CN107065006B - A kind of seismic signal coding method based on online dictionary updating - Google Patents

Abstract

A kind of seismic signal coding method based on online dictionary updating, belong to seismic signal data coding and transmission method, it solves to use dictionary learning and rarefaction representation bring dictionary transmission problem in seismic signal coding, can be applied in the various ground end mappings based on seismic signal measurement.The present invention, which includes: (1), will input seismic signal and is divided into multiple groups sequentially in time, carries out sparse coding using the dictionary in caching to every group of data and calculates sparse coefficient；(2) quantization and entropy coding are carried out to the sparse coefficient in step (1)；(3) the reconstruction data that P a transmission group in front is read from caching carry out dictionary learning in conjunction with the sparse coefficient of current group transmission, thus dictionary needed for updating next group of Sparse expression.The present invention under the precondition for guaranteeing the effective rarefaction representation of signal, is not needed real-time Transmission dictionary information, to effectively reduce data transfer rate amount, can be adapted for various seismic signal high speed acquisitions application by way of online dictionary updating.

Description

Seismic signal coding method based on online dictionary updating

Technical Field

The invention belongs to a seismic signal data transmission method, and particularly relates to a seismic signal lossy coding method.

Background

The mapping technology based on seismic signal measurement is one of the effective methods for measuring the ground structure and mineral resources at present. In each mapping, the seismic signal measurement of the ground will generate data exceeding 100T, and the bandwidth of the signal transmission is very limited, so it is necessary to reduce the data volume of the seismic signal by the seismic signal coding technology before the transmission. In the prior art, a seismic signal coding method based on discrete cosine transform is proposed, which can obtain a compression multiple close to 3 times. And a two-dimensional discrete cosine transform technology based on local seismic signal self-adaptation is also adopted, so that important characteristics of the reconstructed seismic signals can be stored. Furthermore, the seismic signal coding technology adopting the adaptive wavelet packet can obtain higher compression multiple and better reconstruction quality, and is widely applied to the feature extraction of the seismic signal at present due to the better direction keeping characteristic. The main idea of the above method is to use a suitable basis or redundant dictionary to characterize the seismic signal so that the characterization of the signal is sparse. In recent years, the sparse representation through dictionary learning is widely concerned, and particularly, the sparse representation is widely applied to the remote sensing image coding to learn the dictionary through a double sparse model, so that a good coding effect is obtained. These findings all demonstrate the feasibility of applying dictionary learning and sparse representation in seismic signal coding.

The traditional encoding method based on dictionary learning and sparse representation often includes the following two main methods: (1) and performing sparse representation on the online data acquired in real time through the offline learning dictionary. For the method, an off-line training set is required to exist in advance, and the required dictionary information is obtained through the off-line training set. Therefore, whether sparse representation of online data is effective depends heavily on the correlation of offline data and online data. For actual seismic signal measurements, it is difficult to obtain a common off-line training set that is suitable for different situations. (2) And training a dictionary by using the real-time online data, and performing sparse representation on the real-time online data through the dictionary. In this method, it is necessary to transmit the dictionary, so that the sparse representation process at the codec end can be synchronized. Therefore, the transmission of dictionary information increases the size of the encoded code stream, thereby degrading encoding performance.

The invention content is as follows:

the invention provides a seismic signal coding method based on online dictionary updating, belongs to a seismic signal data coding transmission method, solves the problem of how to transmit a dictionary due to dictionary learning and sparse representation in seismic signal coding, and can be applied to various seismic signal measurement-based ground mapping.

The invention relates to a seismic signal coding method based on online dictionary updating, which comprises a coding step and a decoding step; wherein,

the encoding step includes:

step 1, dividing input seismic signals into a plurality of groups according to a time sequence, and carrying out sparse coding on each group of data by adopting a dictionary in a cache, wherein the method specifically comprises the following steps:

step 11, dividing the seismic signal data of the near T traces into a group, and processing each group of data independently; assume that the current set of data is Z set of data, denoted as Y_z(ii) a Equally dividing the data of each trace into a plurality of units, each unit y_iHas a length of M.times.1, and y_iSorting according to a column mode; thus, Y_z＝[y₁,...y_i,...y_N](ii) a Assuming that the length of the data recorded on each track is U, there is the following relation: t × U ═ M × N;

step 12, reading dictionary D in cache_z-1Given a sparse coefficient matrix W_ZIs L, the following formula is optimized and solved:

step 2, quantizing and entropy coding the sparse coefficient in step S1, specifically including:

step 21, quantizing the sparse coefficient matrix by using a uniform quantization method, which specifically comprises the following steps:

w_Z(i, j) represents a sparse coefficient matrix W_ZThe value of the coefficient with the middle coordinate (i, j), delta represents the quantization step,represents the quantization result of the coefficient value of (i, j), round (·) represents the rounding operation;

