filtering or spectral subtraction. The target noise-reduced vibration signal refers to the output waveform data whose signal-to-noise ratio is potentially improved after processing, such as a pulse vibration waveform after removing periodic environmental interference.

[0028] Specifically, based on the operational logic defined by the target noise reduction strategy, the system first analyzes and establishes a statistical model of background noise in the current environment (e.g., noise power spectrum or noise variance distribution) using target noise reference data. Then, the system transforms the target vibration signal to the transform domain (e.g., frequency domain) and calculates the signal-to-noise ratio or posterior probability of each component in the signal using the aforementioned noise statistical model. Based on the calculated statistical indicators, the system applies corresponding gain adjustment or amplitude correction to each component of the target vibration signal to suppress components consistent with the noise statistical model in a way that minimizes estimation errors. Finally, the system transforms the corrected signal back to the time domain to generate the target noise-reduced vibration signal.

[0029] Step S101A3: Input the noise-reduced vibration signal of the target into a neural network classifier to obtain several classification results, wherein each classification result corresponds to a classification probability value; In this context, a neural network classifier refers to a deep learning or machine learning model that has been pre-trained with a large number of samples and can map input signals to specific event categories. Examples include one-dimensional convolutional neural networks (1D-CNN), long short-term memory networks (LSTM), or multilayer perceptrons (MLP). The classification result refers to the predefined event labels corresponding to the model's output layer, such as "third-party mining," "vehicle passing by," "human walking," or "silent background." The classification probability value refers to the confidence score or normalized score calculated by the model for each classification result, used to represent the likelihood that the input signal belongs to that category. For example, the confidence score for the "mining" category is 0.95, and the confidence score for the "walking" category is 0.03.

[0030] Specifically, the system performs necessary format adaptation (such as normalization or dimensionality adjustment) on the target noise-reduced vibration signal generated in step S101A2 before feeding it into the input layer of the neural network classifier. The model extracts features and performs nonlinear transformations on the signal through its internal neural network structure, ultimately generating a vector at the output layer. Each element in this vector represents the probability of different event types preset by the system. The system analyzes this output vector to obtain a set of predictions for the current signal, namely several classification results and the corresponding numerical classification probability value for each result.

[0031] Step S101A4: Take the classification probability value with the largest probability value as the target noise reduction score, and take the classification result corresponding to the target noise reduction score as the first final classification result. Specifically, the system compares or sorts the set of classification probability values ​​output in step S101A3. The system selects the probability value with the largest value and directly assigns it as the target noise reduction score. This score reflects the highest confidence level of the classifier in recognizing the signal under the current target noise reduction strategy. Simultaneously, the system finds the classification result (label) pointed to by this maximum probability value, locks it, and records it as the first final classification result. These two parameters (score and classification result) will serve as the final output of this strategy execution, used in subsequent steps to compare with preset thresholds to determine whether to adopt the result or continue trying other strategies.

[0032] Step S102: When any of the noise reduction scores is greater than the preset score threshold, the corresponding noise-reduced vibration signal is taken as the final vibration signal, and the final vibration signal and the first final classification result corresponding to the final vibration signal are output. The denoised vibration signal refers to waveform data generated after processing by the signal processing algorithm in the current denoising strategy, with some interference components removed. The final vibration signal refers to the version of signal data that has been screened and confirmed and is finally decided to be used for subsequent storage, display or further analysis. The final classification result refers to the event category label determined by the algorithm model corresponding to the final vibration signal, such as "manual excavation", "mechanical construction" or "vehicle passing by".

[0033] Specifically, during the process of the system cyclically executing noise reduction strategies according to priority, once the system detects that the noise reduction score output by a certain strategy execution is greater than a preset score threshold (e.g., the threshold is set to 0.85, and the signal recognition confidence level calculated by the current strategy is 0.90), the system determines that the processing effect of the current strategy has met the high reliability requirements. At this time, the system no longer executes the remaining noise reduction strategies in the priority list, immediately locks the noise-reduced vibration signal generated by the current strategy, and confirms it as the final vibration signal. Simultaneously, the system extracts the event category determination information generated during the strategy processing and confirms it as the first final classification result. The system packages and outputs this confirmed final vibration signal and the first final classification result to the alarm interface or database, completing the processing of this abnormal event.

