CN102129074B - Satellite SAR original data anti-saturation vector compression coding and decoding method - Google Patents

Abstract

The invention discloses a satellite SAR original data anti-saturation vector compression coding and decoding method. The method comprises the following steps of: dividing a cell by a definition domain of input data according to a cell division nearest neighbor rule when generating a codebook; modifying an original cell mass centre equation of a vector quantizer by introducing an analog-to-digital converter (ADC) saturation threshold when the mass center of the cell is solved according to the input data by combining a probability density function; processing to obtain a corresponding anti-saturation codebook according to an anti-saturation vector quantization codebook generation process; when coding, deblocking data; obtaining an ADC input signal standard deviation of each sub-block according to the mapping relationship between an ADC output signal amplitude average value and the ADC input signal standard deviation; normalizing each sub-block; coding by selecting a corresponding vector quantization codebook through the amplitude average value of a sampling signal; when decoding, searching the corresponding vector quantization codebook according to the amplitude average value corresponding to each sub-block; and decoding according to the mapping relationship between the sub-block amplitude average value and the ADC input signal standard deviation.

Description

Satellite-borne SAR raw data antisaturation vector compression coding and coding/decoding method

Technical field

The present invention relates to a kind of satellite-borne SAR raw data antisaturation quantization encoding and coding/decoding method, be applied to satellite-borne SAR compressing original data field.

Background technology

The satellite-borne SAR compressing original data is to solve the effective way that data transfer bandwidth on mass data that satellite-borne SAR obtains and the star is difficult to matching problem.

The conventional vector quantization method passes through training set data, iterate and calculate to mate the statistical property of raw data, when ADC (Analog-to-Digital Converter) is saturated, because the truncation effect of input signal, the vector statistics characteristic of raw data is no longer obeyed Joint Gaussian distribution, greatly reduce the correlativity of Codebook of Vector Quantization and data to be compressed, cause Codebook of Vector Quantization and data statistics characteristic mismatch to be compressed.This will cause the compressing original data algorithm performance obviously to descend, and then worsen image radiation resolution, reduce picture quality, finally affect the use in actual applications of SAR raw data.Therefore, must solve the antisaturation problem in the practicality of satellite-borne SAR compressing original data algorithm.

Summary of the invention

The object of the present invention is to provide a kind of antisaturation vector compression coding and coding/decoding method, to solve the problem of satellite-borne SAR compressing original data algorithm hydraulic performance decline when ADC is saturated.

For achieving the above object, the technical solution used in the present invention is: when generating Codebook of Vector Quantization, do not re-use conventional needle to the quantification code book of multivariate joint probability Gaussian distribution, but construct new antisaturation Codebook of Vector Quantization.

Main operational steps comprises:

1) vector normalization: will input the raw data piecemeal, calculate the amplitude equalizing value of data in the sub-block, utilize the mapping relations of ADC amplitude output signal average and ADC input signal standard deviation, amplitude equalizing value is mapped as ADC input signal standard deviation, scalar data in the sub-block is converted to vector data, and use ADC input signal standard deviation, the vector data in the sub-block is carried out normalization;

2) compressed encoding: will go normalized data to be combined with ADC amplitude output signal average, and use the antisaturation code book, and carry out quantization encoding, the quantization encoding result is exported with the sub-block amplitude equalizing value;

The design procedure of described antisaturation code book comprises:

21) saturated data occur for ADC, introduce saturation threshold, according to the division result of saturation threshold to cell, judge the saturability of each cell;

22) use original barycenter solution formula for unsaturated cell, for saturated cell occurring, use revised cell barycenter solution formula;

23) step 21 and step 22 are carried out loop iteration, the condition that satisfied circulation stops to stop circulation, and obtaining saturation threshold is the antisaturation code book of M;

3) decoding: use antisaturation code book and sub-block amplitude equalizing value, with the result of compressed encoding processings of decoding, acquisition decoded result;

4) vector goes normalization: the mapping relations of utilizing ADC amplitude output signal average and ADC input signal standard deviation, amplitude equalizing value is mapped as ADC input signal standard deviation, and use ADC input signal standard deviation decoded vector is gone normalized, then corresponding vector is converted to scalar and gets final product.

