A signal processing method for quantum-enhanced distributed fiber optic monitoring

Abstract

The application discloses a signal processing method for quantum-enhanced distributed optical fiber monitoring, and belongs to the technical field of optical fiber water seepage detection. Firstly, the application injects quantum-enhanced probe light into an optical fiber and collects echo signals, and obtains echo difference signals by subtracting signal values at adjacent moments; then, the echo difference signals and the echo signals are normalized respectively, and signal enhancement is completed by combining a signal enhancement coefficient to obtain difference-enhanced signals and original-enhanced signals; then, the two types of enhanced signals are divided into multiple sub-signals and normal sub-signals are marked, and then four abnormal change vectors are obtained based on a difference main amplitude, an original main amplitude, and a difference high-frequency energy and an original high-frequency energy; finally, a seepage recognition neural network is used to process the four abnormal change vectors, and a water seepage state result is output. Through the signal processing strategy of quantum enhancement and multi-dimensional abnormal change feature fusion, the application improves the accuracy of water seepage detection in distributed optical fiber monitoring.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of optical fiber water seepage detection, and particularly relates to a signal processing method for quantum-enhanced distributed optical fiber monitoring. BACKGROUND

[0002] Dam water seepage monitoring is one of the important technical means to ensure the safe operation of hydraulic structures. In the prior art, a common method is a water seepage monitoring technology based on the amplitude change of distributed optical fiber echo signals. This method usually lays distributed optical fibers in the dam body or dam foundation, injects probe light into the optical fiber, collects echo signals distributed along the optical fiber, and uses the change of the echo signal amplitude over time to determine whether there is water seepage anomaly at the corresponding position. When the echo signal amplitude at a certain position changes relative to the historical baseline, it is considered that there may be a water seepage risk at this position.

[0003] However, when the water seepage is in the early stage or the seepage intensity is weak, the change amplitude of the echo signal caused by the water seepage is small, and is easily covered by environmental noise, structural vibration and other non-water seepage factors, so that the water seepage characteristics are not obvious, thereby affecting the accurate identification of the water seepage state, and the problem of low water seepage detection accuracy exists. SUMMARY

[0004] In view of the above problems in the prior art, the signal processing method for quantum-enhanced distributed optical fiber monitoring provided by the present application solves the problem of low water seepage detection accuracy in the prior art.

[0005] In order to achieve the above-mentioned application purposes, the technical scheme adopted by the present application is as follows: a signal processing method for quantum-enhanced distributed optical fiber monitoring, comprising:

[0006] injecting quantum-enhanced probe light into the optical fiber, collecting echo signals, subtracting the signal values of adjacent time points in the echo signals to obtain echo difference signals;

[0007] normalizing the echo difference signals and the echo signals respectively, obtaining signal enhancement coefficients, performing signal enhancement to obtain difference-enhanced signals and original-enhanced signals;

[0008] dividing the difference-enhanced signals and the original-enhanced signals into a plurality of sub-signals, and marking normal sub-signals;

[0009] obtaining difference main frequency amplitude and original main frequency amplitude anomaly vectors according to the difference between each sub-signal and the normal sub-signal in the difference main amplitude and the original main amplitude;

[0010] obtaining difference high frequency energy and original high frequency energy anomaly vectors according to the difference between each sub-signal and the normal sub-signal in the difference high frequency energy and the original high frequency energy;

[0011] The four abnormal vectors are processed by the permeation recognition neural network to obtain a water infiltration state result.

[0012] Further, the process of obtaining the differential enhanced signal and the original enhanced signal comprises:

[0013] The echo differential signal and the echo signal are normalized respectively to obtain a normalized echo differential signal and a normalized echo signal.

[0014] The first signal enhancement coefficient is obtained from the normalized echo differential signal, and the normalized echo differential signal is enhanced to obtain the differential enhanced signal.

[0015] The second signal enhancement coefficient is obtained from the normalized echo signal, and the normalized echo signal is enhanced to obtain the original enhanced signal.

[0016] Further, the process of obtaining the differential enhanced signal comprises:

[0017] In the normalized echo differential signal, the local differential energy at each time is obtained.

[0018] The local differential energy at all times is averaged to obtain a global energy average.

[0019] The ratio of the local differential energy at each time to the global energy average is taken as the differential energy ratio.

[0020] The differential energy ratio is mapped to the first signal enhancement coefficient.

[0021] The first signal enhancement coefficient at each time is multiplied by the signal value at the corresponding time of the normalized echo differential signal to obtain the differential enhanced signal.

[0022] Further, the process of obtaining the original enhanced signal comprises:

[0023] In the normalized echo signal, the local mean and the local fluctuation amplitude at each time are obtained.

[0024] The ratio of the local fluctuation amplitude to the local mean is taken as the fluctuation ratio.

[0025] The fluctuation ratio is mapped to the second signal enhancement coefficient.

[0026] The second signal enhancement coefficient at each time is multiplied by the signal value at the corresponding time of the normalized echo signal to obtain the original enhanced signal.

[0027] Further, the process of dividing the differential enhanced signal and the original enhanced signal into a plurality of sub-signals and marking the normal sub-signals comprises:

[0028] The differential enhancement signal is divided into a plurality of sub-signals to obtain a plurality of differential enhancement sub-signals, and the original enhancement signal is divided into a plurality of sub-signals to obtain a plurality of original enhancement sub-signals;

[0029] An average signal value is obtained for the differential enhancement signal to obtain a global differential amplitude;

[0030] An average signal value is obtained for each differential enhancement sub-signal to obtain a segment differential amplitude;

[0031] When the segment differential amplitude is less than the global differential amplitude, the corresponding differential enhancement sub-signal is marked as a normal differential enhancement sub-signal;

[0032] A standard deviation is obtained for the original enhancement signal to obtain a global fluctuation value;

[0033] A standard deviation is obtained for each original enhancement sub-signal to obtain a segment fluctuation value;

[0034] When the segment fluctuation value is less than the global fluctuation value, the corresponding original enhancement sub-signal is marked as a normal original enhancement sub-signal.

