A method for adaptively extracting Rayleigh wave dispersion curve

Abstract

The present application relates to the technical field of geophysical exploration data processing, and discloses a microtremor surface wave dispersion curve adaptive denoising extraction method, which acquires microtremor array data, divides time windows, and constructs a microtremor observation unit; calculates noise disturbance coefficients according to the time-frequency characteristics, phase characteristics, azimuth characteristics and station consistency characteristics of the microtremor observation unit; generates an initial dispersion energy map based on the microtremor observation unit, and generates a dispersion ridge line reliability map in combination with the noise disturbance coefficients; reversely maps the dispersion ridge line reliability map to the microtremor observation unit to generate an adaptive denoising mask; reconstructs a denoising dispersion energy map by using the adaptive denoising mask, and extracts dispersion curves through constraint path search; updates the adaptive denoising mask according to the curve quality score feedback until the final dispersion curve is output. The present application can weaken high-energy false ridge line interference, and improve the continuity and stability of dispersion curve extraction.

Description

Technical Field

[0001] This invention relates to the field of geophysical exploration data processing technology, specifically to an adaptive noise reduction and extraction method for micro-moving surface wave dispersion curves. Background Technology

[0002] Micro-motion surface wave exploration is a geophysical exploration method that utilizes environmental micro-vibration signals to extract the dispersion characteristics of surface waves in subsurface media and further invert the shallow shear wave velocity structure. Existing micro-motion surface wave dispersion curve extraction typically includes steps such as micro-motion array data acquisition, preprocessing, cross-correlation calculation, dispersion imaging, and dispersion curve picking. The continuity, stability, and reliability of the dispersion curve directly affect the subsequent shear wave velocity inversion results. Therefore, in complex urban environments, road traffic environments, or industrial equipment operating environments, accurately extracting the true surface wave dispersion curve from noisy micro-motion data is a key issue in micro-motion surface wave exploration processing.

[0003] Existing methods typically employ fixed filtering, energy thresholds, dispersion map clustering, or preset search parameters to process micro-motion data or dispersion energy maps. Their processing logic is often a one-way flow of "denoising first, then imaging, then picking," lacking a feedback relationship between the denoising parameters and the quality of the final extracted dispersion curve. When narrowband mechanical noise, traffic impulse noise, or local strong vibration source interference exists in the micro-motion data, this noise can easily form high-energy pseudo-ridges in the dispersion energy map. Existing methods often misidentify these pseudo-ridges as true dispersion curves, leading to abrupt changes, breaks, or mode mixing in the dispersion curve, thereby reducing the reliability of the dispersion curve extraction results. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an adaptive noise reduction and extraction method for micro-motion surface wave dispersion curves. This method solves the problem that existing micro-motion surface wave dispersion curve extraction methods lack a feedback relationship between the noise reduction process and the dispersion curve quality, and in complex noise environments, they are prone to mistakenly extracting high-energy pseudo-ridges as real dispersion curves, resulting in insufficient continuity and stability of the dispersion curves.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an adaptive noise reduction and extraction method for micro-motion surface wave dispersion curves, comprising: Acquire micro-motion array data, divide the micro-motion array data into time windows, and construct multiple micro-motion observation units; Based on the time-frequency characteristics, phase characteristics, azimuth characteristics, and station consistency characteristics of the micro-motion observation unit, the noise disturbance coefficient of each micro-motion observation unit is calculated. An initial dispersion energy map is generated based on the micro-motion observation unit, and a dispersion ridge confidence map is generated based on the initial dispersion energy map and the noise disturbance coefficient. The dispersion ridge confidence map is reverse-mapped to the micro-motion observation unit to generate an adaptive noise reduction mask. The micro-motion observation unit is weighted using the adaptive noise reduction mask to obtain noise-reduced micro-motion observation data, and the noise-reduced dispersion energy map is reconstructed based on the noise-reduced micro-motion observation data; Based on the noise reduction dispersion energy map and the dispersion ridge confidence map, a dispersion candidate point set is constructed, and a dispersion curve is extracted from the dispersion candidate point set through constrained path search. A curve quality score is generated based on the dispersion curve. When the curve quality score does not meet the termination condition, the adaptive noise reduction mask is updated and the dispersion curve is extracted again until the final dispersion curve is output.

[0006] Preferably, constructing multiple micro-motion observation units includes: The micro-motion array data is subjected to time synchronization, mean removal, and trend removal processing. The processed micro-motion array data is divided into multiple time window data according to the preset time window length and preset overlap rate; The micro-motion observation unit is constructed based on the station pairs, frequency points, station spacing, phase difference, coherence coefficient, azimuth angle, and energy value in each time window data.

[0007] Preferably, calculating the noise disturbance coefficient of each of the micro-motion observation units includes: The degree of narrowband energy surge is determined based on the aforementioned time-frequency characteristics; The phase dispersion degree is determined based on the phase characteristics; The degree of directional energy concentration is determined based on the aforementioned directional characteristics; The degree of cross-station propagation consistency is determined based on the aforementioned station consistency characteristics; The noise disturbance coefficient is generated based on the degree of narrowband energy surge, phase dispersion, azimuth energy concentration, and cross-station propagation consistency.