In described vector compression coding and the coding/decoding method, the normalization step comprises:

11) with after the raw data piecemeal, the data in the piecemeal are carried out scalar to the conversion of vector;

12) ADC output amplitude average is mapped as ADC input signal standard deviation, uses ADC input signal standard deviation to carry out normalized the vector after the conversion.

In described vector compression coding and the coding/decoding method, use the antisaturation code book during vector quantization coding, on hyperspace take distortion function as cost, for saturated Joint Gaussian distribution Design of Signal, its design process is similar to conventional vector quantization code the design algorithm, and difference is for saturated cell structure antisaturation Codebook of Vector Quantization.

In described vector compression coding and the coding/decoding method, decoding step comprises:

31) judge according to data amplitude average corresponding to coded data place sub-block, determine the saturation threshold that this sub-block is corresponding;

32) determine the antisaturation code book that decoding will be used according to saturation threshold, to the coding codeword processing of decoding.

In described vector compression coding and the coding/decoding method, vector goes the normalization step to comprise:

41) the ADC output amplitude average that coded data place sub-block is corresponding is mapped as ADC input signal standard deviation, decoded data be multiply by the ADC input signal standard deviation of place piecemeal;

42) corresponding vector data is converted to scalar data.

The invention has the beneficial effects as follows, obviously improved the performance of Vector Quantization algorithm when ADC is saturated, obtained having the Codebook of Vector Quantization of antisaturation performance, the present invention has expanded the range of application of vector quantization in satellite-borne SAR compressing original data field, for the reliability application of vector quantization in satellite-borne SAR compressing original data field provides theoretical direction.

Description of drawings

Shown in Figure 1 is antisaturation vector compression coding of the present invention and decoding process.

It is the conventional two-dimensional Gaussian probability-density function shown in Fig. 2-1.

It is the saturated dimensional Gaussian probability density function of the present invention shown in Fig. 2-2.

That conventional vector quantizes two-dimentional cell division shown in Fig. 3-1.

That antisaturation two dimension cell of the present invention is divided shown in Fig. 3-2.

Shown in Figure 4 is antisaturation Codebook of Vector Quantization design cycle of the present invention.

Fig. 5 is ADC amplitude output signal average and ADC input signal standard deviation mapping curve.

Fig. 6 is ADC amplitude output signal average and signal saturation degree mapping curve.

Embodiment

Technical scheme of the present invention is: when generating code book, field of definition by the input data, according to the arest neighbors criterion that cell is divided cell is divided, when finding the solution the barycenter of cell according to input data aggregate probability density function, introduce ADC saturation threshold M, cell barycenter equation to original vector quantizer is revised, and then antisaturation Codebook of Vector Quantization product process according to the present invention is processed, and tries to achieve corresponding antisaturation code book.For specific ADC, according to the relation between the saturation degree of ADC amplitude output signal average and data, generate corresponding code book for different saturation degrees.

When coding, data are carried out piecemeal, to each sub-block, according to the mapping relations of ADC amplitude output signal average and ADC input signal standard deviation, try to achieve ADC input signal standard deviation, sub-block is carried out normalization.Then select corresponding Codebook of Vector Quantization to encode by the amplitude equalizing value of sampled signal, during decoding, the amplitude equalizing value corresponding according to each sub-block searched corresponding Codebook of Vector Quantization, and according to the mapping relations of sub-block amplitude equalizing value and ADC input signal standard deviation, decode.

The present invention is further described below in conjunction with drawings and Examples.

Fig. 1 is antisaturation vector compression coding and decoding process, and wherein dotted portion is the part that the present invention is different from conventional vector compressed encoding and decoding.As shown in Figure 1, during coding, be different from traditional approach, the present invention uses the antisaturation code book to replace traditional Codebook of Vector Quantization, and the same antisaturation code book that uses replaces traditional Codebook of Vector Quantization during decoding.

