A dual-angle DAS data processing method based on multimodal matching

Abstract

The present invention discloses a dual-angle DAS data processing method based on multimodal matching, which relates to the technical field of electronic digital data processing. The dual-angle DAS data processing method based on multimodal matching includes the following steps: DAS feature extraction; DAS data feature matching; dual-angle feature information acquisition; and dual-angle feature interaction efficiency determination. The present invention quantifies the feature matching validity in the DAS data through the acquired feature complementary data and performs feature matching validity determination. At the same time, it quantifies the dual-angle feature interaction efficiency based on the acquired feature interaction data and performs dual-angle feature interaction efficiency determination. This achieves the effect of improving the integrity of the corresponding dual-angle DAS data information extraction during the multimodal dynamic matching process, and solves the problem of incomplete dual-angle DAS data information extraction during the multimodal dynamic matching process in the prior art.

Description

Technical Field

[0001] The present invention relates to the technical field of electronic digital data processing, and in particular to a dual-angle DAS data processing method based on multi-modal matching. Background Art

[0002] With the rapid development of oil and gas exploration and development, geological disaster monitoring, and structural health monitoring, the demand for high-precision, high-resolution distributed acoustic sensing technology is becoming increasingly urgent. Distributed acoustic sensing (DAS) technology achieves distributed sensing of acoustic or vibration signals by measuring changes in backscattered light signals within optical fibers. It offers advantages such as immunity to electromagnetic interference, corrosion resistance, and long-distance monitoring. However, in practical applications, DAS systems face challenges such as complex environmental noise interference, signal attenuation, and difficulties in multimodal data fusion, which compromise the accuracy and reliability of data processing.

[0003] The core of multimodal data fusion lies in the compatibility of data from different modalities. Because data from different modalities differ in time, space, and scale, direct fusion can lead to information distortion or redundancy. Therefore, achieving efficient matching and fusion of multimodal data has become a key issue in DAS data processing.

[0004] Existing technologies use multimodal data fusion techniques, such as feature fusion, spatial fusion, or deep learning fusion methods, to align and integrate data from different modalities. Secondly, dual-angle analysis can provide a more comprehensive understanding of the characteristics and patterns of DAS data, thereby extracting more valuable information.

[0005] For example, the publication number CN117251574A discloses a text classification and extraction method and system based on multi-feature data fusion, which includes: preprocessing the data to be processed to obtain first data; performing data embedding processing on the first data based on preset data embedding rules to obtain second data; performing feature extraction on the second data to obtain multiple feature data; determining the weight of each feature data based on a preset scoring model; determining a feature vector matrix and a weight matrix based on multiple feature data and weights; and determining the input of the classification layer based on the feature vector matrix and the weight matrix.

[0006] For example, the publication number CN119513172A discloses a drawing data extraction method based on multimodal features, which includes: S1, receiving multimodal data and preprocessing the multimodal data; S2, performing feature extraction on the data of each modality, and generating 256-dimensional feature vectors for each modality; S3, performing multi-channel fusion and feature alignment on the features generated by different modal feature generators; S4, using labeled normal sample data and a multimodal autoencoder model to perform model training; S5, inputting a new multimodal data into multiple feature generators to generate multimodal features, fusing them to generate a C*256 feature matrix, and then inputting the C*256 feature matrix into the trained multimodal autoencoder for data extraction.

[0007] However, in the process of implementing the technical solutions of the invention in the embodiments of the present application, the present application found that the above technology has at least the following technical problems:

[0008] In the existing technology, different modal data may differ in quantity and scale, and there may be a lack of effective filtering and denoising algorithms during the fusion process, which is easily affected by information redundancy and noise interference. In addition, in dual-angle analysis, the correlation and complementarity between the two angles are usually not fully considered, and a complete fusion strategy is not established to comprehensively extract information. There is a problem of incomplete information extraction of dual-angle DAS data during multimodal dynamic matching. Summary of the Invention

[0009] The embodiment of the present application solves the problem of incomplete extraction of corresponding dual-angle DAS data information during multimodal dynamic matching of dual-angle DAS data in the prior art by providing a dual-angle DAS data processing method based on multimodal matching, thereby improving the integrity of the extraction of corresponding dual-angle DAS data information during multimodal dynamic matching.

[0010] An embodiment of the present application provides a dual-angle DAS data processing method based on multimodal matching, comprising the following steps: S1, obtaining a DAS waveform and a DAS time-frequency diagram corresponding to the DAS data after denoising, and adding two-dimensional position coding to the DAS waveform and the DAS time-frequency diagram to introduce the spatiotemporal features of the DAS data; S2, obtaining the horizontal features and the vertical features in the spatiotemporal features through directional convolution, and quantifying the feature matching validity in the DAS data based on the obtained feature complementary data, and performing a feature matching validity judgment, which is used to determine whether to reduce the low-frequency components in the DAS signal by increasing the sampling frequency and reducing the distance between sensor points; S3, performing dual-angle feature interaction on the horizontal features and the vertical features that meet the feature matching validity, and obtaining dual-angle feature information, which is used to reflect the feature redundancy suppression effect between the horizontal features and the vertical features; S4, quantifying the dual-angle feature interaction efficiency of the horizontal features and the vertical features based on the obtained feature interaction data, and performing a dual-angle feature interaction efficiency judgment, which is used to determine whether to enhance the linear correlation between the horizontal features and the vertical features by increasing attention entropy and information gain.

[0011] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0012] 1. Through directional convolution, the horizontal features and vertical features in the spatiotemporal features are respectively obtained, and then the feature matching effectiveness in the DAS data is quantified based on the obtained feature complementary data, and the feature matching effectiveness is judged. Finally, the dual-angle feature interaction efficiency is quantified based on the obtained feature interaction data, and the dual-angle feature interaction efficiency is judged, thereby achieving an improvement in the accuracy of the dual-angle feature interaction between the horizontal features and the vertical features in the DAS data, and then achieving an improvement in the integrity of the corresponding dual-angle DAS data information extraction in the multimodal dynamic matching process, effectively solving the problem of incomplete dual-angle DAS data information extraction in the multimodal dynamic matching process in the prior art.

[0013] 2. By judging whether the high-level duty cycle exceeds the set threshold, it is decided whether to increase the filtering frequency to optimize signal processing and reduce noise interference. At the same time, the matching attention mechanism is used to dynamically adjust the filtering parameters to improve the accuracy of signal processing. Through refined feature interaction data acquisition and processing, the processing accuracy and efficiency of DAS data are effectively improved, redundant information is reduced, and the accuracy of feature matching is enhanced. It provides a strong guarantee for the integrity of information extraction in the multimodal dynamic matching process, which helps to achieve more accurate data analysis and early warning in fields such as earthquake monitoring.