[0008] Preferably, generating the initial dispersion energy map includes: Cross-correlation processing is performed on multiple micro-motion observation units to obtain time-window cross-correlation data; The cross-correlation data of the time window is subjected to dispersion imaging processing to obtain the energy value corresponding to each frequency point and phase velocity point; The energy values ​​corresponding to multiple frequency points and multiple phase velocity points are combined to form the initial dispersion energy map.

[0009] Preferably, generating a dispersion ridge confidence map includes: Calculate the energy saliency corresponding to each frequency point and phase velocity point based on the initial dispersion energy map; Calculate the probability of repeated occurrence across multiple time windows based on the energy distribution in multiple time windows; Calculate the continuity of adjacent frequency points based on the phase velocity change between adjacent frequency points; Array coherence is calculated based on the phase relationship between multiple station pairs; The dispersion ridge confidence map is generated by combining the energy significance, the probability of recurrence across time windows, the continuity of adjacent frequency points, the array coherence, and the noise perturbation coefficient.

[0010] Preferably, generating the adaptive noise reduction mask includes: Determine the time window, frequency point, and station pair corresponding to the low confidence grid points in the dispersion ridge confidence map; The time windows, frequency points, and station pairs corresponding to the low-confidence grid points are located as pseudo-ridgeline contribution sources; Increase the suppression weight of the micro-motion observation unit corresponding to the pseudo-ridge contribution source; Reduce the suppression weight of micro-motion observation units that contribute to stability at high-confidence grid points; The adaptive noise reduction mask is generated based on the suppression weights of each micro-motion observation unit.

[0011] Preferably, the noise-reduced micro-motion observation data includes: The participation weight of each of the micro-motion observation units in the cross-correlation superposition, phase matching evaluation, or dispersion imaging process is determined based on the adaptive noise reduction mask. The energy contribution, coherence contribution, and phase matching contribution of the micro-motion observation unit are adjusted based on the participation weights to obtain the noise-reduced micro-motion observation data. Based on the noise-reduced micro-motion observation data, the dispersion imaging process is re-executed to obtain the noise-reduced dispersion energy map.

[0012] Preferably, extracting dispersion curves from the dispersion candidate point set through constrained path search includes: In the noise reduction dispersion energy map, frequency points and phase velocity points that meet the confidence conditions are selected to construct the dispersion candidate point set; Configure path costs for candidate points in the dispersion candidate point set, the path costs including confidence cost, phase velocity jump cost, curvature smoothing cost and noise crossing cost; Based on the path cost, a target path that satisfies the phase velocity continuity constraint and curvature smoothness constraint is searched. The sequence of candidate points corresponding to the target path is determined as the dispersion curve.

[0013] Preferably, generating a curve quality score based on the dispersion curve includes: Calculate the continuity, smoothness, consistency across time windows, array coherence, and effective bandwidth coverage of the dispersion curve; The curve quality score is generated based on the curve continuity, curve smoothness, cross-time window consistency, array coherence, and effective frequency band coverage. When the curve quality score is lower than the quality threshold, a low-quality frequency band is identified, and the adaptive noise reduction mask is updated based on the low-quality frequency band.

[0014] Preferably, the final output dispersion curve includes: Output the frequency, phase velocity, and curve reliability corresponding to the final dispersion curve; Low-confidence frequency bands are marked according to the confidence level of the curve; Output the time window, frequency point, and station pair corresponding to the low-confidence frequency band as the dispersion curve verification information.

[0015] This invention provides an adaptive noise reduction and extraction method for micro-motion surface wave dispersion curves. It has the following beneficial effects: 1. This invention generates an adaptive noise reduction mask by reverse mapping of the dispersion ridge credibility map, which directly links the noise reduction process with the dispersion curve quality. It can adaptively adjust the suppression weight for pseudo ridge contribution sources in micro-motion array data, weaken the high-energy pseudo ridges formed by narrowband mechanical noise, traffic impulse noise and local strong vibration source interference from the source, and improve the continuity and stability of dispersion curve extraction.

[0016] 2. When generating the dispersion ridge confidence map, this invention comprehensively considers energy significance, probability of recurrence across time windows, continuity of adjacent frequency points, array coherence, and noise perturbation coefficient, avoiding the determination of dispersion points based solely on the maximum energy value, and can reduce the probability of high-energy noise peaks being mistakenly extracted as true dispersion curves.

[0017] 3. This invention traces the noise processing target from the noise points in the dispersion energy map to the micro-motion observation unit. By adjusting the participation weight of the micro-motion observation unit in the cross-correlation superposition, phase matching evaluation, or dispersion imaging process, it can weaken the contribution source of false ridges while retaining the effective signal of weak surface waves, thereby reducing the loss of effective weak ridges caused by traditional image filtering or dispersion map post-processing.

[0018] 4. This invention extracts dispersion curves through constrained path search and introduces phase velocity continuity constraints, curvature smoothing constraints, and noise travel costs, so that the extracted dispersion curves conform to the physical continuity law of surface wave dispersion, which can reduce the probability of curve jumps, reversals, and mode aliasing.

