Summary of the invention
The technical problem that the present invention will solve is; Method for distilling calculation of complex, pre-treatment step to pulse signal characteristic point of the prior art be too loaded down with trivial details, can't realize that real-time processing and detection and hardware realize the high shortcoming of difficulty, provides computation complexity to reduce, need not pre-treatment step, is convenient to hardware and realizes and can realize handling in real time and the method for distilling based on the pulse signal characteristic point of second filial generation small echo that detects.
The technical problem that the present invention will solve is achieved through following technical scheme: the method for distilling based on the pulse signal characteristic point of second filial generation small echo is provided; Wherein, said method comprises based on second filial generation wavelet decomposition pulse signal and based on modulus maximum method extraction pulse signal characteristic point.
In the method for distilling of above-mentioned pulse signal characteristic point based on second filial generation small echo, saidly may further comprise the steps based on second filial generation wavelet decomposition pulse signal:
A1: set and decompose the number of plies, under the said decomposition number of plies, first signal is split into odd column signal and even column signal;
A2:, adopt the wavelet coefficient forecast model to predict the wavelet coefficient under the said decomposition layer number based on said odd column signal in the steps A 1 and said even column signal; And
A3: adopt scale coefficient to upgrade the scale coefficient under the said decomposition number of plies of model modification.
In the method for distilling of above-mentioned pulse signal characteristic point based on second filial generation small echo, in said steps A 2, said wavelet coefficient forecast model is:
Wherein, On behalf of the i layer, i decompose the number of plies; L represents l sampled point;
is predicting the outcome of the i+1 layer wavelet coefficient that decomposes l sampled point under the number of plies;
is the odd column signal that the i layer decomposes scale coefficient under the number of plies, and
is the even column signal that the i layer decomposes scale coefficient under the number of plies.
In the method for distilling of above-mentioned pulse signal characteristic point based on second filial generation small echo, in said steps A 3, said scale coefficient more new model is:
Wherein, On behalf of the i layer, i decompose the number of plies; L represents l sampled point;
is the renewal result that the i+1 layer decomposes the scale coefficient of l sampled point under the number of plies;
is the even column signal that the i layer decomposes scale coefficient under the number of plies;
is predicting the outcome of the i+1 layer wavelet coefficient that decomposes l sampled point under the number of plies;
is the predictive value that the i+1 layer decomposes the wavelet coefficient of l+1 sampled point under the number of plies, and
is the predictive value that the i+1 layer decomposes the wavelet coefficient of l-1 sampled point under the number of plies.
In the method for distilling of above-mentioned pulse signal characteristic point based on second filial generation small echo, saidly extract the pulse signal characteristic point based on the modulus maximum method and may further comprise the steps:
B1: confirm that the just a plurality of-negative maximum value of wavelet coefficient is right under the current decomposition number of plies;
B2: adopt the zero crossing computation model to confirm a plurality of zero crossings that just a plurality of described in the step B1-negative maximum value is right;
B3: a plurality of main wave-wave peak scope of confirming pulse signal based on a plurality of zero crossings described in the step B2;
B4: confirm a plurality of main wave-waves peak in the said a plurality of main wave-waves peak scope in step B3, said main wave-wave peak dot is first characteristic point;
B5: setting very first time window, is starting point with a plurality of first characteristic points described in the step B4, in said very first time window, confirms a plurality of single ripple starting points, and said single ripple starting point is second characteristic point;
B6: a plurality of dicrotic wave anchor points of confirming pulse signal based on a plurality of first characteristic points described in the step B4;
B7: set second time window; With a plurality of dicrotic wave anchor points described in the step B6 is starting point; In said second time window, confirm a plurality of dicrotic wave crests and a plurality of dicrotic wave trough respectively, said dicrotic wave crest and dicrotic wave trough are respectively the 3rd characteristic point and four characteristic points.