[0014] 3. By comparing the high-level duty cycle at the end of the feature interaction period with the high-level duty cycle set in the database, a first comparison result is obtained, and at the same time, the first comparison result obtained is corrected in combination with the high-level duty cycle weight value to obtain the high-level duty cycle coefficient. The result of inverse proportional processing of the obtained feature interaction data volume coefficient is coupled with the high-level duty cycle coefficient and the feature interaction redundancy coefficient to obtain the dual-angle feature interaction efficiency value, thereby achieving an improvement in the accuracy of obtaining the dual-angle feature interaction efficiency value, and further achieving an improvement in the accuracy and reliability of the dual-angle feature interaction efficiency evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 A flowchart of a dual-angle DAS data processing method based on multimodal matching provided in an embodiment of the present application;

[0016] Figure 2 Output logic diagram of horizontal features and vertical features provided in the embodiment of the present application;

[0017] Figure 3 A logic diagram of the dual-angle feature interaction provided in an embodiment of the present application;

[0018] Figure 4 A logic diagram of the matching attention mechanism provided in an embodiment of the present application;

[0019] Figure 5 A logic diagram of a dual-angle DAS data processing method based on multimodal matching provided in an embodiment of the present application;

[0020] Figure 6 Flowchart for determining the validity of feature matching provided in an embodiment of the present application;

[0021] Figure 7 A flowchart of the effectiveness determination of the dual-angle feature interaction provided in an embodiment of the present application;

[0022] Figure 8 An interface diagram of an intelligent DAS seismic signal parsing and analysis system for a dual-angle DAS data processing method based on multi-modal matching provided in an embodiment of the present application;

[0023] Figure 9 This is one of the data analysis interface diagrams in the intelligent DAS seismic signal parsing and analysis system provided in an embodiment of the present application;

[0024] Figure 10 Figure 2 of the data analysis interface in the intelligent DAS seismic signal parsing and analysis system provided in an embodiment of the present application;

[0025] Figure 11This is a diagram of the parameter optimization and comparison interface in the intelligent DAS seismic signal parsing and analysis system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0026] The embodiment of the present application solves the problem of incomplete extraction of dual-angle DAS data information corresponding to dual-angle DAS data in the multimodal dynamic matching process in the prior art by providing a dual-angle DAS data processing method based on multimodal matching. The method obtains the DAS waveform and DAS time-frequency diagram corresponding to the DAS data after denoising, and adds two-dimensional position coding to the DAS waveform and DAS time-frequency diagram respectively to introduce the spatiotemporal features of the DAS data. Then, the horizontal features and vertical features in the spatiotemporal features are obtained respectively through directional convolution. At the same time, the feature matching validity in the DAS data is quantified based on the obtained feature complementary data, and the feature matching validity is judged. Then, dual-angle feature interaction is performed on the horizontal features and vertical features with qualified feature matching validity to obtain dual-angle feature information. Finally, the dual-angle feature interaction efficiency of the horizontal features and vertical features is quantified based on the obtained feature interaction data, and the dual-angle feature interaction efficiency is judged, thereby improving the integrity of the corresponding dual-angle DAS data information extraction in the multimodal dynamic matching process.

[0027] The technical solution in the embodiment of the present application is to solve the problem of incomplete extraction of the corresponding dual-angle DAS data information during the multimodal dynamic matching process. The overall idea is as follows:

[0028] Through directional convolution, the horizontal and vertical features in the spatiotemporal features are obtained respectively. Then, the feature matching effectiveness in the DAS data is quantified based on the obtained feature complementary data, and the feature matching effectiveness is judged. Finally, the dual-angle feature interaction efficiency is quantified based on the obtained feature interaction data, and the dual-angle feature interaction efficiency is judged, which achieves the effect of improving the integrity of the corresponding dual-angle DAS data information extraction in the multimodal dynamic matching process.

[0029] In order to better understand the above technical solution, the above technical solution will be described in detail below with reference to the accompanying drawings and specific implementation methods.

[0030] like Figure 1 FIG. 1 is a flowchart of a dual-angle DAS data processing method based on multimodal matching provided by an embodiment of the present application, the method comprising the following steps:

[0031] S1, obtain the DAS waveform and DAS time-frequency diagram corresponding to the DAS data after denoising, and add two-dimensional position coding to the DAS waveform and DAS time-frequency diagram respectively to introduce the spatiotemporal characteristics of the DAS data. The DAS waveform is used to visualize the change of the vibration amplitude of the DAS signal corresponding to the DAS data over time, and the DAS time-frequency diagram is used to visualize the change of the vibration amplitude of the DAS signal corresponding to the DAS data in the two dimensions of time and frequency. Denoising refers to joint spatiotemporal wavelet denoising, which can effectively capture the instantaneous characteristics of the DAS signal.

[0032] S2, through directional convolution, respectively obtains the horizontal and vertical features in the spatiotemporal features. At the same time, based on the obtained feature complementary data, the feature matching effectiveness in the DAS data is quantified, and the feature matching effectiveness judgment is performed. The feature matching effectiveness judgment is used to determine whether to reduce the low-frequency components in the DAS signal by increasing the sampling frequency and reducing the distance between sensor points. The spatiotemporal features include a preset multimodal model to visualize the temporal and spatial characteristics of the DAS data. The horizontal features are used to reflect the periodic changes of the DAS signal in time, and the vertical features are used to reflect the spatial distribution law of the DAS signal at sensor points at different locations.

[0033] S3, performs dual-angle feature interaction on the horizontal features and vertical features that meet the feature matching validity requirements to obtain dual-angle feature information. The dual-angle feature information is used to reflect the feature redundancy suppression effect between the horizontal features and the vertical features. The dual-angle feature interaction is performed in the matching attention mechanism, which usually means taking the horizontal features as queries and the vertical features as keys, and capturing the modulation relationship of space to time by scaling the dot product attention. The matching attention mechanism is used to highlight the differences between the horizontal features and the vertical features to reduce feature redundancy.

[0034] S4, quantify the dual-angle feature interaction efficiency of horizontal features and vertical features based on the acquired feature interaction data, and perform dual-angle feature interaction efficiency judgment. The dual-angle feature interaction efficiency judgment is used to determine whether to enhance the linear correlation between horizontal features and vertical features by increasing attention entropy and information gain.

[0035] The data collected by distributed fiber optic acoustic sensing in real time includes, but is not limited to, feature complementarity data and feature interaction data. DAS data often contains various noises, such as environmental noise (sound wave interference from the surrounding environment, such as traffic noise and wind) and equipment noise (noise generated by the sensor itself, such as thermal noise from electronic components). DAS data represents information related to sound waves (or vibrations) collected by distributed fiber optic sensors. Distributed fiber optic sensing technology uses optical fiber as the sensing medium. When external sound waves or vibrations act on the fiber, they cause light scattering (such as Rayleigh scattering) or phase changes in the fiber. By detecting, demodulating, and analyzing these optical signals, information about the sound waves or vibrations distributed along the fiber can be obtained. This information is stored and processed in digital form. For example, in pipeline safety monitoring, when the pipeline is subjected to external vibrations such as excavation or impact, the optical fiber senses these vibrations and converts them into analyzable DAS data.

[0036] The DAS data in the DAS waveform diagram are arranged in ascending chronological order, with time as the horizontal axis and the vibration amplitude of the DAS data as the vertical axis. The intensity changes of the DAS signal at different time points can be directly observed, helping to identify the occurrence time and duration of the vibration event. For example, when monitoring seismic activity, the DAS waveform diagram can clearly show the arrival time of the seismic wave and the changes in the strength of the vibration; the DAS time-frequency diagram has the horizontal axis as time, the vertical axis as frequency, and the color (or grayscale) as the vibration amplitude of the sound wave signal. It can show the changes in the frequency components of the DAS signal over time, helping to identify the frequency characteristics and time evolution of the vibration event. For example, when monitoring the operating status of mechanical equipment, the DAS time-frequency diagram can be used to observe the changes in the vibration of the equipment at different frequencies over time, thereby determining whether the equipment has a fault.

