Spatially constrained image cross-correlation dynamic time warping method, system, and storage medium

Abstract

The application discloses a spatial constraint image cross-correlation dynamic time warping method, a system and a storage medium, and belongs to the field of image time-lapse processing. The method comprises the following steps: constructing an error matrix containing spatial information based on a two-dimensional Gaussian weighted local cross-correlation difference coefficient to measure the local structure similarity of two images in a space-time window; introducing a spatial constraint term when calculating a three-dimensional cumulative distance matrix, and making the time-lapse path keep continuity in the time and space directions through time and space penalty coefficients; extracting a two-dimensional time-varying time-lapse sequence through a backtracking algorithm, and applying the time-lapse sequence to a monitoring image to complete time difference correction, and generating a corrected monitoring image and a two-image difference profile. The application scheme keeps the advantages of the cross-correlation dynamic time warping method, such as processing large time-lapse and strong noise resistance, and significantly improves the problem of poor lateral continuity in the process of the cross-correlation dynamic time warping method through the introduction of the spatial constraint, so that the image time-lapse field is smooth and continuous in the spatial and time dimensions.

Description

Technical Field

[0001] This invention relates to the field of time-shifted image processing technology, and more specifically to a method, system, and storage medium for dynamic time warping of spatially constrained image cross-correlation. Background Technology

[0002] Currently, the main methods for estimating and correcting the time difference between two images can be divided into two categories. The first category is represented by cross-correlation methods, which estimate the time difference by comparing the waveform similarity of the two image signals within a local time window, exhibiting good stability and noise resistance. However, the local cross-correlation method is inherently limited by small time shifts: for time differences that vary non-uniformly with depth, keeping the window constant only handles small time shifts; if the window is enlarged to accommodate large time shifts, the calculation results are easily dominated by high-amplitude events within the window, leading to bias. To address these limitations, researchers have successively proposed different time difference estimation methods such as Taylor expansion and analysis of variance, but they still struggle to provide accurate point-by-point time-varying time differences when faced with complex situations involving large time shifts.

[0003] The second category is the Dynamic Time Warping (DTW) method. DTW estimates point-to-point time-varying time differences by searching for the globally optimal path to align two phase images within the error matrix, possessing the ability to handle large time shifts and automatically estimate time-varying time differences. However, conventional DTW constructs the error matrix using the Euclidean distance between the amplitude values ​​of sampling points. This approach only considers the amplitude information of the sampling points, completely ignoring the local waveform structure properties around the sampling points. In real-world data, two sampling points may have the same amplitude value but correspond to completely different phases or local waveform characteristics. This can lead DTW to generate incorrect optimal paths, making it highly sensitive to noise and exhibiting poor stability, severely limiting the application of this method in practical data processing.

[0004] To improve the noise resistance of Time-Shift Warping (DTW), Wang Jianhua et al. (2023) introduced the concept of local cross-correlation into DTW and proposed a cross-correlation-based Dynamic Time Warping (CDTW) method. When constructing the error matrix, CDTW uses the difference coefficient calculated by weighted normalized cross-correlation within a local time window to replace the Euclidean distance, effectively overcoming the sensitivity of conventional DTW to noise and amplitude variations. However, the CDTW method still uses a channel-by-channel processing approach when processing two-dimensional or three-dimensional images, completely ignoring the waveform correlation and spatial continuity information between adjacent channels at the same time. When the data signal-to-noise ratio is low, independent channel-by-channel processing easily produces contradictory time difference estimation results between different channels, manifesting as poor lateral continuity of the time-shifted field, spatial jitter, or pseudo-discontinuities, affecting the reliability of subsequent time-shifted image interpretation.