[0035] Further, the process of obtaining the differential dominant frequency amplitude and the original dominant frequency amplitude anomaly vector includes:

[0036] Fourier transform is performed on each differential enhancement sub-signal and original enhancement sub-signal, respectively, to obtain a differential spectrum and an original spectrum;

[0037] The maximum amplitude is extracted from the differential spectrum and the original spectrum, respectively, to obtain a differential dominant amplitude and an original dominant amplitude;

[0038] The differential dominant amplitudes belonging to the normal differential enhancement sub-signals are averaged to obtain a differential comparative dominant amplitude;

[0039] The original dominant amplitudes belonging to the normal original enhancement sub-signals are averaged to obtain an original comparative dominant amplitude;

[0040] The differential dominant frequency amplitude anomaly value is obtained according to the difference between each differential dominant amplitude and the differential comparative dominant amplitude, and the differential dominant frequency amplitude anomaly values are arranged in time sequence to obtain a differential dominant frequency amplitude anomaly vector;

[0041] The original dominant frequency amplitude anomaly value is obtained according to the difference between each original dominant amplitude and the original comparative dominant amplitude, and the original dominant frequency amplitude anomaly values are arranged in time sequence to obtain an original dominant frequency amplitude anomaly vector.

[0042] Further, the process of obtaining the differential high-frequency energy and the original high-frequency energy anomaly vector includes:

[0043] Fourier transform is performed on each differential enhancement sub-signal and original enhancement sub-signal, respectively, to obtain a differential spectrum and an original spectrum;

[0044] extracting high frequency ranges from the difference spectrum and the original spectrum respectively to obtain a difference high frequency spectrum and an original high frequency spectrum;

[0045] adding absolute values of amplitude values corresponding to each frequency point in the difference high frequency spectrum to obtain a difference high frequency energy;

[0046] adding absolute values of amplitude values corresponding to each frequency point in the original high frequency spectrum to obtain an original high frequency energy;

[0047] taking a mean value of each difference high frequency energy belonging to a normal difference enhancer signal to obtain a difference contrast high frequency energy;

[0048] taking a mean value of each original high frequency energy belonging to a normal original enhancer signal to obtain an original contrast high frequency energy;

[0049] obtaining a difference high frequency energy anomaly value according to a difference between each difference high frequency energy and the difference contrast high frequency energy, arranging the difference high frequency energy anomaly values in a time occurrence order to obtain a difference high frequency energy anomaly vector;

[0050] obtaining an original high frequency energy anomaly value according to a difference between each original high frequency energy and the original contrast high frequency energy, arranging the original high frequency energy anomaly values in a time occurrence order to obtain an original high frequency energy anomaly vector.

[0051] Further, the penetration recognition neural network comprises a main frequency amplitude anomaly feature fusion unit, a high frequency energy anomaly feature fusion unit, a feature extraction splicing unit, a double-channel attention enhancement unit and an output layer.

[0052] Further, the main frequency amplitude anomaly feature fusion unit is configured to extract features from the difference main frequency amplitude anomaly vector and the original main frequency amplitude anomaly vector, and to fuse and process the features to obtain main frequency amplitude anomaly fusion features;

[0053] The high frequency energy anomaly feature fusion unit is configured to extract features from the difference high frequency energy anomaly vector and the original high frequency energy anomaly vector, and to fuse and process the features to obtain high frequency energy anomaly fusion features;

[0054] The feature extraction splicing unit is configured to extract mean value features and maximum value features from the main frequency amplitude anomaly fusion features and the high frequency energy anomaly fusion features respectively, and to splice the features to obtain main frequency amplitude-high frequency energy mean value splicing features and main frequency amplitude-high frequency energy maximum value splicing features;

[0055] The double-channel attention enhancement unit is configured to apply attention to the main frequency amplitude-high frequency energy mean value splicing features and the main frequency amplitude-high frequency energy maximum value splicing features respectively to obtain attention enhanced features;

[0056] The output layer is configured to output a water seepage state result according to the attention enhanced feature.

[0057] Further, the main frequency amplitude anomaly feature fusion unit and the high frequency energy anomaly feature fusion unit have the same structure, and each includes a first one-dimensional convolution layer, a second one-dimensional convolution layer and a multiplier M1.

[0058] The first one-dimensional convolution layer is configured to extract a differential shallow layer feature from the differential main frequency amplitude anomaly vector or the differential high frequency energy anomaly vector.

[0059] The second one-dimensional convolution layer is configured to extract an original shallow layer feature from the original main frequency amplitude anomaly vector or the original high frequency energy anomaly vector.

[0060] The multiplier M1 is configured to multiply the differential shallow layer feature and the original shallow layer feature element by element to obtain the main frequency amplitude anomaly fusion feature or the high frequency energy anomaly fusion feature.

[0061] The present application has the following advantages:

[0062] 1. The present application injects quantum enhanced probe light into the optical fiber. Compared with the ordinary probe light used in traditional distributed optical fiber monitoring, the quantum enhanced probe light has higher signal-to-noise ratio and detection sensitivity, and can effectively capture the weak echo signal changes caused by early dam seepage or weak seepage stage.

[0063] 2. The present application converts the original echo signal into a differential signal through differential operation of adjacent time echo signals. Differential processing can effectively filter out environmental noise, structural vibration and other static or slowly changing interference factors, and only retain the dynamic change amount of the signal over time, which can accurately highlight the subtle fluctuations of the signal caused by early seepage, and greatly improve the recognizability of weak seepage features.