[0034] Step S103: When all the noise reduction strategies have been executed, determine the noise-reduced vibration signal with the highest noise reduction score as the final vibration signal, and output the final vibration signal and the first final classification result corresponding to the final vibration signal. Specifically, if the system executes all preset noise reduction strategies in sequence, and the noise reduction score calculated each time is less than or equal to the preset score threshold (e.g., the recognition confidence of all strategies is between 0.5 and 0.7, failing to reach the 0.85 standard), it indicates that the current environment is extremely complex or the signal characteristics are not obvious. To avoid directly discarding the signal and causing missed detections, the system executes a backtracking comparison logic. The system retrieves the noise reduction scores generated by each strategy execution recorded in the cache, sorts them numerically, and selects the score with the largest value. The system locates the processing result corresponding to the highest score, forcibly identifies the noise-reduced vibration signal generated by that strategy as the final vibration signal, and confirms the corresponding category determination as the first final classification result for output. This step ensures that the system can still provide relatively optimal signals and recognition conclusions within the current algorithm's capabilities even under extreme conditions.

[0035] The following is a more detailed description of the process of the method provided in this implementation.

[0036] Optionally, the noise database in this scheme includes a spatial reference data pool, a temporal reference data pool, and a statistical reference data pool; Among them, the spatial reference data pool is used to store the first vibration signals of several first pipeline segments within a first preset range before and after the target pipeline segment at the current moment; Specifically, assuming the target pipeline segment is located at the 500th meter of the optical cable, the spatial reference data pool will include the vibration data of these first pipeline segments within the range of 450 meters to 550 meters before and after the 500th meter (i.e., the first preset range, for example, 50 meters before and after).

[0037] The time reference data pool is used to store the second vibration signal of the target pipeline segment within a first preset time period before the current time, and the noise of the second vibration signal is less than a preset noise threshold. Specifically, for the target pipeline segment at the 500-meter mark, the data pool includes data from a period of time prior to the current moment of the current vibration signal (i.e., a first preset duration, such as the past 20 minutes). The system does not select data blindly, but rather specifically filters segments with lower vibration amplitudes (noise less than a preset noise threshold) as the second vibration signal.

[0038] The statistical reference data pool is used to store the background noise spectrum feature data. The background noise spectrum feature data is calculated based on the third vibration signal collected within a second preset time period before the current moment from several second pipeline segments within a second preset range before and after the target pipeline segment. The second preset time period is longer than the first preset time period, and the second preset range is larger than the first preset range.

[0039] Specifically, the system selects a wider range before and after the 500-meter mark (i.e., the second preset range, for example, 1 kilometer before and after), as well as a longer time period past the current moment (i.e., the second preset duration, for example, the past fifteen days). The system uses this large amount of third vibration signal to perform big data calculations, obtain the background noise spectrum characteristic data of a large area where the target pipeline segment is located, and store it in the statistical reference data pool.

[0040] Optionally, the noise reduction strategy includes steps S101B1-S101B4; Step S101B1: When the target noise reduction strategy is the first noise reduction strategy, the first vibration signal with the least noise is selected as the target noise reference data, and the target vibration signal is noise-reduced using the minimum mean square error noise reduction algorithm based on the target noise reference data. The first denoising strategy refers to the system's processing mode that utilizes spatial correlation to eliminate common-mode interference. The minimum mean square error (MSE) denoising algorithm is an optimal estimation method aimed at minimizing the mean square error between the estimated signal and the original noise-free signal. It is commonly used to recover the original signal from noisy observations; this scheme employs the Log-MMSE algorithm.

[0041] This step is typically performed in scenarios where widespread environmental vibrations are detected, such as thunderstorms, passing trains, or heavy vehicle traffic. Specifically, after determining to implement the first noise reduction strategy, the system iterates through all first vibration signals in the spatial reference data pool at the current moment, calculates the energy value or amplitude variance of these signals, and selects the first vibration signal with the smallest energy value or amplitude variance. This prevents strong signals containing valid intrusion characteristics from being mistakenly used as reference noise, thus avoiding signal cancellation or leakage. Subsequently, the system assigns this selected first vibration signal as the target noise reference data and inputs it, along with the vibration signal to be processed, into the minimum mean square error noise reduction algorithm model. The algorithm model utilizes the statistical characteristics of the target noise reference data to estimate the noise components in the vibration signal and performs adaptive filtering or spectral subtraction operations, thereby outputting a signal with improved signal-to-noise ratio.

