A shallow profile layering extraction method and layering extractor based on multi-level resolution

Abstract

The application discloses a shallow profile layering extraction method and a layering extractor based on multi-level resolution. The layering extraction method comprises the following steps: converting multi-channel shallow stratum echo data into a shallow profile image intensity matrix; performing N-level (N>1) resolution decomposition on the shallow profile image intensity matrix, extracting an edge point matrix and intensity double thresholds from an N-level low-frequency resolution coefficient matrix obtained through the decomposition, and performing edge following and connection processing on the edge point matrix by using the intensity double thresholds to obtain an N-level reflection layer matrix; and compensating the N-level reflection layer matrix by using an N-level high-frequency resolution coefficient matrix obtained through the decomposition to obtain an N-1-level reflection layer matrix with a higher resolution, and in this way, a 0-level target reflection layer matrix containing multi-level high-resolution detail information is finally obtained. The application can continuously and autonomously extract reflection layers from echo data, and can stably improve the resolution of the reflection layers, has strong data adaptability, can obtain more reflection layer detail information, and is highly reliable.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of digital signal processing, and particularly relates to a shallow profile layering method based on multi-level resolution. BACKGROUND

[0002] The shallow seismic profiling technology originated in the 1860s. Whether high-resolution reflection layers can be picked up from the shallow seabed stratum echo data has always been a difficult problem restricting the accurate identification of geological structures, and is also related to the establishment of high-quality sediment databases and the scientific and effective development of seabed sedimentology research. However, the traditional manual picking method is time-consuming and laborious, and it is also difficult to deeply mine the information carried by the data.

[0003] The existing sediment layer edge automatic detection technology can effectively solve the above problems and provide efficient, accurate and high-resolution basic data information for subsequent scientific interpretation of shallow seabed profiling data, which can significantly reduce the workload of data processing and significantly improve the flexibility of data information utilization. In the automatic detection technology, manual setting and single threshold parameters related to intensity cannot realize the extraction of continuous reflection layers, and the extraction method based on the Markov random field theory can ensure the continuity of the sediment layer picking by considering the adjacent information, but it is easily affected by the unbalanced data intensity. The linear detection method with single-level resolution can quickly extract the main reflection layer, but it needs high original resolution and signal-to-noise ratio as a prerequisite, and it is also easy to lose the detail information of the reflection layer. SUMMARY

[0004] In view of the defects of the prior art, the purpose of the present application is to provide a method for robust layering of shallow profiling echo data and continuous and autonomous extraction of high-resolution reflection layer signals. The method has less limitation on data intensity and other factors, can obtain more detail information of the reflection layer, and can be widely used.

[0005] The technical scheme of the present application is as follows:

[0006] A shallow profile layering extraction method based on multi-level resolution, comprising:

[0007] Step 1: converting the multi-channel shallow stratum echo data received in the shallow profiling into a shallow profile image intensity matrix S0 in the range of 0-255 grayscale values;

[0008] Step 2: performing N-level resolution decomposition on the shallow profile image intensity matrix S0 by low-pass filtering and high-pass filtering to obtain N low-frequency resolution coefficient matrices and N high-frequency resolution coefficient matrix groups, wherein the high-frequency resolution coefficient matrix group is composed of high-frequency resolution coefficient matrices in the horizontal direction, the vertical direction and the diagonal direction of the image;

[0009] Step 3: extracting an edge point matrix P from an Nth-level low-frequency resolution coefficient matrix S N in the N low-frequency resolution coefficient matrices N , and a high-intensity threshold T1 and a low-intensity threshold T2;

[0010] Step 4: performing edge following and connection processing on the edge point matrix P N according to the high-intensity threshold T1 and the low-intensity threshold T2, to obtain an Nth-level reflector matrix R N with continuous structure;

[0011] Step 5: reconstructing an Nth-level high-frequency resolution matrix G N from an Nth-level high-frequency resolution coefficient matrix group in the N high-frequency resolution coefficient matrix groups, compensating the Nth-level reflector matrix R N with the Nth-level high-frequency resolution matrix G N , and forming an N-1th-level reflector matrix R N-1 containing Nth-level high-frequency resolution information;

[0012] Step 6: sequentially compensating an N-1th-level reflector matrix R N-1 with an N-1th-level high-frequency resolution coefficient matrix group according to the processes of steps 3-5, to obtain an N-2th-level reflector matrix R N-2 , and finally obtaining a reflector matrix R0 containing multi-level high-frequency resolution information, which is a target reflector.

