CN106073709A - A kind of method and apparatus of rale detection - Google Patents
A kind of method and apparatus of rale detection Download PDFInfo
- Publication number
- CN106073709A CN106073709A CN201610390420.4A CN201610390420A CN106073709A CN 106073709 A CN106073709 A CN 106073709A CN 201610390420 A CN201610390420 A CN 201610390420A CN 106073709 A CN106073709 A CN 106073709A
- Authority
- CN
- China
- Prior art keywords
- singular value
- signal
- level
- characteristic vector
- order
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7253—Details of waveform analysis characterised by using transforms
Abstract
The present invention relates to a kind of rale detection method and device, in one embodiment, the method includes: receive Lung Sounds to be measured, and small echo signal and Lung Sounds convolution to be measured are carried out wavelet transformation, it is thus achieved that first order decomposed signal;Carry out singular value decomposition to first order decomposed signal and obtain corresponding first order singular value;Build the characteristic vector of Lung Sounds to be measured according at least to first order singular value;Sending into the characteristic vector of Lung Sounds to be measured in the grader trained and differentiating, output differentiates result.Rale detection method and device that the embodiment of the present invention is provided have higher Detection accuracy, and its Detection accuracy can reach 100%, has a good application prospect.
Description
Technical field
The present invention relates to processing of biomedical signals technical field, particularly relate to the method and apparatus of a kind of rale detection.
Background technology
Respiratory disease is common disease, for example asthma, COPD (chronic obstructive
Pulmonary disease, is called for short COPD), pneumonia, bronchitis etc..Recently as air pollution and other environmental factors,
Respiratory disease illness rate sustainable growth, Diagnosis of Pulmonary Diseases has obtained more concern.Lung's breath sound contains a large amount of letter
Breath, is an important indicator of physiology and the pathology reacting lung.Wherein, rale is a kind of common pulmonary abnormalities sound, with pneumonia
Closely related etc. disease.At present, the diagnosis to rale for the clinical medicine relies primarily on auscultation, and this is largely dependent upon doctor's
Stethoscopic technique and clinical experience, subjective.Compared to traditional auscultation, study that a kind of precision is higher, the visitor of non-intrusion type
The pulmonary abnormalities sound detection technique seen is particularly important.
Present stage, increasing researcher is devoted to Lung Sounds research both at home and abroad, is analyzed by signal processing technology
Lungs sound, extracts physiology therein and pathological information, sets up a categorizing system, to quantify normal and abnormal lungs sound exactly, from
And it is state estimation and the more foundation of medical diagnosis on disease offer of lung.At present, signal processing technology detection rale is used to pass through
Time frequency analysis (Short Time Fourier Transform, wavelet decomposition etc.) extracts its feature (power, fractal dimension etc.), by machine learning or
The method training grader of person's neutral net, thus rale is identified.
Wavelet analysis is used widely in time frequency analysis due to its transformable time frequency resolution, wavelet transformation
(wavelet transform is called for short WT), compared with Fourier transformation, wavelet transformation was that the localization of time (space) frequency is divided
Analysis, it progressively carries out multi-scale refinement by flexible shift operations to signal (function), is finally reached high frequency treatment time subdivision, low
Frequency segmentation at Pin, can automatically adapt to the requirement that time frequency signal is analyzed, thus can focus on any details of signal, solve Fu
In the difficult problem of leaf transformation, but due to the particularity of rale waveform, do not find more suitable wavelet application at present
In rale detection, thus have impact on Detection accuracy.
Content of the invention
For solving the problems referred to above, first aspect, embodiments provide the classifier training side of a kind of rale detection
Method, the method includes: receives lungs sound training signal, by small echo signal and the lungs sound training signal convolution receiving, carries out small echo
Conversion, it is thus achieved that first order decomposed signal;Carry out singular value decomposition to first order decomposed signal and obtain corresponding first order singular value;
Build the characteristic vector of lungs sound training signal according at least to first order singular value;By the characteristic vector of lungs sound training signal, instruction
Practice grader.
Alternatively, said method also includes: first order decomposed signal is carried out n times down-sampled, it is thus achieved that the first down-sampled letter
Number, down-sampled to small echo signal and first signal convolution is carried out wavelet transformation, it is thus achieved that second level decomposed signal;Wherein, n for more than
The positive integer of 1, such as 2;Second level decomposed signal is carried out n times down-sampled, it is thus achieved that the second down-sampled signal, by small echo signal and
Two down-sampled signal convolution carry out wavelet transformation, it is thus achieved that third level decomposed signal;Second level decomposed signal and the third level are decomposed
Signal carries out singular value decomposition respectively and obtains corresponding second level singular value and third level singular value.
Alternatively, in the above-mentioned methods, small echo signal is rale analog signal.
Alternatively, in the above-mentioned methods, when first order singular value is for carrying out singular value decomposition according to first order decomposed signal
When obtaining corresponding L singular value, step builds the feature of described lungs sound training signal according at least to described first order singular value
Vector, comprising: L singular value is added, it is thus achieved that singular value with;Each singular value is compared with singular value and work, it is thus achieved that L ratio
Value, by L ratio composition first order characteristic vector;Build the spy of described lungs sound training signal according at least to first order characteristic vector
Levy vector.