step 22, creating a non-zero coefficient position matrix PT composed of a value 0 and a value 1, wherein the creating method comprises the following steps:

wherein abs (·) represents an absolute value operation;

step 23, adopting arithmetic coding for the nonzero coefficient position matrix PT;

step 24, for non-zero systemNumber (corresponding to PT (i, j) ═ 1 position)Huffman coding is adopted;

step 3, reading the reconstructed data of the previous P transmitted groups from the cache, and performing dictionary learning by combining the sparse coefficients transmitted by the current group, so as to update the dictionary required by sparse representation of the next group of data, specifically comprising:

step 31, calculating P +1 group reconstruction datap∈[Z-P,Z](the current group data is Z group), the calculation method is as follows:

wherein,unit of

Step 32, solving the required dictionary D according to the following optimization process_Z：

Wherein,a_irepresenting constants that describe the inter-group correlation, equation (2) is solved iteratively by:

step 321, fix D_ZW' can be calculated by the PS method described previously;

step 322, fixing W', D_ZThe updating can be performed according to an MOD method:

step 323, repeating the above steps 321 and 322 to the specified iteration times, and updating the required dictionary D_Z；

The decoding step includes:

step 4, carrying out inverse quantization and entropy decoding on the received sparse coefficient to generate a non-zero coefficient matrix W'_ZThe method comprises the following steps:

step 41, Huffman decoding is carried out on the non-zero coefficient coding code stream to obtain a non-zero coefficient w_c；

Step 42, for non-zero coefficient w_cCarrying out inverse quantization to obtain an inverse quantization coefficient w'_cThe method comprises the following steps:

w'_c＝w_c×Δ

step 43, performing arithmetic decoding on the non-zero coefficient position matrix PT coded stream to obtain a non-zero coefficient position matrix PT, and combining the dequantized coefficient w 'generated in step 42'_cGenerating a non-zero coefficient matrix W'_Z；

And 5, reconstructing the seismic signals, which specifically comprises the following steps:

step 51, reading dictionary D in cache_z-1Generating a reconstructed signal Y'_zThe method comprises the following steps:

Y'_z＝D_z-1×W'_Z

step 52, to reconstructed signal Y'_z＝[y'₁...y'_i...y'_N](y 'per unit'_iLength of M x 1) are rearranged, and a plurality of adjacent units are arranged according toIn a row, the tracks are joined end to end, so that each track has a length ofT traces are shared in total;

step 6, generating a dictionary D in the cache_zFor reconstruction of the next set of data, namely: reading the reconstructed data of the previous P transmitted groups from the cache, and performing dictionary learning by combining the sparse coefficients transmitted by the current group, so as to update the dictionary required by sparse representation of the next group of data, specifically comprising:

step 61, calculating P +1 group reconstruction datap∈[Z-P,Z](the current group data is Z group), the calculation method is as follows:

wherein,unit of

Step 62, solving the required dictionary D according to the following optimization process_Z：

Wherein,a_irepresenting constants that describe the inter-group correlation, equation (2) is solved iteratively by:

step 621, fix D_ZW' can be calculated by the PS method described previously;

step 622, fixing W', D_ZThe updating can be performed according to an MOD method:

step 623, repeating the steps 321 and 322 to the specified iteration times, and updating the required dictionary D_Z。

According to the invention, by means of online dictionary updating, dictionary information does not need to be transmitted in real time under the premise of ensuring effective sparse representation of signals, so that data transmission data volume is effectively reduced, and the method can be suitable for various seismic signal high-speed acquisition application occasions.

Drawings

FIG. 1 is a flow chart of the present invention;

FIG. 2 is a portion of signals in test seismic signal data;

FIG. 3 is a dictionary for learning;

FIG. 4 shows the results of comparing the performance of different methods.

Detailed Description

Example (b):

the invention mainly comprises the following steps:

s1: dividing input seismic signals into a plurality of groups according to a time sequence, and carrying out sparse coding on each group of data by adopting a dictionary in a cache;

s2, quantizing and entropy coding the sparse coefficient in the step S1;

s3: and reading the reconstructed data of the previous P transmitted groups from the cache, and performing dictionary learning by combining the sparse coefficients transmitted by the current group, thereby updating the dictionary required by sparse representation of the next group of data.

Further, step S1 is specifically:

s11: dividing the seismic signal data of T adjacent traces into a group, and processing each group of data independently; assume that the current set of data is Z set of data, denoted as Y_z. Equally dividing the data of each trace into a plurality of units, each unit y_iHas a length of M.times.1, and y_iThe sorting is performed in a columnar manner. Thus, Y_z＝[y₁,...y_i,...y_N]. Assuming that the length of the data recorded on each track is U, there is the following relation: t × U ═ M × N.