[0005] In summary, existing methods face the following main problems in time-shifted image time difference estimation: cross-correlation methods have difficulty handling large time shifts; conventional DTW is sensitive to noise and amplitude changes; although CDTW solves the noise resistance problem, the problem of poor lateral continuity of the time-shifted field caused by channel-by-channel processing has not been effectively solved. Summary of the Invention

[0006] The purpose of this invention is to provide a spatially constrained image cross-correlation dynamic time warping method, system, and storage medium to solve the problem of poor lateral continuity of the time-shift field caused by channel-by-channel processing in time-shift image time difference estimation in existing technologies, while maintaining the advantages of the CDTW method in handling large time shifts and noise resistance.

[0007] To achieve the above objectives, one embodiment of the present invention provides a spatially constrained image cross-correlation dynamic time warping method. The method includes: acquiring a base image and a monitoring image, and establishing a spatial correspondence between the two images; for each spatial-temporal location to be processed within the target processing area, calculating the difference coefficient within a preset spatiotemporal window based on two-dimensional Gaussian weighted local cross-correlation, and constructing a three-dimensional error matrix containing spatial information; based on the three-dimensional error matrix, calculating a three-dimensional cumulative distance matrix by introducing a cumulative distance matrix recursive formula of time smoothing coefficient and spatial penalty coefficient; tracing the optimal path on the three-dimensional cumulative distance matrix using a backtracking algorithm to extract a two-dimensional time-varying time-shift sequence; applying the two-dimensional time-varying time-shift sequence to the monitoring image to complete time difference correction, generating a time difference-corrected monitoring image, and calculating the difference profile between the two images.

[0008] Optionally, the steps for constructing the three-dimensional error matrix include: using spatial points and time point Set half the width of the space as the center. and time half width A two-dimensional spatiotemporal window is constructed; within the spatiotemporal window, a two-dimensional Gaussian weighting function is used to assign weights to each point; the spatiotemporal window is calculated. The weighted normalized cross-correlation value between the baseline image and the monitoring image is converted into a difference coefficient ranging from 0 to 2 to obtain the error matrix. .

[0009] Optionally, the expression for the two-dimensional Gaussian weighting function is: ,in and These represent the Gaussian standard deviations in the spatial and temporal directions, respectively. Spatial half-width. The empirical value range is 1 to 3 questions, with a time half-width. The value is 1.5 to 2 times the wavelet period.

[0010] Optionally, when calculating the 3D cumulative distance matrix, a time penalty coefficient (ranging from 0 to 1) and a spatial penalty coefficient can be used. (Values ​​range from 0 to 1) Introduce constraints in the recursive formula: For each spatial-temporal location, when selecting the predecessor node, simultaneously consider the candidate path of the current path at the previous time step and apply a penalty to the previous time step of the adjacent path. The candidate paths are then used to ensure that the final time-shift field remains continuous in both time and space.

[0011] Optionally, when extracting the two-dimensional time-varying time-shift sequence using the backtracking algorithm, the slope constraint parameter can be used. and The maximum values ​​of the time shift rate of change in the time direction and the spatial direction are respectively limited, so as to allow the path to jump between adjacent paths to achieve spatial continuity while ensuring the physical rationality of the time shift sequence.

[0012] Optionally, when applying the obtained two-dimensional time-varying time-shift sequence to the monitoring image, Sinc interpolation or cubic spline interpolation methods are used to correct the time difference for subsampling accuracy, and the correction effect is quantitatively evaluated by the mean of the normalized cross-correlation coefficient and the root mean square error of the residual time shift.

[0013] A second aspect of the present invention provides a spatially constrained image cross-correlation dynamic time warping system, the system comprising: a data acquisition unit, an error matrix construction unit, a cumulative distance calculation unit, a time shift extraction unit, and a time difference correction unit.

[0014] A third aspect of the present invention provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the above-described spatial constraint cross-correlation dynamic time warping method.

[0015] In summary, the present invention has the following advantages:

[0016] 1. The temporal two-dimensional error matrix of CDTW is extended into an error matrix that includes spatial information. The local structural similarity of two images in the spatiotemporal window is jointly evaluated by two-dimensional Gaussian weights, making the difference coefficient more robust to noise and amplitude changes.