In the method for distilling of above-mentioned pulse signal characteristic point based on second filial generation small echo, said step B1 comprises following substep:
B11: a plurality of positive maximum points and negative maximum value point of confirming wavelet coefficient under the current decomposition number of plies;
B12: adopt maximum point threshold calculations model and minimum point threshold calculations model to confirm the maximum point threshold value and the minimum point threshold value of wavelet coefficient under the current decomposition number of plies respectively;
B13: confirm that with the relation of said minimum point threshold value just said-negative maximum value is right based on said a plurality of positive maximum points and relation and said a plurality of negative maximum value point of said maximum point threshold value.
In the method for distilling of above-mentioned pulse signal characteristic point based on second filial generation small echo, in said step B12, said maximum point threshold calculations model is:
Wherein, th
1Be maximum point threshold value, M
1, M
2, M
3And M
4For said wavelet coefficient being divided into the maximum of the wavelet coefficient of each section after four sections;
Said minimum point threshold calculations model is:
Wherein, th
2Be minimum point threshold value, N
1, N
2, N
3And N
4For said wavelet coefficient being divided into the minima of the wavelet coefficient of each section after four sections.
In the method for distilling of above-mentioned pulse signal characteristic point based on second filial generation small echo, in said step B2, said zero crossing computation model is:
Wherein, l
0Be zero crossing position, l
1For just-positive maximum point position that the negative maximum value is right, d
1For just-wavelet coefficient that the right positive maximum point of negative maximum value is corresponding, l
2For just-negative maximum value point position that the negative maximum value is right, and d
2For just-wavelet coefficient that negative maximum value centering negative maximum value point is corresponding.
In the method for distilling of above-mentioned pulse signal characteristic point based on second filial generation small echo, between said step B2 and B3, saidly extract the pulse signal characteristic point based on the modulus maximum method and also comprise a plurality of zero crossings of confirming among the said step B2 are revised.
In the method for distilling of above-mentioned pulse signal characteristic point based on second filial generation small echo, said step B6 comprises:
Confirm the modulus maximum point between two of a plurality of main wave-waves peak, said modulus maximum point is the dicrotic wave anchor point; Perhaps
Between two of a plurality of main wave-waves peak, adopt omission point compensation model to confirm the dicrotic wave anchor point.
The method for distilling based on the pulse signal characteristic point of second filial generation small echo of embodiment of the present invention; Can obtain following beneficial effect: can effectively reduce computation complexity based on second filial generation wavelet decomposition pulse signal; Reduce hardware simultaneously and realize difficulty, be convenient on embedded platform, realize pulse signal Feature Points Extraction of the present invention; The characteristic point of pulse signal can be accurately extracted when adopting the modulus maximum method that the pulse signal based on second filial generation wavelet decomposition is extracted, and false retrieval point and flase drop point can be deleted pointedly, and compensation omission point, fallout ratio and false drop rate reduced.Method of the present invention can accurately be extracted the pulse signal characteristic point under the prerequisite that effectively reduces computation complexity, realization is convenient and accuracy rate is high.
The specific embodiment
In order to make the object of the invention, technical scheme and advantage clearer,, the present invention is further elaborated below in conjunction with accompanying drawing and embodiment.Should be appreciated that specific embodiment described herein only in order to explanation the present invention, and be not used in qualification the present invention.
Referring to Fig. 1, Fig. 1 is the sketch map at typical pulse signal; In practical application, choose illustrated a, b, c and d usually at 4 as characteristic point.Wherein, a point is single ripple starting point, indicates the beginning of heart phase of maximum ejection, has mainly reflected pressure and the volume situation in the end of term arteries of shrinking.The b point is main wave-wave peak, and the rate of climb of its upstroke is mainly relevant with tube wall elasticity with cardiac output, Ve speed, tremulous pulse resistance, and its characteristic can be represented with the upstroke slope.The c point is a dicrotic pulse wave-wave paddy, mainly reflects aorta static pressure emptying time, is the separation of heart contraction and diastole.The d point is dicrotic pulse wave-wave peak, function status, blood vessel elasticity and the blood flow flow regime of reflection aortic valve.Among the present invention a, b, c and 4 of d are called second characteristic point, first characteristic point, four characteristic points and the 3rd characteristic point respectively.Therefore, the objective of the invention is to, how from typical pulse signal, accurately extract first, second, third with four characteristic points, the practical application of promotion pulse signal aspect the cardiovascular function status analysis.