[0019] 5. This invention generates a curve quality score based on the dispersion curve, and updates the adaptive noise reduction mask when the curve quality score does not meet the termination condition, so that the noise reduction intensity and candidate search range can be automatically adjusted under different sites, different array forms and different noise backgrounds, reducing the dependence on manual parameter tuning.

[0020] 6. This invention outputs the curve confidence level, low confidence band, and corresponding time window, frequency point, and station pair while outputting the final dispersion curve. This provides quality constraint information for subsequent manual verification and shear wave velocity inversion, and reduces the misleading effect of low confidence band on the inversion results. Attached Figure Description

[0021] Figure 1 This is an overall flowchart of the adaptive noise reduction extraction method for micro-motion surface wave dispersion curves according to the present invention; Figure 2 This is a flowchart of the adaptive noise reduction mask generation process in this invention; Figure 3 This is a flowchart of the dispersion curve extraction and feedback update process in this invention. Detailed Implementation

[0022] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] This invention provides an adaptive denoising extraction method for dispersion curves of micro-motion surface waves. The core idea is as follows: Micro-motion array data is divided into micro-motion observation units with time, frequency, spatial, and phase attributes. Instead of directly picking the dispersion curve based on the maximum energy value in the initial dispersion energy map, a dispersion ridge confidence map is generated by combining the noise perturbation coefficient. This confidence map is then mapped back to the micro-motion observation units to identify pseudo-ridge contribution sources and generate an adaptive denoising mask. The adaptive denoising mask is then used to adjust the participation weight of the micro-motion observation units in subsequent dispersion imaging, reconstructing the denoised dispersion energy map, and extracting the dispersion curve through constrained path search. Finally, the adaptive denoising mask is updated based on the curve quality score of the dispersion curve until the final dispersion curve is output. This directly links the denoising process to the dispersion curve quality, reducing the impact of narrowband mechanical noise, traffic impulse noise, and local high-energy pseudo-ridges on the dispersion curve extraction results in urban micro-motion environments.

[0024] like Figure 1 As shown, in one embodiment, an adaptive noise reduction extraction method for micro-motion surface wave dispersion curves includes the following steps: S1. Acquire micro-motion array data, divide the micro-motion array data into time windows, and construct multiple micro-motion observation units.

[0025] Multiple geophones are deployed in the target exploration area, forming a micro-motion array. The micro-motion array can be linear, circular, rectangular, or irregular. Each geophone acquires continuous micro-motion signals, which include at least a vertical component and, depending on the actual exploration needs, a horizontal component. After acquisition, the micro-motion array data is obtained.

[0026] The micro-motion array data undergoes time synchronization, mean removal, and detrending processing. Time synchronization ensures that micro-motion signals acquired from different detector points have a consistent time reference; mean removal eliminates DC bias; and detrending eliminates instrument drift or long-period trend terms. For instantaneous outliers that significantly exceed the normal amplitude range, amplitude truncation or median replacement can be used to avoid single-point anomalies interfering with subsequent dispersive imaging.

[0027] The processed micro-motion array data is divided into multiple time windows according to a preset time window length and a preset overlap rate. Each time window contains micro-motion signals from multiple detector points within that time window. For each time window, station pairs are constructed according to the combination relationship between detector points, and the corresponding station spacing, azimuth, frequency point, phase difference, coherence coefficient, and energy value are calculated for each station pair.

[0028] Therefore, a micromotion observation unit is formed by combining time window number, frequency point, station pair, station spacing, phase difference, coherence coefficient, azimuth angle, and energy value. The micromotion observation unit is used to characterize the micromotion propagation observation results under a specific time window, frequency point, and station pair. By constructing the micromotion observation unit, continuous micromotion array data can be converted into an analysis object with time, frequency, spatial, and phase attributes, providing a data foundation for subsequent noise perturbation coefficient calculation, dispersion ridge confidence map generation, and adaptive noise reduction mask construction.

[0029] S2. Calculate the noise disturbance coefficient of each micro-motion observation unit based on its time-frequency characteristics, phase characteristics, azimuth characteristics, and station consistency characteristics.

[0030] For each micro-motion observation unit, time-frequency features, phase features, azimuth features, and station consistency features are extracted.

[0031] Time-frequency characteristics are used to characterize the energy surge of a micro-motion observation unit within a local frequency band. When a certain frequency point shows a significant narrowband energy enhancement within a few time windows, while adjacent frequency points do not show synchronous enhancement, it indicates that the micro-motion observation unit may be affected by mechanical equipment or a fixed-frequency vibration source.

[0032] Phase characteristics are used to characterize whether the phase difference of a micro-motion observation unit satisfies the surface wave propagation law. The phase difference of a real surface wave signal between adjacent stations has a relatively stable relationship with frequency, station spacing, and phase velocity; if the phase difference between different station pairs at the same frequency point is highly discrete, it indicates that the frequency point may contain strong non-surface wave noise components.

[0033] Orientation features are used to characterize whether micro-motion energy is excessively concentrated in a certain direction. When the energy in a certain time window mainly comes from a single local vibration source direction, and this direction does not match the effective propagation direction of the array, local high-energy pseudo-ridges are easily formed in the dispersion energy map.