Fig. 2-1 is the conventional two-dimensional Gaussian probability-density function, and Fig. 2-2 is saturated dimensional Gaussian probability density function, because ADC's is saturated, causes data to be blocked, and the vector statistics characteristic of raw data no longer satisfies Joint Gaussian distribution.

The design of antisaturation Codebook of Vector Quantization is the improvement to conventional vector quantization code the design, by introducing saturation threshold, cell division, the barycenter equation of code book design is made amendment, and improves the performance of Vector Quantization algorithm when ADC is saturated.

The input vector x of definition vector quantizer and the distortion measure that quantizes between the output Q (x) are:

$D=E\left[d(x,Q\left(x\right))\right]=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}E\left[d(x,y_i)\right]=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{\∫}_{R_i}{|x-y_i|}^{2}p\left(x\right)\mathrm{dx}---\left(1\right)$

X=[x wherein ₁, x ₂... x _k] ^T, y _i=[y _I1, y _I2... y _Ik] ^T, R _iBe i division of correspondence, p (x) is the k dimension priori joint probability density function of input vector x, and N is code book length.Code book Y=[y to be generated ₁, y ₂... y _N] ^T

For making D reach minimum, the cell of code book design is divided to be needed to satisfy:

R _i={ x:||x-y _i|| ²≤ || x-y _j|| ², j ≠ i} and $\underset{i=1}{\overset{N}{\∪}}{R}_i=R$

Finding the solution of barycenter needs to satisfy:

$y_{\mathrm{ij}}=\frac{{\∫}_{R_i}{x}_{j}p\left(x\right)\mathrm{dx}}{{\∫}_{R_i}p\left(x\right)\mathrm{dx}}=\frac{1}{\mathrm{\Δ}{P}_i}{\∫}_{R_i}{x}_{j}p\left(x\right)\mathrm{dx},i=\mathrm{1,2},...N;j=\mathrm{1,2},...k---\left(3\right)$

Use formula (2), (3) to carry out circular treatment and can try to achieve code book corresponding to vector quantization.

When data occur when saturated, the cell of conventional vector quantization algorithm is divided, the barycenter solving equation is no longer applicable, needs to introduce saturation threshold and revises.

Among Fig. 3, Fig. 3-1 divides schematic diagram for conventional vector quantizes two-dimentional cell, and code book length is 16 among the figure, and the code vector dimension is 2 dimensions, and whole field of definition is divided into 16 cells.Fig. 3-2 is that antisaturation two dimension cell of the present invention is divided schematic diagram.Wherein the dotted line among Fig. 3-2 represents to introduce saturation threshold M, saturation threshold is nine regional I-IX with whole field of definition interval division, except regional IX, all the other zones are the zone of saturation, and wherein zone of saturation I-VIII is respectively corresponding different saturated: II, IV, VI, VIII area data only wherein one dimension occur saturated; Every one dimension of I, III, V, VII area data all occurs saturated.From Fig. 3-2 as seen, cell R ₁-R ₈All in regional IX, be unsaturation cell, cell R ₉-R ₁₆The saturated of different situations all appears in each cell, belongs to saturated cell.

After the ADC appearance was saturated, the joint probability density function of input data did not change in non-saturated area inside, and changes at the zone of saturation joint probability density function, needs to revise the barycenter equation of code book design.

Owing to data truncation occurring, saturation threshold is divided into two parts with saturated cell, and a part is the saturated part of cell, and a part is cell unsaturation part, introduces saturation threshold M formula (3) is rewritten as:

$y_{\mathrm{ij}}=\frac{{\∫}_{R_{\mathrm{ins}}}{x}_{j}p\left(x\right)\mathrm{dx}+{\∫}_{R_{\mathrm{is}}}{x}_{j}p\left(x\right)\mathrm{dx}}{{\∫}_{R_i}p\left(x\right)\mathrm{dx}}---\left(4\right)$

Consider truncation effect, need to revise for saturated cell occurring, introduce