[0064] 3. The present application performs normalization processing and signal enhancement on the differential signal and the original signal respectively, so that the seepage signal value is further highlighted and the seepage feature is prominent. At the same time, the differential main frequency amplitude, the original main frequency amplitude, the differential high frequency energy and the original high frequency energy of the sub-signal are extracted, which represents the signal anomaly features at different positions from two dimensions of amplitude change and energy distribution, realizes multi-dimensional accurate description of the seepage state, and avoids the problems of missed judgment and misjudgment caused by single feature dimension analysis.

[0065] 4. The present application uses a seepage recognition neural network to cooperatively process the four anomaly vectors, uses the feature fusion and pattern recognition ability of the neural network, learns the signal anomaly law under the seepage state, can effectively distinguish the signal changes caused by real seepage from the signal fluctuations caused by non-seepage interference factors, and significantly improves the recognition accuracy of the dam seepage state. BRIEF DESCRIPTION OF DRAWINGS

[0066] Figure 1 A flow chart of a signal processing method for quantum-enhanced distributed fiber monitoring;

[0067] Figure 2 A structural diagram of a penetration recognition neural network;

[0068] Figure 3 A structural diagram of a main frequency amplitude anomaly feature fusion unit;

[0069] Figure 4 A structural diagram of a high-frequency energy anomaly feature fusion unit;

[0070] Figure 5 A structural diagram of a feature extraction splicing unit;

[0071] Figure 6 A structural diagram of a dual-channel attention enhancement unit. DETAILED DESCRIPTION

[0072] The specific embodiments of the present application are described below to facilitate the understanding of the present application for those skilled in the art, but it should be clear that the present application is not limited to the scope of the specific embodiments, and for those skilled in the art, it is obvious that various changes are within the spirit and scope of the present application defined and determined by the appended claims, and all the inventions utilizing the concept of the present application are within the scope of protection.

[0073] As shown in Figure 1 A signal processing method for quantum-enhanced distributed fiber monitoring, comprising:

[0074] Injecting quantum-enhanced probe light into the optical fiber, collecting echo signals, subtracting the signal values of adjacent time points in the echo signals to obtain echo difference signals;

[0075] Respectively normalizing the echo difference signals and the echo signals, obtaining signal enhancement coefficients, performing signal enhancement to obtain difference-enhanced signals and original-enhanced signals;

[0076] Dividing the difference-enhanced signals and the original-enhanced signals into a plurality of sub-signals, and marking normal sub-signals;

[0077] According to the difference between each sub-signal and the normal sub-signal in the difference main amplitude and the original main amplitude, obtaining the difference main frequency amplitude and the original main frequency amplitude anomaly vector;

[0078] According to the difference between each sub-signal and the normal sub-signal in the difference high-frequency energy and the original high-frequency energy, obtaining the difference high-frequency energy and the original high-frequency energy anomaly vector;

[0079] The four anomaly vectors are processed by a penetration recognition neural network to obtain a water penetration state result.

[0080] In the embodiment, the quantum enhanced probe light is probe light modulated by quantum state, including squeezed state light, signal light in entangled photon pair, and probe light output by quantum illumination light source, and the core feature is to reduce probe noise and improve signal-to-noise ratio through quantum effect to capture weak echo signal change.

[0081] In the embodiment, the process of obtaining the differential enhanced signal and the original enhanced signal includes:

[0082] The echo differential signal and the echo signal are normalized respectively to obtain normalized echo differential signal and normalized echo signal.

[0083] The first signal enhancement coefficient is obtained from the normalized echo differential signal, and the normalized echo differential signal is enhanced to obtain the differential enhanced signal.

[0084] The second signal enhancement coefficient is obtained from the normalized echo signal, and the normalized echo signal is enhanced to obtain the original enhanced signal.

[0085] In the embodiment, the calculation formula of the echo differential signal is:

[0086] ,

[0087] Wherein, is the signal value of the first moment in the echo differential signal, is the signal value of the first moment in the echo signal, is the signal value of the first moment in the echo signal, is the absolute value.

[0088] In the embodiment, the maximum and minimum values are used for signal normalization, the maximum signal value and the minimum signal value are taken from the signal, the difference between the signal value at each moment and the minimum signal value is calculated to obtain the signal difference value, the maximum signal value is subtracted from the minimum signal value to obtain the maximum signal difference value, and the ratio of the signal difference value of the same signal to the maximum signal difference value is taken as the normalized value.

[0089] In the embodiment, the process of obtaining the differential enhanced signal includes:

[0090] In the normalized echo differential signal, the local differential energy at each moment is obtained:

[0091] ,

[0092] Wherein, is the local differential energy at the first moment, normalizing the signal value of each time point in the echo difference signal, normalizing the signal value of each time point in the echo difference signal, normalizing the signal value of each time point in the echo difference signal, normalizing the signal value of each time point in the echo difference signal, normalizing the signal value of each time point in the echo difference signal,

[0093] taking the average of the local difference energy of all time points to obtain the global energy average; taking the average of the local difference energy of all time points to obtain the global energy average;

[0094] taking the average of the local difference energy of all time points to obtain the global energy average; taking the average of the local difference energy of all time points to obtain the global energy average;

[0095] mapping the difference energy ratio to a first signal enhancement coefficient: mapping the difference energy ratio to a first signal enhancement coefficient:

[0096] mapping the difference energy ratio to a first signal enhancement coefficient: mapping the difference energy ratio to a first signal enhancement coefficient: mapping the difference energy ratio to a first signal enhancement coefficient:

[0097] mapping the difference energy ratio to a first signal enhancement coefficient: the first signal enhancement coefficient of the kth time point, the difference energy ratio of the kth time point, the difference energy ratio of the kth time point, the difference energy ratio of the kth time point, a natural constant; a natural constant;

[0098] multiplying the first signal enhancement coefficient of each time point with the signal value of the corresponding time point of the normalized echo difference signal to obtain a difference enhancement signal. multiplying the first signal enhancement coefficient of each time point with the signal value of the corresponding time point of the normalized echo difference signal to obtain a difference enhancement signal.