[0013] According to some embodiments of the present application, the step 1 comprises:

[0014] Step 1-1: arranging and combining all single-channel signals e1(n), e2(n), …, e M (n) received in the shallow profile detection in time sequence to form an original signal matrix E, E = [e1(n) e2(n) e3(n) … e M (n)], where M is the total number of channels, n ∈ [1, K] represents the sequence number of signal sample points, and K is the total number of sample points of the single-channel signal acquisition;

[0015] Step 1-2: performing normalization processing on the original signal matrix E to obtain a normalized original signal matrix E a ;

[0016] Step 1-3: mapping the normalized original signal matrix E a to the maximum gray value range of 0-255 to obtain a shallow profile image intensity matrix S0.

[0017] According to some embodiments of the present application, the low-frequency resolution coefficient matrix and the high-frequency resolution coefficient matrix group are obtained through the following calculation models, respectively:

[0018] S j = L r L c S j-1

[0019] H j = L r G c S j-1

[0020] V j = G r L c S j-1

[0021] D j = G r G c S j-1

[0022] wherein j represents any decomposition level, j∈[1, N], N represents total decomposition levels; S j represents a low-frequency resolution coefficient matrix obtained by the jth decomposition; H j , V j , D j respectively represent high-frequency resolution coefficient matrices in horizontal direction, vertical direction and diagonal direction obtained by the jth decomposition, which constitute the jth high-frequency resolution coefficient matrix group {H j V j D j}; L r and L c are respectively operators of low-pass filtering on rows and columns of the shallow profile image intensity matrix S0, G r and G c are respectively operators of high-pass filtering on rows and columns of the shallow profile image intensity matrix S0, which are conjugate mirror filters, and satisfy the following conditions in frequency domain:

[0023] L r,c (ω)G r,c (ω) + L r,c (ω+π)G r,c (ω+π) = 0

[0024] wherein L r,c (ω) respectively represents an operator of low-pass filtering on rows and columns of the shallow profile image intensity matrix S0 in frequency domain, G r,c (ω) respectively represents an operator of high-pass filtering on rows and columns of the shallow profile image intensity matrix S0 in frequency domain, and ω represents an angular frequency.

[0025] According to some specific embodiments of the present application, the step 3 comprises:

[0026] Step 3-1: For the Nth level low-frequency resolution coefficient matrix S N Maximum value extraction is performed on each column of data.

[0027] The edge point matrix P is composed of the extracted maxima of each column. N ;

[0028] Step 3-2: Calculate the edge point matrix P N The intensity value data distribution is used to obtain an intensity statistical histogram;

[0029] Step 3-3: Suppress the first peak of the intensity statistical histogram and smooth it to obtain a smoothed histogram;

[0030] Steps 3-4: Extract the first-order statistical features of the smoothed histogram, including its statistical mean x and standard deviation σ, and obtain the high-intensity threshold T1 and low-intensity threshold T2 according to the following formulas:

[0031] T1=x+2σ

[0032] T2 = x.

[0033] According to some specific embodiments of the present invention, step 4 includes:

[0034] Step 4-1: In the edge point matrix P N In the process, edge points with intensity values ​​lower than the low intensity threshold T2 are considered noise and discarded.

[0035] Step 4-2: In the edge point matrix P N In the process, find any point whose intensity value is higher than the intensity threshold T1, and set it as the edge starting point;

[0036] Step 4-3: Search the right neighborhood of the starting point of the edge and make a decision. If the right neighborhood contains an edge point with an intensity value higher than the low threshold T2, find the edge point with the largest intensity value in the right neighborhood and set it as a new starting point to continue searching. Repeat this process until the right neighborhood of the starting point does not contain an edge point with an intensity value higher than the low threshold T2. Extract all starting points to form the right reflection trace of the starting point of the edge.