$M_1=\left\{\begin{array}{cc}M& x_j>M,\\ x_j& -M<x_j\&le;M,j=\mathrm{1,2},...,k\\ -M& x_j\&le;-M\end{array}\right.---\left(5\right)$

Formula (4) can be rewritten as:

$y_{\mathrm{ij}}=\frac{{\&Integral;}_{R_{\mathrm{ins}}}{x}_{j}p\left(x\right)\mathrm{dx}+{\&Integral;}_{R_{\mathrm{is}}}{M}_{1}p\left(x\right)\mathrm{dx}}{{\&Integral;}_{R_i}p\left(x\right)\mathrm{dx}}---\left(6\right)$

First of formula (6) represents respectively the saturated part of cell and the unsaturation part of cell with second.

Divide according to antisaturation Codebook of Vector Quantization design cycle shown in Figure 4 according to revised barycenter solution formula (6) and the definite cell of formula (2), can try to achieve saturation threshold is the antisaturation code book of M.

The below illustrates that as an example of satellite-borne SAR antisaturation two-dimensional vector quantization code the design example the cell of antisaturation Vector Quantization algorithm is divided and barycenter is found the solution.

Given initial codebook Y ₀, code book length is 16, the code vector dimension is 2, Y ₀=

[-0.6461，0.8687；0.4270，0.8238；0.8661，0.2085；0.6360，-0.8514；-0.3865，-0.8254；-0.8371，-0.2111；-0.1668，0.3468；0.1906，-0.2162；-1.5088，1.5645；0.2608，1.8113；1.6949，1.5762；1.7991，0.0495；1.4083，-1.6072；-0.2897，-1.8239；-1.6868，-1.5658；-1.7726，-0.0482]；

The joint probability density function of input data is:

$f(x_1,x_2)=\frac{1}{2\mathrm{\&pi;}}\exp \{-\frac{1}{2}(x_1^2+x_2^2)\}$

In computing, in order to reduce operand, when calculating, constant term 1/2 π in the joint probability density function is ignored, do not affect code book result of calculation.The joint probability density function graph of a correspondence is shown in Fig. 2-1.Corresponding cell is divided shown in Fig. 3-1.

When ADC occurs when saturated, joint probability density is no longer obeyed Joint Gaussian distribution, introduces saturation threshold M, and saturated joint probability density distributes shown in Fig. 2-2, according to cell division methods and the saturation threshold M of formula (2), try to achieve corresponding cell and divide shown in Fig. 3-2.

Find the solution the barycenter operation:

The tradition unsaturation is divided shown in Fig. 3-1, tries to achieve barycenter Y according to unsaturation cell barycenter solution formula (3) ₁For:

[-0.6888，0.8937；0.4542，0.8767；0.9061，0.1951；0.6708，-0.8673；-0.4039，-0.8662；-0.8630，-0.1964；-0.1775，0.3330；0.1814，-0.2280；-1.4589，1.5193；0.1748，1.8390；1.5717，1.4239；1.8209，-0.0270；1.3876，-1.5541；-0.2059，-1.8419；-1.5676，-1.4068；-1.7949，0.0105]；

Saturation threshold is that the cell of M is divided shown in Fig. 3-2, and to unsaturation cell use formula (3) wherein, saturated cell uses formula (6) to try to achieve barycenter Y _1SFor:

[0.6888,0.8937; 0.4542,0.8767; 0.9061,0.1951; 0.6708 ,-0.8673;-0.4039 ,-0.8662;-0.8630 ,-0.1964;-0.1775,0.3330; 0.1814 ,-0.2280;-1.4150,1.4639; 0.1748,1.7291; 1.5159,1.3864; 1.7168 ,-0.0271; 1.3517 ,-1.4957;-0.2059 ,-1.7318;-1.5118 ,-1.3719;-1.6972,0.0105]; Cell R ₁-R ₈So owing to saturated Y do not occur ₁And Y _1SThe barycenter numerical value of trying to achieve is consistent, for cell R ₉-R ₁₆, saturated owing to having occurred, at Y _1SMiddle use correction barycenter solution formula (6) is found the solution, at Y ₁In still use traditional barycenter solution formula (3) to find the solution.