[0037] The preset multimodal model can be a deep learning model, such as a convolutional neural network or a Transformer model, which can simultaneously process DAS data from different modalities (waveform graphs and time-frequency graphs) and extract the spatiotemporal features of the DAS data. That is, the input of the preset multimodal model is the DAS signal corresponding to the DAS data, and the output is the horizontal and vertical features corresponding to the DAS data. The visualization part can display the extracted spatiotemporal features in the form of images or heat maps, helping users to understand the characteristics and laws of DAS data more intuitively.

[0038] like Figure 2As shown in the figure, the output logic diagram of the horizontal and vertical features provided in the embodiment of the present application is shown. The input features enter the vertical and horizontal multi-scale convolution modules respectively. The vertical module uses 3x1, 6x1, and 12x1 convolution kernels to extract vertical features of different scales; the horizontal module uses 1x3, 1x6, 1x12 and other convolution kernels to obtain horizontal features. After convolution, each module fuses features through point-by-point convolution. Finally, the horizontal and vertical fused features are further convolved point-by-point to output horizontal and vertical features. This design can mine feature information from different directions and scales, enhancing feature representation capabilities.

[0039] like Figure 3 As shown, it is a logic diagram of the dual-angle feature interaction provided by the embodiment of the present application, using F A With F B Represents a dual-angle feature, the sizes of which are , after the maximum-minimum normalization, the two enter the two convolutions respectively to obtain: 、 、 、 ,make: and 、 and Perform matrix multiplication respectively to obtain 、 , and finally And remapping is performed to obtain the final deep features with the same size as the input features. This method can fully explore the complementary information between features and solve the problem of interference in information.

[0040] Among them, C represents the number of channels, H represents the height, and W represents the width. 、 、 、 Indicates F A With F B After maximum-minimum normalization, the two convolution operations are performed to obtain the results. R represents the real space, Conv represents the convolution operation, Concate represents the concatenation operation, and ⨂ represents the fusion operation.

[0041] It should be understood that the inputs of the preset multimodal model used in this example are the DAS waveform and DAS time-frequency diagram. First, a short-time Fourier transform (STFT) is performed on each channel of the obtained DAS data to obtain the power spectral density of each channel, and the time-frequency diagram of each channel is restored to 6000 time points through time axis interpolation. Finally, a DAS time-frequency diagram of the same size as the DAS waveform is obtained by frequency averaging and splicing each time-frequency diagram. Each time-frequency diagram is obtained using the following formula:

[0042]

[0043] Where, X k (t,f) represents the time-frequency of the kth signal at time t and frequency f, X k (τ) represents the amplitude of the kth data at time point t, N is the window length, w represents the Hamming window, interp1d represents the linear interpolation operation, t original Indicates the number of points before the linear interpolation operation, t new Indicates the number of points after the linear interpolation operation, P k (t,f) represents the power spectrum density after linear interpolation operation, M final (t,k) represents the time-frequency diagram obtained by splicing the signals of each channel, which means the average power spectrum density of the kth channel at time t.

[0044] Two-dimensional position encoding is used to encode the row and column position information of DAS data into one-dimensional vectors, which are then fused to generate the final two-dimensional features. First, the two-dimensional position (i, j) is decomposed into independent row coordinates i and column coordinates j, and encoded separately to avoid positional relationship confusion caused by mixed row and column calculations. Taking row encoding as an example, the sine function absolute position encoding in the Transformer is used to generate unique vectors for different rows:

[0045]

[0046] Among them, i represents the position of the current row or column, d represents the total dimension of the position encoding vector, u is the index of the position encoding dimension currently calculated, and the column encoding uses the same method to generate the vector PE col (j), row encoding uses the same method to generate vector PE row (i), the final two-dimensional position encoding is: PE(i,j)=PE row (i)+PE col (j), that is, PE(i,j) represents the encoding of the two-dimensional features corresponding to the row coordinate i and the column coordinate j.

[0047] This example effectively extracts spatiotemporal features from DAS data through multimodal data processing. Two-dimensional position encoding is added to comprehensively incorporate spatiotemporal information, enhancing feature richness. Directional convolution is used to obtain horizontal and vertical features, accurately reflecting the temporal and spatial variations of DAS signals. The effectiveness of feature matching is quantified based on feature complementarity data, enabling targeted adjustments to the sampling frequency and sensor spacing to reduce interference from low-frequency components. Dual-angle feature interaction is implemented within a matching attention mechanism to highlight feature differences and reduce redundancy. Quantifying and determining the efficiency of dual-angle feature interaction enhances linear correlation between features. This overall approach improves the accuracy and reliability of DAS data processing, enhancing the integrity of information extracted from dual-angle DAS data during multimodal dynamic matching. This approach provides more accurate data support for fields such as earthquake monitoring and pipeline safety monitoring, helping to promptly detect abnormal events and ensure safe operations.

[0048] Furthermore, the feature complementary data, the specific acquisition steps include: obtaining the transverse vibration amplitude and the longitudinal vibration amplitude, the transverse vibration amplitude is used to reflect the vibration intensity of the DAS signal in the direction perpendicular to the optical fiber axis (i.e., transverse direction), and the longitudinal vibration amplitude is used to reflect the vibration intensity of the DAS signal in the direction of the optical fiber axis (i.e., longitudinal direction); if the obtained transverse vibration amplitude or longitudinal vibration amplitude is greater than the corresponding maximum allowable vibration amplitude in the database, a convolution kernel weight adjustment instruction is sent, otherwise a feature vector is obtained, and when the matching attention mechanism receives the convolution kernel weight adjustment instruction, the weight of each convolution kernel in the directional convolution is dynamically adjusted according to the input vibration amplitude deviation, thereby generating an adaptive dynamic convolution, and the feature vector includes a transverse feature vector and a longitudinal feature vector. The transverse feature vector represents the feature information related to the transverse vibration obtained when it is within the allowable range of the transverse vibration amplitude, and the longitudinal feature vector represents the feature information related to the longitudinal vibration obtained when it is within the allowable range of the longitudinal vibration amplitude. The transverse feature vector indicates that the obtained transverse vibration amplitude is within the allowable range of the transverse vibration amplitude, which is not greater than the maximum allowable transverse vibration amplitude set in the database, and the longitudinal feature vector indicates that the obtained longitudinal vibration amplitude is within the allowable range of the longitudinal vibration amplitude, which is not greater than the maximum allowable longitudinal vibration amplitude set in the database.

[0049] The effectiveness of feature matching in DAS data is quantified based on the acquired feature complementary data, including: comparing the acquired horizontal feature vector and vertical feature vector with the corresponding reference feature vector in the database respectively, and performing cosine similarity calculation based on the comparison results to obtain a feature matching effectiveness value. The feature matching effectiveness value is used to quantify the effectiveness of feature matching of horizontal features and vertical features within the feature matching period. The feature matching effectiveness value represents the result of summing the horizontal feature matching effectiveness value and the vertical feature matching effectiveness value.