[0099] The present application can depict the local difference energy distribution of the signal at different time positions by introducing a local window-based energy calculation method in the normalized echo difference signal, so that the signal enhancement process has time adaptability and avoids indiscriminate amplification of the full time sequence signal. The present application compares the local difference energy with the global energy average to form a difference energy ratio, so that the enhancement coefficient reflects the abnormality degree of the current time signal relative to the overall background level. By mapping the difference energy ratio to a first signal enhancement coefficient and multiplying it with the original difference signal at each time point, the difference feature is enhanced specifically, which helps to highlight the local abnormal change and suppress the influence of the overall noise on the difference signal, thereby improving the distinguishability of the abnormal features in the difference signal.

[0100] In the present embodiment, the process of obtaining the original enhancement signal includes:

[0101] In the normalized echo signal, the local mean and the local fluctuation amplitude of each time point are obtained:

[0102]

[0103] the local mean of the kth time point, the local mean of the kth time point, the length of the local window, the length of the local window, a signal value of the normalized echo signal at the i th time point;

[0104]

[0105] wherein, a local fluctuation amplitude at the i th time point, a signal value of the normalized echo signal at the i th time point, and k is a number of the time point; a ratio of the local fluctuation amplitude to the local mean value as a fluctuation ratio;

[0106] mapping the fluctuation ratio to a second signal enhancement coefficient: wherein,

[0107] a second signal enhancement coefficient at the i th time point, a fluctuation ratio at the i th time point; multiplying the second signal enhancement coefficient at each time point with a signal value of the normalized echo signal at the corresponding time point to obtain an original enhancement signal.

[0108] multiplying the second signal enhancement coefficient at each time point with a signal value of the normalized echo signal at the corresponding time point to obtain an original enhancement signal.

[0109] The present application calculates the local mean value and the local fluctuation amplitude of the normalized echo signal, the local mean value can represent the reference level of the signal in the window, and the local fluctuation amplitude can quantify the discrete degree of the signal relative to the reference, and the fluctuation ratio is constructed by the ratio of the local fluctuation amplitude to the local mean value, which can eliminate the signal reference amplitude difference of different monitoring positions and different time periods, and focus on the relative fluctuation characteristics of the signal. In the early stage of water seepage, the fluctuation degree of the signal amplitude will be significantly improved relative to the normal state, and the fluctuation ratio can amplify this relative change, so that the weak water seepage characteristics hidden by the background amplitude are converted into a quantifiable difference index. The time points with significant fluctuations are enhanced, and the influence of the background smooth part is suppressed.

[0110] In the present embodiment, the process of dividing the difference enhancement signal and the original enhancement signal into a plurality of sub-signals and marking the normal sub-signals includes:

[0111] dividing the difference enhancement signal into a plurality of sub-signals to obtain a plurality of difference enhancement sub-signals, and dividing the original enhancement signal into a plurality of sub-signals to obtain a plurality of original enhancement sub-signals, the division lengths of which are the same;

[0112] obtaining an average signal value of the difference enhancement signal to obtain a global difference amplitude;

[0113] obtaining an average signal value of each difference enhancement sub-signal to obtain a segment difference amplitude;

[0114] ​​​​​When the segment difference amplitude is less than the global difference amplitude, the corresponding difference enhancer signal is marked as a normal difference enhancer signal.

[0115] The standard deviation of the original enhancer signal is obtained, and a global fluctuation value is obtained.

[0116] The standard deviation of each original enhancer signal is obtained, and a segment fluctuation value is obtained.

[0117] When the segment fluctuation value is less than the global fluctuation value, the corresponding original enhancer signal is marked as a normal original enhancer signal.

[0118] The present application divides the difference enhancer signal and the original enhancer signal into a plurality of sub-signals respectively, avoids that the local weak water seepage feature is covered by the overall signal, and selects the normal sub-signal through comparison of the segment difference amplitude and the global difference amplitude, and comparison of the segment fluctuation value and the global fluctuation value.

[0119] In the embodiment, the process of obtaining the difference main frequency amplitude abnormality vector and the original main frequency amplitude abnormality vector includes:

[0120] The Fourier transform is respectively performed on each difference enhancer signal and original enhancer signal to obtain a difference frequency spectrum and an original frequency spectrum.

[0121] The maximum amplitude is extracted from the difference frequency spectrum and the original frequency spectrum respectively to obtain a difference main amplitude and an original main amplitude.

[0122] The average of each difference main amplitude belonging to the normal difference enhancer signal is taken to obtain a difference comparison main amplitude.

[0123] The average of each original main amplitude belonging to the normal original enhancer signal is taken to obtain an original comparison main amplitude.

[0124] According to the difference between each difference main amplitude and the difference comparison main amplitude, a difference main frequency amplitude abnormality value is obtained, and each difference main frequency amplitude abnormality value is arranged in time sequence to obtain a difference main frequency amplitude abnormality vector.

[0125] According to the difference between each original main amplitude and the original comparison main amplitude, an original main frequency amplitude abnormality value is obtained, and each original main frequency amplitude abnormality value is arranged in time sequence to obtain an original main frequency amplitude abnormality vector.

[0126] The process of obtaining the difference main frequency amplitude abnormality value includes: subtracting the difference comparison main amplitude from each difference main amplitude, taking the absolute value of the subtraction result to obtain a difference main amplitude difference, and taking the ratio of the difference main amplitude difference to the difference comparison main amplitude as the difference main frequency amplitude abnormality value.