Alternatively, said method also includes: will carry out singular value decomposition according to second level decomposed signal and obtain corresponding L
Singular value is as second level singular value, and is added L singular value, it is thus achieved that singular value and;By each singular value and singular value and
Work compares, it is thus achieved that L ratio, and L ratio is formed corresponding second level characteristic vector;To carry out very according to third level decomposed signal
Different value is decomposed and is obtained corresponding L singular value as third level singular value, and by L singular value addition, it is thus achieved that singular value with;Will
Each singular value compares with singular value and work, it is thus achieved that L ratio, and L ratio is formed corresponding third level characteristic vector.
Alternatively, in the above-mentioned methods, step according at least to first order singular value build lungs sound training signal feature to
Amount includes: according to first order characteristic vector or first order characteristic vector and second level characteristic vector or first order feature to
Amount and second level characteristic vector and third level characteristic vector build the characteristic vector of lungs sound training signal.
Alternatively, in the above-mentioned methods, grader is support vector machine classifier.
Second aspect, the embodiment of the present invention provides a kind of rale detection method, and the method includes: receive lungs sound to be measured letter
Number, and by small echo signal and Lung Sounds convolution to be measured, carry out wavelet transformation, it is thus achieved that first order decomposed signal;To the first fraction
Solve signal and carry out the corresponding first order singular value of singular value decomposition acquisition;Build lungs sound to be measured letter according at least to first order singular value
Number characteristic vector;Sending into the characteristic vector of Lung Sounds to be measured in the grader trained and differentiating, output differentiates knot
Really.
Alternatively, said method also includes: first order decomposed signal is carried out n times down-sampled, it is thus achieved that the first down-sampled letter
Number, down-sampled to small echo signal and first signal convolution is carried out wavelet transformation, it is thus achieved that second level decomposed signal;The second level is decomposed
Signal carry out n times down-sampled, it is thus achieved that the second down-sampled signal, down-sampled to small echo signal and second signal convolution is carried out small echo change
Change, it is thus achieved that third level decomposed signal;Wherein, n is the positive integer more than 1;To second level decomposed signal and third level decomposed signal
Carry out singular value decomposition respectively and obtain corresponding second level singular value and third level singular value.
Alternatively, in the above-mentioned methods, small echo signal is rale analog signal.
Alternatively, in the above-mentioned methods, when first order singular value is for carrying out singular value decomposition according to first order decomposed signal
When obtaining corresponding L singular value, step builds the feature of described lungs sound training signal according at least to described first order singular value
Vector, comprising: L singular value is added, it is thus achieved that singular value with;Each singular value is compared with singular value and work, it is thus achieved that L ratio
Value, by L ratio composition first order characteristic vector;Build the spy of described lungs sound training signal according at least to first order characteristic vector
Levy vector.
Alternatively, said method also includes, will carry out singular value decomposition according to second level decomposed signal and obtain corresponding L
Singular value is as second level singular value, and is added L singular value, it is thus achieved that singular value and;By each singular value and singular value and
Work compares, it is thus achieved that L ratio, and L ratio is formed corresponding second level characteristic vector;To carry out very according to third level decomposed signal
Different value is decomposed and is obtained corresponding L singular value as third level singular value, and by L singular value addition, it is thus achieved that singular value with;Will
Each singular value compares with singular value and work, it is thus achieved that L ratio, and L ratio is formed corresponding third level characteristic vector.
Alternatively, in the above-mentioned methods, step according at least to first order singular value build the feature of Lung Sounds to be measured to
Amount, comprising: according to first order characteristic vector or first order characteristic vector and second level characteristic vector or first order feature
Vector sum second level characteristic vector and third level characteristic vector build the characteristic vector of lungs sound training signal.
Alternatively, in the above-mentioned methods, grader is support vector machine classifier.
The third aspect, the embodiment of the present invention provides a kind of sieve sound detection device, and this device includes: wavelet transform unit is used for
Receive Lung Sounds to be measured, small echo signal and Lung Sounds convolution to be measured are carried out wavelet transformation, it is thus achieved that at least level of decomposition letter
Number;Singular value decomposition unit is used for receiving at least level of decomposition signal, and carries out singular value at least level of decomposition signal respectively
Decompose and obtain corresponding at least one-level singular value;Characteristic vector construction unit is for building lung to be measured according at least one-level singular value
The characteristic vector of tone signal;Taxon is carried out for sending into the characteristic vector of Lung Sounds to be measured in the grader trained
Differentiating, output differentiates result.
Rale detection method that the embodiment of the present invention is provided and device, by using rale analog signal as mother wavelet,
First wavelet decomposition is carried out to Lung Sounds, extract its signature waveform, then singular value decomposition is carried out to decomposed signal, by unusual
Value composition characteristic vector, chooses SVMs training grader, to whether containing rale in lungs sound detects.The method has
Having higher Detection accuracy, its Detection accuracy can reach 100%, has a good application prospect.
Brief description
It is illustrated more clearly that the technical scheme of the embodiment of the present invention, required use in embodiment being described below
Accompanying drawing is briefly described, it should be apparent that, the accompanying drawing in describing below is only some embodiments of the present invention, for ability
From the point of view of the those of ordinary skill of territory, on the premise of not paying creative work, the attached of other can also be obtained according to these accompanying drawings
Figure.