[0017] 2. Introducing path candidates and spatial penalty coefficients from adjacent paths into the recursive calculation of the cumulative distance matrix allows the search for the optimal path to consider both temporal and spatial continuity constraints, fundamentally solving the problem of lateral discontinuity caused by path-by-path processing.

[0018] 3. By using two adjustable parameters, time and space penalty coefficients, the smoothness of the time-shifted field can be flexibly controlled according to the signal-to-noise ratio of the actual image, balancing processing flexibility and result reliability.

[0019] 4. The proposed method is naturally compatible with parallel computing frameworks (MPI+OpenMP hybrid strategy), can efficiently process large-scale 3D / 4D image data, and has good engineering practicality. Attached Figure Description

[0020] Figure 1 This is a flowchart of the steps of a spatially constrained image cross-correlation dynamic time warping method provided in one embodiment of the present invention;

[0021] Figure 2 This is a schematic diagram illustrating the error matrix construction principle provided by one embodiment of the present invention;

[0022] Figure 3 This is a test result image of low signal-to-noise ratio image data provided by one embodiment of the present invention;

[0023] Figure 4 This is a test result diagram of Monitor1, an anticline horizontal layered model with time shift and phase change, provided by one embodiment of the present invention;

[0024] Figure 5 This is a test result diagram of Monitor2 data, which includes amplitude variation and global time shift, based on Monitor1, provided by one embodiment of the present invention;

[0025] Figure 6 This is a schematic diagram of the baseline cross-section of an actual image provided in one embodiment of the present invention;

[0026] Figure 7 This is a schematic cross-sectional view of the monitor for an actual image provided in one embodiment of the present invention;

[0027] Figure 8 This is a cross-sectional schematic diagram of an actual image monitor corrected using the method of the present invention, provided in one embodiment of the present invention;

[0028] Figure 9 This is a schematic diagram of the cross-sectional differences between the original two actual images provided in one embodiment of the present invention;

[0029] Figure 10 This is a schematic diagram of the actual image difference profile after correction using the present invention, provided by one embodiment of the present invention;

[0030] Figure 11 This is a structural block diagram of a spatially constrained image cross-correlation dynamic time warping system provided in one embodiment of the present invention;

[0031] Figure 12 This is an internal structural diagram of a computer device provided in one embodiment of the present invention.

[0032] Explanation of reference numerals in the attached figures:

[0033] A01 - Processor; A02 - Network Interface; A03 - Internal Memory; A04 - Non-volatile Storage Media; B01 - Operating System; B02 - Computer Program. Detailed Implementation

[0034] This invention provides a spatially constrained image cross-correlation dynamic time warping method (S-CDTW), such as... Figure 1 As shown, the method includes:

[0035] Step A1: Construct the improved error matrix.

[0036] In this embodiment, the principle of error matrix construction in the CDTW method is first reviewed, and the construction process of the error matrix of this invention is explained based on this. In conventional DTW, the error matrix is ​​constructed by the Euclidean distance between the amplitudes of the sampling points. This error matrix only considers the amplitude value of the sampling point and ignores the local waveform structure properties around the point, making DTW highly sensitive to noise and amplitude changes.

[0037] The CDTW method replaces the above Euclidean distance with local cross-correlation difference coefficients: First, define... and Zero-delay cross-correlation within the local time window is performed, then normalized and Gaussian weights are introduced to obtain the CDTW error matrix:

[0038]

[0039] in The weighting function is Gaussian. The time half-window length, The time-direction Gaussian standard deviation is represented by . This error matrix ranges from 0 to 2; a smaller difference coefficient indicates a more similar local structure between the two signals at that location. However, the CDTW error matrix described above only performs local cross-correlation calculations along the time direction, without considering adjacent channel information in the spatial direction.