The invention provides a kind of method for distilling of the pulse signal characteristic point based on second filial generation small echo; As shown in Figure 2; This method starts from step 10, and be included in carry out in the step 11 based on second filial generation wavelet decomposition pulse signal and in step 12, carry out extract the pulse signal characteristic point based on the modulus maximum method.This method ends at step 13.
In step 11, based on second filial generation wavelet decomposition pulse signal; Second filial generation wavelet transformation can effectively be analyzed the low-and high-frequency characteristic of pulse signal as the good signal time frequency analyzing tool, is used for decision signal singular point position.Select Bi-orthogonal Spline Wavelet Transformation bior1.3 as the second filial generation small echo that decomposes pulse signal among the present invention; Because the tight supportive of such small echo; Pulse signal with spike shape facility is had good responding ability, and the signal that receives sound pollution is still had stronger identification ability, when therefore using Bi-orthogonal Spline Wavelet Transformation bior1.3; Even need not original pulse signal is carried out a large amount of pretreatment in early stage, also can obtain ideal detection accuracy rate.Specifically describe the present invention though should be noted that the following Bi-orthogonal Spline Wavelet Transformation bior1.3 that will combine, the spendable second filial generation small echo of method of the present invention is not limited thereto.On the contrary, those skilled in the art can select other second filial generation small echos that is fit to according to actual needs, and for example haar, db1, bior1.1, cdf1.1, cdf1.1, cdf1.3, cdf1.5 etc. perhaps directly make up a new second filial generation small echo.It is understandable that above-mentioned replacement is all in the protection domain of accompanying claims of the present invention.
Referring to Fig. 3-4, Fig. 3-4 has set forth sketch map and the flow chart that decomposes pulse signal based on Bi-orthogonal Spline Wavelet Transformation bior1.3 according to of the present invention.Can know by Fig. 3, of the present inventionly decompose pulse signal based on Bi-orthogonal Spline Wavelet Transformation bior1.3 and comprise division, prediction and 3 steps of renewal, can obtain the scale coefficient and the wavelet coefficient of pulse signal under the different decomposition number of plies through above-mentioned operation splitting.Said division refers to set decomposes the number of plies, under this decomposition number of plies, first signal is split into odd column signal and even column signal; Said prediction refers to that based on the odd column signal and the even column signal that obtain this decomposes the wavelet coefficient under the number of plies to adopt the prediction of wavelet coefficient forecast model; And said renewal refers to adopt scale coefficient to upgrade the scale coefficient under this decomposition number of plies of model modification.First signal here refers to pulse signal.The idiographic flow of this method is as shown in Figure 5.
As shown in Figure 5, this method starts from step 110.In step 111, set and decompose number of plies n, n is the positive integer greater than 1.Each decomposes, and pulse signal can have corresponding wavelet coefficient and yardstick system under the number of plies.In step 112, be written into first signal; In next step 113, first signal that is written into is divided processing subsequently.If first signal is x, handle as follows based on the division of Bi-orthogonal Spline Wavelet Transformation bior1.3:
Wherein,
Be the even column signal after the division of first signal, x
2lBe to be designated as even part under in first signal;
Be the odd column signal after the division of first signal, x
2l+1It is the part that is designated as odd number in first signal down.
In step 114, based on odd column signal and the even column signal that a last step obtains, adopt wavelet coefficient forecast model prediction wavelet coefficient, the wavelet coefficient forecast model is:
Wherein, On behalf of the i layer, i decompose the number of plies (i is the positive integer smaller or equal to n); L represents l sampled point;
is predicting the outcome of the i+1 layer wavelet coefficient that decomposes l sampled point under the number of plies;
is the odd column signal that the i layer decomposes scale coefficient under the number of plies, and
is the even column signal that the i layer decomposes scale coefficient under the number of plies.Can further release the wavelet coefficient model according to the wavelet coefficient forecast model:
Wherein, d
lFor this decomposes the wavelet coefficient under the number of plies (i layer),
Be predicting the outcome of the i+1 layer wavelet coefficient that decomposes l sampled point under the number of plies.