[0042] Step S101B2: When the target noise reduction strategy is the second noise reduction strategy, select the second vibration signal that is closest to the current time of the target vibration signal as the target noise reference data, and use the minimum mean square error noise reduction algorithm to perform noise reduction processing on the target vibration signal based on the target noise reference data. The second noise reduction strategy refers to the system's processing mode that utilizes short-term historical background in the time dimension. The closest point to the current moment of the vibration signal is represented on the time axis by having the smallest absolute time difference between the timestamp of the candidate data and the timestamp of the current signal to be processed.

[0043] This step is typically performed in scenarios with relatively stable environmental backgrounds, but where it is necessary to eliminate low-frequency baseline drift caused by optical path polarization drift or gradual temperature changes. Specifically, after determining to implement the second noise reduction strategy, the system accesses the time reference data pool and retrieves the time tags of all stored second vibration signals. The system compares and calculates to identify the second vibration signal whose time tag is closest to the current moment. The purpose of selecting the most recent data is that the background state of fiber optic sensing systems often drifts slowly over time, and the most recent historical background best represents the ideal reference state under normal circumstances, thus ensuring the accuracy of differential or filtering. The system sets this second vibration signal as the target noise reference data and, combined with a minimum mean square error noise reduction algorithm, suppresses quasi-static background components in the vibration signal that are similar to the target noise reference data, highlighting the abrupt signal characteristics occurring at the current moment.

[0044] Step S101B3: When the target noise reduction strategy is the third noise reduction strategy, the background noise spectrum feature data is selected as the target noise reference data, and the target vibration signal is noise-reduced using the target minimum mean square error noise reduction algorithm based on the target noise reference data. The third noise reduction strategy refers to a processing mode in which the system uses long-term statistical laws to suppress continuous and steady-state background noise at the frequency domain level.

[0045] Specifically, after determining to implement the third noise reduction strategy, the system directly retrieves pre-calculated background noise spectrum feature data from the statistical reference data pool. This data is not a specific time-domain waveform, but rather describes the average energy distribution of background noise in the region across various frequency components. The system uses the background noise spectrum feature data as target noise reference data and inputs it into a frequency-domain minimum mean square error noise reduction algorithm (such as the Log-MMSE algorithm). Based on this prior spectral feature, the algorithm calculates the posterior and prior signal-to-noise ratios of the current vibration signal at various frequency points, and then calculates the optimal spectral gain function. It then weights the vibration signal's spectrum, suppressing the energy of frequency bands matching the background noise spectrum feature data while retaining anomalous spectral components. Finally, it recovers the denoised time-domain signal through inverse Fourier transform.

[0046] Step S101B4: When the target noise reduction strategy is the fourth noise reduction strategy, the second vibration signal closest to the current time of the vibration signal is selected as the target noise reference data, and the gain parameter of the minimum mean square error noise reduction algorithm is adjusted according to the preset strategy, and the target vibration signal is denoised according to the target noise reference data. The fourth noise reduction strategy refers to a processing mode that, based on utilizing historical background for noise reduction, introduces an adaptive parameter adjustment mechanism to address the need for low signal-to-noise ratio or weak signal extraction. The preset strategy refers to a predefined logical rule table in the system that establishes a mapping relationship between signal characteristics and algorithm parameters; for example, a rule for dynamically looking up parameter values ​​based on the total energy level of the current signal. The gain parameter refers to a numerical variable within the minimum mean square error noise reduction algorithm used to control the noise suppression strength, over-subtraction factor, or smoothing coefficient.

[0047] Specifically, after determining to implement the fourth noise reduction strategy, the system first accesses the time reference data pool, retrieves the time tags of all second vibration signals, and locks onto the second vibration signal that is closest in time to the current moment by calculating the time difference. Subsequently, the system performs pre-analysis on the current vibration signal according to a preset strategy (such as calculating a preliminary signal-to-noise ratio estimate) or analyzes the fluctuation characteristics of the target noise reference data, and then dynamically calculates the optimal gain parameter. For example, if the preset strategy detects large fluctuations in background noise, it may automatically reduce the gain parameter; if a weak signal is detected, the parameter is increased to reduce the reduction of the effective signal amplitude, sacrificing some noise reduction strength to maximize the preservation of the integrity of the original signal. Finally, the system uses the adjusted gain parameter and inputs the most recent second vibration signal as the target noise reference data into the algorithm to perform minimum mean square error noise reduction calculation on the vibration signal, achieving adaptive enhancement and extraction of weak signals.

