CN103926575A - Laser radar total wave signal analyzing method - Google Patents

Abstract

The invention provides a laser radar total wave signal analyzing method. According to the method, when updating is carried out every time, one of the parameters including the number k of echo signals, noise signals B incident to a laser radar detector, the peak position t0 of each echo signal and the peak amplitude beta of each echo signal is selected at random so that updating can continue to be carried out; the parameter k is updated based on the Metropolis algorithm thought, the parameter B is updated based on the Metropolis algorithm or the simulation tempering algorithm thought, and the parameter t0 and the parameter beta are respectively updated based on the simulation tempering algorithm thought. According to the laser radar total wave signal analyzing method, total wave signals of targets with unknown characters can be processed quickly without being affected by the signal noise ratio of the total wave signals.

Description

A kind of laser radar Full wave shape signal analysis method

Technical field

The invention belongs to laser radar technique field, be specifically related to a kind of laser radar Full wave shape signal analysis method.

Background technology

Traditional laser radar can only extract a distance feature in laser facula conventionally, cannot process the echoed signal of a plurality of distance feature targets in hot spot.Employing has the intactly Full wave shape laser radar of record reception semiotic function can obtain this Full wave shape signal that comprises a plurality of echoed signals, for Full wave shape signal analysis provides the foundation, so-called Full wave shape signal is produced jointly by the echoed signal of all characteristic distances in the laser facula marking.In order effectively to resolve the distance feature comprising in the single laser facula marking, need to adopt peak, quantity, peak amplitude, noise that Full wave shape signal analysis technology extracts the echoed signal wherein comprising totally 4 crucial Full wave shape parameters, be labeled as respectively t ₀, k, β, B.Wherein, echo quantity and peak are that Full wave shape parameter is of paramount importance two parameters, and for quantity and the mutual alignment relation of marking path feature and complete Full wave shape signal decomposition, peak is the photon flight time on accurate distance.

Full wave shape signal y can regard the posteriority sampling results that receives signal F as, and the energy distribution form with energy on time shaft represents.The object of Full wave shape signal analysis utilizes Full wave shape signal y to solve Full wave shape parameter t exactly ₀, k, β, B, its process can be regarded as utilizes posterior information to estimate prior imformation.Wherein, posterior information is Full wave shape signal y, and corresponding prior imformation is Full wave shape parameter t ₀, k, β, B.Conventionally can adopt posteriority distribution probability density function to evaluate the order of accuarcy of Full wave shape Estimation of Parameters value.

Typical Full wave shape analysis strategy mainly comprises non-linear least square waveform fitting method and the employing EM algorithm that adopts LM algorithm and the Full wave shape parameter maximum likelihood value method of estimation of improving algorithm thereof.The former needs the valid data model of precognition target in advance, is only applicable to the scene of single characteristic, such as: vegetation, building etc.; The latter's constringency performance depends on the default of initial value strongly, cannot realize no manual intervention's Full wave shape parameter completely and extract, and sometimes has the risk losing efficacy.

Recently, a kind of Reversible Jump based on Bayesian Estimation framework Er Kefu Monte Carlo algorithm (RJMCMC) of spurring the horse on (refers to IEEE Transactions on Pattern Analysis and Machine Intelligence2007,29 (12), 2170-2180) be suggested as a kind of Full wave shape analysis strategy.In this strategy, its crucial part be adopt RJMCMC algorithm according to fixing order with circulation form Full wave shape parameter t to be separated one by one ₀, k, β and B, and by all overall approximate solution (t that formed by parameter approximate solution of posteriority distribution probability density function assessment ₀, k, β, B), select the most suitable overall approximate solution (t ₀, k, β, B) as globally optimal solution.The method allows Full wave shape parameter to be separated in renewal process between the default distributed model of different scale, to have jumped the detection of distribution space is not relied on to prior imformation, thereby realizes the Full wave shape signal analysis to target property the unknown.But the method mainly depends on effectively default distribution to the quick detection of distribution space, and the transition probability of each parameter is conventionally lower, thereby causes slower arithmetic speed.

Summary of the invention

The present invention proposes a kind of laser radar Full wave shape signal analysis method, Full wave shape signal that can fast processing characteristic unknown object, and be not subject to the impact of Full wave shape Signal-to-Noise.

In order to solve the problems of the technologies described above, the invention provides a kind of laser radar Full wave shape signal analysis method, comprise the following steps:

Step 1, a laser radar Full wave shape signal y of input, parameter clearly to be asked is: t ₀, k, β, B, wherein:

K is the quantity of echoed signal, and its span is: { 0,1,2Kk _max, k _maxit is the maximal value of default k;

B is the noise signal that incides laser radar detection device, and its span is: (0, max (y)), and max (y) is the maximal value of Full wave shape signal y;

T ₀for the k dimensional vector of the accurate flight time formation of the photon in each echoed signal, the peak of corresponding k the echoed signal of its element, t ₀the span of all elements be: (0, i _max), i _maxthe length of Full wave shape signal y;

β is the k dimensional vector that the peak amplitude of each echoed signal forms, the peak amplitude of corresponding k the echoed signal of its element, and the span of all elements of β is: (0, max (y));

From choose in the span of parameter to be asked respectively arbitrary value respectively assignment to parameter t ₀, k, β and B, complete the respectively initialization of parameter to be asked; In completing parameter initialized to be asked, choose any one parameter, with selected parameter, start once to upgrade operation;

Step 2, need to be asked parameter to form an overall approximate solution (t institute upgrading after operation ₀, k, β, B), at each overall approximate solution (t forming after upgrading operation ₀, k, β, B) in random any one parameter of selecting proceed to upgrade operation; Repeat this step, until stop upgrading operation after upgrading the maximal value that the summation of number of operations reaches predefined renewal number of operations;

The all overall approximate solution obtaining after step 3, renewal operation is each time substitution posteriority distribution probability density function respectively, obtain the posteriority distribution probability value of each overall approximate solution, will be worth overall approximate solution corresponding to maximum posteriority distribution probability value as globally optimal solution; Described posteriority distribution probability density function is as shown in Equation (1):

$\mathrm{\π}(t_0,k,\mathrm{\β},B|y)=\left(\underset{i=1}{\overset{i_{\max }}{\mathrm{\Π}}}\exp (-(\underset{j=1}{\overset{k}{\mathrm{\Σ}}}f(i,{\mathrm{\β}}_j,t_{0j})+B))\frac{{(\underset{j=1}{\overset{k}{\mathrm{\Σ}}}f(i,{\mathrm{\β}}_j,t_{0j})+B)}^{y_i}}{y_i!}\right)$

$\×\frac{1}{k_{\max }}{\left(\frac{1}{i_{\max }}\right)}^k{f}_G\left(B\right|c,d)\underset{n=1}{\overset{k}{\mathrm{\Π}}}{f}_G\left({\mathrm{\β}}_n\right|a,b)---\left(1\right)$

In formula (1), π (t ₀, k, β, B|y) and be posteriority distribution probability value; f _gfor Gamma distribution probability density function, a and c are f _gyardstick parameter, b and d are f _gshape parameter, and a, b, c, d are default positive number; y _ithe discrete representation of y on time i; F (i, β _j, t _0j) be the energy distribution of j echoed signal on time i, j is not more than k _maxnonnegative integer, i gets 0 conventionally to i _maxbetween integer, t _0jand β _jbe respectively accurate flight time of photon of j echoed signal and peak amplitude, be also t ₀with j the element of β, f (i, β _j, t _0j) function expression as shown in Equation (2):

$f(i,{\mathrm{\&beta;}}_j,t_{0j})={\mathrm{\&beta;}}_j\left\{\begin{array}{cc}{e}^{-{(t_{1j}-t_{0j})}^2/{2\mathrm{\&sigma;}}^2}{e}^{(i-t_{1j})/{\mathrm{\&tau;}}_1},& i<t_{1j}\\ e^{-{(i-t_{0j})}^2/{2\mathrm{\&sigma;}}^2},& t_{1j}\&le;i<t_{2j}\\ e^{-{(t_{2j}-t_{0j})}^2/{2\mathrm{\&sigma;}}^2}{e}^{-(i-t_{2j})/{\mathrm{\&tau;}}_2},& t_{2j}\&le;i<t_{3j}\\ e^{-{(t_{2j}-t_{0j})}^2/{2\mathrm{\&sigma;}}^2}{e}^{-(t_{3j}-t_{2j})/{\mathrm{\&tau;}}_2}{e}^{-(i-t_{3j})/{\mathrm{\&tau;}}_3},& i\&GreaterEqual;t_{3j}\end{array}\right.---\left(2\right)$

In formula (2), t _0j, t _1j, t _2j, t _3jbetween there is stable difference magnitude relationship, i.e. t _0j-t _1j=c ₁, t _0j-t _2j=c ₂, t _0j-t _3j=c ₃, c wherein ₁, c ₂, c ₃for default constant; τ ₁, τ ₂, τ ₃be default constant with σ, and 0 < τ ₁< τ ₂< τ ₃, σ=0.