In next step 115, adopt scale coefficient to upgrade the model modification scale coefficient, and further release the scale coefficient under this decomposition number of plies according to the renewal result of scale coefficient, the scale coefficient that this step relates to more new model is following:
Wherein, On behalf of the i layer, i decompose the number of plies; L represents l sampled point;
is the renewal result that the i+1 layer decomposes the scale coefficient of l sampled point under the number of plies;
is the even column signal that the i layer decomposes scale coefficient under the number of plies;
is predicting the outcome of the i+1 layer wavelet coefficient that decomposes l sampled point under the number of plies;
is the predictive value that the i+1 layer decomposes the wavelet coefficient of l+1 sampled point under the number of plies, and
is the predictive value that the i+1 layer decomposes the wavelet coefficient of l-1 sampled point under the number of plies;
This decomposes the scale coefficient under the number of plies to adopt following formula to derive:
Wherein, s
lFor this decomposes the scale coefficient under the number of plies (i layer),
Be the renewal result that the i+1 layer decomposes the scale coefficient of l sampled point under the number of plies.
Subsequently, at next step 116, judge whether to decompose to the required decomposition number of plies, whether particularly, calculating n-1 is 0, if, showing that then this decomposition number of plies has been the required decomposition number of plies, can finish the resolution process of pulse signal this moment; If not, then in step 117, be written into secondary signal, and repeat above-mentioned steps 113-116, begin next pulse signal that decomposes under the number of plies and decompose.Wherein, secondary signal is the scale coefficient under this decomposition number of plies that calculates in the formula (5).This method ends at step 118.
Among the present invention, pulse signal is that the pulse transducer of 200Hz collects by sample frequency, and sampling length can be divided into 2048 points on time domain.Utilize above-mentioned Bi-orthogonal Spline Wavelet Transformation bior1.3 to decompose pulse signal; Obtain wherein i=1 of each layer wavelet coefficient
; 2; 3,4 ... n.Because Bi-orthogonal Spline Wavelet Transformation bior1.3 can reduce by half the last layer signal length in the catabolic process of each layer, the location of pulse signal characteristic point is stretched to the wavelet coefficient length of decomposing each decomposition number of plies of back and the primary signal equal in length for ease.
Referring to Fig. 5-8, it has shown according to the method that the present invention is based on modulus maximum method extraction pulse signal characteristic point.Among Fig. 5, extract the pulse signal characteristic point based on the modulus maximum method and may further comprise the steps: B1: confirm that the just a plurality of-negative maximum value of wavelet coefficient is right under the current decomposition number of plies; B2: adopt the zero crossing computation model to confirm a plurality of zero crossings that just a plurality of among the step B2-negative maximum value is right; B3: a plurality of main wave-wave peak scope of confirming pulse signal based on a plurality of zero crossings among the step B2; B4: in step B3, confirm a plurality of main wave-waves peak in a plurality of main wave-waves peak scope, wherein main wave-wave peak is first characteristic point; B5: setting very first time window, is starting point with a plurality of first characteristic points among the step B4, in very first time window, confirms a plurality of single ripple starting points, and wherein single ripple starting point is second characteristic point; B6: a plurality of dicrotic wave anchor points of confirming pulse signal based on a plurality of first characteristic points among the step B4; B7: setting second time window, is starting point with a plurality of dicrotic wave anchor points among the step B6, in second time window, confirms a plurality of dicrotic wave crests and a plurality of dicrotic wave trough respectively, and dicrotic wave crest and dicrotic wave trough are respectively the 3rd characteristic point and four characteristic points.First, second, third and the four characteristic points that extracts can be used for the analysis of cardiovascular disease, has practical significance.
Below will combine Fig. 6-8 to introduce the method for extracting the pulse signal characteristic point based on the modulus maximum method of the present invention in detail.This method starts from step 1200, and advances to step 1201.