[0048] Optionally, this solution may also execute steps S104-S105; Step S104: The cumulative number of times the denoised vibration signal of each of the denoising strategies is determined as the final vibration signal within a preset statistical period. The preset statistical period represents the length of the time window for the system to perform performance evaluation and self-optimization. It refers to a fixed time interval, such as setting it to every 24 hours or every 7 days. The cumulative number represents the total number of frequencies within the current preset statistical period where the signal output by a specific noise reduction strategy outperforms other strategies and is successfully selected as the final vibration signal.

[0049] This step is performed during the background monitoring or periodic maintenance phase of the system operation, aiming to quantitatively evaluate the adaptability of each noise reduction strategy in the current real-world environment. Specifically, the system continuously logs the results of each signal processing step during operation. Whenever the system completes a round of noise reduction processing of vibration signals and selects a final vibration signal, it traces back to which noise reduction strategy (e.g., the first noise reduction strategy) generated that final vibration signal. The system maintains a set of counters in memory or a database, corresponding to each noise reduction strategy. Once the winning strategy is determined, the system increments the counter corresponding to that strategy. At the end of the preset statistical period, the system summarizes all processing events within that period, obtaining the cumulative number of wins for each noise reduction strategy. This statistical data objectively reflects which noise reduction logic is most effective under the current specific geological environment, climatic conditions, and interference type.

[0050] Step S105: Reorder each noise reduction strategy according to the cumulative number of times of each strategy from high to low, and use the sorting result as the updated strategy execution priority order. The order of cumulative counts from highest to lowest represents a performance-based ranking rule, where strategies with higher cumulative counts are ranked first, and those with lower counts are ranked last. Reordering refers to adjusting the index of the strategy list or execution queue stored internally by the system. The strategy execution priority order indicates the order in which the system activates various noise reduction strategies when processing new vibration signals, or, when system resources are limited (e.g., computational power is restricted) and full parallel processing is not possible, the preferred strategy sequence to execute.

[0051] This step is performed during the system configuration update phase after the preset statistical period ends, aiming to enable the system to adaptively evolve in response to environmental changes. Specifically, the system reads the cumulative frequency data of each noise reduction strategy obtained in step S104. Based on this data, the system sorts all noise reduction strategies. For example, if the statistics show that the first noise reduction strategy utilizing spatial correlation was selected most frequently as the final vibration signal in the past 24 hours, while the third noise reduction strategy utilizing the background noise spectrum was selected least frequently, the system will move the first noise reduction strategy to the top of the strategy list and the third noise reduction strategy to the bottom. After sorting, the system saves this new sequence as the updated strategy execution priority order. In subsequent operating cycles, when the system receives a new vibration signal, it will prioritize the use of higher-priority strategies according to this updated order, or allocate resources to higher-priority strategies, thereby improving the overall computational efficiency and noise reduction success rate of the system under specific environments.

[0052] Optionally, when all noise reduction strategies have been executed, this scheme may also execute steps S106-S109. Step S106: When all the noise reduction strategies have been executed, obtain the noise-reduced vibration signal and noise reduction score output by each noise reduction strategy. This step is executed at the end of the parallel or serial signal processing flow, that is, when the system has completed the calculation tasks of all preset algorithm paths for the same original input signal. Specifically, the system checks the execution status flag of each noise reduction strategy module to confirm that there are no suspended or ongoing computation tasks. Subsequently, the system reads the data packets output by each strategy module in batches from the memory cache or video memory. This data packet contains two core parts: first, multiple sets of noise-reduced vibration signal waveform arrays after processing by different algorithm logics; second, the noise reduction scores corresponding to each signal synchronously calculated by the system during the noise reduction process. For example, the system will simultaneously acquire the "spatial difference waveform" and its corresponding "signal-to-noise ratio score" output by the first noise reduction strategy, and the "spectral subtraction waveform" and its corresponding "spectral flatness score" output by the third noise reduction strategy, and collect these data to the main control unit for subsequent fusion calculation.