[0034] Station consistency characteristics are used to characterize whether there are consistent propagation characteristics among multiple station pairs. True surface wave signals can usually exhibit continuous phase relationship and stable coherence among multiple station pairs, while local noise or sporadic impulse noise usually only manifests as high energy in some station pairs.

[0035] In one embodiment, the noise disturbance factor is calculated according to the following formula: ; in, This represents the noise disturbance coefficient of the i-th micro-motion observation unit; Indicates the degree of narrowband energy surge; Indicates the degree of phase dispersion; Indicates the degree of energy concentration in a given location; Indicates the degree of consistency in transmission across different stations; , , , These represent the corresponding weights.

[0036] Among them, the degree of narrowband energy surge can be determined based on the difference between the energy at the current frequency point and the background energy of the adjacent frequency band; the degree of phase dispersion can be determined based on the degree of dispersion of the phase difference between multiple station pairs at the same frequency point; the degree of azimuth energy concentration can be determined based on the degree of concentration of energy distribution corresponding to different azimuth angles; and the degree of cross-station propagation consistency can be determined based on the coherence coefficient and phase change law between multiple station pairs.

[0037] A larger noise disturbance coefficient indicates that the corresponding micro-motion observation unit is more likely to be a source of pseudo-ridge contribution; a smaller noise disturbance coefficient indicates that the corresponding micro-motion observation unit is more likely to contain stable surface wave effective components.

[0038] S3. Generate an initial dispersion energy map based on the micro-motion observation unit, and generate a dispersion ridge confidence map based on the initial dispersion energy map and the noise disturbance coefficient.

[0039] Cross-correlation processing is performed on multiple micro-motion observation units to obtain time-window cross-correlation data. Cross-correlation processing can be performed on different station pairs within each time window, or it can be superimposed and averaged across multiple time windows. Through cross-correlation processing, the surface wave components with stable propagation characteristics can be enhanced, and the influence of random uncorrelated noise can be reduced.

[0040] Dispersion imaging processing is performed on the cross-correlation data of the time window to obtain the energy value corresponding to each frequency point and phase velocity point. Dispersion imaging processing can employ any one of the following methods: spatial autocorrelation method, frequency-wavenumber method, phase shift method, Radon transform, or a combination of these methods. The energy values ​​corresponding to multiple frequency points and multiple phase velocity points are combined to form an initial dispersion energy map.

[0041] Each grid point in the initial dispersion energy map corresponds to a frequency point and a phase velocity point. Traditionally, the phase velocity point with the highest energy at the same frequency is directly selected as the dispersion point. However, in urban micro-motion environments, narrowband mechanical noise, traffic impulse noise, or localized strong noise may generate high-energy pseudo-ridges, leading to mis-picking when relying solely on the maximum energy value. Therefore, this implementation does not directly pick dispersion curves based on the initial dispersion energy map, but instead generates a dispersion ridge confidence map based on the initial dispersion energy map.

[0042] Specifically, for each frequency point and phase velocity point in the initial dispersion energy map, the energy saliency, the probability of recurrence across time windows, the continuity of adjacent frequency points, the array coherence, and the noise projection penalty are calculated.

[0043] Energy saliency characterizes the prominence of the current grid point's energy relative to the background energy within the same frequency search range. Cross-time-window recurrence probability characterizes whether the current grid point or its neighborhood consistently appears across multiple time windows. Adjacent frequency continuity characterizes the smoothness of phase velocity changes between the current grid point and adjacent frequency candidate grid points. Array coherence characterizes whether the micro-motion observation cell corresponding to this grid point exhibits consistent propagation characteristics across multiple station pairs. Noise projection penalty characterizes whether the micro-motion observation cell upon which this grid point depends has a high noise perturbation coefficient.

[0044] During dispersive imaging, the contribution of each micro-motion observation unit to the grid energy corresponding to each frequency point and phase velocity point is recorded. The contribution value can be determined by the energy value, coherence coefficient, and phase matching degree of the micro-motion observation unit at the corresponding frequency point. For any grid point corresponding to a frequency point and phase velocity point, the set of micro-motion observation units that participate in forming the energy of that grid point is counted, and the contribution value of each micro-motion observation unit is normalized to obtain the contribution weight of that grid point to the micro-motion observation unit.

[0045] In one embodiment, the contribution of the i-th micro-motion observation unit to the g-th grid point is calculated according to the following formula: ; in, This represents the contribution value of the i-th micro-motion observation unit to the g-th grid point; This represents the energy value of the i-th micro-motion observation unit at the corresponding frequency point; This represents the coherence coefficient of the i-th micro-motion observation unit at the corresponding frequency point; This represents the phase matching deviation between the actual phase difference of the i-th micro-motion observation unit and the theoretical phase difference corresponding to the g-th grid point.

[0046] In one implementation, the contribution weight of the i-th micro-motion observation unit to the g-th grid point is calculated according to the following formula: ; in, This represents the contribution weight of the i-th micro-motion observation unit to the g-th grid point; This represents the set of micro-motion observation units that participate in the formation of the energy of the g-th grid point.

[0047] Based on contribution weights, the reliability information of grid points can be traced back to the corresponding micro-motion observation unit, and the degree of noise disturbance to the grid point can also be calculated. In one embodiment, the noise projection penalty of the g-th grid point is calculated according to the following formula: ; in, This represents the noise projection penalty for the g-th grid point; This represents the noise disturbance coefficient of the i-th micro-motion observation unit.