[0040] The three-dimensional error matrix of this invention extends the cross-correlation calculation from a one-dimensional time window to a two-dimensional spatiotemporal window, such as... Figure 2 As shown, using spatial points and time point Set half the width of the space as the center. and time half width Construct an improved error matrix:

[0041]

[0042] in, Represents a three-dimensional error matrix. Represents a point in space. Represents a point in time. Represents time shift, and These represent the half-lengths of the spatial and temporal windows, respectively, in the experiments conducted in this paper. The experience value is 1-3 questions. The value of must include a complete wavelet shape to ensure the stability of the cross-correlation, and should be 1.5 to 2 times the wavelet period. and Represents a two-dimensional input signal. and Representing respectively and The spatial and temporal range centered on the center. and These are the standard deviations of the Gaussian function in the spatial and temporal directions, respectively, and their ranges are respectively... and .

[0043] The aforementioned error matrix not only contains the temporal local structure information at the location to be processed, but also integrates the waveform similarity information of all neighboring channels within the spatial neighborhood at the corresponding time. Therefore, it has stronger robustness to local noise and can introduce spatial consistency constraints at the error matrix level.

[0044] Step A2: Optimized 3D cumulative distance matrix.

[0045] In conventional DTW and CDTW, the cumulative distance matrix Calculated using the following recursive formula:

[0046]

[0047]

[0048] The above recursion only considers the three candidate predecessor nodes of the same track at the previous time step (time shift changes by ±1 or remains unchanged), without considering the path information of adjacent tracks. Therefore, processing each track independently can easily lead to lateral discontinuous time shift estimates.

[0049] In the recursive calculation of the three-dimensional cumulative distance matrix of this invention, candidate predecessor paths from adjacent paths are simultaneously introduced. Three-dimensional cumulative distance matrix The recursive formula is as follows:

[0050]

[0051]

[0052] in The time penalty coefficient (ranging from 0 to 1) is mainly used to suppress the violent jitter of the time-shift path in the time direction and generate a time-shift curve that changes smoothly with time. This is a spatial penalty coefficient (ranging from 0 to 1), which is applied when a path jumps from an adjacent path. At the cost of controlling the path to jump in spatial direction, its core effect is to introduce spatial continuity constraints into the algorithm, making the final calculated time-shift field smoother and more continuous in the horizontal direction.

[0053] In practical applications, and The value needs to be adjusted based on the data signal-to-noise ratio: when the data signal-to-noise ratio is low, increase it appropriately. and This helps enhance the smoothness of time-shifted fields; when the image undergoes significant lateral changes, it is necessary to appropriately reduce the [value / size]. This allows for reasonable lateral variations in the time-shifted field in space.

[0054] Step A3: Extract the two-dimensional time-varying time-shift sequence using a backtracking algorithm.

[0055] In obtaining the three-dimensional cumulative distance matrix Then, the optimal path is traced through a backtracking algorithm, and the optimal time shift estimate of each spatial channel at each moment is extracted to form a two-dimensional time-varying time shift sequence.

[0056] The backtracking process specifically includes: First, at the last time sampling point... For each space path Search separately exist The global minimum value in the direction determines the backtracking starting point:

[0057]

[0058] Then, starting from the backtracking point of each path, trace backward step by step according to the recursive formula:

[0059]

[0060] This includes candidate nodes with the smallest cumulative distance from the current path and from adjacent paths.

[0061] At the same time, through slope constraint parameters and The maximum values ​​of the time shift rate of change in both the time and spatial directions are limited to ensure the physical rationality of the time shift sequence. Control the time shift change between adjacent time points to not exceed One sampling point, Control the time shift difference between adjacent channels to not exceed Each sampling point. In this embodiment of the invention, Set to 1; The value is dynamically set based on whether the lateral changes in the geological structure are drastic. Smaller values ​​can be set for areas with gentle geological structures, while larger values ​​can be set for areas with complex geological structures.

[0062] The above backtracking process is performed in parallel on all spatial channels to finally obtain a complete two-dimensional time-varying time-shift sequence with the same dimension as the input two-dimensional image.

[0063] Step A4: Apply the two-dimensional time-varying time-shift sequence to the monitoring image to complete the time difference correction.