The classifier training method of a kind of rale detection that Fig. 1 provides for the embodiment of the present invention;
A kind of rale detection method flow chart that Fig. 2 provides for the embodiment of the present invention;
A kind of sieve sound detection device that Fig. 3 provides for the embodiment of the present invention.
Detailed description of the invention
Below by drawings and Examples, technical scheme is described in further detail.
Rale has obvious characteristic at time-frequency domain: in time domain, rale start from a unexpected sharp-pointed concussion then by
Gradually broadening, the duration is generally less than 20ms;On frequency domain, its frequency is typically in 150-1800Hz, the frequency with normal lungs sound
There is bigger difference.
The detection process of rale includes training process and identification process.Training process includes receiving normal lungs sound and rale two
Class some training Lung Sounds, extracts the characteristic information of training Lung Sounds, builds the characteristic vector of training Lung Sounds, utilize
This feature vector training grader;Identification process includes receiving lungs sound to be measured, extracts the characteristic information of lungs sound to be measured, builds to be measured
The characteristic vector of Lung Sounds, and utilize the grader that training process obtains to carry out it point according to the characteristic vector of lungs sound to be measured
Class identification, exports recognition result.
The classifier training method of a kind of rale detection that Fig. 1 provides for the embodiment of the present invention, as it is shown in figure 1, the method
Including step S101-step S104:
Step S101, receives lungs sound training signal, by small echo signal and the lungs sound training signal convolution receiving, carries out little
Wave conversion, it is thus achieved that first order decomposed signal;
Alternatively, " step S101 " also includes: first order decomposed signal is carried out n times down-sampled, it is thus achieved that first is down-sampled
Down-sampled to small echo signal and first signal convolution is carried out wavelet transformation, it is thus achieved that second level decomposed signal by signal;To the second fraction
Solve signal carry out n times down-sampled, it is thus achieved that the second down-sampled signal, down-sampled to small echo signal and second signal convolution is carried out small echo
Conversion, it is thus achieved that third level decomposed signal;Wherein, described n is the positive integer more than 1, and n is 2 in the present embodiment.
Alternatively, rale analog signal is used to carry out wavelet transformation, rale ripple to lungs sound training signal as small echo signal
Shape simulation formula is as follows:
Wherein, a1=0.5, a2=1.49, a3=0.78, f0=2.0.
It should be noted that the lungs sound training signal receiving is with noises such as heart sound, alimentary canal sound and environmental noises
Signal.Mother wavelet signal is rale analog signal, the rale waveform of this maximum approaching to reality of rale analog signal, so
Decomposing lungs sound training signal based on rale analog signal, its discomposing effect is better than general wavelet transformation.Need explanation
Being that rale analog signal is not proper small echo, it is unsatisfactory for the normalizing condition that small echo is possessed.
Step S102, carries out singular value decomposition and obtains corresponding first order singular value to first order decomposed signal.
Specifically, carry out singular value decomposition to first order decomposed signal and obtain corresponding L singular value σ1, σ2..., σLFor
First order singular value, wherein, σ1≥σ2≥…≥σr> σr+1=σr+2=σL=0 (r≤L).
Alternatively, " step S102 " also includes: carry out the singular value decomposition acquisition second level to second level decomposed signal unusual
Value, and carry out singular value decomposition acquisition third level singular value to third level decomposed signal, it is thus achieved that second level singular value, the third level
The concrete mode of singular value is identical with obtaining first order singular value, and it will not go into details herein.
Alternatively, length L of singular value is 20 in embodiments of the present invention.It should be noted that can be according to detection essence
The specific requirement of degree, determines the concrete length of singular value.
It it should be noted that singular value decomposition (Singular Value Decomposition is called for short SVD), is linear
In algebraically, a kind of important matrix decomposition, is the popularization of normal matrix unitarily diagonalizable in matrix analysis, is that modern numerical is analyzed
One of most basic and most important instrument.
Step S103, builds the characteristic vector of lungs sound training signal according at least to first order singular value.
Specifically, L first order singular value is added, it is thus achieved that singular value and;By each singular value and singular value with make ratio,
The vector that L ratio after making ratio forms is first order characteristic vector, in order to describe first order characteristic vector more accurately
Operation principle and process, citing illustrates, such as: calculate the sum of L first order singular value: S=σ1+σ2+…+σr;Final acquisition
First order characteristic vector be:
Alternatively, " step S103 " also includes: will carry out singular value decomposition according to second level decomposed signal and obtain corresponding L
Individual singular value is as second level singular value, and is added L singular value, it is thus achieved that singular value and;By each singular value with described very
Different value and work compare, it is thus achieved that L ratio, and L ratio is formed corresponding second level characteristic vector;And will decompose according to the third level
Signal carries out singular value decomposition and obtains corresponding L singular value as third level singular value, and by L singular value addition, it is thus achieved that
Singular value and;Each singular value is compared with described singular value and work, it is thus achieved that L ratio, L ratio is formed the corresponding third level
Characteristic vector.
It is alternatively possible to according to first order characteristic vector or first order characteristic vector and second level characteristic vector or
First order characteristic vector and second level characteristic vector and third level characteristic vector build the characteristic vector of described lungs sound training signal.