[0048] In one embodiment, the confidence level of the dispersion ridge is calculated according to the following formula: ; in, This represents the confidence level of the dispersion ridge at the g-th grid point; Indicates the significance of energy; Indicates the probability of repeated occurrences across a time window; Indicates the continuity of adjacent frequency points; Indicates array coherence; Indicates noise projection penalty; , , , , These represent the corresponding weights.

[0049] If a grid point has high energy, but is mainly contributed by micro-motion observation units with high noise perturbation coefficients, the reliability of the dispersion ridge line of that grid point will be reduced; if a grid point does not have the highest energy, but appears repeatedly in multiple time windows and has good array coherence and continuity of adjacent frequency points, then that grid point can still obtain a high reliability of the dispersion ridge line.

[0050] After calculating the dispersion ridge confidence level for all grid points in the initial dispersion energy map, a dispersion ridge confidence level map is formed. The dispersion ridge confidence level map has the same frequency-phase velocity grid structure as the initial dispersion energy map, but its value does not represent the energy magnitude, but rather the confidence level of each grid point belonging to a true surface wave dispersion ridge.

[0051] S4. Reverse map the dispersion ridge confidence map to the micro-motion observation unit to generate an adaptive noise reduction mask.

[0052] like Figure 2 As shown, after the dispersion ridge confidence map is generated, a contribution mapping relationship is established between the grid points in the dispersion ridge confidence map and the micro-motion observation units that participated in forming the energy of that grid point. The contribution mapping relationship is determined by the aforementioned contribution weights.

[0053] Low-confidence grid points can be determined by a confidence threshold or by confidence ranking within the same frequency range. A grid point is identified as low-confidence when its dispersion ridge confidence is below the confidence threshold or when it falls within the later interval of the confidence ranking within the same frequency range. High-confidence grid points can be determined by a high-confidence threshold or by the earlier interval of the confidence ranking within the same frequency range.

[0054] For low-confidence grid points in the dispersion ridge confidence map, the corresponding time window, frequency point, and station pair are determined. The confidence loss of the low-confidence grid point is then allocated to the corresponding micro-motion observation unit according to its contribution weight, thereby determining the degree of low-confidence contribution of each micro-motion observation unit. If a certain micro-motion observation unit has a high contribution weight for multiple low-confidence grid points, it indicates that the micro-motion observation unit is more likely to be a source of pseudo-ridge contribution.

[0055] For high-confidence grid points, the contribution of each micro-motion observation unit to the high-confidence grid point is statistically analyzed according to its contribution weight. If a certain micro-motion observation unit consistently contributes to the high-confidence grid point across multiple time windows, it indicates that the micro-motion observation unit is more likely to contain effective surface wave components, and its suppression level should be reduced.

[0056] In one implementation, the low confidence contribution of the i-th micro-motion observation unit is calculated according to the following formula: ; in, This indicates the degree of low confidence contribution of the i-th micro-motion observation unit; Represents a set of low-confidence grid points; This represents the contribution weight of the i-th micro-motion observation unit to the g-th grid point; This represents the confidence level of the dispersion ridge at the g-th grid point.

[0057] In one implementation, the high confidence contribution of the i-th micro-motion observation unit is calculated according to the following formula: ; in, Z represents the high confidence contribution of the i-th micro-motion observation unit; Z represents the set of high confidence grid points.

[0058] An adaptive noise reduction mask is generated based on the noise perturbation coefficient, the low-confidence contribution level, and the high-confidence contribution level. In one embodiment, the suppression weights in the adaptive noise reduction mask are calculated according to the following formula: ; in, This represents the suppression weight of the i-th micro-motion observation unit; Indicates the noise disturbance coefficient; Indicates a low level of credibility in contribution; Indicates a high level of credibility in contribution; , , These represent the corresponding weights; This represents the normalized mapping function.

[0059] The normalization mapping function can be a sigmoid function, a piecewise linear function, or a max-min normalization function, used to restrict the suppression weights between 0 and 1. When the suppression weight is close to 1, it means that the participation weight of the corresponding micro-motion observation unit in subsequent dispersive imaging is reduced; when the suppression weight is close to 0, it means that the corresponding micro-motion observation unit is retained.

[0060] The adaptive noise reduction mask differs from fixed-band filters and also from directly deleting noise points from the dispersion energy map. This adaptive noise reduction mask is obtained by back-mapping the dispersion ridge confidence map to the micro-motion observation cell. Its target is the micro-motion observation cell, thus weakening the pseudo-ridge contribution source at the source while retaining the weak but stable effective surface wave component.

[0061] S5. The micro-motion observation unit is weighted using an adaptive noise reduction mask to obtain noise-reduced micro-motion observation data, and the noise-reduced dispersion energy map is reconstructed based on the noise-reduced micro-motion observation data.

[0062] The participation weight of each micro-motion observation unit in subsequent cross-correlation superposition, phase matching evaluation, and dispersion imaging processes is determined based on an adaptive noise reduction mask. In one embodiment, the participation weight of the i-th micro-motion observation unit is determined according to the following formula: ; in, This represents the participation weight of the i-th micro-motion observation unit; This represents the suppression weight of the i-th micro-motion observation unit.