[0064] After obtaining the two-dimensional time-varying time-shift sequence, the time shift estimation with integer sampling point precision is improved to sub-sampling precision by using Sinc interpolation or cubic spline interpolation. Then, the interpolated time shift is applied point by point to the corresponding channel and time of the monitoring image. The time difference between the two images is eliminated by time shift compensation, and the monitoring image after time difference correction is obtained.

[0065] Finally, the amplitude difference profiles of the base image and the monitoring image before and after time difference correction are calculated to intuitively reflect the temporal dynamic changes of the image; at the same time, the mean of the normalized cross-correlation coefficient and the root mean square error of the residual time shift are calculated to quantitatively evaluate the correction effect.

[0066] Example: First, a simple model is used as an example to verify the method of this invention. Figure 3 Baseline in Figure 3 (a) In this context, a known time shift is applied and random noise is added to generate Monitor data. Figure 3 (b) of the dataset has a signal-to-noise ratio (SNR) of 2dB. The dataset contains 11 channels, each with 750 sampling points and a sampling interval of 4ms.

[0067] Figure 3 (c) shows the single-channel waveform record of channel 6 for a visual comparison of the differences between the two phases. The S-CDTW method is applied to process the model data pair to estimate the time-varying time shift. Figure 3 Figure (e) shows the distance matrix calculated by this invention and its corresponding time-varying time shift estimation result. It can be seen that the distance matrix calculated using the S-CDTW method proposed in this invention and the estimated time-varying time shift are consistent with the known time shift (…). Figure 3 The results (d) are almost identical. In noisy environments, the S-CDTW method exhibits strong robustness in distance matrix construction.

[0068] Secondly, the method of the present invention is verified using a horizontal layered model containing an anticline as an example. The number of spatial channels in the Baseline data and Monitor data is 99, the number of temporal sampling points is 876, the spatial sampling interval is 25m, and the temporal sampling interval is 2ms.

[0069] (a) Monitor1 data test: same as Baseline data ( Figure 4 Compared to (a) in the previous data, the Monitor1 data shows a significant increase in anticline structural amplitude between 1.0s and 1.25s, indicating time lag and phase reversal in the underlying strata. Figure 4 (b) Red and black arrows in the diagram). Using the S-CDTW method of this invention to match and correct Monitor1 data to Baseline data, it can be seen that the S-CDTW method of this invention effectively eliminates the lateral discontinuity problem and accurately corrects the polarity reversal layer. Figure 4 (c) in the middle), the time-shifted image residual ( Figure 4 (d) in the figure accurately reflects the time-shifting changes of the image.

[0070] (II) Monitor2 Data Testing: Compared with Baseline Data ( Figure 5 Compared to (a) in the previous example, Monitor2 data also exhibits a slight global time lag compared to Monitor1 data. Figure 5 (b) Red arrow), while there is a clear energy change in the first layer ( Figure 5 (b) Blue arrow in the diagram). The S-CDTW method of this invention can effectively correct erroneous stratigraphic responses caused by overall time shift ( Figure 5 (c) verifies the robustness of the method in handling systematic time-shift biases and complex amplitude variations, and the resulting time-shifted image residuals ( Figure 5 (d) in the figure accurately reflects the time-shifting changes of the image.

[0071] (III) Actual Image Testing: The S-CDTW method of this invention was applied to a real time-shifted image ( Figure 6 , Figure 7 The image has 688 spatial channels, 2001 temporal sampling points, and a sampling interval of 1 ms. After applying the method of this invention, the corrected detection image is as follows: Figure 8 As shown. Compare the differences between the original image and the cross-section ( Figure 9 The amplitude difference between the corrected monitoring image and the base image was significantly reduced. Figure 10 This effectively eliminates spurious amplitude differences caused by time differences, accurately reflects the temporal dynamic changes of the Monitor image, and verifies the effectiveness of the method of the present invention on actual images.