[0037] Step 4-4: Repeat the process of Step 4-3 for the left neighborhood of the edge starting point in Step 4-2 to obtain the left reflection trace. The right reflection trace and the left reflection trace together form the complete reflection trace of the edge starting point.

[0038] Steps 4-5: Determine the length of the complete reflection trace at the starting point of the edge. If it is less than the preset minimum reflection layer length, discard the complete reflection trace; otherwise, retain the complete reflection trace.

[0039] Step 4-6: Traverse all points with intensity value higher than the intensity high threshold T1 according to the procedures of steps 4-2 to 4-5, extract all complete reflection traces from all edge starting points, and form a structure-continuous Nth-level reflection layer matrix R N .

[0040] According to some embodiments of the present application, in step 5, the reconstruction of the high-frequency resolution matrix is realized according to the following calculation model:

[0041]

[0042] wherein G j is the jth-level high-frequency resolution matrix, H j , V j , and D j are the high-frequency resolution coefficient matrices in the image horizontal direction, vertical direction, and diagonal direction in the jth-level high-frequency resolution coefficient matrix group {H j V j D j}, L r and L c are operators for low-pass filtering the rows and columns of the shallow profile image intensity matrix S0, G r and G c are operators for high-pass filtering the rows and columns of the shallow profile image intensity matrix S0, L*r, L*c, G*r, and G*c are the conjugate transpose matrices of L r , L c , G r , and G c .

[0043] According to some embodiments of the present application, step 5 includes:

[0044] Step 5-1: reconstruct the Nth-level high-frequency resolution matrix G N using the Nth-level high-frequency resolution coefficient matrix group;

[0045] Step 5-2: perform same-dimension addition of the Nth-level reflection layer matrix R N and the Nth-level high-frequency resolution matrix G N to form an N-1th-level noisy reflection layer matrix R' N-1 ;

[0046] Step 5-3: suppress non-maximum points in the N-1th-level noisy reflection layer matrix R' N-1 to realize the refinement of the edges of the N-1th-level noisy reflection layer matrix R' N-1 ;

[0047] Step 5-4: perform edge refinement on the N-1th-level noisy reflection layer matrix R' N-1In the middle, the edge point with the intensity value lower than the low intensity threshold T2 is regarded as noise and discarded, and then two points adjacent to the reflection layer are bridged to obtain the N-1 level reflection layer matrix R containing the N level high frequency resolution information N-1 .

[0048] According to some embodiments of the present application, the shallow profile image intensity matrix S0 is subjected to two-level resolution decomposition, that is, N is 2.

[0049] Based on the shallow profile layer extraction based on multi-level resolution described above, the present application can further obtain a shallow profile layer extractor based on multi-level resolution, which comprises a storage medium storing computer executable instructions, the executable instructions can implement the shallow profile layer extraction method described above when executed.

[0050] According to some embodiments of the present application, the shallow profile layer extractor further comprises a processor for executing the executable instructions.

[0051] The shallow profile layer extraction method based on multi-level resolution can perform multi-level resolution decomposition on the original echo data of the seabed shallow profile, compensate the reflection layer using the high frequency resolution information obtained by decomposition, and finally obtain the shallow profile reflection layer output containing multi-level high resolution detail information.

[0052] The layer extraction method of the present application can track the reflection layer in the shallow profile intensity image obtained from the original shallow profile signal without interruption by using low frequency resolution information and multi-level high frequency resolution information, and accurately and highly resolvedly obtain the layer information of the shallow profile, wherein the reflection layer details are compensated using multi-level high frequency resolution information, which can effectively improve the fineness of the shallow profile layer information, and the double threshold decision method can reduce the false detection probability of the extraction process; and the multi-level resolution decomposition process only needs to use high-pass and low-pass filters with fixed and unchanged filter coefficients, which is easier to implement.

[0053] The extraction method of the present application can automatically extract the layer information of the high resolution shallow profile, and through multi-level resolution decomposition, fusion application and processing of low frequency resolution coefficients and high frequency resolution coefficients, it can accurately and highly resolvedly extract the reflection layer information autonomously under low signal-to-noise ratio, and has high robustness and reliability, and strong data adaptability; at the same time, through reconstruction of the high frequency resolution coefficient, the multi-level resolution local detail feature information can be captured, and the detail information of the sediment layer is more accurately positioned and displayed.