Compared with prior art, its remarkable advantage is in the present invention: (1) can select parameter to carry out parameter renewal at random, has reduced to a certain extent operand, has promoted the speed of solving; (2) peak, peak amplitude, three parameters of noise have increased initiatively tempering step in upgrading, realized the interim expansion of approximate solution space, realized the quick renewal of parameter, improved the detection efficiency of approximate solution, reached the quick object of upgrading of Full wave shape parameter approximate solution to be separated and kept robustness; (3) parameter of different nature is taked different update algorithm, has effectively reduced operand, has improved travelling speed.

Accompanying drawing explanation

Fig. 1 is a kind of laser radar Full wave shape of the present invention signal analysis method process flow diagram.

Fig. 2 is that to test the detection of a target in be frosted glass and carton in the present invention, frosted glass and the parallel placement of carton different spacing, obtain the signal decomposition figure of the Full wave shape signal of its seven different spacing, wherein, Fig. 2 (a) is that spacing is the signal decomposition figure of the Full wave shape signal of 5cm, Fig. 2 (b) is that spacing is the signal decomposition figure of the Full wave shape signal of 8cm, Fig. 2 (c) is that spacing is the signal decomposition figure of the Full wave shape signal of 10cm, Fig. 2 (d) is that spacing is the signal decomposition figure of the Full wave shape signal of 20cm, Fig. 2 (e) is that spacing is the signal decomposition figure of the Full wave shape signal of 30cm, Fig. 2 (f) is that spacing is the signal decomposition figure of the Full wave shape signal of 50cm, Fig. 2 (g) is that spacing is the signal decomposition figure of the Full wave shape signal of 100cm.

Fig. 3 is the signal decomposition figure of the present invention's low signal-to-noise ratio Full wave shape signal of acquisition in experiment two, the detection of a target is frosted glass and carton, frosted glass and the parallel placement of carton different spacing, wherein, Fig. 3 (a) is that spacing is that signal decomposition figure, Fig. 3 (b) of the low signal-to-noise ratio Full wave shape signal of 10cm is that spacing is the signal decomposition figure of the low signal-to-noise ratio Full wave shape signal of 20cm.

The distribution of Fig. 4 target that to be the present invention rebuild in experiment three in distance, the detection of a target is frosted glass, N shape letter and the carton that equidistant parallel is placed, and has reached the level of centimetre-sized.

embodiment

As shown in Figure 1, laser radar Full wave shape signal analysis method of the present invention, comprises the following steps:

Step 1, a laser radar Full wave shape signal y of input, parameter clearly to be asked is: t ₀, k, β, B, wherein:

K is the quantity of echoed signal, and its span is: { 0,1,2Kk _max, k _maxbe the maximal value of default k, according to the echo quantity in the target Full wave shape signal of understanding in advance, can set suitable k _maxvalue, to reduce calculated amount;

B is the noise signal that incides laser radar detection device, and its span is: (0, max (y)), and max (y) is the maximal value of Full wave shape signal y;

T ₀for the k dimensional vector of the accurate flight time formation of the photon in each echoed signal, the peak of corresponding k the echoed signal of its element, t ₀the span of all elements be: (0, i _max), i _maxthe length of Full wave shape signal y;

β is the k dimensional vector that the peak amplitude of each echoed signal forms, the peak amplitude of corresponding k the echoed signal of its element, and the span of all elements of β is: (0, max (y));

From choose in the span of parameter to be asked respectively arbitrary value respectively assignment to parameter t ₀, k, β and B, complete the respectively initialization of parameter to be asked; In completing parameter initialized to be asked, choose any one parameter, with selected parameter, start once to upgrade operation, according to different update method of operating corresponding to different parameters of choosing, carry out this and upgrade operation, other not selected parameter participates in this and upgrades operation, and all parameters are likely updated also and likely remain unchanged after once upgrading operation;

Step 2, need to be asked parameter to form an overall approximate solution (t institute upgrading after operation ₀, k, β, B), at each overall approximate solution (t forming after upgrading operation ₀, k, β, B) in random any one parameter of selecting proceed to upgrade operation; Repeat this step, until stop upgrading operation after upgrading the maximal value that the summation of number of operations reaches predefined renewal number of operations;

The all overall approximate solution obtaining after step 3, renewal operation is each time substitution posteriority distribution probability density function respectively, obtain the posteriority distribution probability value of each overall approximate solution, will be worth overall approximate solution corresponding to maximum posteriority distribution probability value as globally optimal solution; Described posteriority distribution probability density function is as shown in Equation (1):

$\mathrm{\&pi;}(t_0,k,\mathrm{\&beta;},B|y)=\left(\underset{i=1}{\overset{i_{\max }}{\mathrm{\&Pi;}}}\exp (-(\underset{j=1}{\overset{k}{\mathrm{\&Sigma;}}}f(i,{\mathrm{\&beta;}}_j,t_{0j})+B))\frac{{(\underset{j=1}{\overset{k}{\mathrm{\&Sigma;}}}f(i,{\mathrm{\&beta;}}_j,t_{0j})+B)}^{y_i}}{y_i!}\right)$

$\&times;\frac{1}{k_{\max }}{\left(\frac{1}{i_{\max }}\right)}^k{f}_G\left(B\right|c,d)\underset{n=1}{\overset{k}{\mathrm{\&Pi;}}}{f}_G\left({\mathrm{\&beta;}}_n\right|a,b)---\left(1\right)$

In formula (1), π (t ₀, k, β, B|y) and be posteriority distribution probability value; f _gfor Gamma distribution probability density function, a and c are f _gyardstick parameter, b and d are f _gshape parameter, and a, b, c, d are default positive number; y _ithe discrete representation of y on time i; F (i, β _j, t _0j) be the energy distribution of j echoed signal on time i, j is not more than k _maxnonnegative integer, i gets 0 conventionally to i _maxbetween integer, t _0jand β _jbe respectively accurate flight time of photon of j echoed signal and peak amplitude, be also t ₀with j the element of β, f (i, β _j, t _0j) function representation can be referring to document (S.Pellegrini, G.Buller, J.Smith, A.Wallace, S.Cova.Laser-Based Distance Measurement Using Picosecond Resolution TCSPC, Measurement Science and Technology, vol.11, pp.712-716,2000), f (i, β _j, t _0j) concrete expression as shown in Equation (2):

$f(i,{\mathrm{\&beta;}}_j,t_{0j})={\mathrm{\&beta;}}_j\left\{\begin{array}{cc}{e}^{-{(t_{1j}-t_{0j})}^2/{2\mathrm{\&sigma;}}^2}{e}^{(i-t_{1j})/{\mathrm{\&tau;}}_1},& i<t_{1j}\\ e^{-{(i-t_{0j})}^2/{2\mathrm{\&sigma;}}^2},& t_{1j}\&le;i<t_{2j}\\ e^{-{(t_{2j}-t_{0j})}^2/{2\mathrm{\&sigma;}}^2}{e}^{-(i-t_{2j})/{\mathrm{\&tau;}}_2},& t_{2j}\&le;i<t_{3j}\\ e^{-{(t_{2j}-t_{0j})}^2/{2\mathrm{\&sigma;}}^2}{e}^{-(t_{3j}-t_{2j})/{\mathrm{\&tau;}}_2}{e}^{-(i-t_{3j})/{\mathrm{\&tau;}}_3},& i\&GreaterEqual;t_{3j}\end{array}\right.---\left(2\right)$

In formula (2), t _0j, t _1j, t _2j, t _3jbetween there is stable difference magnitude relationship, i.e. t _0j-t _1j=c ₁, t _0j-t _2j=c ₂, t _0j-t _3j=c ₃, c wherein ₁, c ₂, c ₃in calculating process, are all constants, need to before processing Full wave shape signal y, arrange; τ ₁, τ ₂, τ ₃be default constant with σ, and 0 < τ ₁< τ ₂< τ ₃, σ=0, need to before processing Full wave shape signal y, arrange.In actual applications, t _0j, t _1j, t _2j, t _3j, τ ₁, τ ₂, τ ₃can affect to a certain extent arithmetic speed and the computing accuracy of the inventive method with σ, so need to adjust t according to the waveform of the Full wave shape signal being formed by single echo _0j, t _1j, t _2j, t _3j, τ ₁, τ ₂, τ ₃to adapt to the echo waveform of various shapes, these parameters all can arrange before processing Full wave shape signal y with σ.

As previously mentioned, need to be at parameter t before each renewal operation ₀, select a parameter to start this to upgrade operation, the different update method of operating that different parameters are corresponding at random arbitrarily between k, β and B.

Choose at random the update method after parameter k:

Parameter k is a special case in four parameters to be asked, and k, through once upgrading after operation, likely reaches state k ', also likely reaches state k ' ', k '=k+1 wherein, and k ' '=k-1(k is present in { 0,1,2Kk _max, k is parameter t simultaneously ₀with the dimension of β, if k upgrades, parameter t ₀must upgrade with β, so k adopts the mode of slower gradual change to reduce calculated amount.）。When once the selected parameter of renewal operation is k, the state that can further select at random k may reach after this upgrades operation in k ' and this two states of k ' ', then according to the random result of selecting (choosing the probability of k ' or k ' ' to equate), according to k ' and k ' ' respectively corresponding different update method carry out this and upgrade and operate.In addition also can be by k ' and k ' ' and parameter t ₀, β and B put together and select this method of upgrading operation, at k ', k ' ', t ₀, select any one parameter in β and B, the update method that this parameter is corresponding is upgraded to the method for operation as this, wherein, choose the probability of k ' or k ' ' to equate.