In step 1201, be written into based on Bi-orthogonal Spline Wavelet Transformation bior1.3 pulse signal is carried out the 1st layer and n layer wavelet coefficient of n layer wavelet decomposition, wherein n is the positive integer greater than 1.Referring to Fig. 7; (a) the untreated original pulse signal of representative (b) is represented the 1st layer of wavelet coefficient
and (c) represent n layer wavelet coefficient
At next step 1202; Confirm the positive maximum point and the negative maximum value point of n layer wavelet coefficient
through the first derivative zero crossing of n yardstick wavelet coefficient
; Wherein the positive and negative maximum point of n yardstick wavelet coefficient
is shown in (e) among Fig. 7, and the positive and negative maximum point of the 1st yardstick wavelet coefficient
is shown in (d) among Fig. 7.
At next step 1203, adopt maximum point threshold calculations model and minimum point threshold calculations model to confirm the maximum point threshold value and the minimum point threshold value of n yardstick wavelet coefficient.Particularly, will
Be equally divided into 4 sections by time domain, every section 1024 points, wherein every section wavelet coefficient maximum is respectively M
1, M
2, M
3, M
4, every section wavelet coefficient minima is respectively N
1, N
2, N
3, N
4, maximum point threshold calculations model and minimum point computation model are respectively:
Wherein, th
1Be maximum point threshold value, M
1, M
2, M
3And M
4For said wavelet coefficient being divided into after four sections the maximum of every section wavelet coefficient; Th
2Be minimum point threshold value, N
1, N
2, N
3And N
4For said wavelet coefficient being divided into the minima of the wavelet coefficient of each section after four sections.
Proceed to step 1204 subsequently; Wherein, based on the relation of the relation of a plurality of positive maximum points and maximum point threshold value and a plurality of negative maximum value point and minimum point threshold value confirm n yardstick wavelet coefficient
just-the negative maximum value is right.Particularly, in substep 1204a, the size of more positive and negative maximum and very big/little value point threshold value keeps positive maximum>th
1With negative maximum value<th
2The point; At substep 1204b; Judge that whether adjacent 2 time domain interval that satisfy above-mentioned condition are less than 80 points (being that express time is at interval less than 0.4s under this sampling condition) and these 2 contrary signs; If, then substep 1204c keep these 2 o'clock as one just-the negative maximum value is right.Wherein, Fig. 7 (f) shown satisfy above-mentioned condition one just-the negative maximum value is right.At substep 1204b; If adjacent 2 time domain interval that satisfy above-mentioned condition are greater than 80 points or 2 jack per lines; Then proceed to substep 1204d, continue the next point of search n yardstick wavelet coefficient
.
At next step 1205, to adopt the zero crossing computation model to confirm respectively just a plurality of-zero crossing that the negative maximum value is right.1. and 2., the zero crossing computation model is like Fig. 8 step:
Wherein, l
0Be zero crossing position, l
1For just-positive maximum point position that the negative maximum value is right, d
1For just-wavelet coefficient that the right positive maximum point of negative maximum value is corresponding, l
2For just-negative maximum value point position that the negative maximum value is right, and d
2For just-wavelet coefficient that negative maximum value centering negative maximum value point is corresponding.
At next step 1206, zero crossing is revised, specifically comprise remove many cautious and to greatly/little value point threshold value revises.Wherein, calculate between two zero crossings distance and zero crossing at substep 1206a and judge distance and the magnitude relationship of zero crossing between two zero crossings at substep 1206b and substep 1206c apart from average apart from average.Judge according to following formula:
In substep 1206b, as D>1.6 * m
1Preset threshold is excessive in the description of step 1203, and go to step 1206d this moment, with greatly/little value point threshold value is reduced to original 1/4; And return execution in step 1204 and 1205 successively, new greatly/in little value point threshold range definite again just-the negative maximum value to and zero crossing; Otherwise this method proceeds to step 1206c.
In substep 1206c, as D<0.7 * m
1, the wavelet coefficient of n yardstick is described
Exist and examine more, execution in step 1206e deleted and examined more this moment, went to next step then.