[0050] Among them, the conditions for obtaining the transverse eigenvector and the longitudinal eigenvector are: the obtained transverse vibration amplitude and the longitudinal vibration amplitude are not greater than the maximum allowable vibration amplitude set in the database; the set maximum allowable transverse vibration amplitude is represented by the result of summing and averaging the maximum values ​​of the historical transverse vibration amplitudes in the database, and the set maximum allowable longitudinal vibration amplitude is represented by the result of summing and averaging the maximum values ​​of the historical longitudinal vibration amplitudes in the database.

[0051] In this embodiment, if Figure 4 The figure below shows the logic diagram of the matching attention mechanism provided by an embodiment of the present application. By performing specific processing and transformations on input features, the matching relationships between features are captured, thereby generating more representative and targeted output features. This mechanism primarily involves processing two input features, and ultimately outputs two output features through operations such as normalization, feature interaction (based on multiplication operations), and multi-head attention mechanisms (MAP and 1-MAP).

[0052] This example dynamically adjusts the convolution kernel weights to adapt the processing to different vibration conditions, improving the accuracy and adaptability of feature extraction. The acquired feature vector contains rich lateral and longitudinal vibration feature information, providing comprehensive data support for subsequent analysis. Quantifying the effectiveness of feature matching can intuitively evaluate the degree of feature matching, helping to determine the reliability of data processing, thereby effectively improving the accuracy and efficiency of DAS data processing, providing more accurate and reliable data for monitoring and analysis in related fields, and helping to promptly identify potential problems and take corresponding measures.

[0053] It should be added that, Figure 5 The figure shows a logic diagram of a dual-angle DAS data processing method based on multimodal matching provided by an embodiment of the present application. It starts with preprocessing multimodal data such as time domain waveforms, including operations such as segmentation and filtering. It then performs feature extraction, such as extracting waveform features and spectral features. Subsequently, multimodal joint analysis is carried out, and different modal features are matched and fused. In the process, a matching attention mechanism is used to enhance feature relevance. Directional convolution and other processing are also performed to mine feature information from both the horizontal and vertical directions. Finally, after processing such as linear layers, the fusion matching result is output. This method effectively integrates the multimodal information of dual-angle DAS data through multi-step and multi-module collaboration, improving the accuracy and effectiveness of data processing.

[0054] like Figure 6The figure shows a flowchart for determining the validity of feature matching provided by an embodiment of the present application. The specific design logic is as follows: first, horizontal and vertical features are acquired and the matching validity is quantified. Then, a determination is made as to whether the first condition (i.e., the first determination condition) is met. If so, feature interaction is performed directly. If not, a further determination is made as to whether the second condition (i.e., the second determination condition) is met. If the second condition is met, the sensor point spacing is adjusted and low-frequency noise is suppressed. Then, a determination is made as to whether the amplitude meets the standard. If so, feature interaction is performed. If not, the sampling frequency is adjusted. After adjusting the sampling frequency, a determination is made as to whether the adjusted condition is met. If so, feature interaction is performed. If not, an alarm is issued.

[0055] Specifically, the feature matching validity judgment is performed, including: judging whether the feature matching process within the feature matching period is valid based on the acquired feature matching validity value: if the acquired feature matching validity value meets the first judgment condition, it is recorded as the feature matching is valid and dual-angle feature interaction is performed, and the first judgment condition indicates that the acquired feature matching validity value is not less than the feature matching validity value set in the database; if the acquired feature matching validity value meets the second judgment condition, it is recorded as the feature matching is invalid and the sensor point spacing is adjusted, and the second judgment condition indicates that the acquired feature matching validity value is less than the feature matching validity value set in the database.

[0056] Among them, the specific steps of adjusting the distance between the sensor points include: obtaining the upper limit frequency of the low-frequency signal at the end of the feature matching period, and obtaining the sampling frequency increase in combination with the actual sampling frequency of the sampling device. The sampling frequency increase represents the difference between twice the upper limit frequency of the low-frequency signal and the actual sampling frequency of the corresponding sampling device at the end of the feature matching period. According to the Nyquist sampling theorem, in order to be able to restore the original signal from the sampled signal without distortion, the sampling frequency must be at least twice the upper limit frequency of the signal. This is the basic theory in the field of signal processing; the obtained feature matching validity value deviation and the sensor point spacing deviation (the result after de-unitization) are harmonically averaged and the result is used as the sensor point spacing reduction. The feature matching validity value deviation is used to quantify the degree of difference between the set feature matching validity value and the obtained feature matching validity value, that is, the difference between the set feature matching validity value and the obtained feature matching validity value. The sensor point spacing deviation represents the absolute value of the difference between the sensor point spacing of the corresponding sampling device at the end of the feature matching period and the sensor point spacing set in the database. The set sensor point spacing represents the initial sensor point spacing set by the preset personnel. The obtained sampling frequency increase and sensor point spacing decrease are summed and input into the adaptive morphological filtering algorithm to suppress low-frequency noise in the DAS signal. After one sensor point spacing adjustment, the effect of the sensor point spacing adjustment is judged.

[0057] The effect of the sensor point spacing adjustment is determined, and the specific steps include: obtaining the high-frequency signal ratio of the corresponding horizontal features and longitudinal features after the sensor point spacing is adjusted once, and generating a corresponding sinusoidal wave signal. If the sinusoidal wave signal amplitude is not less than the sinusoidal wave signal amplitude set in the database, the sensor point spacing adjustment is completed and dual-angle feature interaction is performed, otherwise the sampling frequency is adjusted, and the high-frequency signal ratio represents the ratio of high-frequency signal energy to total signal energy; the sampling frequency is adjusted, and the specific steps include: harmonically averaging the feature matching validity value deviation and the sinusoidal wave signal amplitude deviation obtained after the sensor point spacing is adjusted once as the secondary sampling frequency increase; if the feature matching validity value obtained after the preset number of sampling frequency adjustments meets the second judgment condition, an alarm instruction for the sampling device is sent, otherwise it is recorded as a valid feature match and dual-angle feature interaction is performed, and the sinusoidal wave signal amplitude deviation represents the difference between the sinusoidal wave signal amplitude obtained after the sensor point spacing is adjusted once and the sinusoidal wave signal amplitude set in the database. The set sinusoidal wave signal amplitude is represented by the result of summing and averaging the historical sinusoidal wave signal amplitudes in the database.

[0058] The aforementioned database is a database for storing various types of setting data established before the design of the dual-angle DAS data processing method based on multimodal matching. The database includes but is not limited to the set feature matching validity value, the set sinusoidal wave signal amplitude, the set dual-angle feature interaction efficiency value, the feature matching period, and the feature interaction period. The various values ​​are directly set by technical personnel. The setting basis of the set feature matching validity value can be determined according to the actual application scenario of DAS data processing. For example, the set feature matching validity value is represented by the result of summing and averaging the historical feature matching validity values ​​at the end of the historical feature matching period in the database. In addition, the various values ​​in the database can be set and fine-tuned by technical personnel according to actual debugging.