It should be noted that the concrete length of characteristic vector according to the specific requirement of accuracy of detection, can be determined.
Step S104, according to the characteristic vector of lungs sound training signal, trains grader.
Alternatively, grader is SVMs (Support Vector Machine, SVM) grader.Support vector
Machine be based on structural risk minimization and VC tie up (Vapnik-Chervonenkis Dimension) theoretical based on, select suitable
When subsets of functions and decision function so that the practical risk of machine learning reach minimize, there is preferable performance, can
Preferably solve small sample, the problem in the presence of non-linear and high dimensional pattern identification.Choose linear letter in the present embodiment
Number, as the kernel function of SVMs, trains grader.
It should be noted that during concrete training, can according to the specific requirement of accuracy of detection, and select to adapt to
Wavelet decomposition number of times, it is thus achieved that at least level of decomposition signal, such as: rale analog signal is carried out little with lungs sound training signal convolution
Wave conversion, it is thus achieved that first order decomposed signal, carries out singular value decomposition and obtains first order singular value to first order decomposed signal, according to
First order singular value obtains first order characteristic vector, builds the characteristic vector of Lung Sounds, and root according to first order characteristic vector
According to this feature vector training grader.
After said method training grader, it is possible to use this grader trained carries out rale detection and identifies, under
State embodiment description is a kind of rale detection method.
A kind of rale detection method flow chart that Fig. 2 provides for the embodiment of the present invention.As in figure 2 it is shown, the method include with
Lower step S201-step S204:
Step S201, receives Lung Sounds to be measured, and small echo signal and Lung Sounds convolution to be measured is carried out wavelet transformation,
Obtain first order decomposed signal.
Alternatively, " step S201 " also includes: first order decomposed signal is carried out n times down-sampled, it is thus achieved that first is down-sampled
Rale analog signal and the first down-sampled signal convolution are carried out wavelet transformation, it is thus achieved that second level decomposed signal by signal;To second
Level decomposed signal carry out n times down-sampled, it is thus achieved that the second down-sampled signal, by rale analog signal and the second down-sampled signal convolution
Carry out wavelet transformation, it is thus achieved that third level decomposed signal;Wherein, described n is the positive integer more than 1, and n is 2 in the present embodiment.
Alternatively, rale analog signal is used to carry out wavelet transformation, rale ripple to lungs sound training signal as small echo signal
Shape simulation formula is as follows:
Wherein, a1=0.5, a2=1.49, a3=0.78, f0=2.0.
It should be noted that the lungs sound training signal receiving is with noises such as heart sound, alimentary canal sound and environmental noises
Signal.Mother wavelet signal is rale analog signal, the rale waveform of this maximum approaching to reality of rale analog signal, so
Decomposing lungs sound training signal based on rale analog signal, its discomposing effect is better than general wavelet transformation.Need explanation
Being that rale analog signal is not proper small echo, it is unsatisfactory for the normalizing condition that small echo is possessed.
Step S202, carries out singular value decomposition and obtains corresponding first order singular value to first order decomposed signal.
Specifically, carry out singular value decomposition to first order decomposed signal and obtain corresponding L singular value σ1, σ2..., σLFor
First order singular value, wherein, σ1≥σ2≥…≥σr> σr+1=σr+2=σL=0 (r≤L).
Alternatively, " step S102 " also includes: carry out the singular value decomposition acquisition second level to second level decomposed signal unusual
Value, third level decomposed signal is carried out singular value decomposition obtain third level singular value, it is thus achieved that second level singular value, the third level are unusual
The concrete mode of value is identical with acquisition first order singular value, and it will not go into details herein.Alternatively, singular value in embodiments of the present invention
Length L be 20.It should be noted that the concrete length of singular value according to the specific requirement of accuracy of detection, can be determined.
It it should be noted that singular value decomposition (Singular Value Decomposition is called for short SVD), is linear
In algebraically, a kind of important matrix decomposition, is the popularization of normal matrix unitarily diagonalizable in matrix analysis, is that modern numerical is analyzed
One of most basic and most important instrument.
Step S203, builds the characteristic vector of Lung Sounds to be measured according at least to first order singular value.
Specifically, L first order singular value is added, it is thus achieved that singular value and;By each singular value and singular value with make ratio,
The vector that L ratio after making ratio forms is first order characteristic vector, in order to describe first order characteristic vector more accurately
Operation principle and process, citing illustrates, such as: calculate the sum of L first order singular value: S=σ1+σ2+…+σr;Final acquisition
First order characteristic vector be:
Alternatively, " step S103 " also includes: will carry out singular value decomposition according to second level decomposed signal and obtain corresponding L
Individual singular value is as second level singular value, and is added L singular value, it is thus achieved that singular value and;By each singular value with described very
Different value and work compare, it is thus achieved that L ratio, and L ratio is formed corresponding second level characteristic vector;And will decompose according to the third level
Signal carries out singular value decomposition and obtains corresponding L singular value as third level singular value, and by L singular value addition, it is thus achieved that
Singular value and;Each singular value is compared with described singular value and work, it is thus achieved that L ratio, L ratio is formed the corresponding third level
Characteristic vector.