ADC occurs when saturated, with above-mentioned cell divide, the cell barycenter finds the solution operation, carries out loop iteration according to saturated Codebook of Vector Quantization design cycle shown in Figure 4, can try to achieve the antisaturation code book under the corresponding saturation threshold.Can obtain after the same method the antisaturation code book in whole saturation degree.

ADC (8-bit) amplitude output signal average and ADC input signal standard deviation Nonlinear Mapping relation shown in formula (7),

$\left|\stackrel{\&OverBar;}{I}\right|=\left|\stackrel{\&OverBar;}{Q}\right|=127.5-\underset{n=0}{\overset{126}{\mathrm{\&Sigma;}}}\mathrm{erf}\left(\frac{n+1}{\sqrt{2}\mathrm{\&sigma;}}\right)---\left(7\right)$

Wherein,

σ is ADC input signal standard deviation.Mapping curve is seen Fig. 5.

The continuous random variable saturation degree is defined as:

$\mathrm{SD}=2{\&Integral;}_M^{\&infin;}f\left(x\right)\mathrm{dx}---\left(8\right)$

Wherein, f (x) is the probability density function of input signal, and M is saturation threshold.

Can be in the hope of the relation of the implicit function between ADC amplitude output signal average and the signal saturation degree by formula (7) and (8), its relation curve is seen Fig. 6.In the situation that ADC amplitude output signal average is definite, can obtain ADC input signal standard deviation and corresponding input data saturation by mapping curve.

When quantization encoding, according to the scrambler flow process shown in the accompanying drawing 1 with deblocking, try to achieve the signal amplitude average in each sub-block, according to the mapping relations of ADC amplitude output signal average and ADC input signal standard deviation, the data of each sub-block are carried out normalization.Then according to the relation of ADC amplitude output signal average and saturation threshold (for definite ADC, saturation threshold is corresponding one by one with saturation degree), select corresponding code book to encode.

During decoding, demoder according to Fig. 1, mapping relations according to ADC amplitude output signal average and saturation threshold, select corresponding code book, according to code vector corresponding to call number output, and then according to the mapping relations of ADC amplitude output signal average and ADC input signal standard deviation, the code vector that obtains is gone normalized, obtain reconstruction signal.

Being respectively 20.75 and 49.96 two-dimensional vector data [40,50] with ADC amplitude output signal average is example, explanation antisaturation vector compression coding and decoding, and concrete steps are as follows:

The vector quantization coding process:

The block data amplitude equalizing value is 20.75 according to the mapping relations curve shown in the formula (7), is mapped as input signal standard deviation 26, data [40,50] is used 26 be normalized to [40,50]/26=[1.5385,1.9231]; According to the relation of amplitude equalizing value with code and thresholding, the coding code book of its correspondence is Y ₁, according to the arest neighbors criterion, it is quantified as code vector [1.5717,1.4239], the call number of correspondence is encoded to binary coding Code_A=1011, will pass to ground under the coding of correspondence and the amplitude equalizing value ASM_A=20.75.

The block data amplitude equalizing value is 49.96 according to the mapping relations curve shown in the formula (7), is mapped as input signal standard deviation 64, data [40,50] is used 64 be normalized to [40,50]/64=[0.6250,0.7813]; According to the relation of amplitude equalizing value with code and thresholding, the coding code book of its correspondence is Y _1S, according to the arest neighbors criterion, it is quantified as code vector [0.4542,0.8767], the call number of correspondence is encoded to binary coding Code_B=0010, will pass to ground under the coding of correspondence and the amplitude equalizing value ASM_B=49.96.