No matter selection mode k ' or state k ' ', k all adopts Metropolis algorithm idea to upgrade, but concrete steps are different.

If the state that the random k selecting may reach after this upgrades operation is k ' time, this computation process of upgrading behaviour is:

(1) if k=k _max, each parameter t ₀, k, β and B remain unchanged, and jumps out this and upgrade operation, reselect parameter and upgrade operation next time; Otherwise continue step (2);

(2), establish t ₀' expression t ₀the state that may arrive after this upgrades operation, and t ₀' be k ' dimensional vector, t ₀' front k element be followed successively by t ₀front k element;

If the state that β ' expression β may arrive after this upgrades operation, and β ' is k ' dimensional vector, front k the element of β ' is followed successively by t ₀front k element;

At interval (0, i _max) in select any one to obey equally distributed random number assignment to t ₀' the individual element of k ', in interval (0, max (y)), select any one to obey equally distributed random number assignment to the individual element of k ' of β ';

(3), by y, t ₀solve according to y, t with β substitution formula (2) ₀energy distribution { f (i, the β of the echoed signal of estimating with β _j, t _0j), j is any nonnegative integer that is not more than k, β _jand t _0jrespectively vectorial β and t ₀j element; By y, t ₀' and β ' substitution formula (2) solve according to y, t ₀' and energy distribution { f (i, the β of the echoed signal estimated of β ' _m', t _0m'), m is any nonnegative integer that is not more than k ', β _m' and t _0m' be respectively vectorial β ' and t ₀' m element;

Then by { f (i, β _j, t _0j), k and B substitution formula (3) solve signal F (i, β, the t that incides laser radar detection device ₀, k, B), by { f (i, β _m', t _0m'), k ' and B substitution formula (3) solve the signal F (i, β ', the t that incide laser radar detection device estimating ₀', k ', B);

Again by F (i, β, t ₀, k, B) and y substitution formula (4) solve Full wave shape signal y at signal F (i, β, t ₀, k, B) joint distribution probability L (y|t ₀, k, β, B), by F (i, β ', t ₀', k ', B) solves Full wave shape signal y at signal F (i, β ', t with y substitution formula (4) ₀', k ', joint distribution probability L (y|t B) ₀', k ', β ', B);

Finally by L (y|t ₀, k, β, B), L (y|t ₀', k ', β ', B) substitution formula (5) solve k ' transition probability α (k ', k), shown in formula (3), (4), (5) are specific as follows:

$F(i,\mathrm{\&beta;},t_0,k,B)=\underset{j=1}{\overset{k}{\mathrm{\&Sigma;}}}f(i,{\mathrm{\&beta;}}_j,t_{0j})+B---\left(3\right)$

$L\left(y\right|t_0,k,\mathrm{\&beta;},B)=\underset{i=1}{\overset{i_{\max }}{\mathrm{\&Pi;}}}\exp (-F\left(i\right))\frac{F{\left(i\right)}^{y_i}}{y_i!}---\left(4\right)$

In formula (4), y ₁, y ₂k for y discrete representation in time,

$\mathrm{\&alpha;}(k^{\&prime;},k)=L\left(y\right|t_0^{\&prime;},k^{\&prime;},{\mathrm{\&beta;}}^{\&prime;},B)-L\left(y\right|t_0,k,\mathrm{\&beta;},B)---\left(5\right)$

(4) if α (k ', k) > 0, by t ₀', β ' and k ' difference assignment be to t ₀, β and k, parameter B remains unchanged, and completes this and upgrades operation, by t ₀, the new overall approximate solution (t that forms of k, β and B ₀, k, β, B) in select any parameter to upgrade operation at random next time;

(5) if α (k ', k)≤0, further whether judgement now arrives the maximum update times N that attempts _k(setting in advance), wherein step (2) to the process of step (4) is once to attempt upgrading; If arrive the maximum update times N that attempts _k, each parameter t ₀, k, β and B remain unchanged, and jumps out this and upgrade operation, reselect parameter and upgrade operation next time; If do not arrive the maximum update times N that attempts _k, jump to step (2) and continue this renewal operation.

When if the state that the random k selecting may reach after this upgrades operation is k ' ', this computation process of upgrading behaviour is:

(1) if k=0, each parameter t ₀, k, β and B remain unchanged, and jumps out this and upgrade operation, reselect parameter and upgrade operation next time; Otherwise, continue step (2);

(2), suppose it is this time attempt to upgrade (step (2) to step (4) is once to attempt upgrading) for the l time, establish t ₀', β ' is respectively that this upgrades t after operation ₀, the β state that may be updated to, and t ₀', β ' is k ' ' dimensional vector, will reject vectorial t after l element ₀distinguish successively in order assignment to vectorial t with the surplus element of β ₀' and the individual element of k ' ' of β ';

(3), by y, t ₀solve according to y, t with β substitution formula (2) ₀energy distribution { f (i, the β of the echoed signal of estimating with β _j, t _0j), j is any nonnegative integer that is not more than k, β _jand t _0jrespectively vectorial β and t ₀j element; By y, t ₀' and β ' substitution formula (2) solve according to y, t ₀' and energy distribution { f (i, the β of the echoed signal estimated of β ' _m', t _0m'), m is any nonnegative integer that is not more than k ' ', β _m' and t _0m' be respectively vectorial β ' and t ₀' m element;

Then by { f (i, β _j, t _0j), k and B substitution formula (3) solve according to { f (i, β _j, t _0j), k and B the signal F (i, β, the t that incide laser radar detection device that estimate ₀, k, B), by { f (i, β _j', t _0j'), k ' ' and B substitution formula (3) solve according to { f (i, β _m', t _0m'), k ' ' and B the signal F (i, β ', the t that incide laser radar detection device that estimate ₀', k ' ', B);

Again by F (i, β, t ₀, k, B) and y substitution formula (4) solve y at signal F (i, β, t ₀, k, B) joint distribution probability L (y|t ₀, k, β, B), by F (i, β ', t ₀', k ' ', B) solves y at signal F (i, β ', t with y substitution formula (4) ₀', k ' ', joint distribution probability L (y|t B) ₀', k ' ', β ', B);

Finally by L (y|t ₀, k, β, B), L (y|t ₀', k ' ', β ', B) substitution formula (6) solve k ' ' transition probability α (k ' ', k);

$\mathrm{\&alpha;}(k^{\&prime;\&prime;},k)=L\left(y\right|t_0^{\&prime;},k^{\&prime;\&prime;},{\mathrm{\&beta;}}^{\&prime;},B)-L\left(y\right|t_0,k,\mathrm{\&beta;},B)---\left(6\right)$

(4) if α (k ' ', k) > 0, by t ₀', the value of β ' and k ' ' respectively assignment to t ₀, β and k, parameter B remains unchanged, and completes this and upgrades operation, by t ₀, the new overall approximate solution (t that forms of k, β and B ₀, k, β, B) in select any parameter to upgrade operation at random next time;

(5) if α (k ' ', k)≤0, further judges whether to arrive the maximum update times k that attempts, if arrive maximum update times k, each parameter t of attempting ₀, k, β and B remain unchanged, and jumps out this and upgrade operation, reselect parameter and upgrade operation next time; If do not arrive the maximum update times k that attempts, jump to step (2) and continue this renewal operation.

Choose at random the update method after parameter B:

If at parameter t ₀, it is random arbitrarily between k, β and B that select is parameter B, by parameter B, start this and upgrade operation, the renewal operation of parameter B can take two kinds of diverse ways to upgrade, and these two kinds of diverse ways are respectively based on Metropolis algorithm and simulated tempering algorithm

Renewal operating process based on Metropolis algorithm is:

(1), establishing B ' is the state that B may be updated to, interval (0, max (y)) select a random number assignment of obeying Gamma distribution to B ', the yardstick parameter of Gamma distribution can not be excessive or too small, excessive or too small meeting caused slow arithmetic speed, the yardstick parameter selection 1.5 of Gamma distribution rule of thumb, the shape parameter of Gamma distribution is chosen as B value;

(2), first by y, t ₀solve according to y, t with β substitution formula (2) ₀energy distribution { f (i, the β of the echoed signal of estimating with β _j, t _0j), j is any nonnegative integer that is not more than k, β _jand t _0jβ and t ₀j element;

Then by { f (i, β _j, t _0j), k and B substitution formula (3) solve according to { f (i, β _j, t _0j), k and B the signal F (i, β, the t that incide laser radar detection device that estimate ₀, k, B), by { f (i, β _j, t _0j), k and B ' substitution formula (3) solve according to { f (i, β _j, t _0j), k and B ' the signal F (i, β, the t that incide laser radar detection device that estimate ₀, k, B ');

Again by F (i, β, t ₀, k, B) and y substitution formula (4) solve y at signal F (i, β, t ₀, k, B) joint distribution probability L (y|t ₀, k, β, B), by F (i, β, t ₀, k, B ') and y substitution formula (4) solve y at signal F (i, β, t ₀, k, B ') joint distribution probability L (y|t ₀, k, β, B ');

Finally by L (y|t ₀, k, β, B), L (y|t ₀, k, β, B ') substitution formula (7) solve B ' transition probability α (B ', B);

α(B′,B)＝L(y|t ₀,k,β,B′)-L(y|t ₀,k,β,B) （7）

(3) if α (B ', B) > 0, by B ' assignment to B, other parameter t ₀, k, β remain unchanged, complete this and upgrade operation, by t ₀, the new overall approximate solution (t that forms of k, β and B ₀, k, β, B) in select any parameter to upgrade operation at random next time;

(4) if α (B ', B)≤0, further judges whether to arrive the maximum update times N that attempts _b1(setting in advance), wherein step (1) to step (3) is once to attempt upgrading, if reach the maximum update times N that attempts _b1, all parameter t ₀, k, β and B remain unchanged, and jumps out this and upgrade operation, reselect parameter and upgrade operation next time; If do not arrive maximum update times N _b1, jump to step (1) and continue this renewal operation.