[0054] Further, in some embodiments, the intensity double threshold in the layered extraction method of the present application is obtained automatically, solving the problem that manual setting of single threshold cannot robustly and effectively extract continuous reflection layers. BRIEF DESCRIPTION OF DRAWINGS

[0055] Figure 1 Flow chart of the multi-level resolution based shallow profile layered extraction method described in the embodiments.

[0056] Figure 2 Flow chart of the multi-level resolution decomposition in the embodiments.

[0057] Figure 3 Gray scale image of the obtained second level low frequency resolution coefficient matrix in the embodiments.

[0058] Figure 4 Gray scale image of the obtained second level high frequency resolution coefficient matrix group in the embodiments.

[0059] Figure 5 Smoothed histogram obtained in the embodiments.

[0060] Figure 6 Edge following and connecting result image of the obtained edge point matrix P2 in the embodiments.

[0061] Figure 7 Flow chart of the high frequency resolution coefficient matrix group reconstruction in the embodiments.

[0062] Figure 8 First level shallow profile layered image obtained in the embodiments.

[0063] Figure 9 Comparison of the high resolution shallow profile layered image obtained by processing in the embodiments and the processing result of the traditional method. DETAILED DESCRIPTION

[0064] The present application is described in detail below with reference to the embodiments and drawings, but it should be understood that the embodiments and drawings are only used to exemplarily describe the present application and cannot constitute any limitation on the protection scope of the present application. All reasonable transformations and combinations within the scope of the inventive concept of the present application fall within the protection scope of the present application.

[0065] Referring to the drawings Figure 1 According to the technical scheme of the present application, the multi-level resolution based shallow profile layered extraction method comprises the following steps:

[0066] Step 1: converting the multi-channel shallow stratum echo data received in the shallow profile detection into a shallow profile image intensity matrix S0 in the range of 0-255 gray scale values.

[0067] Further, it can comprise the following implementation steps:

[0068] Step 1-1: Arrange all the single-channel signals received in the shallow profile detection, e1(n), e2(n), …, eM(n) in time sequence to form a raw signal matrix E, E = [e1(n) e2(n) e3(n) … eM(n)], where M is the total number of channels, n ∈ [1, K] represents the sequence number of signal sample points, and K is the total number of sample points of single-channel signal collection. M M (n)], where M is the total number of channels, n ∈ [1, K] represents the sequence number of signal sample points, and K is the total number of sample points of single-channel signal collection.

[0069] Step 1-2: Normalize the obtained raw signal matrix E to obtain a normalized raw signal matrix E a .

[0070] Step 1-3: Map the normalized raw signal matrix E a to the maximum gray value range of 0-255 to obtain a shallow profile image intensity matrix S0.

[0071] Step 2: Perform N-level resolution decomposition (N > 1) on the shallow profile image intensity matrix S0 by low-pass filtering and high-pass filtering to obtain N low-frequency resolution coefficient matrices and N high-frequency resolution coefficient matrix groups, wherein the high-frequency resolution coefficient matrix group is composed of high-frequency resolution coefficient matrices in the horizontal direction, the vertical direction and the diagonal direction of the image, and N can be determined optimally according to the correlation between the current low-frequency resolution coefficient matrix and the shallow profile image intensity matrix.

[0072] Further, in some specific embodiments, the low-frequency resolution coefficient matrix and the high-frequency resolution coefficient matrix group are obtained by the following calculation models, respectively:

[0073] S j = L r L c S j-1

[0074] H j = L r G c S j-1

[0075] V j = G r L c S j-1

[0076] D j = G r G c S j-1

[0077] wherein j represents any decomposition level, j ∈ [1, N], N represents the total decomposition level; S j ​denote the low-frequency resolution coefficient matrix obtained by the jth decomposition (when j = 1, it is the shallow profile image intensity matrix S0); H j j j denote the high-frequency resolution coefficient matrix of horizontal direction, vertical direction and diagonal direction obtained by the jth decomposition, which constitutes the jth high-frequency resolution coefficient matrix group {H j j j}; L r and L c are the low-pass filter operators for the row and column of the shallow profile image intensity matrix S0, G r and G c are the high-pass filter operators for the row and column of the shallow profile image intensity matrix S0, which are conjugate mirror filters and satisfy the following conditions in the frequency domain:

[0078] L r,c (ω)G r,c (ω) + L r,c (ω + π)G r,c (ω + π) = 0

[0079] wherein, L r,c (ω) denote the low-pass filter operators for the row and column of the shallow profile image intensity matrix S0 in the frequency domain, G r,c (ω) denote the high-pass filter operators for the row and column of the shallow profile image intensity matrix S0 in the frequency domain, and ω denotes the angular frequency.