[0053] Step S107: Calculate the fusion weight value corresponding to each of the noise reduction scores for each of the noise-reduced vibration signals. This step is performed after acquiring the quality assessment data of all strategies, aiming to establish a mathematical benchmark for the weighted synthesis of multi-source signals. Specifically, the system reads the noise reduction scores obtained in step S106 and performs normalization processing using a preset weight allocation algorithm. Common processing methods include Softmax function mapping or simple proportional calculation. The system uses each noise reduction score as an input variable, amplifies the advantage of high-scoring strategies through exponential operations, and suppresses the influence of low-scoring strategies, thereby deriving the normalization coefficients corresponding to each strategy. For example, if the noise reduction scores of strategies A, B, and C are 10, 20, and 70, respectively, the fusion weight values ​​calculated by the system may be 0.1, 0.2, and 0.7, respectively. This process ensures that the higher-quality denoised vibration signal dominates in the final fusion result, while the lower-quality signal provides only a small amount of supplementary information or is completely ignored.

[0054] Step S108: Based on each of the fusion weight values, perform weighted superposition processing on each of the noise-reduced vibration signals to obtain a fused noise-reduced signal; This step, performed after weight allocation, is the core computational step for achieving complementary advantages across multiple strategies. Specifically, the system initializes a zero vector with the same length as the original signal as a buffer. Subsequently, the system iterates through the denoised vibration signal output by each denoising strategy, multiplying the amplitude of each sampling point by the fusion weight value corresponding to that strategy calculated in step S107. After calculation, the system adds the weighted waveform data of each strategy point-by-point in the time domain. For example, the final output value of the Nth sampling point is equal to: the Nth point value of strategy A × the weight of strategy A + the Nth point value of strategy B × the weight of strategy B + ... In this way, the fused denoised signal can simultaneously take into account the ability of the first strategy to suppress spatial common-mode noise and the ability of the third strategy to filter out steady-state noise in the frequency domain, thereby obtaining waveform data that is more robust and has more significant characteristics than a single strategy.

[0055] Step S109: Input the fused and denoised signal into the neural network classifier to obtain several fused classification results, wherein each fused classification result corresponds to a fused classification probability value, and take the fused classification result with the highest fused classification probability value as the second final classification result; output the fused and denoised signal and the second final classification result; Specifically, the system inputs the fused and denoised signal into a neural network for forward inference. The network's output layer (Softmax layer) generates a probability vector, where each element corresponds to a predefined event type and its fused classification probability value. The system scans this probability vector, compares and finds the element with the largest value, and locks its corresponding event category as the second final classification result. Finally, the system synchronously outputs the waveform of the fused and enhanced denoised signal, as well as the second final classification result (e.g., "Intrusion detected, confidence level 98%)", for monitoring personnel to review or trigger alarm logic.

[0056] Optionally, this solution may also perform step S110.

[0057] Step S110: Store the final vibration signal with the classification result being a preset normal background noise category and the fused noise-reduced signal with the fusion classification result being the preset normal background noise category into the time reference data pool. Among them, the preset normal background noise category refers to a specific category in the classifier label system that represents "no intrusion", "quiet environment" or "normal natural interference" (such as normal wind and grass), which means that the system determines that no threatening intrusion event has occurred in the current fiber optic defense zone.

[0058] This step is performed during the data maintenance and closed-loop update phase after the classification decision is completed. Specifically, after the system completes the classification and identification of the currently acquired signal, it immediately performs logical verification on the decision result. The system checks the classification results corresponding to the "final vibration signal" or "fused noise reduction signal". Only when the classification result clearly points to the "preset normal background noise category" will the system determine that the signal segment is qualified as a background reference. If the classification result shows an intrusion event such as "digging", "climbing", or "vehicle passing", the system will strictly prohibit storing the signal segment in the time reference data pool, because mistaking the intrusion signal as a background sample will cause the real intrusion signal to be canceled out as background during subsequent processing, thus causing false negatives. Once the verification passes and the current signal is confirmed to be pure background noise, the system will write the waveform data into the time reference data pool. This data pool usually follows the "first-in, first-out" (FIFO) update principle, that is, when storing the latest background signal, the oldest data in the pool is automatically removed to ensure that the pool always retains the data most recent to the current time. Through this continuous update mechanism, the system can provide a highly accurate dynamic reference benchmark for the aforementioned adaptive noise reduction strategy, thereby effectively dealing with the slow drift of the environment.