Renewal operating process based on simulated tempering algorithm is:

(1), according to the requirement of simulated tempering algorithm, set function T=g (T ₀, u), 1 < T < T _max, and T meets formula (8) and (9) simultaneously;

T′＞0 （8）

(T′) ²-2T(T′) ²+T′′(T) ²≤0 （9）

In formula (8) and (9), T ' and T ' ' are respectively T about derived function and the second derived function of u, and this function of initialization is T=T ₀(T ₀> 1);

(2), establish T _v=g (T ₀, v), v is the arbitrary value in the scope of u, is arranged on T _v=g (T ₀, the maximum under v) is attempted update times N _b2(setting in advance);

(3), establishing B ' is the state that B may be updated to, interval (0, max (y)) select a random number assignment of obeying Gamma distribution to B ', the yardstick parameter of Gamma distribution can not be excessive or too small, excessive or too small meeting causes slower arithmetic speed, the yardstick parameter selection 1.5 of Gamma distribution rule of thumb, the shape parameter of Gamma distribution is chosen as B value;

(4), first by y, t ₀solve according to y, t with β substitution formula (2) ₀energy distribution { f (i, the β of the echoed signal of estimating with β _j, t _0j), j is any nonnegative integer that is not more than k, β _jand t _0jβ and t ₀j element;

Then by { f (i, β _j, t _0j), k and B substitution formula (3) solve the signal F (i, β, the t that incide laser radar detection device estimating ₀, k, B), by { f (i, β _j, t _0j), k and B ' substitution formula (3) solve the signal F (i, β, the t that incide laser radar detection device estimating ₀, k, B ');

Again by F (i, β, t ₀, k, B) and y substitution formula (4) solve y at signal F (i, β, t ₀, k, B) joint distribution probability L (y|t ₀, k, β, B), by F (i, β, t ₀, k, B ') and y substitution formula (4) solve y at signal F (i, β, t ₀, k, B ') joint distribution probability L (y|t ₀, k, β, B ');

Finally by L (y|t ₀, k, β, B), L (y|t ₀, k, β, B ') substitution formula (7) solve B ' transition probability α (B ', B);

(5) if α (B ', B) > 0, by B ' assignment to B, other parameter t ₀, k and β remain unchanged, complete this and upgrade operation, by t ₀, the new overall approximate solution (t that forms of k, β and B ₀, k, β, B) in select any parameter to upgrade operation at random next time;

(6) if α (B ', B)≤0, calculate to improve transition probability η (B ', B), account form as shown in Equation (10),

η(B′,B)＝exp(α(B′,B)/T _v) （10）

(7) if η (B ', B) > ε, ε is that in interval (0,1) meets equally distributed random number, by B ' assignment to B, other parameter t ₀, k, β remain unchanged, complete this and upgrade operation, by t ₀, the new overall approximate solution (t that forms of k, β and B ₀, k, β, B) in select any parameter to upgrade operation at random next time;

(8) if η (B ', B)≤ε, further judges whether to arrive at T _v=g (T ₀, v) the maximum update times N that attempts under condition _b2, wherein step (3) to step (7) is once to attempt upgrading, if do not arrive at T _v=g (T ₀, v) the maximum update times N that attempts under condition _b2, jump to step (3) and continue this renewal operation; If arrive at T _v=g (T ₀, v) the maximum update times N that attempts under condition _b2, according to mode shown in formula (11), regulate T:

T _v+1＝g(T ₀,v+1) （11）

In formula (11), T _v+1for the T after regulating;

(9) if T _v+1< T _max, forward step (2) to and continue this renewal operation; If T _v+1>=T _max, parameter t ₀, k, β and B all remain unchanged, and jumps out this and upgrade operation, reselect parameter and upgrade operation next time.

Choose at random parameter t ₀after update method:

If at parameter t ₀, it is random arbitrarily between k, β and B that select is parameter t ₀, by parameter t ₀start this and upgrade operation, parameter t ₀renewal operation based on simulated tempering algorithm, upgrade, the steps include:

(1), according to the requirement of simulated tempering algorithm, set function T=g (T ₀, u), and meet formula (8) (9); Wherein, the span of restriction u makes 1 < T < T _max, T ' and T ' ' are respectively T about derived function and the second derived function of u, this function of initialization T=T ₀(T ₀> 1);

(2), establish T _v=g (T ₀, v), v is the arbitrary value within the scope of u, is arranged on T _v=g (T ₀, the maximum under v) is attempted update times (setting in advance);

(3), establish k dimensional vector t ₀' be t ₀the state that may be updated to, at interval (0, i _max) select k to obey equally distributed random number assignment to k dimensional vector t ₀' all elements;

(4), by y, t ₀solve according to y, t with β substitution formula (2) ₀energy distribution { f (i, the β of the echoed signal of estimating with β _j, t _0j), by y, t ₀' solve according to y, t with β substitution formula (2) ₀' and energy distribution { f (i, the β of the echoed signal estimated of β _j, t _0j'), j is any nonnegative integer that is not more than k, t _0j, β _jand t _0j' be respectively vectorial t ₀, β and _t0' j element;

Then by { f (i, β _j, t _0j), k and B substitution formula (3) solve the signal F (i, β, the t that incide laser radar detection device estimating ₀, k, B), by { f (i, β _j, t _0j'), k and B substitution formula (3) solve the signal F (i, β, the t that incide laser radar detection device estimating ₀', k, B);

Again by F (i, β, t ₀, k, B) and y substitution formula (4) solve y at signal F (i, β, t ₀, k, B) joint distribution probability L (y|t ₀, k, β, B), by F (i, β, t ₀', k, B) and y substitution formula (4) solve y at signal F (i, β, t ₀', k, B) joint distribution probability L (y|t ₀', k, β, B);

Finally by L (y|t ₀, k, β, B), L (y|t ₀', k, β, B) substitution formula (12) solves t ₀' transition probability α (t ₀', t ₀);

$\mathrm{\&alpha;}(t_0^{\&prime;},t_0)=L\left(y\right|t_0^{\&prime;},k,\mathrm{\&beta;},B)-L\left(y\right|t_0,k,\mathrm{\&beta;},B)---\left(12\right)$

(5) if α is (t ₀', t ₀) > 0, by t ₀' assignment is to t ₀, other parameters k, β and B remain unchanged, and complete this and upgrade operation, by t ₀, the new overall approximate solution (t that forms of k, β and B ₀, k, β, B) in select any parameter to upgrade operation at random next time;

(6) if α is (t ₀', t ₀)≤0, calculates and improves transition probability η (t ₀', t ₀), account form as shown in Equation (13),

$\mathrm{\&eta;}(t_0^{\&prime;},t)=\exp (\mathrm{\&alpha;}(t_0^{\&prime;},t)/T_v)---\left(13\right)$

(7) if η is (t ₀', t) > ε, ε is that interval (0,1) one meets uniform random number, by t ₀' assignment is to t ₀, other parameter k, β and B remain unchanged, and complete this and upgrade operation, by t ₀, the new overall approximate solution (t that forms of k, β and B ₀, k, β, B) in select any parameter to upgrade operation at random next time;

(8) if η is (t ₀', t)≤ε, further judges whether to arrive at T _v=g (T ₀, v) the maximum update times of attempting under condition wherein step (3) to step (7) is once to attempt upgrading, if do not arrive at T _v=g (T ₀, v) the maximum update times of attempting under condition forward step (3) to and continue this renewal operation; If arrive at T _v=g (T ₀, v) the maximum update times of attempting under condition according to mode shown in formula (11), regulate T;

(9) if T _v+1< T _max, T wherein _v+1be the T after regulating, forward step (2) to and continue this renewal operation; If T _v+1>=T _max, parameter t ₀, k, β and B all remain unchanged, and jumps out this and upgrade operation, reselect parameter and upgrade operation next time.