[0059] It should be added that the adaptive filtering of the adaptive morphological filtering algorithm is specifically as follows: first, dynamic structural elements are generated, and in the time dimension, based on the local signal variance σ 2 (t) Adjust the length L of the structural element t , according to the adjacent channel correlation ρ in the spatial dimension k,k+1 To select the shape of the structural element, that is:

[0060]

[0061]

[0062] Where, f s represents the sampling rate, σ th Through the statistical setting of the data set, the filtered signal is finally obtained through the corrosion-expansion cascade:

[0063]

[0064] In the formula, γ represents the opening operation, Φ represents the closing operation, and L t and B shape are the space-time structural elements, D raw Represents a matrix in the space-time dimension.

[0065] In this embodiment, the adaptive morphological filtering algorithm can dynamically adjust the size and shape of the structural elements of the morphological operation according to the sum of the input sampling frequency increase and the reduction in the sensor point spacing, accurately match the signal characteristics, and effectively suppress the low-frequency noise in the DAS signal; the secondary sampling frequency increase can assist the algorithm in more flexibly optimizing the processing parameters, strengthen the targeted suppression of complex low-frequency noise, and improve the purity and quality of the DAS signal.

[0066] This example uses feature matching validity determination to accurately identify the effectiveness of the feature matching process and ensure data processing accuracy. When adjusting sensor point spacing, factors such as the upper frequency limit of the low-frequency signal and the deviation of the feature matching validity value are comprehensively considered to scientifically calculate the reduction in sensor point spacing and effectively suppress low-frequency noise. The effectiveness of the sensor point spacing adjustment is determined by combining the proportion of high-frequency signals and the amplitude of the sine wave signal to ensure that the adjustment achieves the desired effect. When adjusting the sampling frequency, the increase in the secondary sampling frequency is rationally determined based on the deviation of the feature matching validity value and the amplitude of the sine wave signal, improving data processing quality. This overall approach improves the stability and reliability of DAS data processing, providing high-quality data support for subsequent analysis.

[0067] Furthermore, the feature interaction data, the acquisition step includes: L1, at the end of the feature interaction period, judging whether the acquired high-level duty cycle is greater than the high-level duty cycle set in the database, if so, obtaining the high-level duty cycle deviation and sending a filter frequency increase instruction, otherwise executing L2 (the condition for executing L2 is: the acquired high-level duty cycle is not greater than the high-level duty cycle set in the database), the high-level duty cycle deviation represents the difference between the acquired high-level duty cycle and the set high-level duty cycle, and the set high-level duty cycle is represented by the result of summing and averaging the historical high-level proportions at the end of the historical feature matching period in the database, and filtering The wave frequency increase instruction is used by the matching attention mechanism to increase the filtering frequency of the corresponding amplitude according to the input high-level duty cycle deviation; L2, obtains the amount of data successfully uploaded and input into the matching attention mechanism for storage, recorded as the feature interaction data amount; L3, obtains the redundant part of the interaction between the horizontal features and the vertical features in the matching attention mechanism, recorded as the redundant interaction data amount, obtains the total interaction intensity in the matching attention mechanism, recorded as the feature interaction intensity, compares the obtained redundant interaction data amount with the feature interaction intensity, and obtains the feature interaction redundancy, that is, the ratio of the redundant interaction data amount to the feature interaction intensity.

[0068] In this embodiment, by determining the high-level duty cycle and adjusting the filtering frequency, signal processing can be optimized, noise interference can be reduced, and data quality can be improved. Calculating feature interaction redundancy can intuitively reflect the proportion of redundant information in the feature interaction process, which helps evaluate the effect of feature interaction. This overall approach helps to more accurately extract effective features, reduce redundant information interference, improve the efficiency and accuracy of feature interaction, and provide more reliable feature data for subsequent data processing and analysis.

[0069] Furthermore, the dual-angle feature interaction efficiency of the horizontal feature and the vertical feature is quantified based on the acquired feature interaction data. The specific steps include:

[0070] First, the high-level duty cycle at the end of the characteristic interaction period is compared with the high-level duty cycle set in the database to obtain the first comparison result (i.e., KB / KB0). At the same time, the first comparison result is corrected in combination with the high-level duty cycle weight value to obtain the high-level duty cycle coefficient. The specific restriction expression of the high-level duty cycle coefficient TJH1 is: , where TJH1 represents the high-level duty cycle coefficient of the horizontal feature and the vertical feature at the end of the feature interaction period, b1 represents the high-level duty cycle weight value, KB represents the high-level duty cycle of the horizontal feature and the vertical feature at the end of the feature interaction period, and KB0 represents the set high-level duty cycle.

[0071] Then, the feature interaction data volume at the end of the feature interaction period is compared with the initial feature interaction data volume at the beginning of the feature interaction period to obtain a second comparison result (i.e., JH / JH0). At the same time, the second comparison result is corrected in combination with the feature interaction data volume weight value to obtain the feature interaction data volume coefficient. The specific restriction expression of the feature interaction data volume coefficient TJH2 is: , TJH2 represents the coefficient of feature interaction data volume between horizontal features and vertical features at the end of the feature interaction period, b2 represents the weight value of feature interaction data volume, JH represents the feature interaction data volume between horizontal features and vertical features at the end of the feature interaction period, JH0 represents the set feature interaction data volume, and the set feature interaction data volume is represented by the sum and average of the historical feature interaction data volumes at the end of the historical feature interaction period in the database.

[0072] Next, the feature interaction redundancy at the end of the feature interaction period is compared with the initial feature interaction redundancy at the beginning of the feature interaction period to obtain a third comparison result (i.e., RY / RY0). At the same time, the third comparison result is corrected based on the feature interaction redundancy weight value to obtain the feature interaction redundancy coefficient. The specific restricted expression of the feature interaction redundancy coefficient TJH3 is: , TJH3 represents the feature interaction redundancy coefficient of the horizontal feature and the vertical feature at the end of the feature interaction period, b3 represents the feature interaction redundancy weight value, RY represents the feature interaction redundancy of the horizontal feature and the vertical feature at the end of the feature interaction period, RY0 represents the set feature interaction redundancy, and the set feature interaction redundancy is represented by the sum and average of the historical feature interaction redundancies at the end of the historical feature interaction period in the database.

[0073] Finally, the result of inverse proportional processing of the acquired feature interaction data volume coefficient is coupled with the high-level duty cycle coefficient and the feature interaction redundancy coefficient to obtain the dual-angle feature interaction efficiency value. The dual-angle feature interaction efficiency value represents the quantitative data of the influence of the feature interaction data on the dual-angle feature interaction efficiency. The feature interaction data includes the high-level duty cycle, the feature interaction data volume and the feature interaction redundancy. The specific restriction expression of the dual-angle feature interaction efficiency value TJH is: , TJH represents the dual-angle feature interaction efficiency value of the horizontal feature and the longitudinal feature at the end of the feature interaction period.

[0074] In the database, a series of weight values ​​have been pre-set for the dual-angle feature interaction efficiency values. There is a clear, predefined correlation pattern between these weight values ​​and the high-level duty cycle, the amount of feature interaction data, and the feature interaction redundancy. This correlation is not formed randomly, and its expression can be a one-to-one precise correspondence, or a many-to-one mapping of multiple groups of parameters pointing to a weight value. Specifically, in practical applications, the high-level duty cycle, the amount of feature interaction data, and the feature interaction redundancy monitored in real time can be directly input into this preset correlation system, and the weight values ​​that match the high-level duty cycle, the amount of feature interaction data, and the feature interaction redundancy can be quickly and accurately parsed out, namely the high-level duty cycle weight value, the feature interaction data weight value, and the feature interaction redundancy weight value.