[0059] Optionally, this solution may also execute steps S111-S112; Step S111: Calculate the energy variance of the target vibration signal; This step involves a rapid "fluctuation detection" of the signal. Energy variance is a statistical value used to measure the severity of signal energy fluctuations. The system performs mathematical calculations on the received target vibration signal to determine the dispersion of its energy distribution. If the calculated variance value is small, it indicates that the signal is stable, possibly just a calm background environment; if the variance value is large, it indicates that the signal energy fluctuates wildly, suggesting sudden vibrations or severe disturbances. The purpose of this step is not to analyze the specific shape of the signal, but to quantify the severity of the signal changes.

[0060] Step S112: When the energy variance is greater than or equal to the preset noise threshold, it is determined that the target vibration signal satisfies the preset abnormal characteristics; This step determines whether signal noise reduction gating is necessary. A preset noise threshold serves as a baseline, representing the system's tolerance level for normal environmental quietness; it's typically an empirical value obtained from long-term operation of the target pipeline segment. The system compares the energy variance calculated in the previous step with this baseline. If the variance exceeds the threshold, it means the current vibration intensity exceeds the range of ordinary background noise, and is highly likely caused by external events such as excavation or walking. In this case, the system marks the signal as "meets anomalous characteristics" (i.e., "it is active / suspicious"), triggering subsequent complex noise reduction and classification processes. Conversely, if the threshold is not exceeded, the system ignores the signal to conserve computational resources.

[0061] The DAS signal noise reduction device in the embodiments of this invention is described below from the perspective of hardware processing. Please refer to [link / reference needed]. Figure 3 This is a schematic diagram of the physical device structure of a DAS signal noise reduction device in the embodiments of this application.

[0062] It should be noted that, Figure 3 The structure of the DAS signal noise reduction device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0063] like Figure 3 As shown, the DAS signal noise reduction device includes a CPU 301, which can perform various appropriate actions and processes according to a program stored in the read-only memory ROM 302 or a program loaded from the storage section 308 into the random access memory RAM 303, such as performing the methods described in the above embodiments. The RAM 303 also stores various programs and data required for system operation. The CPU 301, ROM 302, and RAM 303 are interconnected via a bus 304. An I / O interface 305 is also connected to the bus 304.

[0064] The following components are connected to I / O interface 305: input section 306 including audio input devices, push-button switches, etc.; output section 307 including a liquid crystal display (LCD) and audio output devices, indicator lights, etc.; storage section 308 including a hard disk, etc.; and communication section 309 including a network interface card such as a LAN (Local Area Network) card, modem, etc. Communication section 309 performs communication processing via a network such as the Internet. Drive 310 is also connected to I / O interface 305 as needed. Removable media 311, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on drive 310 as needed so that computer programs read from them can be installed into storage section 308 as needed.

[0065] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing computer programs for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 309, and / or installed from removable medium 311. When the computer program is executed by CPU 301, it performs the various functions defined in the present invention.

[0066] It should be noted that specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0067] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. Each block in a flowchart or block diagram may represent a module, program segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those shown in the drawings.

[0068] Specifically, the DAS signal noise reduction device in this embodiment includes a processor and a memory. The memory stores a computer program, and when the computer program is executed by the processor, it implements the DAS signal noise reduction method provided in the above embodiment.

[0069] In another aspect, the present invention also provides a computer-readable storage medium, which may be included in the DAS signal noise reduction device described in the above embodiments; or it may exist independently and not assembled into the DAS signal noise reduction device. The storage medium carries one or more computer programs, which, when executed by a processor of the DAS signal noise reduction device, cause the DAS signal noise reduction device to implement the DAS signal noise reduction method provided in the above embodiments.

[0070] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM or random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A method for denoising DAS signals, characterized in that, The method includes: When the target vibration signal of the target pipeline segment meets the preset abnormal characteristics, several noise reduction strategies are executed in sequence according to the priority of strategy execution, until all the noise reduction strategies are executed or the noise reduction score output by any of the noise reduction strategies is greater than the preset score threshold. Executing the noise reduction strategy includes: Select target noise reference data corresponding to the current target noise reduction strategy from the reference noise database corresponding to the target pipeline segment. The target noise reduction strategy is any one of the several noise reduction strategies. Based on the target noise reference data and the target noise reduction strategy, perform noise reduction processing on the target vibration signal to obtain the target noise-reduced vibration signal. Input the target noise-reduced vibration signal into a neural network classifier to obtain several classification results, wherein each classification result corresponds to a classification probability value. Take the classification probability value with the largest probability value as the noise reduction score, and take the classification result corresponding to the noise reduction score as the first final classification result. When any of the noise reduction scores is greater than the preset score threshold, the corresponding noise-reduced vibration signal is taken as the final vibration signal, and the final vibration signal and the first final classification result corresponding to the final vibration signal are output. When all the noise reduction strategies have been executed, the noise-reduced vibration signal with the highest noise reduction score is determined as the final vibration signal, and the final vibration signal and the first final classification result corresponding to the final vibration signal are output.