S12: reading dictionary D in cache_z-1Given a sparse coefficient matrix W_ZIs L, the following formula is optimized and solved:

for the solution of equation (1), we intend to use the PS method ("Partial search vector selection for search signal representation," in NORSIG-03). The PS method is based on the OMP algorithm (orthogonal matching pursuit, "computer of basis selection methods," in Signals, Systems and Computers,1996.Conference Record of the third inertia orthogonal Conference on), and therefore, the flow of the OMP algorithm is given first:

the PS method only modifies the searching process of the maximum correlation dictionary unit in the step (1) into the searching process of a plurality of maximum correlation dictionary units, thereby providing more searching decisions and obtaining better sparse vectors.

Step S2 specifically includes:

s21: the sparse coefficient matrix is quantized by adopting a uniform quantization method, which specifically comprises the following steps:

w_Z(i, j) represents a sparse coefficient matrix W_ZThe value of the coefficient with the middle coordinate (i, j), delta represents the quantization step,represents the quantization result of the coefficient value of (i, j), and round (·) represents the rounding operation.

S22: creating a non-zero coefficient matrix PT composed of a value 0 and a value 1, wherein the creating method comprises the following steps:

where abs (·) represents an absolute value operation.

S23: arithmetic coding is applied to the non-zero coefficient matrix PT.

S24: for non-zero coefficients (corresponding to positions PT (i, j) ═ 1 w^r _i,j) Huffman coding is used.

Step S3 specifically includes:

s31: computing P +1 set of reconstructed datap∈[Z-P,Z](the current group data is Z group), the calculation method is as follows:

wherein,unit of

S32, solving the required dictionary D according to the following optimization process_Z：

Wherein,a_irepresents a constant describing inter-group correlation.

Equation (2) can be solved iteratively by the following steps:

1) fixed D_ZW' can be calculated by the PS method described previously;

2) fixing W', D_ZUpdates may be made according to MOD methods ("Method of Optimal orientations for FrameDesign," in 1999 IEEE International Conference on Acoustics, Speech and SignalProcessing (ICASSP)):

3) repeating the steps 1) and 2) to the specified iteration times to generate the required updated dictionary D_Z。

Example 1:

1. the test seismic signal data come from a UTAM image database (http:// UTAM. gg. utah. edu/SeismiData. html), the Find-tracked-minsers data are selected as test data, the test data comprise 72 sensors, and each sensor comprises 135 traces;

2. each trace takes 1600 time length samples, the data of each 10 traces is 1 group, and partial test data is shown in fig. 1;

1. assuming that the current group is the 3 rd group (the first 2 groups of data are all encoded and the related data are output to the decoding end and the buffer), reading the dictionary D in the buffer₂Sparse coding is carried out, the size of the dictionary in the cache is 16 multiplied by 64 in the embodiment, the sparsity is 1/16 (the proportion of non-zero coefficients to the overall coefficients), and W obtained by calculation is₃Part of the data is as follows;

0	0	0	0	0	0	0	0
								0	0	0	0	0	191086.69	0	0
0	0	0	0	0	0	0	69516.63
								0	0	0	0	0	0	0	0
0	0	0	0	0	0	0	0
								0	0	263371.03	0	0	0	0	0
-275961.58	248225.21	0	0	0	0	0	0
								0	0	0	0	0	0	0	0
0	0	0	0	0	0	0	0

2. to W₃Quantization is carried out, and the quantization step length is 1024; huffman coding is adopted for non-zero data after quantization, arithmetic coding is adopted for non-zero element positions, and the difference between an original signal and a reconstructed signal is measured through an SNR (signal to noise ratio), wherein the SNR is 21.2 dB.

3. By reading the first two sets of reconstructed data and W₃Performing dictionary update, the updated dictionary D₃As shown in fig. 3.

Example 2:

1. dictionary D updated with third group data for fourth group data₃Performing sparse representation and updating dictionary D₄；

2. Dictionary D updated with fourth group data for fifth group data₄Performing sparse representation and updating dictionary D₅；

3. Quantizing and coding each group of data, calculating code rate and measuring distortion through SNR;

4. and adjusting different sparsity, and repeating the 3 steps to obtain the distortion conditions under different code rates.

5. In order to prove the effectiveness of the algorithm, the seismic signal coding algorithm based on DCT, Curvelet and an off-line dictionary learning method (K-SVD + ORMP) is simultaneously tested to obtain the rate distortion conditions of different algorithms, and the related comparison test result is shown in FIG. 4.

The specific embodiments described in the art are merely illustrative of the spirit of the invention. Various modifications or additions may be made to the described embodiments or alternatives may be employed by those skilled in the art without departing from the spirit or ambit of the invention as defined in the appended claims.