Wherein, D is a distance between two zero crossings, m
1For zero crossing apart from average.Step 1206 can be expressed as, as 0.7 * m
1<D<1.6 * m
1The time, this zero crossing is available zero crossing, at this moment, this method goes to next step.
At next step 1207, on the maximum of the 1st yardstick, be starting point with the zero crossing, (l
0-i) a position sweep forward j non-zero points is as starting point, (l
0-i) j non-zero points searched for backward as terminal point (Fig. 8 step is 3.) in the position, confirms main wave-wave peak scope (Fig. 8 step is 4.).Exemplary, j is set at 3.Forward direction described here and back are to being meant that early stage sampled point and the later stage sampled point to the zero crossing position moves in the time domain scope.
At next step 1208, confirm main wave-wave peak (being called first characteristic point again) based on the main wave-wave peak scope of confirming.Particularly,, in untreated original pulse signal, locate maximum of points, be the main wave-wave peak or first characteristic point, in Fig. 1, be expressed as the b point according to main wave-wave peak scope.
At next step 1209, confirm single ripple starting point, i.e. second characteristic point.Particularly, calculate the range averaging value of a plurality of first characteristic points, set very first time window according to following formula:
TW
1=0.25×m
2;
Wherein, m
2Be the range averaging value of a plurality of first characteristic points between in twos.Then, be starting point with first characteristic point, the minimum point in untreated original pulse signal in the sweep forward very first time window, this minimum point is second characteristic point; In Fig. 1, be expressed as a point.
Subsequently, at next step 1210, confirm a plurality of dicrotic wave anchor points of pulse signal based on first characteristic point.Particularly, at substep 1210a, at n yardstick wavelet coefficient
In detect except that just above-mentioned-negative maximum value to the modulus maximum point; At substep 1210b, the time-domain position of establishing immediate first characteristic point of forward direction of this modulus maximum point is l, and whether the time-domain position of judging this modulus maximum point is at (l+0.15 * m
2, l+0.35 * m
2) in the scope, if not, then at substep 1210c this modulus maximum point of deletion and return execution substep 1210a; If, then judge whether there is the modulus maximum point between two first characteristic points at substep 1210d, if not, be (l+0.25 * m then at substep 1210e compensation time-domain position
2) point, and with this point as the dicrotic wave anchor point, thereby avoid occurring omission point phenomenon; If; Then further judge whether there are a plurality of modulus maximum points that meet above-mentioned time-domain position requirement between two first characteristic points at substep 1210f; If then keep second modulus maximum point near the forward direction first characteristic point direction at next substep 1210g; This point is detected dicrotic wave anchor point, among Fig. 7 shown in (g).
Subsequently; At next step 1211; Set second time window; In untreated original pulse signal, the search minimum point is a four characteristic points in the forward direction second time window scope of a plurality of dicrotic wave anchor points of in step 1210, confirming, and back search maximum of points in second time window is the 3rd characteristic point.Wherein, second time window is confirmed according to following formula:
TW
2=0.1×m
2;
Wherein, m
2Be the range averaging value of a plurality of first characteristic points between in twos.
Subsequently, this method ends at step 1212.At this moment, the pulse signal that has four characteristic points that extraction obtains based on the modulus maximum method is shown in Fig. 7 (h).
In sum, can when reducing computation complexity, accurately obtain a plurality of characteristic points of pulse signal through the method for distilling of the pulse signal characteristic point based on second filial generation small echo provided by the present invention; Computation complexity has promoted the application of this method at embedded platform when reducing; Owing to can take measures areput further to guarantee the accuracy and the effectiveness of feature point extraction to many cautious, flase drop point and omission point; Selecting for use of suitable second filial generation small echo can be avoided original pulse signal is carried out a large amount of pretreatment in early stage, and computation complexity is further controlled.
The above is merely the preferred embodiments of the present invention, not in order to restriction the present invention, all any modifications of in spirit of the present invention and principle, being done, is equal to and replaces or improvement etc., all should be included in protection scope of the present invention.