[0127] The process of obtaining the original main frequency amplitude anomaly value includes: subtracting each original main amplitude from the original contrast main amplitude, taking the absolute value of the subtraction result to obtain the original main amplitude difference, and taking the ratio of the original main amplitude difference and the original contrast main amplitude as the original main frequency amplitude anomaly value.

[0128] The present application can accurately reflect the main energy characteristics of the signal in the frequency domain by performing Fourier transform on the difference enhancer signal and the original enhancer signal and extracting the maximum amplitude, can highlight the abnormal deviation of the signal in the frequency domain from the normal state by calculating the difference between the main amplitude of each sub-signal and the contrast main amplitude of the normal sub-signal and constructing an anomaly vector in time sequence, can enhance the sensitivity to weak anomalies or local mutations, and can improve the detection accuracy and reliability of water infiltration or other weak disturbances.

[0129] In the present embodiment, the process of obtaining the difference high-frequency energy and the original high-frequency energy anomaly vector includes:

[0130] Each difference enhancer signal and original enhancer signal is subjected to Fourier transform to obtain a difference frequency spectrum and an original frequency spectrum;

[0131] The high-frequency range of the difference frequency spectrum and the original frequency spectrum is extracted to obtain a difference high-frequency spectrum and an original high-frequency spectrum;

[0132] The absolute values of the amplitudes corresponding to each frequency point in the difference high-frequency spectrum are added to obtain the difference high-frequency energy;

[0133] The absolute values of the amplitudes corresponding to each frequency point in the original high-frequency spectrum are added to obtain the original high-frequency energy;

[0134] The difference high-frequency energies belonging to normal difference enhancer signals are averaged to obtain a difference contrast high-frequency energy;

[0135] The original high-frequency energies belonging to normal original enhancer signals are averaged to obtain an original contrast high-frequency energy;

[0136] The difference high-frequency energy anomaly value is obtained according to the difference between each difference high-frequency energy and the difference contrast high-frequency energy, and each difference high-frequency energy anomaly value is arranged in time sequence to obtain a difference high-frequency energy anomaly vector;

[0137] The original high-frequency energy anomaly value is obtained according to the difference between each original high-frequency energy and the original contrast high-frequency energy, and each original high-frequency energy anomaly value is arranged in time sequence to obtain an original high-frequency energy anomaly vector.

[0138] Each sub-signal corresponds to a frequency spectrum, and each frequency spectrum extracts a main amplitude and a high-frequency energy.

[0139] The process of obtaining the differential high-frequency energy anomaly value includes: subtracting each differential high-frequency energy from the differential contrast high-frequency energy, taking the absolute value of the subtraction result to obtain a differential high-frequency energy difference, and taking the ratio of each differential high-frequency energy difference to the differential contrast high-frequency energy as the differential high-frequency energy anomaly value.

[0140] The process of obtaining the original high-frequency energy anomaly value includes: subtracting each original high-frequency energy from the original contrast high-frequency energy, taking the absolute value of the subtraction result to obtain an original high-frequency energy difference, and taking the ratio of each original high-frequency energy difference to the original contrast high-frequency energy as the original high-frequency energy anomaly value.

[0141] Early seepage or weak seepage of the dam will cause weak, rapid and local dynamic changes in the refractive index, temperature and stress of the medium around the optical fiber, and such changes will be reflected in the echo signal as an increase in the energy of high-frequency components. The present application can capture the short-time and rapid change characteristics caused by seepage by performing Fourier transform on the differential enhancement sub-signals and the original enhancement sub-signals and extracting high-frequency spectra. The high-frequency components mainly reflect the formation of local seepage or small seepage channels. By calculating the difference between the high-frequency energy of each sub-signal and the contrast high-frequency energy of the normal sub-signal and constructing an anomaly vector in time sequence, the abnormal fluctuations caused by seepage can be highlighted, and the recognizability of small or early seepage states can be significantly enhanced.

[0142] In the present embodiment, the high-frequency range is [M / 2, M], and M represents the number of effective frequency points after removing the direct current component and the mirror component after Fourier transform, i.e., the high-frequency range is a frequency band located above half of the Nyquist frequency in the frequency domain effective spectrum. The high-frequency range can be adjusted and selected according to actual conditions, and is not limited to the range of the present embodiment.

[0143] As shown in Figure 2 The seepage recognition neural network includes a main frequency amplitude anomaly feature fusion unit, a high-frequency energy anomaly feature fusion unit, a feature extraction and splicing unit, a double-channel attention enhancement unit and an output layer.

[0144] The main frequency amplitude anomaly feature fusion unit is configured to extract features from the differential main frequency amplitude anomaly vector and the original main frequency amplitude anomaly vector, and perform fusion processing to obtain main frequency amplitude anomaly fusion features.

[0145] The high-frequency energy anomaly feature fusion unit is configured to extract features from the differential high-frequency energy anomaly vector and the original high-frequency energy anomaly vector, and perform fusion processing to obtain high-frequency energy anomaly fusion features.

[0146] The feature extraction and splicing unit is configured to extract mean features and maximum features from the main frequency amplitude anomaly fusion features and the high-frequency energy anomaly fusion features, respectively, and splice them to obtain main frequency amplitude-high-frequency energy mean splicing features and main frequency amplitude-high-frequency energy maximum splicing features.

[0147] The double-channel attention enhancement unit is used for applying attention to the main frequency amplitude-high frequency energy mean value splicing feature and the main frequency amplitude-high frequency energy maximum value splicing feature respectively, so as to obtain an attention enhanced feature;

[0148] The output layer is used for outputting a water seepage state result according to the attention enhanced feature.