Choose at random the update method after parameter β:

If at parameter t ₀, random arbitrarily between k, β and B that select is parameter β, by parameter β, starts this and upgrades operation, the renewal operation of parameter β is upgraded based on simulated tempering algorithm, the steps include:

(1), according to the requirement of simulated tempering algorithm, set function T=g (T ₀, u), and meet formula (8) (9); Wherein, the span of restriction u makes 1 < T < T _max, T ' and T ' ' are respectively T about derived function and the second derived function of u, this function of initialization T=T ₀(T ₀> 1);

(2), establish T _v=g (T ₀, v), v is arbitrary value within the scope of u, is arranged on T _v=g (T ₀, the maximum under v) is attempted update times N _β(setting in advance);

(3), establish k dimensional vector t ₀' be t ₀the state that may be updated to, selects k to obey equally distributed random number assignment to all the elements of β ' at interval (0, max (y));

(4), first by y, t ₀solve according to y, t with β substitution formula (2) ₀energy distribution { f (i, the β of the echoed signal of estimating with β _j, t _0j), by y, t ₀and β ' substitution formula (2) solves according to y, t ₀and energy distribution { f (i, the β of the echoed signal estimated of β ' _j', t _0j), j is any nonnegative integer that is not more than k, t _0j, β _jand β _j' be respectively vectorial t ₀, β and β ' j element;

Then by { f (i, β _j, t _0j), k and B substitution formula (3) solve signal F (i, β, the t that incides laser radar detection device ₀, k, B), by { f (i, β _j', t _0j), k and B substitution formula (3) solve signal F (i, β ', the t that incides laser radar detection device ₀, k, B);

Again by F (i, β, t ₀, k, B) and y respectively substitution formula (4) solve y at signal F (i, β, t ₀, k, B) joint distribution probability L (y|t ₀, k, β, B), by F (i, β ', t ₀, k, B) and y respectively substitution formula (4) solve y at signal F (i, β ', t ₀, k, B) joint distribution probability L (y|t ₀, k, β ', B);

Finally by L (y|t ₀, k, β, B), L (y|t ₀, k, β ', B) substitution formula (15) solve β ' transition probability α (β ', β);

α(β′,β)＝L(y|t ₀,k,β′,B)-L(y|t ₀,k,β,B) （15）

(5) if α (β ', β) > 0, by β ' assignment to β, other parameter t ₀, k and B remain unchanged, complete this and upgrade operation, by t ₀, the new overall approximate solution (t that forms of k, β and B ₀, k, β, B) in select any parameter to upgrade operation at random next time;

(6) if α (β ', β)≤0, the transition probability η calculate improving (β ', β), account form as shown in Equation (16),

η(β′,β)＝exp(α(β′,β)/T _v) （16）

(7) if η (β ', β) > ε, ε is that interval [(0,1)] one meets uniform random number, by β ' assignment to β, other parameter t ₀, k and B remain unchanged, complete this and upgrade operation, by t ₀, the new overall approximate solution (t that forms of k, β and B ₀, k, β, B) in select any parameter to upgrade operation at random next time;

(8) if η (β ', β)≤ε, further judges whether to arrive at T _v=g (T ₀, v) the maximum update times N that attempts under condition _β, wherein step (3) to step (7) is once to attempt upgrading, if do not arrive at T _v=g (T ₀, v) the maximum update times N that attempts under condition _β, forward step (3) to and continue this renewal operation; If arrive at T _v=g (T ₀, v) the maximum update times N that attempts under condition _β, according to mode shown in formula (11), regulate T;

(9) if T _v+1< T _max, T wherein _v+1be the T after regulating, forward step (2) to and continue this renewal operation; If T _v+1>=T _max, parameter t ₀, k, β and B all remain unchanged, and jumps out this and upgrade operation, reselect parameter and upgrade operation next time.

Metropolis of the present invention calculates ratio juris can be referring to document (Journal of Computer and System Science 1,998 57 (1): 20-36), ratio juris can (Europhysics Letters 2,007 29 (12): 2170-2180) referring to document in simulated tempering calculation.

Beneficial effect of the present invention can further illustrate by following experiment:

The present invention has done four experiments under same experimental situation, experimental situation is: use the photon counting radar of development voluntarily to obtain Full wave shape signal y, system minimum time is distinguished as 24ps, use the inventive method to process this Full wave shape signal, extract echoed signal quantity k and peak t in Full wave shape parameter ₀, the peak t that computing is obtained ₀with the echoed signal shown in echoed signal peak amplitude β substitution formula (2) energy distribution function, obtain according to peak t ₀with the echo of echoed signal peak amplitude β, complete the signal decomposition of Full wave shape signal, thereby realize the extraction of the distance in laser facula.

According to the inventive method requirement, the starting condition of setting is: c ₁=10, c ₂=-5, c ₁=-20, τ ₁=5, τ ₂=10, τ ₃=20, gamma distribution probability density function f _gyardstick parameter a, c and shape parameter b, d respectively value be: a=10, b=1, c=1.1, d=100; β=max (y); B=0; K=1; Select (0, i _max) in max (y) position assignment to t ₀; t ₀adopt the renewal operating process based on simulated tempering algorithm of the present invention with β, wherein, T=T ₀* 1.2 ^v, T ₀=2, by controlling v, change T, and the trial update times under T is arbitrarily all 20, and N _βbe 20; K and B adopt the renewal operating process based on Metropolis algorithm of the present invention, and attempt update times and be 20, i.e. N _kand N _b1be 20; It is 600 that all parameter update times summations are set; With matlab instrument coding, process related data.

Experiment one:

This experimental detection target is the carton of frosted glass and placement in parallel thereafter.The detection of a target is about 40 meters apart from laser radar.The spacing of adjusting carton and frosted glass is respectively: 5 centimetres, 8 centimetres, 10 centimetres, 20 centimetres, 30 centimetres, 50 centimetres and 100 centimetres, obtain 7 groups of Full wave shape experimental datas.By the t that uses the inventive method to obtain ₀corresponding element substitution formula (2) with β, restores echo waveform, and Full wave shape and the echo waveform that restores are as shown in Figure 2.In Fig. 2, with the curve of burr, be Full wave shape number, other two curves are the target echo signals that restore by the inventive method.Fig. 2 (a) is that spacing is the signal decomposition figure of the Full wave shape signal of 5cm, Fig. 2 (b) is that spacing is the signal decomposition figure of the Full wave shape signal of 8cm, Fig. 2 (c) is that spacing is the signal decomposition figure of the Full wave shape signal of 10cm, Fig. 2 (d) is that spacing is the signal decomposition figure of the Full wave shape signal of 20cm, Fig. 2 (e) is that spacing is the signal decomposition figure of the Full wave shape signal of 30cm, Fig. 2 (f) is that spacing is the signal decomposition figure of the Full wave shape signal of 50cm, Fig. 2 (g) is that spacing is the signal decomposition figure of the Full wave shape signal of 100cm.

At Fig. 2, can find out, adopt identical photon counting laser radar, for the detection of a target of different distance, identical algorithm initial setting up, all extractions of energy realization character distance, realize the robotization of signal and process, and this is the advantage that some other algorithm does not have.The non-linear least square waveform fitting method of LM algorithm is only applicable to the scene of single characteristic; The operation result convergence of the Full wave shape parameter maximum likelihood value method of estimation of EM algorithm and improvement algorithm thereof depends on the default of initial value strongly, cannot realize no manual intervention's Full wave shape parameter completely and extract, and sometimes has the risk losing efficacy.

From Fig. 2 (a), Fig. 2 (b), Fig. 2 (c), Fig. 2 (d), can find out, frosted glass and carton spacing are less, and waveform is overlapping more serious, and its decomposition is more difficult, and this also proves the advantage of this algorithm in precision.From Fig. 2 (e), Fig. 2 (f), Fig. 2 (g), can find out, frosted glass and carton spacing are larger, overlapping less between echo, algorithm operation result calculate obtain peak t ₀very approaching with the echo signal peak detecting of observing roughly, this and actual distance are quite approaching, can prove the accuracy of this algorithm.

Experiment two:

In order further to check the present invention to process the ability of Full wave shape signal, this experiment reduces Laser emission energy to obtain the Full wave shape signal of low signal-to-noise ratio.The detection of a target of this experiment is the carton of frosted glass and subsequent placement in parallel.The detection of a target is apart from approximately 40 meters of laser radars.The spacing rough measure of adjusting carton and frosted glass is respectively 10cm and 20cm.Processing through the present invention to Full wave shape signal, will obtain t by computing ₀be updated to the corresponding element of β the echo waveform that formula (2) obtains decompositing, the waveform that same group of data is decomposited and original waveform are presented at a figure, wherein the curve with burr is Full wave shape number, and other two curves are the target echo signals that restore by the inventive method.Fig. 3 (a) is that spacing is the exploded view of the low signal-to-noise ratio Full wave shape signal of 10cm, and by operation result, obtaining its spacing is 9.76cm; Fig. 3 (b) is that spacing is the exploded view of the low signal-to-noise ratio Full wave shape signal of 20cm, and by operation result, obtaining its spacing is 21.6cm.

Experiment three:

The Full wave shape signal of the present invention being processed to the face battle array target that laser radar obtains by spot scan, restores the distribution of target in distance.As shown in Figure 4, the detection of a target is frosted glass, N shape letter and the carton of parallel placement successively, and the distance between carton and N shape letter and frosted glass is respectively 20cm and 10cm(Fig. 4 (a) is the vertical view of target, the front view that Fig. 4 (b) is target).Every beam of laser is all irradiated on N shape letter or carton behind through frosted glass, processes so every a branch of laser echo signal, and rebuilds three-dimensional configuration as Fig. 4 (c).Fig. 4 (c) has provided the target morphology of reduction, in figure, black region is all positions that laser can be irradiated to, the corresponding actual frosted glass of black boxed area, middle black N shape letter is N shape letter cardboard, remaining black region is the region that on carton, laser can be irradiated to, the white N shape letter at rear be on carton Ear Mucosa Treated by He Ne Laser Irradiation less than region, from the range coordinate in left side, can find out the target that the restores information on distance, close with actual value.