[0063] When using an adaptive noise reduction mask to weight the micro-motion observation units, the phase difference value within each micro-motion observation unit is not changed. Instead, the participation weight of the micro-motion observation unit in cross-correlation superposition, phase matching evaluation, or dispersive imaging processes is adjusted. For micro-motion observation units with high suppression weights, their energy contribution, coherence contribution, and phase matching contribution are reduced; for micro-motion observation units with low suppression weights, their contribution in the dispersive imaging process is maintained.

[0064] Specifically, when the adaptive denoising mask is applied to the energy value, it reduces the energy contribution of micro-motion observation units with high suppression weights; when the adaptive denoising mask is applied to the coherence coefficient, it reduces the coherence contribution of micro-motion observation units with high suppression weights; and when the adaptive denoising mask is applied to phase matching evaluation, it reduces the weight of the phase matching result in dispersive imaging for micro-motion observation units with high suppression weights. After weighting, the data for each micro-motion observation unit is normalized to obtain denoised micro-motion observation data.

[0065] Based on the denoised micro-motion observation data, cross-correlation processing and dispersion imaging processing were re-performed to obtain the denoised dispersion energy map. Since micro-motion observation cells with high noise perturbation and low confidence contribution are suppressed during reconstruction, isolated high-energy points, narrow-band vertical noise bands, and discontinuous pseudo-ridges in the denoised dispersion energy map are weakened. At the same time, the continuity of the true surface wave dispersion ridge is enhanced by reducing the suppression weight of micro-motion observation cells that contribute stably to high-confidence grid points.

[0066] S6. Construct a set of dispersion candidate points based on the noise reduction dispersion energy map and the dispersion ridge credibility map, and extract the dispersion curve from the set of dispersion candidate points through constrained path search.

[0067] like Figure 3 As shown, frequency points and phase velocity points that meet the confidence criteria are selected in the denoised dispersion energy map to construct a dispersion candidate point set. Each candidate point in the dispersion candidate point set includes frequency, phase velocity, energy value, dispersion ridge confidence, array coherence, and noise crossover information.

[0068] Configure path costs for candidate points in the dispersion candidate point set. Path costs include confidence cost, phase velocity jump cost, curvature smoothing cost, and noise traversal cost. Confidence cost is used to reduce the probability of low-confidence candidate points being selected; phase velocity jump cost is used to constrain abrupt changes in phase velocity between adjacent frequency points; curvature smoothing cost is used to constrain the smoothness of the overall shape of the dispersion curve; and noise traversal cost is used to reduce the probability of the path traversing low-confidence noise regions.

[0069] In one embodiment, the total path cost of the dispersion curve is calculated according to the following formula: ; in, This represents the total path cost of candidate path P; This represents the confidence level of the dispersion ridge line at the k-th candidate point in the candidate path; This represents the change in phase velocity between adjacent frequency points; The second-order difference representing the phase velocity change; This represents the noise traversal cost corresponding to the k-th candidate point; , , , These represent the corresponding weights.

[0070] The path cost search algorithm finds the target path that satisfies both phase velocity continuity and curvature smoothness constraints. The path search can be implemented using dynamic programming, shortest path search, or graph search algorithms. The phase velocity continuity constraint limits the phase velocity variation between adjacent frequency points to an adaptive velocity variation limit; the curvature smoothness constraint limits the dispersion curve from unreasonably reversing within local frequency bands.

[0071] When the confidence of candidate points within a certain frequency band is low, the target path is allowed to be interrupted in that frequency band, and this frequency band is marked as a low-confidence band, instead of forcibly connecting to form a false continuous curve. The low-confidence band is used to represent a frequency range with insufficient quality in the dispersion curve, which is different from the low-confidence grid points in the dispersion ridge confidence map.

[0072] The candidate point sequence corresponding to the target path is determined as a dispersion curve. If multiple dispersion curves need to be extracted, after extracting the first dispersion curve, the neighborhood of the extracted path can be suppressed, and the construction of the dispersion candidate point set and the constraint path search can be repeated to obtain higher-order dispersion curves. A minimum velocity interval can be set between dispersion curves of different modes to reduce the risk of mode aliasing.

[0073] S7. Generate a curve quality score based on the dispersion curve. When the curve quality score does not meet the termination condition, update the adaptive noise reduction mask and re-extract the dispersion curve until the final dispersion curve is output.

[0074] The curve continuity, curve smoothness, time window consistency, array coherence, and effective frequency band coverage are calculated for the extracted dispersion curves.

[0075] Curve continuity is used to characterize the degree to which the dispersion curve exists continuously within the target frequency band; curve smoothness is used to characterize the stability of the phase velocity as a function of frequency; cross-time window consistency is used to characterize whether the candidate points corresponding to the dispersion curve appear stably in multiple time windows; array coherence is used to characterize whether the micro-motion observation units corresponding to the dispersion curve satisfy the propagation consistency relationship among multiple stations; effective frequency band coverage is used to characterize the proportion of the available frequency band covered by the dispersion curve.