Specifically, can according to first order characteristic vector or first order characteristic vector and second level characteristic vector or
First order characteristic vector and second level characteristic vector and third level characteristic vector build the characteristic vector of described Lung Sounds to be measured.
It should be noted that the concrete length of characteristic vector according to the specific requirement of accuracy of detection, can be determined.
Step S204, sends into the characteristic vector of Lung Sounds to be measured in the grader trained and carries out Classification and Identification, defeated
Going out recognition result, recognition result will point out whether rale occur in this Lung Sounds to be measured.
Alternatively, grader is support vector machine classifier.SVMs is to tie up based on structural risk minimization and VC
Based on Li Lun, select suitable subsets of functions and decision function, so that the practical risk of machine learning reaches to minimize, tool
There is preferable performance, can preferably solve small sample, the problem in the presence of non-linear and high dimensional pattern identification.In this reality
Execute the kernel function choosing linear function in example as SVMs.
It should be noted that above-described embodiment is only a kind of specific embodiment of technical solution of the present invention, do not limit
Determine the present invention, during concrete identification, can according to the specific requirement of accuracy of detection, and select suitable wavelet function and
Wavelet decomposition number of times, it is thus achieved that at least level of decomposition signal, such as: rale analog signal and lungs sound training signal convolution are carried out small echo
Conversion, it is thus achieved that first order decomposed signal, carries out singular value decomposition and obtains first order singular value, according to the to first order decomposed signal
One-level singular value obtains first order characteristic vector, builds the characteristic vector of Lung Sounds to be measured according to first order characteristic vector, and
This feature vector is sent into the grader trained carries out Classification and Identification, export recognition result.Wherein, Lung Sounds to be measured
Characteristic vector length is consistent with the characteristic vector length of lungs sound training signal.
Accordingly, detection device corresponding with the detection method in above-described embodiment is embodiments provided.
A kind of sieve sound detection device that Fig. 3 provides for the embodiment of the present invention, as it is shown on figure 3, this device 3 includes: small echo becomes
Change unit the 31st, singular value decomposition unit the 32nd, characteristic vector construction unit the 33rd, taxon 34;Wherein,
Wavelet transform unit 31, is used for receiving Lung Sounds to be measured, by small echo signal and described Lung Sounds convolution to be measured
Carry out wavelet transformation, it is thus achieved that one or more levels decomposed signal.
Specifically, wavelet transform unit 31 obtains the concrete level of decomposed signal according to needed for concrete accuracy of detection requires to determine
Number;
If only obtaining level of decomposition signal, then wavelet transform unit 31 is after receiving Lung Sounds to be measured, simulates rale
Signal and Lung Sounds convolution to be measured carry out wavelet transformation, it is thus achieved that first order decomposed signal;
If obtaining multi-level decomposition signal, such as three grades, (first order decomposed signal, second level decomposed signal and the third level decompose letter
Number), then wavelet transform unit 31 is after receiving Lung Sounds to be measured, and small echo signal and Lung Sounds convolution to be measured are carried out small echo
Conversion, it is thus achieved that first order decomposed signal;First order decomposed signal is carried out n times down-sampled, it is thus achieved that the first down-sampled signal, by little
Ripple signal and the first down-sampled signal convolution carry out wavelet transformation, it is thus achieved that second level decomposed signal;Second level decomposed signal is entered
Row n times is down-sampled, it is thus achieved that the second down-sampled signal, down-sampled to small echo signal and second signal convolution is carried out wavelet transformation, obtains
Obtain third level decomposed signal;Wherein, described n is the positive integer more than 1, and n is 2 in the present embodiment.
Alternatively, rale analog signal is used to carry out wavelet transformation, rale ripple to lungs sound training signal as small echo signal
Shape simulation formula is as follows:
Wherein, a1=0.5, a2=1.49, a3=0.78, f0=2.0.
It should be noted that the lungs sound training signal receiving is with noises such as heart sound, alimentary canal sound and environmental noises
Signal.Mother wavelet signal is rale analog signal, the rale waveform of this maximum approaching to reality of rale analog signal, so
Decomposing lungs sound training signal based on rale analog signal, its discomposing effect is better than general wavelet transformation.Need explanation
Being that rale analog signal is not proper small echo, it is unsatisfactory for the normalizing condition that small echo is possessed.
Singular value decomposition unit 32, is used for receiving one or more levels decomposed signal, and decomposes letter to one or more levels respectively
Number carry out singular value decomposition and obtain one or more levels singular value corresponding;As received first order decomposed signal, the second level is decomposed
Signal and third level decomposed signal, then first order decomposed signal is carried out singular value decomposition obtain corresponding first order singular value,
Carry out singular value decomposition to obtain second level singular value to second level decomposed signal, carry out singular value decomposition to third level decomposed signal
Obtain third level singular value.
Specifically, first and second or three grades of decomposed signals are carried out singular value decomposition and obtain corresponding L singular value σ1,
σ2..., σLIt is first and second or three grades of singular values, wherein, σ1≥σ2≥…≥σr> σr+1=σr+2=σL=0 (r≤L).Optional
Ground, length L of singular value is 20 in embodiments of the present invention.It should be noted that can according to the specific requirement of accuracy of detection,
Determine the concrete length of singular value.