Experiment four:

This experiment has been done comparative experiments by the inventive method and RJMCMC method, further the present invention is evaluated.Two kinds of methods are processed same test figure under the same conditions.Two kinds of methods are processed respectively (being mainly to extract two distance feature) 100 times to testing 7 groups of test figures obtaining in one, RJMCMC algorithm circulates in a fixed order and upgrades all parameters, each parameter upgrades 600 times, and the update times summation of all parameters of the present invention is 600.Difference statistical average working time and 100 operation results, the inventive method and RJMCMC method average operating time and 100 operation results (each spacing average, and standard deviation) added up as shown in table 1.In spacing average and poor these two indexs of separation criteria, the stability of RJMCMC method operation result is better than the present invention; And comparison in program runtime average, the present invention has huge advantage in speed.

Table 1 the present invention and RJMCMC method are at distance extraction and the statistical form of operation time

Claims (6)

1. a laser radar Full wave shape signal analysis method, is characterized in that, comprises the following steps:

Step 1, a laser radar Full wave shape signal y of input, parameter clearly to be asked is: t ₀, k, β, B, wherein:

K is the quantity of echoed signal, and its span is: { 0,1,2Kk _max, k _maxit is the maximal value of default k;

B is the noise signal that incides laser radar detection device, and its span is: (0, max (y)), and max (y) is the maximal value of Full wave shape signal y;

T ₀for the k dimensional vector of the accurate flight time formation of the photon in each echoed signal, the peak of corresponding k the echoed signal of its element, t ₀the span of all elements be: (0, i _max), i _maxthe length of Full wave shape signal y;

β is the k dimensional vector that the peak amplitude of each echoed signal forms, the peak amplitude of corresponding k the echoed signal of its element, and the span of all elements of β is: (0, max (y));

From choose in the span of parameter to be asked respectively arbitrary value respectively assignment to parameter t ₀, k, β and B, complete the respectively initialization of parameter to be asked; In completing parameter initialized to be asked, choose any one parameter, with selected parameter, start once to upgrade operation;

Step 2, need to be asked parameter to form an overall approximate solution (t institute upgrading after operation ₀, k, β, B), at each overall approximate solution (t forming after upgrading operation ₀, k, β, B) in random any one parameter of selecting proceed to upgrade operation; Repeat this step, until stop upgrading operation after upgrading the maximal value that the summation of number of operations reaches predefined renewal number of operations;

The all overall approximate solution obtaining after step 3, renewal operation is each time substitution posteriority distribution probability density function respectively, obtain the posteriority distribution probability value of each overall approximate solution, will be worth overall approximate solution corresponding to maximum posteriority distribution probability value as globally optimal solution; Described posteriority distribution probability density function is as shown in Equation (1):

$\mathrm{\&pi;}(t_0,k,\mathrm{\&beta;},B|y)=\left(\underset{i=1}{\overset{i_{\max }}{\mathrm{\&Pi;}}}\exp (-(\underset{j=1}{\overset{k}{\mathrm{\&Sigma;}}}f(i,{\mathrm{\&beta;}}_j,t_{0j})+B))\frac{{(\underset{j=1}{\overset{k}{\mathrm{\&Sigma;}}}f(i,{\mathrm{\&beta;}}_j,t_{0j})+B)}^{y_i}}{y_i!}\right)$

$\&times;\frac{1}{k_{\max }}{\left(\frac{1}{i_{\max }}\right)}^k{f}_G\left(B\right|c,d)\underset{n=1}{\overset{k}{\mathrm{\&Pi;}}}{f}_G\left({\mathrm{\&beta;}}_n\right|a,b)---\left(1\right)$

In formula (1), π (t ₀, k, β, B|y) and be posteriority distribution probability value; f _gfor Gamma distribution probability density function, a and c are f _gyardstick parameter, b and d are f _gshape parameter, and a, b, c, d are default positive number; y _iit is y discrete representation in time; F (i, β _j, t _0j) be the energy distribution of j echoed signal on time i, j is not more than k _maxnonnegative integer, i gets 0 conventionally to i _maxbetween integer, t _0jand β _jbe respectively accurate flight time of photon of j echoed signal and peak amplitude, be also t ₀with j the element of β, f (i, β _j, t _0j) function expression as shown in Equation (2):

$f(i,{\mathrm{\&beta;}}_j,t_{0j})={\mathrm{\&beta;}}_j\left\{\begin{array}{cc}{e}^{-{(t_{1j}-t_{0j})}^2/{2\mathrm{\&sigma;}}^2}{e}^{(i-t_{1j})/{\mathrm{\&tau;}}_1},& i<t_{1j}\\ e^{-{(i-t_{0j})}^2/{2\mathrm{\&sigma;}}^2},& t_{1j}\&le;i<t_{2j}\\ e^{-{(t_{2j}-t_{0j})}^2/{2\mathrm{\&sigma;}}^2}{e}^{-(i-t_{2j})/{\mathrm{\&tau;}}_2},& t_{2j}\&le;i<t_{3j}\\ e^{-{(t_{2j}-t_{0j})}^2/{2\mathrm{\&sigma;}}^2}{e}^{-(t_{3j}-t_{2j})/{\mathrm{\&tau;}}_2}{e}^{-(i-t_{3j})/{\mathrm{\&tau;}}_3},& i\&GreaterEqual;t_{3j}\end{array}\right.---\left(2\right)$

In formula (2), t _0j, t _1j, t _2j, t _3jbetween there is stable difference magnitude relationship, i.e. t _0j-t _1j=c ₁, t _0j-t _2j=c ₂, t _0j-t _3j=c ₃, c wherein ₁, c ₂, c ₃for default constant; τ ₁, τ ₂, τ ₃be default constant with σ, and 0 < τ ₁< τ ₂< τ ₃, σ=0.

2. laser radar Full wave shape signal analysis method as claimed in claim 1, it is characterized in that, when once the selected parameter of renewal operation is k, the state that can further select at random k may reach after this upgrades operation in k ' and this two states of k ' ', then according to the random result of selecting, according to k ' and k ' ' respectively corresponding different update method carry out this and upgrade and operate; Or at k ', k ' ', t ₀, select any one parameter in β and B, the update method that this parameter is corresponding is upgraded to the method for operation as this, k '=k+1 wherein, k ' '=k-1;

Described random selection, the state that k may reach after this upgrades operation is k ' time, this computation process of upgrading behaviour is:

If 2.1.1 k=k _max, each parameter t ₀, k, β and B remain unchanged, and jumps out this and upgrade operation, reselect parameter and upgrade operation next time; Otherwise continue step 2.1.2;

2.1.2, establish t ₀' expression t ₀the state that may arrive after this upgrades operation, and t ₀' be k ' dimensional vector, t ₀' front k element be followed successively by t ₀front k element;

If the state that β ' expression β may arrive after this upgrades operation, and β ' is k ' dimensional vector, front k the element of β ' is followed successively by t ₀front k element;

At interval (0, i _max) in select any one to obey equally distributed random number assignment to t ₀' the individual element of k ', in interval (0, max (y)), select any one to obey equally distributed random number assignment to the individual element of k ' of β ';

2.1.3, by y, t ₀solve according to y, t with β substitution formula (2) ₀energy distribution { f (i, the β of the echoed signal of estimating with β _j, t _0j), j is any nonnegative integer that is not more than k, β _jand t _0jrespectively vectorial β and t ₀j element; By y, t ₀' and β ' substitution formula (2) solve according to y, t ₀' and energy distribution { f (i, the β of the echoed signal estimated of β ' _m', t _0m'), m is any nonnegative integer that is not more than k ', β _m' and t _0m' be respectively vectorial β ' and t ₀' m element;

Then by { f (i, β _j, t _0j), k and B substitution formula (3) solve the signal F (i, β, the t that incide laser radar detection device estimating ₀, k, B), by { f (i, β _m', t _0m'), k ' and B substitution formula (3) solve the signal F (i, β ', the t that incide laser radar detection device estimating ₀', k ', B);

Again by F (i, β, t ₀, k, B) and y substitution formula (4) solve Full wave shape signal y at signal F (i, β, t ₀, k, B) joint distribution probability L (y|t ₀, k, β, B), by F (i, β ', t ₀', k ', B) solves Full wave shape signal y at signal F (i, β ', t with y substitution formula (4) ₀', k ', joint distribution probability L (y|t B) ₀', k ', β ', B);

Finally by L (y|t ₀, k, β, B), L (y|t ₀', k ', β ', B) substitution formula (5) solve k ' transition probability α (k ', k), shown in formula (3), (4), (5) are specific as follows:

$F(i,\mathrm{\&beta;},t_0,k,B)=\underset{j=1}{\overset{k}{\mathrm{\&Sigma;}}}f(i,{\mathrm{\&beta;}}_j,t_{0j})+B---\left(3\right)$