[0072] Parallel computing implementation:

[0073] For 3D / 4D image data and massive trace data, a hybrid parallel strategy combining MPI and OpenMP can be adopted to improve computational efficiency: MPI is responsible for data decomposition, dividing the data into trace groups and distributing them to different processes for parallel processing; within each process, OpenMP is used to perform parallel computation of core calculation steps such as the 3D error matrix, the 3D cumulative distance matrix, and backtracking. When constructing the 3D error matrix, the computation of different spatial traces is independent and can be fully parallelized; however, when calculating the 3D cumulative distance matrix, there are data dependencies between adjacent traces, requiring boundary data exchange via MPI to maintain the spatial constraint information of adjacent traces. This parallel strategy can significantly improve computational efficiency and meet the speed requirements of engineering practice.

[0074] like Figure 11 As shown, this invention provides a spatially constrained image cross-correlation dynamic time warping system. The system includes: a data acquisition unit for acquiring a base image and a monitoring image, and establishing a spatial correspondence between the two images; an error matrix construction unit for constructing an error matrix containing spatial information based on a two-dimensional Gaussian weighted local cross-correlation difference coefficient; a cumulative distance calculation unit for calculating a three-dimensional cumulative distance matrix based on the error matrix by introducing a recursive formula for time and space penalty coefficients; a time shift extraction unit for tracing the optimal path on the three-dimensional cumulative distance matrix using a backtracking algorithm to extract a two-dimensional time-varying time shift sequence; and a time difference correction unit for applying the two-dimensional time-varying time shift sequence to the monitoring image to complete time difference correction, and generating a time difference-corrected monitoring image and a difference profile between the two images.

[0075] The present invention also provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the above-described spatially constrained image cross-correlation dynamic time warping method.

[0076] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 12 As shown, the computer device includes a processor A01, a network interface A02, internal memory A03, and a non-volatile storage medium A04 connected via a system bus. The non-volatile storage medium A04 stores an operating system B01 and a computer program B02; the internal memory A03 provides the environment for the operation of the operating system and the computer program; the processor A01 provides computing and control capabilities; and the network interface A02 is used for communication with external terminals via a network connection. When the computer program B02 is executed by the processor A01, it implements a spatially constrained image cross-correlation dynamic time warping method.

[0077] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a microcontroller, chip, or processor to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, portable hard drive, read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0078] The optional embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the embodiments of the present invention are not limited to the specific details described above. Within the scope of the technical concept of the embodiments of the present invention, various simple modifications can be made to the technical solutions of the embodiments of the present invention, and these simple modifications all fall within the protection scope of the embodiments of the present invention. It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, the embodiments of the present invention will not further describe the various possible combinations.

[0079] Furthermore, various different embodiments of the present invention can be combined in any way, as long as they do not violate the spirit of the embodiments of the present invention, they should also be regarded as the content disclosed by the embodiments of the present invention.

Claims

1. A method for dynamic time warping of spatially constrained image cross-correlation, characterized in that, The method includes: The difference coefficient is calculated within the spatiotemporal window based on two-dimensional Gaussian weighted local cross-correlation, and an error matrix containing spatial information is constructed. Based on the error matrix, a three-dimensional cumulative distance matrix is ​​formed by introducing a recursive formula for the cumulative distance matrix with time and space penalty coefficients. The optimal path is traced on the three-dimensional cumulative distance matrix using a backtracking algorithm to extract a two-dimensional time-varying time-shift sequence; The two-dimensional time-varying time-shift sequence is applied to the monitoring image to complete the time difference correction, generating the time difference-corrected monitoring image, and the difference profile between the two images is calculated.

2. The method according to claim 1, characterized in that, The steps to construct an error matrix containing spatial information include: In terms of spatial points and time point Set half the width of the space as the center. and time half width Construct a two-dimensional spacetime window; Within the spatiotemporal window, the basic image and monitoring images Construct respectively with and The dataset is a neighborhood of the center; a two-dimensional Gaussian weighting function is used. Each point within the spatiotemporal window is assigned a weight, which decreases as the distance from the point to the center increases; Calculate the weighted normalized cross-correlation value between the base image and the monitoring image within the spatiotemporal window, and convert the normalized cross-correlation value into the difference coefficient to obtain the error matrix. Its range is between 0 and 2.