[0080] Step 3: extracting the edge point matrix P N and an intensity high threshold T1 and an intensity low threshold T2 from the Nth low-frequency resolution coefficient matrix S N of the N low-frequency resolution coefficient matrices.

[0081] Further, step 3 can include:

[0082] Step 3-1: performing maximum value extraction on each column data of the low-frequency resolution coefficient matrix S N , and the edge point matrix P N is composed of the extracted maximum values of each column.

[0083] Step 3-2: counting the intensity value data distribution of the edge point matrix P N to obtain an intensity statistical histogram.

[0084] Step 3-3: suppressing the first peak of the intensity statistical histogram, i.e. resetting the value of the first peak, so that the starting point of the intensity statistical histogram starts from zero, is linearly connected to the second peak, and is smoothed to obtain a smoothed histogram.

[0085] ​​​​Step 3-4: Extract the first order statistical features of the smoothed histogram, including its statistical mean x and mean square deviation σ, and obtain the high threshold T1 and low threshold T2 according to the following formulas:

[0086] T1 = x + 2σ

[0087] T2 = x.

[0088] In the above embodiment, the present application can automatically extract two thresholds according to the data distribution as the decision conditions for the next step, so that detection errors rarely occur, effectively improving the continuity and authenticity of the reflection layer.

[0089] Step 4: Perform edge following and connection processing on the edge point matrix P N to obtain a structure-continuous reflection layer matrix R N .

[0090] Further, step 4 can include:

[0091] Step 4-1: In the edge point matrix P N , edge points with intensity values lower than the low threshold T2 are discarded as noise;

[0092] Step 4-2: In the edge point matrix P N , find any point with an intensity value higher than the high threshold T1 and set it as an edge starting point;

[0093] Step 4-3: Search the right neighborhood of the edge starting point and make a decision. If the right neighborhood contains edge points with intensity values higher than the low threshold T2, find the edge point with the maximum intensity value in the right neighborhood and set it as a new starting point to continue searching. Repeat this process until the right neighborhood of the starting point does not contain edge points with intensity values higher than the low threshold T2, and extract all starting points to form the right reflection trace of the edge starting point;

[0094] Step 4-4: Repeat the process of step 4-3 for the left neighborhood of the edge starting point of step 4-2 to obtain the left reflection trace, and the complete reflection trace of the edge starting point is composed of the right reflection trace and the left reflection trace;

[0095] Step 4-5: Determine the length of the complete reflection trace of the edge starting point. If it is less than a pre-set minimum reflection layer length, discard the complete reflection trace, otherwise retain the complete reflection trace;

[0096] Step 4-6: Traverse all points with intensity values higher than the high threshold T1 according to the process of steps 4-2 to 4-5, and extract all complete reflection traces of the edge starting points to form the structure-continuous reflection layer matrix R N .

[0097] Step 5: reconstructing the high-frequency resolution matrix G N V N D N from the N high-frequency resolution coefficient matrix groups {H N Compensate the reflection layer matrix R N , and form the N-1 reflection layer matrix R N-1 containing the N high-frequency resolution information.

[0098] Further, step 5 can include:

[0099] Step 5-1: reconstruct the j high-frequency resolution coefficient matrix group {H j V j D j} by the following calculation model to obtain the j high-frequency resolution matrix G j :

[0100]

[0101] Wherein, L*r, L*c, G*r, G*c are the conjugate transpose matrix of L r , L c , G r , G c .

[0102] Step 5-2: add the reflection layer matrix R N and the N high-frequency resolution matrix G N to form the N-1 noisy reflection layer matrix R' N-1 .