[0076] In one embodiment, the curve quality score is calculated according to the following formula: ; in, Indicates the quality score of the curve; Indicates the continuity of the curve; Indicates curve smoothness; T c Indicates consistency across time windows; Indicates array coherence; Indicates effective frequency band coverage; , , , , These represent the corresponding weights.

[0077] When the curve quality score meets the termination condition, the current dispersion curve is determined as the final dispersion curve. The termination condition can be that the curve quality score reaches a quality threshold, or that the improvement in the curve quality score after consecutive iterations is less than the improvement threshold.

[0078] When the curve quality score does not meet the termination criteria, a low-quality frequency band is identified. Low-quality frequency bands can be determined based on frequency bands with low curve reliability, large phase velocity jumps, low array coherence, or poor consistency across time windows. The corresponding time window, frequency point, and station pair are then located in reverse order based on the low-quality frequency band, and the adaptive noise reduction mask is updated.

[0079] Specifically, when a low-quality frequency band is caused by curve breaks or phase velocity jumps, the participation weight of high-noise-disturbance-coefficient micro-motion observation units within that frequency band is preferentially reduced, and the candidate search range for adjacent frequency points is expanded. When a low-quality frequency band is caused by insufficient array coherence, the suppression weight of micro-motion observation units with low cross-station propagation consistency is preferentially increased. When a low-quality frequency band is caused by insufficient effective energy, the suppression weight of micro-motion observation units that contribute stably to adjacent high-confidence grid points is reduced. Subsequently, the generation of denoised micro-motion observation data, reconstruction of the denoised dispersion energy map, and constraint path search are re-executed until the final dispersion curve is output.

[0080] The final output includes the frequency, phase velocity, and curve confidence level corresponding to the final dispersion curve; low-confidence frequency bands are marked based on the curve confidence level; and the time window, frequency point, and station pair corresponding to the low-confidence frequency bands are output as dispersion curve verification information. This verification information can be used for weight control during subsequent manual verification or shear wave velocity inversion.

[0081] In a specific application example, multiple vertical geophones are deployed along urban roads to collect continuous micro-motion array data. After acquisition, the micro-motion array data is divided into multiple overlapping time windows, and micro-motion observation units are constructed. Due to vehicle traffic and nearby mechanical equipment operation, some frequency points form narrowband high-energy peaks within a few time windows. Noise perturbation coefficient calculations reveal that these micro-motion observation units exhibit a high degree of narrowband energy surge and a low degree of cross-station propagation consistency. Further generation of an initial dispersion energy map reveals that these high-energy peaks appear as local vertical pseudo-ridges. Dispersion ridge confidence map calculations show that these pseudo-ridges are marked as low-confidence grid points due to their low probability of recurrence across time windows and poor continuity between adjacent frequency points. Subsequently, the dispersion ridge confidence map is back-mapped to the micro-motion observation units to generate an adaptive noise reduction mask, suppressing the time windows, frequency points, and station pairs that form pseudo-ridges. After reconstructing the noise-reduced dispersion energy map, the continuity of the true surface wave dispersion ridges is enhanced, and a continuous dispersion curve is obtained through constrained path search. If low-confidence points still exist in local frequency bands, the adaptive noise reduction mask is updated based on the curve quality score feedback until the final dispersion curve that meets the termination condition is obtained.

[0082] As can be seen from the above embodiments, this invention does not simply add filtering before processing the micro-motion array data, nor does it directly perform noise point clustering and deletion on the dispersion energy map. Instead, it forms an adaptive denoising mask through reverse mapping of the dispersion ridge confidence map, allowing the denoising object to be accurately traced back to the corresponding micro-motion observation unit. Then, the adaptive denoising mask is iteratively updated through curve quality score feedback. This processing method can weaken the contribution source of high-energy pseudo-ridges while retaining the effective signal of weak surface waves, improving the continuity, stability, and reliability of micro-motion surface wave dispersion curve extraction.

[0083] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for adaptive noise reduction and extraction of micro-motion surface wave dispersion curves, characterized in that, include: Acquire micro-motion array data, divide the micro-motion array data into time windows, and construct multiple micro-motion observation units; Based on the time-frequency characteristics, phase characteristics, azimuth characteristics, and station consistency characteristics of the micro-motion observation unit, the noise disturbance coefficient of each micro-motion observation unit is calculated. An initial dispersion energy map is generated based on the micro-motion observation unit, and a dispersion ridge confidence map is generated based on the initial dispersion energy map and the noise disturbance coefficient. The dispersion ridge confidence map is reverse-mapped to the micro-motion observation unit to generate an adaptive noise reduction mask. The micro-motion observation unit is weighted using the adaptive noise reduction mask to obtain noise-reduced micro-motion observation data, and the noise-reduced dispersion energy map is reconstructed based on the noise-reduced micro-motion observation data; Based on the noise reduction dispersion energy map and the dispersion ridge confidence map, a dispersion candidate point set is constructed, and a dispersion curve is extracted from the dispersion candidate point set through constrained path search. A curve quality score is generated based on the dispersion curve. When the curve quality score does not meet the termination condition, the adaptive noise reduction mask is updated and the dispersion curve is extracted again until the final dispersion curve is output.