[0149] The present application can comprehensively reflect the main amplitude change characteristics and high frequency disturbance characteristics caused by water seepage by respectively extracting and fusing the differential main frequency amplitude anomaly vector and the original main frequency amplitude anomaly vector, and the differential high frequency energy anomaly vector and the original high frequency energy anomaly vector, avoiding the problem that a single feature is insufficient to represent the water seepage state; by further extracting the mean value feature and the maximum value feature from the fused feature and splicing them, the overall change trend and local mutation characteristics of the water seepage process can be described at the same time; on this basis, the double-channel attention enhancement mechanism is introduced, the importance of different statistical features is adaptively weighted, the features with higher correlation with the water seepage state are strengthened, so as to improve the accuracy and stability of the water seepage state recognition and enhance the discrimination ability for early or weak water seepage state.

[0150] As shown in Figure 3 and 4 , the main frequency amplitude anomaly feature fusion unit and the high frequency energy anomaly feature fusion unit have the same structure, both of which include a first one-dimensional convolution layer, a second one-dimensional convolution layer and a multiplier M1.

[0151] The first one-dimensional convolution layer is used for extracting differential shallow features from the differential main frequency amplitude anomaly vector or the differential high frequency energy anomaly vector.

[0152] The second one-dimensional convolution layer is used for extracting original shallow features from the original main frequency amplitude anomaly vector or the original high frequency energy anomaly vector.

[0153] The multiplier M1 is used for multiplying the differential shallow features and the original shallow features element by element to obtain the main frequency amplitude anomaly fusion feature or the high frequency energy anomaly fusion feature.

[0154] As shown in Figure 3 , in the main frequency amplitude anomaly feature fusion unit, the first one-dimensional convolution layer is used for inputting the differential main frequency amplitude anomaly vector, the second one-dimensional convolution layer is used for inputting the original main frequency amplitude anomaly vector, and the multiplier M1 outputs the main frequency amplitude anomaly fusion feature.

[0155] As shown in Figure 4 , in the high frequency energy anomaly feature fusion unit, the first one-dimensional convolution layer is used for inputting the differential high frequency energy anomaly vector, the second one-dimensional convolution layer is used for inputting the original high frequency energy anomaly vector, and the multiplier M1 outputs the high frequency energy anomaly fusion feature.

[0156] The multiplier M1 of the application performs element-wise multiplication on the same type of features, which can effectively highlight the correlation response between the difference and the original features (such as the cooperative change of the two under the water seepage state), and strengthen the complementarity and distinguishability of multi-dimensional features.

[0157] The convolution kernel size of the first one-dimensional convolution layer and the second one-dimensional convolution layer is 1x3.

[0158] As shown in Figure 5 The feature extraction and splicing unit includes a first one-dimensional average pooling layer, a second one-dimensional average pooling layer, a first one-dimensional maximum pooling layer, a second one-dimensional maximum pooling layer, a first Concat layer and a second Concat layer.

[0159] The first one-dimensional average pooling layer is used for extracting the mean value feature of the main frequency amplitude anomaly fusion feature to obtain the main frequency amplitude anomaly mean value feature.

[0160] The first one-dimensional maximum pooling layer is used for extracting the maximum value feature of the main frequency amplitude anomaly fusion feature to obtain the main frequency amplitude anomaly maximum value feature.

[0161] The second one-dimensional average pooling layer is used for extracting the mean value feature of the high-frequency energy anomaly fusion feature to obtain the high-frequency energy anomaly mean value feature.

[0162] The second one-dimensional maximum pooling layer is used for extracting the maximum value feature of the high-frequency energy anomaly fusion feature to obtain the high-frequency energy anomaly maximum value feature.

[0163] The first Concat layer is used for splicing the main frequency amplitude anomaly mean value feature and the high-frequency energy anomaly mean value feature to obtain the main frequency amplitude-high-frequency energy mean value splicing feature.

[0164] The second Concat layer is used for splicing the main frequency amplitude anomaly maximum value feature and the high-frequency energy anomaly maximum value feature to obtain the main frequency amplitude-high-frequency energy maximum value splicing feature.

[0165] The application can simultaneously extract the overall change level and local extreme value change feature of the frequency domain feature under the water seepage state by performing one-dimensional average pooling and one-dimensional maximum pooling on the main frequency amplitude anomaly fusion feature and the high-frequency energy anomaly fusion feature respectively, wherein the mean value feature is used to reflect the continuity and overall trend of the water seepage process, and the maximum value feature is used to describe the instantaneous enhancement or local mutation caused by water seepage; by splicing the main frequency amplitude anomaly feature and the high-frequency energy anomaly feature at the mean value level and the maximum value level respectively, the joint expression of different frequency domain water seepage features can be realized.

[0166] In the embodiment, the pooling window size of the first one-dimensional average pooling layer, the second one-dimensional average pooling layer, the first one-dimensional maximum pooling layer and the second one-dimensional maximum pooling layer is set to 1x5, and the step size is set to 2.

[0167] As shown in Figure 6 The dual-channel attention enhancement unit includes a first Sigmoid layer, a second Sigmoid layer, a multiplier M2, a multiplier M3, a third one-dimensional convolution layer, a fourth one-dimensional convolution layer, a fifth one-dimensional convolution layer, and an adder A1.