3. The method according to claim 2, characterized in that, The expression for the two-dimensional Gaussian weighting function is: ;in and Let be the Gaussian standard deviations in the spatial and temporal directions, respectively, and let their ranges satisfy . and ; space half width The empirical value range is 1 to 3 questions, with a time half-width. The value of must contain at least one complete wavelet, and is taken as 1.5 to 2 times the wavelet period.

4. The method according to claim 1, characterized in that, The improved recursive formula for the 3D cumulative distance matrix is ​​as follows: ； ； Path initialization is performed for the first time sampling point. For subsequent time sampling points, during the recursive calculation, the predecessor node from the current channel in the previous time and the predecessor node from the adjacent channel in the previous time are considered simultaneously. A time penalty coefficient is introduced into the recursive formula. and space penalty coefficient The time penalty coefficient A constraint term acting in the time direction suppresses severe jitter in the time-shift path; its value ranges from 0 to 1. The spatial penalty coefficient... As a penalty imposed when a path jumps from an adjacent path, a spatial continuity constraint is introduced to smooth the time-shift field laterally; its value ranges from 0 to 1. Represents the error matrix. This represents the optimized 3D cumulative distance matrix. Indicates the space channel number, Indicates the time sampling point. Indicates the time shift.

5. The method according to claim 1, characterized in that, The steps for extracting a two-dimensional time-varying time-shift sequence using a backtracking algorithm include: At the last time sampling point At each location, search for each spatial path separately. exist The global minimum value in the direction is used to determine the backtracking starting point; starting from the backtracking starting point, the optimal time shift for each time step is traced backward according to the recursive formula; the maximum value of the time shift change rate in the time direction and the spatial direction is limited by the slope constraint parameter to ensure the physical rationality of the time shift sequence; backtracking is performed in parallel on all spatial channels to obtain the complete two-dimensional time shift field.

6. The method according to claim 1, characterized in that, Before applying the two-dimensional time-shift field to the monitoring image, the process includes: using the Sinc interpolation method to perform subsampling accuracy time-shift correction on the estimated two-dimensional time-shift sequence; applying the interpolated time-shift amount to each channel and time point of the monitoring image, and eliminating the time difference between the two images through time difference correction.

7. The method according to claim 1, characterized in that, The steps for acquiring base images and monitoring images and establishing the spatial correspondence between the two data periods include: The base image and monitoring image are unified under the same coordinate system to ensure that the spatial sampling position, channel spacing and time sampling interval of the two phases are consistent; the amplitude normalization and phase consistency processing of the two phases are performed to eliminate the amplitude ratio difference caused by non-repetitive acquisition factors; a one-to-one correspondence between the two phases of data is established according to the same spatial channel number to form a pairing of base and monitoring image sets.

8. A spatially constrained image cross-correlation dynamic time warping system, characterized in that, The system includes: a data acquisition unit, used to acquire basic images and monitoring images, and establish a spatial correspondence between the two images; The error matrix construction unit is used to construct an error matrix containing spatial information based on the two-dimensional Gaussian weighted local cross-correlation difference coefficient. The cumulative distance calculation unit is used to calculate the three-dimensional cumulative distance matrix based on the three-dimensional error matrix by introducing a recursive formula with time and space penalty coefficients. The time-shift extraction unit is used to trace the optimal path on the three-dimensional cumulative distance matrix using a backtracking algorithm to extract the two-dimensional time-varying time-shift sequence; The time difference correction unit is used to apply the two-dimensional time-varying time-shift sequence to the monitoring image to complete the time difference correction, and generate the time difference corrected monitoring image and the difference profile between the two images.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the spatially constrained image cross-correlation dynamic time warping method according to any one of claims 1 to 7.

Images