$L\left(y\right|t_0,k,\mathrm{\&beta;},B)=\underset{i=1}{\overset{i_{\max }}{\mathrm{\&Pi;}}}\exp (-F\left(i\right))\frac{F{\left(i\right)}^{y_i}}{y_i!}---\left(4\right)$

In formula (4), y ₁, y ₂k for y discrete representation in time,

$\mathrm{\&alpha;}(k^{\&prime;},k)=L\left(y\right|t_0^{\&prime;},k^{\&prime;},{\mathrm{\&beta;}}^{\&prime;},B)-L\left(y\right|t_0,k,\mathrm{\&beta;},B)---\left(5\right)$

If 2.1.4 α (k ', k) > 0, by t ₀', β ' and k ' difference assignment be to t ₀, β and k, parameter B remains unchanged, and completes this and upgrades operation, by t ₀, the new overall approximate solution (t that forms of k, β and B ₀, k, β, B) in select any parameter to upgrade operation at random next time;

If 2.1.5 α (k ', k)≤0, further whether judgement now arrives the maximum update times N that attempts _k(setting in advance), wherein step 2.1.2 is once to attempt upgrading to the process of step 2.1.4; If arrive the maximum update times N that attempts _k, each parameter t ₀, k, β and B remain unchanged, and jumps out this and upgrade operation, reselect parameter and upgrade operation next time; If do not arrive the maximum update times N that attempts _k, jump to step 2.1.2 and continue this renewal operation;

Described random selection, when the state that k may reach after this upgrades operation is k ' ', this computation process of upgrading behaviour is:

If 2.2.1 k=0, each parameter t ₀, k, β and B remain unchanged, and jumps out this and upgrade operation, reselect parameter and upgrade operation next time; Otherwise, continue step 2.2.2;

2.2.2, suppose it is this time attempt to upgrade for the l time, step 2.2.2 is once to attempt upgrading to step 2.2.4, establishes t ₀', β ' is respectively that this upgrades t after operation ₀, the β state that may be updated to, and t ₀', β ' is k ' ' dimensional vector, will reject vectorial t after l element ₀distinguish successively in order assignment to vectorial t with the surplus element of β ₀' and the individual element of k ' ' of β ';

2.2.3, by y, t ₀solve according to y, t with β substitution formula (2) ₀energy distribution { f (i, the β of the echoed signal of estimating with β _j, t _0j), j is any nonnegative integer that is not more than k, β _jand t _0jrespectively vectorial β and t ₀j element; By y, t ₀' and β ' substitution formula (2) solve according to y, t ₀' and energy distribution { f (i, the β of the echoed signal estimated of β ' _m', t _0m'), m is any nonnegative integer that is not more than k ' ', β _m' and t _0m' be respectively vectorial β ' and t ₀' m element;

Then by { f (i, β _j, t _0j), k and B substitution formula (3) solve according to { f (i, β _j, t _0j), k and B the signal F (i, β, the t that incide laser radar detection device that estimate ₀, k, B), by { f (i, β _j', t _0j'), k ' ' and B substitution formula (3) solve according to { f (i, β _m', t _0m'), k ' ' and B the signal F (i, β ', the t that incide laser radar detection device that estimate ₀', k ' ', B);

Again by F (i, β, t ₀, k, B) and y substitution formula (4) solve y at signal F (i, β, t ₀, k, B) joint distribution probability L (y|t ₀, k, β, B), by F (i, β ', t ₀', k ' ', B) solves y at signal F (i, β ', t with y substitution formula (4) ₀', k ' ', joint distribution probability L (y|t B) ₀', k ' ', β ', B);

Finally by L (y|t ₀, k, β, B), L (y|t ₀', k ' ', β ', B) substitution formula (6) solve k ' ' transition probability α (k ' ', k);

$\mathrm{\&alpha;}(k^{\&prime;\&prime;},k)=L\left(y\right|t_0^{\&prime;},k^{\&prime;\&prime;},{\mathrm{\&beta;}}^{\&prime;},B)-L\left(y\right|t_0,k,\mathrm{\&beta;},B)---\left(6\right)$

If 2.2.4 α (k ' ', k) > 0, by t ₀', the value of β ' and k ' ' respectively assignment to t ₀, β and k, parameter B remains unchanged, and completes this and upgrades operation, by t ₀, the new overall approximate solution (t that forms of k, β and B ₀, k, β, B) in select any parameter to upgrade operation at random next time;

If 2.2.5 α (k ' ', k)≤0, further judges whether to arrive the maximum update times k that attempts, if arrive maximum update times k, each parameter t of attempting ₀, k, β and B remain unchanged, and jumps out this and upgrade operation, reselect parameter and upgrade operation next time; If do not arrive the maximum update times k that attempts, jump to step 2.2.2 and continue this renewal operation.

3. laser radar Full wave shape signal analysis method as claimed in claim 1, is characterized in that, when once the selected parameter of renewal operation is B, this process of upgrading operation is:

3.1, establishing B ' is the state that B may be updated to, and at interval (0, max (y)), selects a random number assignment of obeying Gamma distribution to B ';

3.2, first by y, t ₀solve according to y, t with β substitution formula (2) ₀energy distribution { f (i, the β of the echoed signal of estimating with β _j, t _0j), j is any nonnegative integer that is not more than k, β _jand t _0jβ and t ₀j element;

Then by { f (i, β _j, t _0j), k and B substitution formula (3) solve according to { f (i, β _j, t _0j), k and B the signal F (i, β, the t that incide laser radar detection device that estimate ₀, k, B), by { f (i, β _j, t _0j), k and B ' substitution formula (3) solve according to { f (i, β _j, t _0j), k and B ' the signal F (i, β, the t that incide laser radar detection device that estimate ₀, k, B ');

Again by F (i, β, t ₀, k, B) and y substitution formula (4) solve y at signal F (i, β, t ₀, k, B) joint distribution probability L (y|t ₀, k, β, B), by F (i, β, t ₀, k, B ') and y substitution formula (4) solve y at signal F (i, β, t ₀, k, B ') joint distribution probability L (y|t ₀, k, β, B ');

Finally by L (y|t ₀, k, β, B), L (y|t ₀, k, β, B ') substitution formula (7) solve B ' transition probability α (B ', B);

α(B′,B)＝L(y|t ₀,k,β,B′)-L(y|t ₀,k,β,B) （7）

If 3.3 α (B ', B) > 0, by B ' assignment to B, other parameter t ₀, k, β remain unchanged, complete this and upgrade operation, by t ₀, the new overall approximate solution (t that forms of k, β and B ₀, k, β, B) in select any parameter to upgrade operation at random next time;

If 3.4 α (B ', B)≤0, further judges whether to arrive the maximum update times N that attempts _b1, wherein step 3.1 to step 3.3 is once to attempt upgrading, if reach the maximum update times N that attempts _b1, all parameter t ₀, k, β and B remain unchanged, and jumps out this and upgrade operation, reselect parameter and upgrade operation next time; If do not arrive maximum update times N _b1, jump to step 3.1 and continue this renewal operation.

4. laser radar Full wave shape signal analysis method as claimed in claim 1, is characterized in that, when once the selected parameter of renewal operation is B, this process of upgrading operation is:

4.1,, according to the requirement of simulated tempering algorithm, set function T=g (T ₀, u), 1 < T < T _max, and T meets formula (8) and (9) simultaneously;

T′＞0 （8）

(T′) ²-2T(T′) ²+T′′(T) ²≤0 （9）

In formula (8) and (9), T ' and T ' ' are respectively T about derived function and the second derived function of u, and this function of initialization is T=T ₀(T ₀> 1);

4.2, establish T _v=g (T ₀, v), v is the arbitrary value in the scope of u, is arranged on T _v=g (T ₀, the maximum under v) is attempted update times N _b2;

4.3, establishing B ' is the state that B may be updated to, and at interval (0, max (y)), selects a random number assignment of obeying Gamma distribution to B ';

4.4, first by y, t ₀solve according to y, t with β substitution formula (2) ₀energy distribution { f (i, the β of the echoed signal of estimating with β _j, t _0j), j is any nonnegative integer that is not more than k, β _jand t _0jβ and t ₀j element;

Then by { f (i, β _j, t _0j), k and B substitution formula (3) solve the signal F (i, β, the t that incide laser radar detection device estimating ₀, k, B), by { f (i, β _j, t _0j), k and B ' substitution formula (3) solve the signal F (i, β, the t that incide laser radar detection device estimating ₀, k, B ');

Again by F (i, β, t ₀, k, B) and y substitution formula (4) solve y at signal F (i, β, t ₀, k, B) joint distribution probability L (y|t ₀, k, β, B), by F (i, β, t ₀, k, B ') and y substitution formula (4) solve y at signal F (i, β, t ₀, k, B ') joint distribution probability L (y|t ₀, k, β, B ');

Finally by L (y|t ₀, k, β, B), L (y|t ₀, k, β, B ') substitution formula (7) solve B ' transition probability α (B ', B);

If 4.5 α (B ', B) > 0, by B ' assignment to B, other parameter t ₀, k and β remain unchanged, complete this and upgrade operation, by t ₀, the new overall approximate solution (t that forms of k, β and B ₀, k, β, B) in select any parameter to upgrade operation at random next time;