2. The method according to claim 1, characterized in that, The reference noise database includes a spatial reference data pool, a temporal reference data pool, and a statistical reference data pool; The spatial reference data pool is used to store the first vibration signals of several first pipeline segments within a first preset range before and after the target pipeline segment at the current moment. The time reference data pool is used to store the second vibration signal of the target pipeline segment within a first preset time period before the current time, and the noise of the second vibration signal is less than a preset noise threshold. The statistical reference data pool is used to store the background noise spectrum feature data. The background noise spectrum feature data is calculated based on the third vibration signal collected within a second preset time period before the current moment, based on a number of second pipeline segments within a second preset range before and after the target pipeline segment. The second preset time period is longer than the first preset time period, and the second preset range is larger than the first preset range.

3. The method according to claim 2, characterized in that, The noise reduction strategy specifically includes: When the target noise reduction strategy is the first noise reduction strategy, the first vibration signal with the least noise is selected as the target noise reference data, and the target vibration signal is noise-reduced using the minimum mean square error noise reduction algorithm based on the target noise reference data. When the target noise reduction strategy is the second noise reduction strategy, the second vibration signal that is closest to the current time of the target vibration signal is selected as the target noise reference data, and the target vibration signal is noise-reduced by the minimum mean square error noise reduction algorithm according to the target noise reference data. When the target noise reduction strategy is the third noise reduction strategy, the background noise spectrum feature data is selected as the target noise reference data, and the target vibration signal is noise-reduced using the minimum mean square error noise reduction algorithm based on the target noise reference data. When the target noise reduction strategy is the fourth noise reduction strategy, the second vibration signal closest to the current time of the vibration signal is selected as the target noise reference data, and the gain parameter of the minimum mean square error noise reduction algorithm is adjusted according to the preset strategy. Then, the target vibration signal is denoised according to the target noise reference data.

4. The method according to claim 1, characterized in that, The method further includes: Within a preset statistical period, the denoised vibration signal of each of the denoising strategies is determined as the cumulative number of times the final vibration signal is obtained. Based on the cumulative number of each noise reduction strategy in descending order, the noise reduction strategies are reordered, and the sorting result is used as the updated strategy execution priority order.

5. The method according to claim 2, characterized in that, The method further includes: When all the noise reduction strategies have been executed, the noise-reduced vibration signal and noise reduction score output by each noise reduction strategy are obtained. Based on the noise reduction scores, calculate the fusion weight value corresponding to each of the noise-reduced vibration signals; Based on the fusion weight values, the denoised vibration signals are weighted and superimposed to obtain the fused denoised signal. The fused and denoised signal is input into the neural network classifier to obtain several fused classification results, wherein each fused classification result corresponds to a fused classification probability value, and the fused classification result with the highest fused classification probability value is taken as the second final classification result. Output the fused and denoised signal and the second final classification result.

6. The method according to claim 5, characterized in that, The method further includes: The final vibration signal with the classification result belonging to the preset normal background noise category and the fused noise-reduced signal with the fusion classification result belonging to the preset normal background noise category are stored in the time reference data pool.

7. The method according to claim 2, characterized in that, The method further includes: Calculate the energy variance of the target vibration signal; When the energy variance is greater than or equal to the preset noise threshold, the target vibration signal is determined to meet the preset abnormal characteristics.

8. A DAS signal noise reduction device, characterized in that, The DAS signal noise reduction device includes: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code including computer instructions, and the one or more processors call the computer instructions to cause the DAS signal noise reduction device to perform the method as described in any one of claims 1-7.

9. A computer-readable storage medium comprising instructions, characterized in that, When the instruction is executed on the DAS signal noise reduction device, the DAS signal noise reduction device performs the method as described in any one of claims 1-7.

10. A computer program product, characterized in that, When the computer program product is run on the DAS signal noise reduction device, the DAS signal noise reduction device performs the method as described in any one of claims 1-7.

Images