[0168] The first Sigmoid layer is configured to perform Sigmoid activation processing on the inputted main frequency amplitude-high frequency energy mean value spliced features, map the feature values to the interval [0, 1], and generate attention weight coefficients corresponding to the mean value spliced features;

[0169] The second Sigmoid layer is configured to perform Sigmoid activation processing on the inputted main frequency amplitude-high frequency energy maximum value spliced features, map the feature values to the interval [0, 1], and generate attention weight coefficients corresponding to the maximum value spliced features;

[0170] The multiplier M2 is configured to multiply the main frequency amplitude-high frequency energy mean value spliced features and the attention weight coefficients outputted by the first Sigmoid layer element by element, and realize weighted enhancement of the mean value spliced features;

[0171] The multiplier M3 is configured to multiply the main frequency amplitude-high frequency energy maximum value spliced features and the attention weight coefficients outputted by the second Sigmoid layer element by element, and realize weighted enhancement of the maximum value spliced features;

[0172] The third one-dimensional convolution layer is configured to perform one-dimensional convolution operation on the weighted mean value spliced features outputted by the multiplier M2, and extract local correlation features thereof;

[0173] The fourth one-dimensional convolution layer is configured to perform one-dimensional convolution operation on the weighted maximum value spliced features outputted by the multiplier M3, and extract local correlation features thereof;

[0174] The adder A1 is configured to add the output features of the third one-dimensional convolution layer and the fourth one-dimensional convolution layer element by element, and realize fusion of the two types of weighted features;

[0175] The fifth one-dimensional convolution layer is configured to perform one-dimensional convolution operation on the fusion features outputted by the adder A1, further refine the feature information, and finally output attention enhanced features.

[0176] In this embodiment, the convolution kernel size of the third one-dimensional convolution layer and the fourth one-dimensional convolution layer is 1x3, and the convolution kernel size of the fifth one-dimensional convolution layer is 1x5.

[0177] The application can adaptively strengthen the feature dimensions with high correlation with the water seepage state by applying Sigmoid-based channel attention weights to the main frequency amplitude-high frequency energy mean value splicing feature and the main frequency amplitude-high frequency energy maximum value splicing feature respectively, and can inhibit the feature components irrelevant to water seepage; by performing one-dimensional convolution processing on the two types of weighted features respectively and weighted fusion, the local correlation relationship of the mean value feature and the maximum value feature in the time dimension can be effectively mined, and the persistent change and instantaneous mutation features in the water seepage process can be comprehensively described; further, the attention enhanced features extracted by convolution can make the water seepage related features more concentrated and prominent, thereby improving the stability and accuracy of water seepage state recognition and enhancing the discrimination ability for early or weak water seepage state.

[0178] In the embodiment, the water seepage state result includes: no water seepage state, slight water seepage state, obvious water seepage state and serious water seepage state.

[0179] The application improves the signal-to-noise ratio of weak signals by injecting quantum enhanced probe light, first performs difference operation on echo signals at adjacent time to filter environmental noise and structural vibration interference, then performs normalization enhancement processing on the difference signals and the original signals respectively, extracts four-dimensional abnormal vectors of difference main frequency amplitude, original main frequency amplitude, difference high frequency energy and original high frequency energy, finally intelligently fuses and recognizes the multi-dimensional features by means of the penetration recognition neural network, accurately distinguishes the real water seepage signal change and the non-water seepage interference fluctuation, effectively captures the weak features of early and weak intensity water seepage, and solves the problem of low water seepage detection accuracy.

[0180] The above is only the preferred embodiment of the application and is not used to limit the application, and the application can have various changes and variations for those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the application shall be included in the protection scope of the application.

Claims

1. A signal processing method for quantum-enhanced distributed fiber-optic monitoring, characterized in that, The method comprises the following steps: Injecting quantum enhanced probe light into the optical fiber, collecting echo signals, subtracting the signal values of adjacent time points in the echo signals to obtain echo difference signals; Respectively normalizing the echo difference signals and the echo signals, and obtaining signal enhancement coefficients to perform signal enhancement to obtain difference enhanced signals and original enhanced signals; Dividing the difference enhanced signals and the original enhanced signals into multiple sub-signals and marking normal sub-signals; Dividing the difference enhanced signals into multiple sub-signals to obtain multiple difference enhanced sub-signals, and dividing the original enhanced signals into multiple sub-signals to obtain multiple original enhanced sub-signals; Obtaining average signal values of the difference enhanced signals to obtain global difference amplitudes; Obtaining average signal values of each difference enhanced sub-signal to obtain segment difference amplitudes; When the segment difference amplitude is smaller than the global difference amplitude, the corresponding difference enhanced sub-signal is marked as a normal difference enhanced sub-signal; Obtaining standard deviations of the original enhanced signals to obtain global fluctuation values; Obtaining standard deviations of each original enhanced sub-signal to obtain segment fluctuation values; When the segment fluctuation value is smaller than the global fluctuation value, the corresponding original enhanced sub-signal is marked as a normal original enhanced sub-signal; According to the difference between each sub-signal and the normal sub-signal in the difference main amplitude and the original main amplitude, difference main frequency amplitude and original main frequency amplitude abnormal change vectors are obtained; According to the difference between each sub-signal and the normal sub-signal in the difference high frequency energy and the original high frequency energy, difference high frequency energy and original high frequency energy abnormal change vectors are obtained; Four abnormal change vectors are processed by using a penetration recognition neural network to obtain a water penetration state result.

2. The signal processing method for quantum-enhanced distributed fiber optic monitoring according to claim 1, wherein, The process of obtaining the difference enhanced signals and the original enhanced signals comprises: Respectively normalizing the echo difference signals and the echo signals to obtain normalized echo difference signals and normalized echo signals; Obtaining first signal enhancement coefficients of the normalized echo difference signals, and enhancing the normalized echo difference signals to obtain the difference enhanced signals; Obtaining second signal enhancement coefficients of the normalized echo signals, and enhancing the normalized echo signals to obtain the original enhanced signals.