If 4.6 α (B ', B)≤0, calculate to improve transition probability η (B ', B), account form as shown in Equation (10),

η(B′,B)＝exp(α(B′,B)/T _v) （10）

If 4.7 η (B ', B) > ε, ε is that in interval (0,1) meets equally distributed random number, by B ' assignment to B, other parameter t ₀, k, β remain unchanged, complete this and upgrade operation, by t ₀, the new overall approximate solution (t that forms of k, β and B ₀, k, β, B) in select any parameter to upgrade operation at random next time;

If 4.8 η (B ', B)≤ε, further judges whether to arrive at T _v=g (T ₀, v) the maximum update times N that attempts under condition _b2, wherein step 4.3 to step 4.7 is once to attempt upgrading, if do not arrive at T _v=g (T ₀, v) the maximum update times N that attempts under condition _b2, jump to step 4.3 and continue this renewal operation; If arrive at T _v=g (T ₀, v) the maximum update times N that attempts under condition _b2, according to mode shown in formula (11), regulate T:

T _v+1＝g(T ₀,v+1) （11）

In formula (11), T _v+1for the T after regulating;

If 4.9 T _v+1< T _max, forward step 4.2 to and continue this renewal operation; If T _v+1>=T _max, parameter t ₀, k, β and B all remain unchanged, and jumps out this and upgrade operation, reselect parameter and upgrade operation next time.

5. laser radar Full wave shape signal analysis method as claimed in claim 1, is characterized in that, when once upgrading the selected parameter of operation, is t ₀time, this process of upgrading operation is:

5.1,, according to the requirement of simulated tempering algorithm, set function T=g (T ₀, u), and meet formula (8) (9); Wherein, the span of restriction u makes 1 < T < T _max, T ' and T ' ' are respectively T about derived function and the second derived function of u, this function of initialization T=T ₀(T ₀> 1);

5.2, establish T _v=g (T ₀, v), v is the arbitrary value within the scope of u, is arranged on T _v=g (T ₀, the maximum under v) is attempted update times

5.3, establish k dimensional vector t ₀' be t ₀the state that may be updated to, at interval (0, i _max) select k to obey equally distributed random number assignment to k dimensional vector t ₀' all elements;

5.4, by y, t ₀solve according to y, t with β substitution formula (2) ₀energy distribution { f (i, the β of the echoed signal of estimating with β _j, t _0j), by y, t ₀' solve according to y, t with β substitution formula (2) ₀' and energy distribution { f (i, the β of the echoed signal estimated of β _j, t _0j'), j is any nonnegative integer that is not more than k, t _0j, β _jand t _0j' be respectively vectorial t ₀, β and t ₀' j element;

Then by { f (i, β _j, t _0j), k and B substitution formula (3) solve the signal F (i, β, the t that incide laser radar detection device estimating ₀, k, B), by { f (i, β _j, t _0j'), k and B substitution formula (3) solve the signal F (i, β, the t that incide laser radar detection device estimating ₀', k, B);

Again by F (i, β, t ₀, k, B) and y substitution formula (4) solve y at signal F (i, β, t ₀, k, B) joint distribution probability L (y|t ₀, k, β, B), by F (i, β, t ₀', k, B) and y substitution formula (4) solve y at signal F (i, β, t ₀', k, B) joint distribution probability L (y|t ₀', k, β, B);

Finally by L (y|t ₀, k, β, B), L (y|t ₀', k, β, B) substitution formula (12) solves t ₀' transition probability α (t ₀', t ₀);

$\mathrm{\&alpha;}(t_0^{\&prime;},t_0)=L\left(y\right|t_0^{\&prime;},k,\mathrm{\&beta;},B)-L\left(y\right|t_0,k,\mathrm{\&beta;},B)---\left(12\right)$

If 5.5 α (t ₀', t ₀) > 0, by t ₀' assignment is to t ₀, other parameters k, β and B remain unchanged, and complete this and upgrade operation, by t ₀, the new overall approximate solution (t that forms of k, β and B ₀, k, β, B) in select any parameter to upgrade operation at random next time;

If 5.6 α (t ₀', t ₀)≤0, calculates and improves transition probability η (t ₀', t ₀), account form as shown in Equation (13),

$\mathrm{\&eta;}(t_0^{\&prime;},t)=\exp (\mathrm{\&alpha;}(t_0^{\&prime;},t)/T_v)---\left(13\right)$

If 5.7 η (t ₀', t) > ε, ε is that interval (0,1) one meets uniform random number, by t ₀' assignment is to t ₀, other parameter k, β and B remain unchanged, and complete this and upgrade operation, by t ₀, the new overall approximate solution (t that forms of k, β and B ₀, k, β, B) in select any parameter to upgrade operation at random next time;

If 5.8 η (t ₀', t)≤ε, further judges whether to arrive at T _v=g (T ₀, v) the maximum update times of attempting under condition wherein step 5.3 to step 5.7 is once to attempt upgrading, if do not arrive at T _v=g (T ₀, v) the maximum update times of attempting under condition forward step 5.3 to and continue this renewal operation; If arrive at T _v=g (T ₀, v) the maximum update times of attempting under condition according to mode shown in formula (11), regulate T;

If 5.9 T _v+1< T _max, T wherein _v+1be the T after regulating, forward step 5.2 to and continue this renewal operation; If T _v+1>=T _max, parameter t ₀, k, β and B all remain unchanged, and jumps out this and upgrade operation, reselect parameter and upgrade operation next time.

6. laser radar Full wave shape signal analysis method as claimed in claim 1, is characterized in that, when once the selected parameter of renewal operation is β, this process of upgrading operation is:

6.1,, according to the requirement of simulated tempering algorithm, set function T=g (T ₀, u), and meet formula (8) (9); Wherein, the span of restriction u makes 1 < T < T _max, T ' and T ' ' are respectively T about derived function and the second derived function of u, this function of initialization T=T ₀(T ₀> 1);

6.2, establish T _v=g (T ₀, v), v is arbitrary value within the scope of u, is arranged on T _v=g (T ₀, the maximum under v) is attempted update times N _β;

6.3, establish k dimensional vector t ₀' be t ₀the state that may be updated to, selects k to obey equally distributed random number assignment to all the elements of β ' at interval (0, max (y));

6.4, first by y, t ₀solve according to y, t with β substitution formula (2) ₀energy distribution { f (i, the β of the echoed signal of estimating with β _j, t _0j), by y, t ₀and β ' substitution formula (2) solves according to y, t ₀and energy distribution { f (i, the β of the echoed signal estimated of β ' _j', t _0j), j is any nonnegative integer that is not more than k, t _0j, β _jand β _j' be respectively vectorial t ₀, β and β ' j element;

Then by { f (i, β _j, t _0j), k and B substitution formula (3) solve signal F (i, β, the t that incides laser radar detection device ₀, k, B), by { f (i, β _j', t _0j), k and B substitution formula (3) solve signal F (i, β ', the t that incides laser radar detection device ₀, k, B);

Again by F (i, β, t ₀, k, B) and y respectively substitution formula (4) solve y at signal F (i, β, t ₀, k, B) joint distribution probability L (y|t ₀, k, β, B), by F (i, β ', t ₀, k, B) and y respectively substitution formula (4) solve y at signal F (i, β ', t ₀, k, B) joint distribution probability L (y|t ₀, k, β ', B);

Finally by L (y|t ₀, k, β, B), L (y|t ₀, k, β ', B) substitution formula (15) solve β ' transition probability α (β ', β);

α(β′,β)＝L(y|t ₀,k,β′,B)-L(y|t ₀,k,β,B) （15）

If 6.5 α (β ', β) > 0, by β ' assignment to β, other parameter t ₀, k and B remain unchanged, complete this and upgrade operation, by t ₀, the new overall approximate solution (t that forms of k, β and B ₀, k, β, B) in select any parameter to upgrade operation at random next time;

If 6.6 α (β ', β)≤0, the transition probability η calculate improving (β ', β), account form as shown in Equation (16),

η(β′,β)＝exp(α(β′,β)/T _v) （16）

If 6.7 η (β ', β) > ε, ε is that interval (0,1) one meets uniform random number, by β ' assignment to β, other parameter t ₀, k and B remain unchanged, complete this and upgrade operation, by t ₀, the new overall approximate solution (t that forms of k, β and B ₀, k, β, B) in select any parameter to upgrade operation at random next time;

If 6.8 η (β ', β)≤ε, further judges whether to arrive at T _v=g (T ₀, v) the maximum update times N that attempts under condition _β, wherein step 6.3 to step 6.7 is once to attempt upgrading, if do not arrive at T _v=g (T ₀, v) the maximum update times N that attempts under condition _β, forward step 6.3 to and continue this renewal operation; If arrive at T _v=g (T ₀, v) the maximum update times N that attempts under condition _β, according to mode shown in formula (11), regulate T;

If 6.9 T _v+1< T _max, T wherein _v+1be the T after regulating, forward step 6.2 to and continue this renewal operation; If T _v+1>=T _max, parameter t ₀, k, β and B all remain unchanged, and jumps out this and upgrade operation, reselect parameter and upgrade operation next time.