[0103] Step 5-3: suppress the non-maximum points in the N-1 noisy reflection layer matrix R' N-1 in the width range, that is, only one edge point with the maximum intensity value is reserved in the width range of the reflection layer, so as to realize the edge refinement of the noisy reflection layer matrix R' N-1 .

[0104] Step 5-4: in the N-1 noisy reflection layer matrix R' N-1 after edge refinement, the edge point with an intensity value lower than the low threshold T2 is regarded as noise and discarded, and then the two adjacent points of the bridge reflection layer are bridged to obtain the N-1 reflection layer matrix R N-1 containing the N high-frequency resolution information.

[0105] Step 6: according to the process of step 5, the N-1 high-frequency resolution coefficient matrix group {H N-1 V N-1 D N-1} is used to compensate the N-1 reflection layer matrix R N-1, to obtain the N-2 level reflection layer matrix R N-2 , and so on, finally obtaining the reflection layer matrix R0 containing multi-level high frequency resolution information, that is, the target reflection layer, so as to realize shallow profile layering processing with high resolution.

[0106] The application further provides the following embodiments:

[0107] Embodiments

[0108] Referring to the drawings Figure 1 According to the above specific embodiments, the reflection layer signal is extracted in shallow profile detection.

[0109] Wherein:

[0110] The total number of channels M is 500, and the total number of single-channel signal samples K is 2600, that is, the single-channel signals received in detection are e1(n), e2(n), e3(n),..., e 500 (n), the original signal matrix E is [e1(n)e2(n)e3(n)...e 500 (n)], n∈[1,2600].

[0111] The total decomposition level N is set to 2, that is, the 2-level resolution decomposition is performed on the shallow profile image intensity matrix S0 as shown in the drawing Figure 2 , to obtain the 2-level low-frequency resolution coefficient matrix S2 and two high-frequency resolution coefficient matrix groups {H1V1D1} and {H2V2D2}, as follows:

[0112] S j = L r L c S j-1

[0113] H j = L r G c S j-1

[0114] V j = G r L c S j-1

[0115] D j = G r G c S j-1

[0116] Wherein, j∈[1,2].

[0117] Correspondingly, the 2-level low-frequency resolution coefficient grayscale diagram as shown in the drawing Figure 3 is obtained; and the 2-level high-frequency resolution coefficient grayscale diagram as shown in the drawing Figure 4The three directions of the 2nd level high frequency resolution coefficient matrix gray scale diagram are shown in the figure; as attached Figure 5 The smooth histogram and the intensity high threshold T1 and the intensity low threshold T2 are shown in the figure, wherein T1 = 39, T2 = 27; as attached Figure 6 The reflection layer matrix R2 is shown.

[0118] And according to the attached Figure 7 The flow chart specifically reconstructs the 2nd level high frequency resolution coefficient matrix group {H2V2D2} into a high frequency resolution matrix G2, which is added to the reflection layer matrix R2 to obtain a noisy reflection layer matrix R'1, and then the corresponding obtained 1st level reflection layer matrix R1 is shown in the figure as attached Figure 8

[0119] Similarly, the 1st level high frequency resolution coefficient matrix group {H1V1D1} is used to compensate the 1st level reflection layer matrix R1 to obtain a target reflection layer matrix R0 containing 2nd level high frequency resolution information, and high resolution shallow profile layer processing is realized.

[0120] Compared with the traditional single level resolution detection method, the resolution and signal-to-noise ratio of the shallow stratum profile obtained by the multi-level resolution processing of the present application are obviously improved, as shown in the figure as attached Figure 9

[0121] The above embodiments are only preferred embodiments of the present application, and the protection scope of the present application is not limited to the above embodiments. Any technical solution falling within the concept of the present application belongs to the protection scope of the present application. It should be pointed out that, for ordinary skilled in the art, the improvements and refinements without departing from the principles of the present application should also be considered as the protection scope of the present application.​​