Claims (1)

1. A seismic signal coding method based on online dictionary updating is characterized by comprising a coding step and a decoding step; wherein,

the encoding step includes:

step 1, dividing input seismic signals into a plurality of groups according to a time sequence, and carrying out sparse coding on each group of data by adopting a dictionary in a cache, wherein the method specifically comprises the following steps:

step 11, dividing the seismic signal data of the near T traces into a group, and processing each group of data independently; assume that the current set of data is Z set of data, denoted as Y_z(ii) a Equally dividing the data of each trace into a plurality of units, each unit y_iHas a length of M.times.1, and y_iSorting according to a column mode; thus, Y_z＝[y₁,...y_i,...y_N](ii) a Assuming that the length of the data recorded on each track is U, there is the following relation: t × U ═ M × N;

step 12, reading dictionary D in cache_z-1Given a sparse coefficient matrix W_ZIs L, the following formula is optimized and solved:

step 2, quantizing and entropy coding the sparse coefficient in step S1, specifically including:

step 21, quantizing the sparse coefficient matrix by using a uniform quantization method, which specifically comprises the following steps:

w_Z(i, j) represents a sparse coefficient matrix W_ZThe value of the coefficient with the middle coordinate (i, j), delta represents the quantization step,represents the quantization result of the coefficient value of (i, j), round (·) represents the rounding operation;

step 22, creating a non-zero coefficient position matrix PT composed of a value 0 and a value 1, wherein the creating method comprises the following steps:

wherein abs (·) represents an absolute value operation;

step 23, adopting arithmetic coding for the nonzero coefficient position matrix PT;

step 24, Huffman coding is adopted for the nonzero coefficient,with non-zero coefficients corresponding to positions PT (i, j) ═ 1

Step 3, reading the reconstructed data of the previous P transmitted groups from the cache, and performing dictionary learning by combining the sparse coefficients transmitted by the current group, so as to update the dictionary required by sparse representation of the next group of data, specifically comprising:

step 31, calculating P +1 group reconstruction datap∈[Z-P,Z]And the current group data is Z group, and the calculation method is as follows:

wherein,unit of

Step 32, solving the required dictionary D according to the following optimization process_Z：

Wherein,a_irepresenting constants that describe the inter-group correlation, equation (2) is solved iteratively by:

step 321, fix D_ZW' is calculated by the PS method;

step 322, fixing W', D_ZThe updating can be performed according to an MOD method:

step 323, repeating the above steps 321 and 322 to the specified iteration times, and updating the required dictionary D_Z；

The decoding step includes:

step 4, carrying out inverse quantization and entropy decoding on the received sparse coefficient to generate a non-zero coefficient matrix W'_ZThe method comprises the following steps:

step 41, Huffman decoding is carried out on the non-zero coefficient coding code stream to obtain a non-zero coefficient w_c；

Step 42, for non-zero coefficient w_cCarrying out inverse quantization to obtain an inverse quantization coefficient w'_cThe method comprises the following steps:

w'_c＝w_c×Δ

step 43, performing arithmetic decoding on the non-zero coefficient position matrix PT coded stream to obtain a non-zero coefficient position matrix PT, and combining the dequantized coefficient w 'generated in step 42'_cGenerating a non-zero coefficient matrix W'_Z；

And 5, reconstructing the seismic signals, which specifically comprises the following steps:

step 51, reading dictionary D in cache_z-1Generating a reconstructed signal Y'_zThe method comprises the following steps:

Y'_z＝D_z-1×W'_Z

step 52, to reconstructed signal Y'_z＝[y'₁...y'_i...y'_N]Are rearranged, each unit y'_iIs Mx 1, adjacent units are spliced together end to end in a column manner to form a trace, so that the length of each trace isT traces are shared in total;

step 6, generating a dictionary D in the cache_zFor reconstruction of the next set of data, namely: reading the reconstructed data of the previous P transmitted groups from the buffer, and combining the sparse coefficient transmitted by the current group to carry out word processingPerforming dictionary learning so as to update a dictionary required by a next group of data sparse representation, specifically comprising:

step 61, calculating P +1 group reconstruction datap∈[Z-P,Z]And the current group data is Z group, and the calculation method is as follows:

wherein,unit of

Step 62, solving the required dictionary D according to the following optimization process_Z：

Wherein,a_irepresenting constants that describe the inter-group correlation, equation (2) is solved iteratively by:

step 621, fix D_ZW' is calculated by the PS method;

step 622, fixing W', D_ZThe updating can be performed according to an MOD method:

step 623, repeating the steps 321 and 322 to the specified iteration times, and updating the required dictionary D_Z。