The vector quantization decoder process:

For Code_A=1011, ASM_A=20.75; During decoding, should select code book Y when decoding according to amplitude equalizing value ₁, search obtains exporting code vector [1.5717,1.4239] through call number, then according to formula (7) amplitude equalizing value 20.75 is mapped as input signal standard deviation 26, multiplies each other with the output code vector again, obtain net result [1.5717,1.4239] * 26=[40.8642,37.0214];

For Code_B=0010, during the ASM_B=49.96 decoding, should select code book Y when decoding according to amplitude equalizing value _1S, search obtains exporting code vector [0.4542,0.8767] through call number, then according to formula (7) amplitude equalizing value 49.96 is mapped as input signal standard deviation 64, multiplies each other with the output code vector again, obtain net result [0.4542,0.8767] * 64=[29.0688,56.1088];

The above used embodiment is not that the present invention is done any pro forma restriction, and the related amendments that every foundation technical spirit of the present invention is carried out all still belongs in the present invention program's the scope.

Claims (4)

1. a satellite-borne SAR raw data antisaturation vector compression coding and coding/decoding method, key step comprises:

1) vector normalization: will input the raw data piecemeal, calculate the amplitude equalizing value of data in the sub-block, utilize the mapping relations of ADC amplitude output signal average and ADC input signal standard deviation, amplitude equalizing value is mapped as ADC input signal standard deviation, scalar data in the sub-block is converted to vector data, and use ADC input signal standard deviation, the vector data in the sub-block is carried out normalization;

2) compressed encoding: normalized data are combined with ADC amplitude output signal average, use the antisaturation code book, carry out quantization encoding, the quantization encoding result is exported with the sub-block amplitude equalizing value;

The design procedure of described antisaturation code book comprises:

21) saturated data occur for ADC, introduce saturation threshold, according to the division result of saturation threshold to cell, judge the saturability of each cell;

22) use original barycenter solution formula for unsaturated cell, for saturated cell occurring, use the revised cell barycenter solution formula that is shown below

$y_{\mathrm{ij}}=\frac{{\&Integral;}_{R_{\mathrm{ins}}}{x}_{j}p\left(x\right)\mathrm{dx}+{\&Integral;}_{R_{\mathrm{is}}}{M}_{1}p\left(x\right)\mathrm{dx}}{{\&Integral;}_{R_i}p\left(x\right)\mathrm{dx}};$

Y in the formula _IjBe the j dimension value of barycenter corresponding to i cell, x _jWherein one dimension for input k n dimensional vector n x; R _iFor i cell of correspondence divided R _IsBe cell R _iSaturated part, R _InsBe cell R _iThe unsaturation part, M ₁Be revised saturation threshold, p (x) is the k dimension priori joint probability density function of input vector x;

23) step 21 and step 22 are carried out loop iteration, the condition that satisfied circulation stops to stop circulation, and obtaining saturation threshold is the antisaturation code book of M;

3) decoding: use antisaturation code book and sub-block amplitude equalizing value, with the result of compressed encoding processings of decoding, acquisition decoded result;

4) vector goes normalization: the mapping relations of utilizing ADC amplitude output signal average and ADC input signal standard deviation, amplitude equalizing value is mapped as ADC input signal standard deviation, and use ADC input signal standard deviation decoded vector is gone normalized, then corresponding vector is converted to scalar and gets final product.

2. vector compression coding according to claim 1 and coding/decoding method is characterized in that, described normalization step comprises:

11) with after the raw data piecemeal, the data in the piecemeal are carried out scalar to the conversion of vector;

12) ADC output amplitude average is mapped as ADC input signal standard deviation, uses ADC input signal standard deviation to carry out normalized the vector after the conversion.

3. vector compression coding according to claim 1 and coding/decoding method is characterized in that, its described decoding step comprises:

31) judge according to data amplitude average corresponding to coded data place sub-block, determine the saturation threshold that this sub-block is corresponding;

32) determine the antisaturation code book that decoding will be used according to saturation threshold, to the coding codeword processing of decoding.

4. vector compression coding according to claim 1 and coding/decoding method is characterized in that, described vector goes the normalization step to comprise:

41) the ADC output amplitude average that coded data place sub-block is corresponding is mapped as ADC input signal standard deviation, decoded data be multiply by the ADC input signal standard deviation of place piecemeal;

42) corresponding vector data is converted to scalar data.

Images