Claims

1. A method for shallow profile layer extraction based on multi-level resolution, characterized in that, It comprises: Step 1: converting the multi-channel shallow stratigraphic echo data received in the shallow profile detection into a shallow profile image intensity matrix S0 in the range of 0-255 grayscale values; Step 2: performing N-level resolution decomposition on the shallow profile image intensity matrix S0 through low-pass filtering and high-pass filtering to obtain N low-frequency resolution coefficient matrices and N high-frequency resolution coefficient matrix groups composed of high-frequency resolution coefficient matrices in the horizontal direction, the vertical direction and the diagonal direction of the image; Step 3: extracting its edge point matrix P from the Nth level low frequency resolution coefficient matrix S N of the N low frequency resolution coefficient matrices N and a high intensity threshold T1 and a low intensity threshold T2; Step 4: According to the high intensity threshold T1 and the low intensity threshold T2, the edge point matrix P is processed to obtain a structure-continuous Nth-level reflection layer matrix R N edge following and connection processing to obtain a structure-continuous Nth-level reflection layer matrix R N ; Step 5: reconstructing the Nth level high frequency resolution matrix G from the Nth level high frequency resolution coefficient matrix group in the N high frequency resolution coefficient matrix groups N , using the Nth level high frequency resolution matrix G N , compensating the Nth level reflection layer matrix R N , forming the (N-1)th level reflection layer matrix R N-1 containing the Nth level high frequency resolution information Step 6: According to the process of steps 3-5, the N-1 level high frequency resolution coefficient matrix group is used in turn to compensate the N-1 level reflection layer matrix R N-1 , to obtain the N-2 level reflection layer matrix R N-2 , and finally obtain the reflection layer matrix R0 containing multi-level high frequency resolution information, which is the target reflection layer.

2. The shallow profiling extraction method of claim 1, wherein, The step 1 comprises: Step 1-1: Arrange all the single-channel signals received in the shallow profile detection e1(n), e2(n), …, eM(n) in time sequence to form a raw signal matrix E, E = [e1(n) e2(n) e3(n) … eM(n)], where M is the total number of channels, n ∈ [1, K] represents the sequence number of signal sample points, and K is the total number of sample points of single-channel signal acquisition. M Step 1-2: Calculate the average value of each channel of the raw signal matrix E, and subtract the average value from each channel to obtain the average value-corrected signal matrix E0, E0 = E - Eavg, where Eavg = [e1avg e2avg e3avg … eMavg]T, e1avg, e2avg, …, eMavg are the average values of each channel. M Step 1-3: Calculate the average value of each channel of the average value-corrected signal matrix E0, and subtract the average value from each channel to obtain the average value-corrected signal Step 1-2: Normalization is performed on the original signal matrix E to obtain a normalized original signal matrix E a ; Step 1-3: The normalized original signal matrix E a is mapped to the range of maximum grey value 0-255 to obtain the shallow profile image intensity matrix So.

3. The method of claim 1, wherein, The low-frequency resolution coefficient matrix and the high-frequency resolution coefficient matrix group are respectively obtained through the following calculation models: S j = L r L c S j-1 H j = L r G c S j-1 V j = G r L c S j-1 D j = G r G c S j-1 wherein j represents any decomposition level, j ∈ [1, N], N represents the total decomposition level; S j represents the low-frequency resolution coefficient matrix obtained by the jth level decomposition, when j = 1, it is the shallow profile image intensity matrix S0; H j , V j , D j respectively represent the high-frequency resolution coefficient matrix of the horizontal direction, the vertical direction and the diagonal direction obtained by the jth level decomposition, which constitutes the jth level high-frequency resolution coefficient matrix group {H j V j D j}; L r and L c are respectively the low-pass filter operators for the rows and columns of the shallow profile image intensity matrix S0, G r and G c are respectively the high-pass filter operators for the rows and columns of the shallow profile image intensity matrix S0, both are conjugate mirror filters, and satisfy the following conditions in the frequency domain: L r,c (ω)G r,c (ω)+L r,c (ω+π)G r,c (ω+π)=0 where L r,c (ω) represent the row and column low-pass filter operators, respectively, of the shallow profile image intensity matrix S0in the frequency domain, G r,c (ω) represent the row and column high-pass filter operators, respectively, of the shallow profile image intensity matrix S0in the frequency domain, and ω represents the angular frequency.