It should be noted that singular value decomposition, be a kind of important matrix decomposition in linear algebra, be in matrix analysis just
The popularization of rule battle array unitarily diagonalizable, is one of the most basic and most important instrument that modern numerical is analyzed.
Characteristic vector construction unit 33, for building the spy of Lung Sounds to be measured according at least one-level singular value receiving
Levy vector.
Specifically, by L first order singular value σ1, σ2..., σL(σ1≥σ2≥…≥σr> σr+1=σr+2=σL=0, r <
=L) be added, it is thus achieved that singular value and;Each singular value is compared with singular value and work, it is thus achieved that make than after L ratio form to
Amount is first order characteristic vector, such as: calculate the sum of L first order singular value: S=σ1+σ2+…+σr;The final first order obtaining
Characteristic vector is:
Alternatively, obtained second level characteristic vector by second level singular value, third level singular value is obtained third level feature
Vector, it is thus achieved that second level characteristic vector, the concrete mode of third level characteristic vector are identical with obtaining first order characteristic vector, herein
It will not go into details.
Specifically, can according to first order characteristic vector or first order characteristic vector and second level characteristic vector or
First order characteristic vector and second level characteristic vector and third level characteristic vector build the one-dimensional characteristic of described lungs sound training signal
Vector.It should be noted that the concrete length of characteristic vector according to the specific requirement of accuracy of detection, can be determined.
The characteristic vector of Lung Sounds to be measured is sent into the grader trained for receiving characteristic vector by taxon 34
In differentiate, output differentiate result, it determines result will point out whether rale occur in this Lung Sounds to be measured.Wherein, to be measured
The characteristic vector length of Lung Sounds is consistent with the characteristic vector length of lungs sound training signal.
Alternatively, grader is support vector machine classifier.SVMs is to tie up based on structural risk minimization and VC
Based on Li Lun, select suitable subsets of functions and decision function, so that the practical risk of machine learning reaches to minimize, tool
There is preferable performance, can preferably solve small sample, the problem in the presence of non-linear and high dimensional pattern identification.In this reality
Execute the kernel function choosing linear function in example as SVMs.
Above-described detailed description of the invention, has been carried out to the purpose of the present invention, technical scheme and beneficial effect further
Describe in detail, be it should be understood that the detailed description of the invention that the foregoing is only the present invention, be not intended to limit the present invention
Protection domain, all within the spirit and principles in the present invention, any modification, equivalent substitution and improvement etc. done, all should comprise
Within protection scope of the present invention.
Claims (15)
1. the classifier training method of a rale detection, it is characterised in that described method includes:
Receive lungs sound training signal, and by small echo signal and described lungs sound training signal convolution, carry out wavelet transformation, it is thus achieved that first
Level decomposed signal;
Carry out singular value decomposition to described first order decomposed signal respectively and obtain corresponding first order singular value;
Build the characteristic vector of described lungs sound training signal according at least to described first order singular value;
According to the characteristic vector of described lungs sound training signal, train grader.
2. method according to claim 1, it is characterised in that described method also includes:
Described first order decomposed signal is carried out n times down-sampled, it is thus achieved that the first down-sampled signal, by small echo signal and described first
Down-sampled signal convolution carries out wavelet transformation, it is thus achieved that second level decomposed signal;
Described second level decomposed signal is carried out n times down-sampled, it is thus achieved that the second down-sampled signal, by small echo signal and described second
Down-sampled signal convolution carries out wavelet transformation, it is thus achieved that third level decomposed signal;Wherein, described n is the positive integer more than 1;
Carry out the corresponding second level of singular value decomposition acquisition to described second level decomposed signal and third level decomposed signal respectively strange
Different value and third level singular value.
3. method according to claim 1 and 2, it is characterised in that described small echo signal is rale analog signal.
4. method according to claim 1, it is characterised in that when described first order singular value is for according to described first fraction
Solve signal carry out singular value decomposition obtain corresponding L singular value when, described according at least to the described first order singular value structure institute
State the characteristic vector of lungs sound training signal, comprising:
Described L singular value is added, it is thus achieved that singular value and;
Each described singular value is compared with described singular value and work, it is thus achieved that L ratio, by described L ratio composition first order spy
Levy vector;
Build the characteristic vector of described lungs sound training signal according at least to first order characteristic vector.
5. method according to claim 1 and 2, it is characterised in that described method also includes:
Corresponding L the singular value of singular value decomposition acquisition will be carried out according to described second level decomposed signal strange as the described second level
Different value, and described L singular value is added, it is thus achieved that singular value and;By each described singular value and described singular value with make ratio, obtain
Obtain L ratio, described L ratio is formed corresponding second level characteristic vector;
Corresponding L the singular value of singular value decomposition acquisition will be carried out according to described third level decomposed signal strange as the described third level
Different value, and described L singular value is added, it is thus achieved that singular value and;By each described singular value and described singular value with make ratio, obtain
Obtain L ratio, by corresponding for described L ratio composition third level characteristic vector.
6. the method according to claim the 1st, 4 or 5, it is characterised in that described according at least to described first order singular value structure
Build the characteristic vector of described lungs sound training signal, comprising:
According to described first order characteristic vector or first order characteristic vector and second level characteristic vector or first order feature
Vector sum second level characteristic vector and third level characteristic vector build the characteristic vector of described lungs sound training signal.