[0075] What is particularly critical is that in order to ensure the uniformity and comparability of the evaluation results, the value range of the three weight values ​​in this system is limited to the interval of 0 to 1, and the sum of these three weight values ​​is always equal to 1.

[0076] In this embodiment, the dual-angle feature interaction efficiency value increases with the increase of the high-level duty cycle and the feature interaction redundancy, and decreases with the increase of the feature interaction data volume. Among them, the increase of the high-level duty cycle means that the duration of the high level in the signal becomes longer, which may increase the redundant information in the feature interaction process, that is, the feature interaction redundancy increases. The two are positively correlated to a certain extent.

[0077] At the same time, as the high-level duty cycle increases, the effective data carried by the high level per unit time may increase, thereby increasing the amount of feature interaction data, showing a positive correlation trend. However, if there is too much redundant information, it may also limit the growth of the effective data volume.

[0078] The increase in feature interaction redundancy will reduce the proportion of valid data in the total interaction data, that is, the amount of feature interaction data will be "relatively" reduced; and when the amount of valid feature interaction data increases, the corresponding processing response time will increase, which may also lead to an increase in redundancy. The two are mutually restrictive and interrelated.

[0079] By considering the above-mentioned mutual influence mechanism, it is helpful to accurately control the three key parameters of high-level duty cycle, feature interaction data volume and feature interaction redundancy, so that they can reach a dynamic equilibrium state in the dual-angle feature interaction process. In this equilibrium state, the system can efficiently process feature interaction tasks in an optimal resource allocation manner, avoiding the negative impact on the overall feature interaction efficiency due to excessive changes in a certain parameter, thereby achieving an improvement in the integrity of the corresponding dual-angle DAS data information extraction in the multimodal dynamic matching process, and effectively solving the problem of incomplete dual-angle DAS data information extraction in the multimodal dynamic matching process in the existing technology.

[0080] like Figure 7 As shown, it is a flowchart of the dual-angle feature interaction efficiency determination provided by the embodiment of the present application. The specific design logic is: first, the dual-angle feature interaction efficiency is quantified based on the feature interaction data. Then, it is determined whether the dual-angle feature interaction efficiency value is greater than the set value. If it is greater than, the dual-angle feature interaction is determined to be invalid, the efficiency value deviation is obtained and the attention parameter optimization is performed; if it is not greater than, the interaction is determined to be valid, and the deep feature extraction of the DAS data is completed. After that, it is determined again whether the increase in the horizontal and vertical interaction efficiency is greater than the set range. If so, the attention parameter optimization is completed and recorded as the dual-angle feature interaction is valid; if not, the preset personnel are prompted to correct the matching attention mechanism.

[0081] Specifically, the dual-angle feature interaction efficiency is determined, and the specific steps include: if the obtained dual-angle feature interaction efficiency value is greater than the dual-angle feature interaction efficiency value set in the database, it is recorded as invalid dual-angle feature interaction and the dual-angle feature interaction efficiency value deviation is obtained, and the attention parameters are optimized at the same time; if the obtained dual-angle feature interaction efficiency value is not greater than the dual-angle feature interaction efficiency value set in the database, it is recorded as valid dual-angle feature interaction and the deep feature extraction of DAS data is completed, and the set dual-angle feature interaction efficiency value is represented by the sum and average of the historical dual-angle feature interaction efficiency values ​​at the end of the historical feature interaction period in the database; the dual-angle feature interaction efficiency value deviation is used to quantify the degree of difference between the obtained dual-angle feature interaction efficiency value and the set dual-angle feature interaction efficiency value, that is, the difference between the obtained dual-angle feature interaction efficiency value and the set dual-angle feature interaction efficiency value; the attention parameters include attention entropy and information gain; attention entropy is used to measure the accuracy of feature interaction between horizontal features and vertical features in the matching attention mechanism; information gain is used to measure the attention of the matching attention mechanism to the corresponding feature interaction between horizontal features and vertical features.

[0082] Among them, the attention parameter optimization, the specific steps include: taking the deviation of the result of the reconciliation processing of the obtained dual-angle feature interaction efficiency value deviation and the learning temperature parameter deviation as the increase of attention entropy, and inputting it into the matching attention mechanism for orthogonal initialization, the learning temperature parameter deviation represents the difference between the reference learning temperature parameter in the database and the learning temperature parameter of the matching attention mechanism at the end of the feature interaction period, and the reference learning temperature parameter is represented by the result of summing and averaging the historical learning temperature parameters of the matching attention mechanism at the end of the historical feature interaction period in the database; taking the deviation of the result of the reconciliation processing of the obtained dual-angle feature interaction efficiency value deviation and the multi-target loss value deviation as the increase of information gain, and inputting it into the matching attention mechanism to suppress the low-level duty cycle, the multi-target loss value deviation represents the difference between the reference multi-target loss value in the database and the multi-target loss value of the matching attention mechanism at the end of the feature interaction period, and the reference multi-target loss value is represented by the difference between the reference multi-target loss value in the database and the multi-target loss value of the matching attention mechanism at the end of the feature interaction period. The target loss value is represented by the sum and average of the historical multi-target loss values ​​of the matching attention mechanism in the database at the end of the historical feature interaction period; after one-time attention parameter optimization, the increase in horizontal interaction efficiency and the increase in vertical interaction efficiency are obtained. If the obtained increase in horizontal interaction efficiency and the increase in vertical interaction efficiency are both greater than the increase in interaction efficiency set in the database, the attention parameter optimization is completed and recorded as the dual-angle feature interaction is valid, otherwise the preset personnel are prompted to correct the matching attention mechanism. One-time attention parameter optimization includes one-time increase in attention entropy and one-time increase in information gain. The increase in horizontal interaction efficiency represents the difference between the horizontal interaction efficiency regained after one-time attention parameter optimization and the horizontal interaction efficiency gained at the end of the feature interaction period. The increase in vertical interaction efficiency represents the difference between the vertical interaction efficiency regained after one-time attention parameter optimization and the vertical interaction efficiency gained at the end of the feature interaction period.

[0083] In this embodiment, the matching attention mechanism dynamically adjusts its own attention weight distribution through the increase in attention entropy and information gain of the input. During the feature interaction process, different features contribute differently to the interaction results. The matching attention mechanism can assign reasonable attention weights to each feature based on the input information, paying more attention to features that have a greater impact on the interaction results, thereby improving the efficiency and accuracy of feature interaction. For example, in some scenarios, certain features may have less information at a low duty cycle, but have important value at a high duty cycle. The matching attention mechanism can reduce the focus on low-value features at a low duty cycle and increase the focus on high-value features at a high duty cycle by adjusting the increase in information gain, thereby optimizing the entire dual-angle feature interaction process.