4. The method of claim 1, wherein, The step 3 comprises: Step 3-1: maximum value extraction is performed on each column data of the Nth level low frequency resolution coefficient matrix S N , and an edge point matrix P N is composed of the extracted maximum values of each column. Step 3-2: Statistics the intensity value data distribution of the edge point matrix P, and get the intensity statistics histogram. N Step 3-2: Statistics the intensity value data distribution of the edge point matrix P, and get the intensity statistics histogram. Step 3-3: suppressing the first peak of the intensity statistical histogram and performing smoothing processing to obtain a smoothed histogram; Step 3-4: extracting the first-order statistical feature quantity of the smoothed histogram including the statistical mean x and the mean square deviation σ, and obtaining the intensity high threshold T1 and the intensity low threshold T2 according to the following calculation formulae: T1=x+2σ T2=x.

5. The method of claim 1, wherein, The step 4 comprises: Step 4-1: In the edge point matrix P N the edge points with intensity values lower than the intensity low threshold T2 are discarded as noise; Step 4-2: In the edge point matrix P N , find any point whose intensity value is higher than the high intensity threshold T1, and set it as the edge starting point; Step 4-3: searching for the right neighborhood of the edge starting point and making a decision, if the right neighborhood contains an edge point with an intensity value higher than the low threshold T2, finding the edge point with the maximum intensity value in the right neighborhood and setting it as a new starting point for continuous searching; repeating the process until the right neighborhood of the starting point does not contain an edge point with an intensity value higher than the low threshold T2, and extracting all starting points to form the right reflection trace of the edge starting point; Step 4-4: repeating the process of step 4-3 for the left neighborhood of the edge starting point of step 4-2 to obtain the left reflection trace, and the complete reflection trace of the edge starting point is composed of the right reflection trace and the left reflection trace; Step 4-5: determining the length of the complete reflection trace of the edge starting point, if it is less than a preset minimum reflection layer length, discarding the complete reflection trace, otherwise, retaining the complete reflection trace; Step 4-6: Traverse all the points with intensity value higher than intensity high threshold Tl according to the procedures of steps 4-2 to 4-5, and extract all the complete reflection traces from all the edge starting points to form a structure-continuous Nth-level reflection layer matrix R N .

6. The method of claim 1, wherein, In the step 5, the reconstruction of the high-frequency resolution matrix is realized according to the following calculation model: wherein G j is the jth high-frequency resolution matrix, H j , V j , D j is the high-frequency resolution coefficient matrix of the image horizontal direction, vertical direction and diagonal direction in the jth high-frequency resolution coefficient matrix group {H j V j D j}, L r and L c are operators for low-pass filtering the row and column of the shallow profile image intensity matrix S0, G r and G c are operators for high-pass filtering the row and column of the shallow profile image intensity matrix S0, L*r, L*c, G*r, G*c are the conjugate transpose matrices of L r , L c , G r , G c .

7. The method of claim 1, wherein, The step 5 comprises: Step 5-1: reconstructing the Nth level high frequency resolution matrix G with the Nth level high frequency resolution coefficient matrix group N ; Step 5-2: the Nth level reflection layer matrix R N with the Nth level high frequency resolution matrix G N homodimensional addition synthesis, forming the N-1th level noisy reflection layer matrix R' N-1 ; Step 5-3: Inhibiting the N-1st level noise-containing reflection layer matrix R' N-1 non-maximum points in a wide range of widths, implementing a noise-containing reflection layer matrix R' N-1 edge thinning; Step 5-4: N-1 level reflection layer matrix R' after edge refinement N-1 In the method, the edge point with intensity value lower than the intensity low threshold T2 is regarded as noise and discarded, and then two points adjacent to the reflection layer are bridged to obtain the N-1 level reflection layer matrix R containing the N level high frequency resolution information N-1 .

8. The method of claim 1, wherein, The shallow profile image intensity matrix S0 is decomposed in two levels of resolution, i.e. the value of N is 2.

9. A shallow profile layer extractor based on multi-level resolution, comprising a storage medium storing computer executable instructions, the executable instructions being executable to implement the shallow profile layer extraction method according to any one of claims 1-4 when executed.

10. The shallow profiling extractor of claim 9, wherein, It further comprises a processor for executing the executable instructions.

Images