7. method according to claim 1, it is characterised in that described grader is support vector machine classifier.
8. a rale detection method, it is characterised in that described method includes:
Receive Lung Sounds to be measured, and small echo signal and described Lung Sounds convolution to be measured are carried out wavelet transformation, it is thus achieved that first
Level decomposed signal;
Carry out singular value decomposition to described first order decomposed signal and obtain corresponding first order singular value;
Build the characteristic vector of described Lung Sounds to be measured according at least to described first order singular value;
Sending into the characteristic vector of described Lung Sounds to be measured in the grader trained and differentiating, output differentiates result.
9. method according to claim 8, it is characterised in that described method also includes:
Described first order decomposed signal is carried out n times down-sampled, it is thus achieved that the first down-sampled signal, by small echo signal and described first
Down-sampled signal convolution carries out wavelet transformation, it is thus achieved that second level decomposed signal;
Described second level decomposed signal is carried out n times down-sampled, it is thus achieved that the second down-sampled signal, by small echo signal and described second
Down-sampled signal convolution carries out wavelet transformation, it is thus achieved that third level decomposed signal;Wherein, described n is the positive integer more than 1;
Carry out the corresponding second level of singular value decomposition acquisition to described second level decomposed signal and third level decomposed signal respectively strange
Different value and third level singular value.
10. method according to claim 8 or claim 9, it is characterised in that described small echo signal is rale analog signal.
11. methods according to claim 8, it is characterised in that when described first order singular value is for according to the described first order
When decomposed signal carries out singular value decomposition acquisition corresponding L singular value, described according at least to described first order singular value structure
The characteristic vector of described lungs sound training signal, comprising:
Described L singular value is added, it is thus achieved that singular value and;
Each described singular value is compared with described singular value and work, it is thus achieved that L ratio, by described L ratio composition first order spy
Levy vector;
Build the characteristic vector of described lungs sound training signal according at least to first order characteristic vector.
12. methods according to claim 8 or claim 9, it is characterised in that described method also includes:
Corresponding L the singular value of singular value decomposition acquisition will be carried out according to described second level decomposed signal strange as the described second level
Different value, and described L singular value is added, it is thus achieved that singular value and;By each described singular value and described singular value with make ratio, obtain
Obtain L ratio, described L ratio is formed corresponding second level characteristic vector;
Corresponding L the singular value of singular value decomposition acquisition will be carried out according to described third level decomposed signal strange as the described third level
Different value, and described L singular value is added, it is thus achieved that singular value and;By each described singular value and described singular value with make ratio, obtain
Obtain L ratio, by corresponding for described L ratio composition third level characteristic vector.
Method described in 13. according to Claim 8,11 or 12, it is characterised in that described according at least to described first order singular value
Build the characteristic vector of described Lung Sounds to be measured, comprising:
According to described first order characteristic vector or first order characteristic vector and second level characteristic vector or first order feature
Vector sum second level characteristic vector and third level characteristic vector build the characteristic vector of described lungs sound training signal.
14. methods according to claim 8, it is characterised in that described grader is support vector machine classifier.
15. 1 kinds of sieve sound detection devices, it is characterised in that described device includes:
Wavelet transform unit, is used for receiving Lung Sounds to be measured, carries out little by small echo signal with described Lung Sounds convolution to be measured
Wave conversion, it is thus achieved that at least level of decomposition signal;
Singular value decomposition unit, be used for receiving described at least level of decomposition signal, and respectively to described at least level of decomposition signal
Carry out singular value decomposition and obtain corresponding at least one-level singular value;
Characteristic vector construction unit, for build according to described at least one-level singular value the feature of described Lung Sounds to be measured to
Amount;
Taxon, differentiates for sending into the characteristic vector of described Lung Sounds to be measured in the grader trained, defeated
Go out to differentiate result.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610390420.4A CN106073709A (en) | 2016-06-03 | 2016-06-03 | A kind of method and apparatus of rale detection |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610390420.4A CN106073709A (en) | 2016-06-03 | 2016-06-03 | A kind of method and apparatus of rale detection |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106073709A true CN106073709A (en) | 2016-11-09 |
Family
ID=57447323