2. The adaptive noise reduction and extraction method for micro-motion surface wave dispersion curves according to claim 1, characterized in that, The construction of multiple micro-motion observation units includes: The micro-motion array data is subjected to time synchronization, mean removal, and trend removal processing. The processed micro-motion array data is divided into multiple time window data according to the preset time window length and preset overlap rate; The micro-motion observation unit is constructed based on the station pairs, frequency points, station spacing, phase difference, coherence coefficient, azimuth angle, and energy value in each time window data.

3. The adaptive noise reduction and extraction method for micro-motion surface wave dispersion curves according to claim 1, characterized in that, The calculation of the noise disturbance coefficient for each of the micro-motion observation units includes: The degree of narrowband energy surge is determined based on the aforementioned time-frequency characteristics; The phase dispersion degree is determined based on the phase characteristics; The degree of directional energy concentration is determined based on the aforementioned directional characteristics; The degree of cross-station propagation consistency is determined based on the aforementioned station consistency characteristics; The noise disturbance coefficient is generated based on the degree of narrowband energy surge, phase dispersion, azimuth energy concentration, and cross-station propagation consistency.

4. The adaptive noise reduction and extraction method for micro-motion surface wave dispersion curves according to claim 1, characterized in that, Generating the initial dispersion energy map includes: Cross-correlation processing is performed on multiple micro-motion observation units to obtain time-window cross-correlation data; The cross-correlation data of the time window is subjected to dispersion imaging processing to obtain the energy value corresponding to each frequency point and phase velocity point; The energy values ​​corresponding to multiple frequency points and multiple phase velocity points are combined to form the initial dispersion energy map.

5. The adaptive noise reduction and extraction method for micro-motion surface wave dispersion curves according to claim 1, characterized in that, The generation of the dispersion ridge confidence map includes: Calculate the energy saliency corresponding to each frequency point and phase velocity point based on the initial dispersion energy map; Calculate the probability of repeated occurrence across multiple time windows based on the energy distribution in multiple time windows; Calculate the continuity of adjacent frequency points based on the phase velocity change between adjacent frequency points; Array coherence is calculated based on the phase relationship between multiple station pairs; The dispersion ridge confidence map is generated by combining the energy significance, the probability of recurrence across time windows, the continuity of adjacent frequency points, the array coherence, and the noise perturbation coefficient.

6. The adaptive noise reduction and extraction method for micro-motion surface wave dispersion curves according to claim 1, characterized in that, Generating an adaptive noise reduction mask includes: Determine the time window, frequency point, and station pair corresponding to the low confidence grid points in the dispersion ridge confidence map; The time windows, frequency points, and station pairs corresponding to the low-confidence grid points are located as pseudo-ridgeline contribution sources; Increase the suppression weight of the micro-motion observation unit corresponding to the pseudo-ridge contribution source; Reduce the suppression weight of micro-motion observation units that contribute to stability at high-confidence grid points; The adaptive noise reduction mask is generated based on the suppression weights of each micro-motion observation unit.

7. The adaptive noise reduction and extraction method for micro-motion surface wave dispersion curves according to claim 1, characterized in that, The noise-reduced micro-motion observation data obtained include: The participation weight of each of the micro-motion observation units in the cross-correlation superposition, phase matching evaluation, or dispersion imaging process is determined based on the adaptive noise reduction mask. The energy contribution, coherence contribution, and phase matching contribution of the micro-motion observation unit are adjusted based on the participation weights to obtain the noise-reduced micro-motion observation data. Based on the noise-reduced micro-motion observation data, the dispersion imaging process is re-executed to obtain the noise-reduced dispersion energy map.

8. The adaptive noise reduction and extraction method for micro-motion surface wave dispersion curves according to claim 1, characterized in that, Extracting dispersion curves from the set of dispersion candidate points through constrained path search includes: In the noise reduction dispersion energy map, frequency points and phase velocity points that meet the confidence conditions are selected to construct the dispersion candidate point set; Configure path costs for candidate points in the dispersion candidate point set, the path costs including confidence cost, phase velocity jump cost, curvature smoothing cost and noise crossing cost; Based on the path cost, a target path that satisfies the phase velocity continuity constraint and curvature smoothness constraint is searched. The sequence of candidate points corresponding to the target path is determined as the dispersion curve.

9. The adaptive noise reduction and extraction method for micro-motion surface wave dispersion curves according to claim 1, characterized in that, The curve quality score generated based on the dispersion curve includes: Calculate the continuity, smoothness, consistency across time windows, array coherence, and effective bandwidth coverage of the dispersion curve; The curve quality score is generated based on the curve continuity, curve smoothness, cross-time window consistency, array coherence, and effective frequency band coverage. When the curve quality score is lower than the quality threshold, a low-quality frequency band is identified, and the adaptive noise reduction mask is updated based on the low-quality frequency band.

10. The adaptive noise reduction and extraction method for micro-motion surface wave dispersion curves according to claim 1, characterized in that, The final output dispersion curve includes: Output the frequency, phase velocity, and curve reliability corresponding to the final dispersion curve; Low-confidence frequency bands are marked according to the confidence level of the curve; Output the time window, frequency point, and station pair corresponding to the low-confidence frequency band as the dispersion curve verification information.

Images