[0084] It is necessary to add that, if Figure 8As shown, this is an interface diagram of an intelligent DAS seismic signal parsing and analysis system for a dual-angle DAS data processing method based on multimodal matching provided in an embodiment of the present application, that is, the homepage of the intelligent DAS seismic signal parsing and analysis system. The top navigation bar is provided with options such as "Homepage", "Data Processing", "Data Analysis", and "Parameter Optimization" to facilitate users to jump quickly. Among them, the "Underground Pipeline Monitoring" module on the left can monitor and analyze underground pipeline-related data; the "Graphical Data Analysis" module in the middle supports visualization of seismic signals and other data; the "Dual-Angle Feature Analysis" module on the right focuses on parsing DAS data features from dual angles. Each module is equipped with an icon and a "View Analysis" button, and users can click to enter the detailed operation page. The interface design is intuitive and easy to use, with clear functional divisions. As Figure 9 As shown in FIG. 1 , it is one of the data analysis interface diagrams in the intelligent DAS seismic signal parsing and analysis system provided in the embodiment of the present application, as shown in FIG. Figure 10 As shown, this is the second diagram of the data analysis interface in the intelligent DAS seismic signal parsing and analysis system provided in the embodiment of the present application, with "Refresh Results" and "Export All" buttons at the top. The page is divided into four sections: the upper left is a line graph of "Underground Pipeline Vibration Monitoring", which shows the changes in vibration amplitude at different times, and the corresponding list below is marked with specific values; the upper right is a bar graph of "DAS Time-Frequency Graph Analysis", which shows the distribution and characteristics of frequency components over time; the lower left is a line graph of "Feature Matching Evaluation", which shows the changes in feature matching with serial number, and the matching evaluation results are given below; the lower right is a line graph of "Dual-Angle Feature Interaction Analysis", which shows the changes in interaction efficiency, and the interaction efficiency evaluation results and alarm prompts are given below. Each section is equipped with function buttons such as "Update" and "Export" for easy operation.

[0085] like Figure 11 As shown, this is a parameter optimization and comparison interface diagram in the intelligent DAS seismic signal parsing and analysis system provided by an embodiment of the present application. The interface is divided into two parts: the left side is the "Feature Matching Parameter Optimization" area, which shows that the current feature matching degree is 0.7 (greater than the threshold value 0.6), and the feature matching is valid and does not need to be optimized. At the same time, example optimization suggestions are given, such as increasing the sampling frequency when the feature matching degree is low, etc. The original and recommended sampling frequencies and sensor point spacing are also listed, and there is an "Apply Optimization Parameters" button below; the right side is the "Dual-Angle Feature Interaction Efficiency Optimization" area, the current interaction efficiency is 0.4 (less than the threshold value 0.5), the system alarms and recommends optimization, gives optimization suggestions, and compares the interaction efficiency before and after optimization through a bar chart. There are attention entropy and information gain sliders below, as well as "Apply Optimization" and "Reset Parameters" buttons.

[0086] The interface design of the intelligent DAS seismic signal parsing and analysis system is intuitive and the functional divisions are clear. The homepage navigation bar allows users to quickly switch between functional modules, and the icons and buttons of each monitoring and analysis module are set for easy operation. The data analysis interface displays information such as vibration amplitude, time-frequency distribution, feature matching, and interaction efficiency in a variety of charts. It also has functions such as refresh and export, which facilitate data viewing and processing. The parameter optimization and comparison interface can intuitively present the feature matching and interaction efficiency, provide optimization suggestions and before-and-after comparisons, and also provide parameter adjustment and reset functions. Through a clear interface and rich functions, the overall system effectively improves the efficiency of parsing and analysis of DAS seismic signals, and enhances the accuracy and practicality of applications such as underground pipeline monitoring.

[0087] In summary, the embodiment of the present application obtains the horizontal features and vertical features in the spatiotemporal features respectively through directional convolution, and then quantifies the feature matching effectiveness in the DAS data based on the obtained feature complementary data, and performs feature matching effectiveness judgment, and finally quantifies the dual-angle feature interaction efficiency based on the obtained feature interaction data, and performs dual-angle feature interaction efficiency judgment, thereby achieving an improvement in the accuracy of dual-angle feature interaction between the horizontal features and the vertical features in the DAS data, and then achieving an improvement in the integrity of the corresponding dual-angle DAS data information extraction in the multimodal dynamic matching process, effectively solving the problem of incomplete dual-angle DAS data information extraction in the multimodal dynamic matching process in the prior art.

[0088] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

Claims

1. A dual-angle DAS data processing method based on multimodal matching, characterized in that: The following steps are involved: S1, obtaining the DAS waveform and DAS time-frequency diagram corresponding to the DAS data after denoising, and adding two-dimensional position coding to the DAS waveform and DAS time-frequency diagram respectively to introduce the spatiotemporal characteristics of the DAS data. The DAS waveform is used to visualize the change of the vibration amplitude of the DAS signal corresponding to the DAS data over time, and the DAS time-frequency diagram is used to visualize the change of the vibration amplitude of the DAS signal corresponding to the DAS data in two dimensions of time and frequency. The two-dimensional position coding is used to encode the position information of the rows and columns of the DAS data into one-dimensional vectors respectively, and then generate the final two-dimensional features through fusion. The spatiotemporal characteristics include the temporal characteristics and spatial characteristics of the DAS data visualized by the preset multimodal model; S2, respectively obtains the horizontal and vertical features of the spatiotemporal features through directional convolution, and quantifies the feature matching validity in the DAS data based on the acquired feature complementary data, and performs feature matching validity judgment. The horizontal features are used to reflect the periodic changes of the DAS signal in time, and the vertical features are used to reflect the spatial distribution pattern of the DAS signal at different sensor points. The feature matching validity judgment is used to determine whether to reduce the low-frequency components in the DAS signal by increasing the sampling frequency and reducing the distance between the sensor points; S3, performing dual-angle feature interaction on the horizontal features and vertical features that meet the feature matching validity requirements to obtain dual-angle feature information, wherein the dual-angle feature interaction is performed in a matching attention mechanism, and the dual-angle feature information is used to reflect the feature redundancy suppression effect between the horizontal features and the vertical features; S4, quantifying the dual-angle feature interaction efficiency of the horizontal features and the vertical features based on the acquired feature interaction data, and performing a dual-angle feature interaction efficiency judgment, wherein the dual-angle feature interaction efficiency judgment is used to determine whether to enhance the linear correlation between the horizontal features and the vertical features by increasing the attention entropy and the information gain.

2. The dual-angle DAS data processing method based on multimodal matching according to claim 1, characterized in that: The specific steps of obtaining the complementary feature data include: Obtaining a transverse vibration amplitude and a longitudinal vibration amplitude, wherein the transverse vibration amplitude is used to reflect the vibration intensity of the DAS signal in a direction perpendicular to the optical fiber axis, and the longitudinal vibration amplitude is used to reflect the vibration intensity of the DAS signal in the direction of the optical fiber axis; If the obtained lateral vibration amplitude or longitudinal vibration amplitude is greater than the corresponding maximum allowable vibration amplitude in the database, a convolution kernel weight adjustment instruction is sent; otherwise, a feature vector is obtained. The convolution kernel weight adjustment instruction is used to dynamically adjust the weight of each convolution kernel in the directional convolution. The feature vector includes a lateral feature vector and a longitudinal feature vector.