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610390420.4A Pending CN106073709A (en) | 2016-06-03 | 2016-06-03 | A kind of method and apparatus of rale detection |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106073709A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110970042A (en) * | 2019-12-13 | 2020-04-07 | 苏州美糯爱医疗科技有限公司 | Artificial intelligent real-time classification method, system and device for pulmonary rales of electronic stethoscope and readable storage medium |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1463675A (en) * | 2002-06-28 | 2003-12-31 | 香港中文大学 | Stethoscope sound plug capable of separating cardiopulmonary murmur and stethoscope containing the sound plug |
CN2865564Y (en) * | 2006-01-18 | 2007-02-07 | 胡晓兵 | Respirotory detecting probe and respirotory detector |
WO2010044452A1 (en) * | 2008-10-16 | 2010-04-22 | 国立大学法人長崎大学 | Information judgment aiding method, sound information judging method, sound information judgment aiding device, sound information judging device, sound information judgment aiding system, and program |
CN101766493A (en) * | 2010-02-09 | 2010-07-07 | 杭州电子科技大学 | Portable and visual Bluetooth electronic stethoscope capable of separating heart and lung sounds |
CN103932733A (en) * | 2014-04-11 | 2014-07-23 | 中国人民解放军第三军医大学第三附属医院 | Digitalized detecting analysis method for pulmonary interstitial fibrosis based on lung sound |
CN104382614A (en) * | 2014-12-09 | 2015-03-04 | 重庆邮电大学 | Heart and lung sound signal separation method and device based on fast ICA (independent component analysis) |
-
2016
- 2016-06-03 CN CN201610390420.4A patent/CN106073709A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1463675A (en) * | 2002-06-28 | 2003-12-31 | 香港中文大学 | Stethoscope sound plug capable of separating cardiopulmonary murmur and stethoscope containing the sound plug |
CN2865564Y (en) * | 2006-01-18 | 2007-02-07 | 胡晓兵 | Respirotory detecting probe and respirotory detector |
WO2010044452A1 (en) * | 2008-10-16 | 2010-04-22 | 国立大学法人長崎大学 | Information judgment aiding method, sound information judging method, sound information judgment aiding device, sound information judging device, sound information judgment aiding system, and program |
CN101766493A (en) * | 2010-02-09 | 2010-07-07 | 杭州电子科技大学 | Portable and visual Bluetooth electronic stethoscope capable of separating heart and lung sounds |
CN103932733A (en) * | 2014-04-11 | 2014-07-23 | 中国人民解放军第三军医大学第三附属医院 | Digitalized detecting analysis method for pulmonary interstitial fibrosis based on lung sound |
CN104382614A (en) * | 2014-12-09 | 2015-03-04 | 重庆邮电大学 | Heart and lung sound signal separation method and device based on fast ICA (independent component analysis) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110970042A (en) * | 2019-12-13 | 2020-04-07 | 苏州美糯爱医疗科技有限公司 | Artificial intelligent real-time classification method, system and device for pulmonary rales of electronic stethoscope and readable storage medium |
WO2021114761A1 (en) * | 2019-12-13 | 2021-06-17 | 苏州美糯爱医疗科技有限公司 | Lung rale artificial intelligence real-time classification method, system and device of electronic stethoscope, and readable storage medium |
CN110970042B (en) * | 2019-12-13 | 2023-04-18 | 苏州美糯爱医疗科技有限公司 | Pulmonary ralated artificial intelligence real-time classification method, system and device of electronic stethoscope and readable storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN102697493B (en) | Method for rapidly and automatically identifying and removing ocular artifacts in electroencephalogram signal | |
CN104523266B (en) | A kind of electrocardiosignal automatic classification method | |
CN110786850B (en) | Electrocardiosignal identity recognition method and system based on multi-feature sparse representation | |
CN105030279B (en) | A kind of tissue characterization method based on ultrasonic radio frequency time series | |
Zhang et al. | Perturbation and nonlinear dynamic analyses of voices from patients with unilateral laryngeal paralysis | |
CN105825516B (en) | A kind of DICOM images blood flow analysis system | |
CN106328150A (en) | Bowel sound detection method, device and system under noisy environment | |
Quiceno-Manrique et al. | Detection of obstructive sleep apnea in ECG recordings using time-frequency distributions and dynamic features | |
CN102512158A (en) | Electrocardiosignal quality evaluation method and device based on high-dimensional fuzzy recognition | |
CN105411565A (en) | Heart rate variability feature classification method based on generalized scale wavelet entropy | |
CN104473631B (en) | A kind of based on non-negative blind separation Fetal ECG instantaneous heart rate recognition methods and system | |
CN109330584B (en) | Electrocardiosignal identity recognition method and system based on dictionary learning and sparse representation | |
CN101751555A (en) | Deformation fingerprint identification method and system | |
CN109589114A (en) | Myoelectricity noise-eliminating method based on CEEMD and interval threshold | |
CN105147252A (en) | Heart disease recognition and assessment method | |
CN105448291A (en) | Parkinsonism detection method and detection system based on voice | |
CN102258368B (en) | Time-domain sparsity linear aliasing blind separation model discrimination method in fetal electrocardiogram detection | |
CN110786849B (en) | Electrocardiosignal identity recognition method and system based on multi-view discriminant analysis | |
CN113273978B (en) | Ultra-wideband radar-based human body respiration and heartbeat frequency detection method | |
CN103637821A (en) | Hepatic fibrosis degree Fisher identification method based on ultrasonic radio frequency (RF) time sequence | |
CN107346541A (en) | A kind of tissue characterization method based on ultrasonic radio frequency time series wavelet analysis | |
CN105662454A (en) | Rale detection method and device | |
CN104622446A (en) | Heart rate variability signal optimization method based on KHM clustering algorithm | |
CN114492519B (en) | Lung ultrasound special sign B line identification and classification method based on ultrasound echo radio frequency signals | |
CN106236075B (en) | A kind of noise-reduction method applied to portable electrocardiograph institute thought-read electrograph |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
AD01 | Patent right deemed abandoned |
Effective date of abandoning: 20200310 |
|
AD01 | Patent right deemed abandoned |