3. The signal processing method for quantum-enhanced distributed fibre optic monitoring according to claim 1 or 2, characterised in that, The process of obtaining the difference enhanced signals comprises: In the normalized echo difference signals, local difference energies of each time point are obtained; Taking the average of all the local difference energies to obtain a global energy average value; Taking the ratio of each time point local difference energy to the global energy average value as a difference energy ratio; Mapping the difference energy ratio to the first signal enhancement coefficient; Multiplying the first signal enhancement coefficient of each time point with the signal value of the corresponding time point of the normalized echo difference signal to obtain the difference enhanced signal.

4. The signal processing method for quantum-enhanced distributed fiber optic monitoring according to claim 1 or 2, characterized in that, The process of obtaining the original enhanced signals comprises: In the normalized echo signals, local mean values and local fluctuation amplitudes of each time point are obtained; Taking the ratio of the local fluctuation amplitude to the local mean value as a fluctuation ratio; Mapping the fluctuation ratio to the second signal enhancement coefficient; Multiplying the second signal enhancement coefficient of each time point with the signal value of the corresponding time point of the normalized echo signal to obtain the original enhanced signal.

5. The signal processing method for quantum-enhanced distributed fiber optic monitoring of claim 1, wherein, The process of obtaining the difference main frequency amplitude and the original main frequency amplitude abnormal change vector comprises: Respectively performing Fourier transform on each difference enhanced sub-signal and original enhanced sub-signal to obtain difference frequency spectrum and original frequency spectrum; The maximum amplitudes of the differential spectrum and the original spectrum are extracted respectively to obtain a differential main amplitude and an original main amplitude; The differential main amplitudes belonging to normal differential enhancer signals are averaged to obtain a differential contrast main amplitude; The original main amplitudes belonging to normal original enhancer signals are averaged to obtain an original contrast main amplitude; The differential main frequency amplitude anomaly values are obtained according to the difference between each differential main amplitude and the differential contrast main amplitude, and the differential main frequency amplitude anomaly values are arranged in time sequence to obtain a differential main frequency amplitude anomaly vector; The original main frequency amplitude anomaly values are obtained according to the difference between each original main amplitude and the original contrast main amplitude, and the original main frequency amplitude anomaly values are arranged in time sequence to obtain an original main frequency amplitude anomaly vector.

6. The signal processing method for quantum-enhanced distributed fiber optic monitoring of claim 1, wherein, The process of obtaining the differential high-frequency energy and the original high-frequency energy anomaly vector includes: The Fourier transform is performed on each differential enhancer signal and original enhancer signal respectively to obtain a differential spectrum and an original spectrum; The high-frequency ranges of the differential spectrum and the original spectrum are extracted respectively to obtain a differential high-frequency spectrum and an original high-frequency spectrum; The absolute values of the amplitudes corresponding to each frequency point in the differential high-frequency spectrum are added to obtain a differential high-frequency energy; The absolute values of the amplitudes corresponding to each frequency point in the original high-frequency spectrum are added to obtain an original high-frequency energy; The differential high-frequency energies belonging to normal differential enhancer signals are averaged to obtain a differential contrast high-frequency energy; The original high-frequency energies belonging to normal original enhancer signals are averaged to obtain an original contrast high-frequency energy; The differential high-frequency energy anomaly values are obtained according to the difference between each differential high-frequency energy and the differential contrast high-frequency energy, and the differential high-frequency energy anomaly values are arranged in time sequence to obtain a differential high-frequency energy anomaly vector; The original high-frequency energy anomaly values are obtained according to the difference between each original high-frequency energy and the original contrast high-frequency energy, and the original high-frequency energy anomaly values are arranged in time sequence to obtain an original high-frequency energy anomaly vector.

7. The signal processing method for quantum-enhanced distributed fiber optic monitoring of claim 1, wherein, The penetration recognition neural network includes a main frequency amplitude anomaly feature fusion unit, a high-frequency energy anomaly feature fusion unit, a feature extraction splicing unit, a double-channel attention enhancement unit, and an output layer.

8. The signal processing method for quantum-enhanced distributed fiber optic monitoring according to claim 7, wherein, The main frequency amplitude anomaly feature fusion unit is used to extract features from the differential main frequency amplitude anomaly vector and the original main frequency amplitude anomaly vector, and to fuse and process them to obtain main frequency amplitude anomaly fusion features; The high-frequency energy anomaly feature fusion unit is used to extract features from the differential high-frequency energy anomaly vector and the original high-frequency energy anomaly vector, and to fuse and process them to obtain high-frequency energy anomaly fusion features; The feature extraction splicing unit is used to extract mean features and maximum value features from the main frequency amplitude anomaly fusion features and the high-frequency energy anomaly fusion features respectively, and to splice them to obtain main frequency amplitude-high-frequency energy mean splicing features and main frequency amplitude-high-frequency energy maximum value splicing features; The double-channel attention enhancement unit is used to apply attention to the main frequency amplitude-high-frequency energy mean splicing features and the main frequency amplitude-high-frequency energy maximum value splicing features respectively to obtain attention enhanced features; The output layer is used to output a water penetration state result according to the attention enhanced features.

9. The signal processing method for quantum-enhanced distributed fiber optic monitoring according to claim 8, wherein, The main frequency amplitude anomaly feature fusion unit and the high frequency energy anomaly feature fusion unit are identical in structure, and each includes a first one-dimensional convolution layer, a second one-dimensional convolution layer, and a multiplier M1. The first one-dimensional convolution layer is configured to extract a differential shallow layer feature from the differential main frequency amplitude anomaly vector or the differential high frequency energy anomaly vector. The second one-dimensional convolution layer is configured to extract an original shallow layer feature from the original main frequency amplitude anomaly vector or the original high frequency energy anomaly vector. The multiplier M1 is configured to multiply the differential shallow layer feature and the original shallow layer feature element by element to obtain the main frequency amplitude anomaly fusion feature or the high frequency energy anomaly fusion feature.

Images