3. The dual-angle DAS data processing method based on multimodal matching according to claim 2, characterized in that: The quantifying the feature matching validity in the DAS data based on the acquired feature complementary data and performing feature matching validity determination includes: Step 1: Compare the acquired horizontal feature vector and vertical feature vector with the corresponding reference feature vector in the database, and calculate the cosine similarity based on the comparison results to obtain a feature matching validity value, which is used to quantify the feature matching validity of the horizontal feature and the vertical feature within the feature matching period; Step 2: Determine whether the feature matching process within the feature matching period is valid based on the acquired feature matching validity value: If the obtained feature matching validity value meets the first determination condition, the feature matching is recorded as valid and dual-angle feature interaction is performed, wherein the first determination condition indicates that the obtained feature matching validity value is not less than the feature matching validity value set in the database; If the acquired feature matching validity value meets the second determination condition, the feature matching is recorded as invalid and the sensing point spacing is adjusted. The second determination condition indicates that the acquired feature matching validity value is less than the feature matching validity value set in the database.

4. The dual-angle DAS data processing method based on multimodal matching according to claim 3, characterized in that: The specific steps of adjusting the distance between sensing points include: Obtain the upper limit frequency of the low-frequency signal at the end of the feature matching period, and simultaneously obtain the sampling frequency increase in combination with the actual sampling frequency of the sampling device; The obtained feature matching validity value deviation and the sensor point spacing deviation are harmonically averaged to obtain a result as the sensor point spacing reduction amount, wherein the feature matching validity value deviation is used to quantify the degree of difference between the set feature matching validity value and the obtained feature matching validity value; The obtained sampling frequency increase and sensor point spacing decrease are summed and input into the adaptive morphological filtering algorithm to suppress the low-frequency noise in the DAS signal. The effect of the sensor point spacing adjustment is then judged after one sensor point spacing adjustment.

5. The dual-angle DAS data processing method based on multimodal matching according to claim 4, characterized in that: The specific steps of determining the effect of adjusting the distance between the sensing points include: Obtain the high-frequency signal proportions of the horizontal and vertical features corresponding to the sensor point spacing adjustment, and generate corresponding sinusoidal signals. If the amplitude of the sinusoidal signal is not less than the amplitude of the sinusoidal signal set in the database, the sensor point spacing adjustment is completed and dual-angle feature interaction is performed. Otherwise, adjust the sampling frequency. The specific steps of adjusting the sampling frequency include: The result of harmonic averaging of the feature matching validity value deviation and the sine wave signal amplitude deviation obtained after the primary sensor point spacing adjustment is used as the secondary sampling frequency increase; If the feature matching validity value reacquired after the preset number of sampling frequency adjustments meets the second judgment condition, a sampling device alarm instruction is sent; otherwise, the feature matching is recorded as valid and dual-angle feature interaction is performed.

6. The dual-angle DAS data processing method based on multimodal matching according to claim 1, characterized in that: The step of acquiring the feature interaction data includes: L1, at the end of the feature interaction period, determines whether the acquired high-level duty cycle is greater than the high-level duty cycle set in the database. If so, obtain the high-level duty cycle deviation and send a filter frequency increase instruction. Otherwise, execute L2. The filter frequency increase instruction is used for the matching attention mechanism to increase the filter frequency of the corresponding amplitude according to the input high-level duty cycle deviation; L2, obtains the amount of data successfully uploaded and input into the matching attention mechanism for storage, recorded as the feature interaction data volume; L3, obtain the redundant part of the interaction between horizontal features and vertical features in the matching attention mechanism, recorded as the redundant interaction data volume, obtain the total interaction intensity in the matching attention mechanism, recorded as the feature interaction intensity, compare the obtained redundant interaction data volume with the feature interaction intensity to obtain the feature interaction redundancy.

7. The dual-angle DAS data processing method based on multimodal matching according to claim 6, characterized in that: The method of quantifying the dual-angle feature interaction efficiency of the horizontal feature and the vertical feature based on the acquired feature interaction data specifically comprises the following steps: Comparing the high-level duty cycle at the end of the characteristic interaction period with the high-level duty cycle set in the database to obtain a first comparison result, and correcting the obtained first comparison result in combination with the high-level duty cycle weight value to obtain a high-level duty cycle coefficient; Comparing the feature interaction data volume at the end of the feature interaction period with the initial feature interaction data volume at the beginning of the feature interaction period to obtain a second comparison result, and simultaneously correcting the obtained second comparison result in combination with the feature interaction data volume weight value to obtain a feature interaction data volume coefficient; Comparing the feature interaction redundancy at the end of the feature interaction period with the initial feature interaction redundancy at the beginning of the feature interaction period to obtain a third comparison result, and correcting the obtained third comparison result in combination with the feature interaction redundancy weight value to obtain a feature interaction redundancy coefficient; The result of inverse proportional processing of the acquired feature interaction data volume coefficient is coupled with the high-level duty cycle coefficient and the feature interaction redundancy coefficient to obtain a dual-angle feature interaction efficiency value. The dual-angle feature interaction efficiency value represents quantitative data of the degree of influence of the feature interaction data on the dual-angle feature interaction efficiency. The feature interaction data includes the high-level duty cycle, the feature interaction data volume and the feature interaction redundancy.

8. The dual-angle DAS data processing method based on multimodal matching according to claim 7, characterized in that: The specific steps of determining the efficiency of the dual-angle feature interaction include: If the obtained dual-angle feature interaction efficiency value is greater than the dual-angle feature interaction efficiency value set in the database, it is recorded as invalid dual-angle feature interaction and the dual-angle feature interaction efficiency value deviation is obtained, and the attention parameter optimization is performed at the same time; If the obtained dual-angle feature interaction efficiency value is not greater than the dual-angle feature interaction efficiency value set in the database, it is recorded as the dual-angle feature interaction is valid and the deep feature extraction of DAS data is completed; The dual-angle feature interaction efficiency value deviation is used to quantify the difference between the obtained dual-angle feature interaction efficiency value and the set dual-angle feature interaction efficiency value; The attention parameters include attention entropy and information gain; The attention entropy is used to measure the accuracy of feature interaction between horizontal features and vertical features in the matching attention mechanism; The information gain is used to measure the attention paid by the matching attention mechanism to the interaction between the corresponding features of the horizontal features and the vertical features.

9. The dual-angle DAS data processing method based on multimodal matching according to claim 8, characterized in that: The attention parameter optimization specifically includes the following steps: The obtained deviation of the dual-angle feature interaction efficiency value and the learning temperature parameter deviation are reconciled and the resulting deviation is used as the increase in attention entropy and input into the matching attention mechanism for orthogonal initialization; The obtained deviation of the dual-angle feature interaction efficiency value and the multi-target loss value is reconciled and the resulting deviation is used as the information gain increase, which is input into the matching attention mechanism to suppress the low-level duty cycle. After one-time attention parameter optimization, the increase in horizontal interaction efficiency and the increase in vertical interaction efficiency are obtained. If the obtained increase in horizontal interaction efficiency and the increase in vertical interaction efficiency are both greater than the increase in interaction efficiency set in the database, the attention parameter optimization is completed and recorded as the dual-angle feature interaction is valid. Otherwise, the preset personnel are prompted to correct the matching attention mechanism. The one-time attention parameter optimization includes one increase in attention entropy and one increase in information gain.

Images