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
- signal
- stage
- singular value
- decomposition
- feature vector
- 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
- 208000037656 Respiratory Sounds Diseases 0.000 title claims abstract description 112
- 238000000034 method Methods 0.000 title claims abstract description 58
- 238000001514 detection method Methods 0.000 title claims abstract description 31
- 206010037833 rales Diseases 0.000 title claims abstract description 15
- 239000013598 vector Substances 0.000 claims abstract description 143
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 142
- 230000009466 transformation Effects 0.000 claims abstract description 33
- 238000012549 training Methods 0.000 claims description 57
- 230000005236 sound signal Effects 0.000 claims description 42
- 238000012706 support-vector machine Methods 0.000 claims description 18
- 238000005070 sampling Methods 0.000 claims description 14
- 238000010276 construction Methods 0.000 claims description 3
- 230000006870 function Effects 0.000 description 11
- CIWBSHSKHKDKBQ-JLAZNSOCSA-N Ascorbic acid Chemical compound OC[C@H](O)[C@H]1OC(=O)C(O)=C1O CIWBSHSKHKDKBQ-JLAZNSOCSA-N 0.000 description 10
- 230000008569 process Effects 0.000 description 10
- 239000011159 matrix material Substances 0.000 description 9
- 210000004072 lung Anatomy 0.000 description 8
- ZZZCUOFIHGPKAK-UHFFFAOYSA-N D-erythro-ascorbic acid Natural products OCC1OC(=O)C(O)=C1O ZZZCUOFIHGPKAK-UHFFFAOYSA-N 0.000 description 5
- 229930003268 Vitamin C Natural products 0.000 description 5
- 235000019154 vitamin C Nutrition 0.000 description 5
- 239000011718 vitamin C Substances 0.000 description 5
- 230000007613 environmental effect Effects 0.000 description 4
- 238000010801 machine learning Methods 0.000 description 4
- 230000002159 abnormal effect Effects 0.000 description 3
- 238000013459 approach Methods 0.000 description 3
- 238000002555 auscultation Methods 0.000 description 3
- 238000003745 diagnosis Methods 0.000 description 3
- 201000010099 disease Diseases 0.000 description 3
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 210000001035 gastrointestinal tract Anatomy 0.000 description 3
- 238000012886 linear function Methods 0.000 description 3
- 238000010606 normalization Methods 0.000 description 3
- 238000003909 pattern recognition Methods 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 206010035664 Pneumonia Diseases 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 208000023504 respiratory system disease Diseases 0.000 description 2
- 238000004088 simulation Methods 0.000 description 2
- 208000006545 Chronic Obstructive Pulmonary Disease Diseases 0.000 description 1
- 208000019693 Lung disease Diseases 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000003915 air pollution Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 208000006673 asthma Diseases 0.000 description 1
- 206010006451 bronchitis Diseases 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000010355 oscillation Effects 0.000 description 1
- 230000001575 pathological effect Effects 0.000 description 1
- 230000007170 pathology Effects 0.000 description 1
- 230000035479 physiological effects, processes and functions Effects 0.000 description 1
- 238000011895 specific detection Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
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
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Pathology (AREA)
- Heart & Thoracic Surgery (AREA)
- Psychiatry (AREA)
- Signal Processing (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Biophysics (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Physiology (AREA)
- Medical Informatics (AREA)
- Molecular Biology (AREA)
- Surgery (AREA)
- Animal Behavior & Ethology (AREA)
- General Health & Medical Sciences (AREA)
- Public Health (AREA)
- Veterinary Medicine (AREA)
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
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 invention relates to the technical field of biomedical signal processing, in particular to a method and a device for detecting a Roots tone.
Background
Respiratory diseases are common diseases, such as asthma, Chronic Obstructive Pulmonary Disease (COPD), pneumonia, bronchitis, and the like. The prevalence of respiratory diseases has continued to increase in recent years with air pollution and other environmental factors, and lung disease diagnosis has received increased attention. The lung respiratory sound contains a great deal of information and is an important index reflecting the physiology and pathology of the lung. Among them, Luo Yin is a common abnormal sound of lung and is closely related to diseases such as pneumonia. At present, clinical medicine mainly relies on auscultation for the diagnosis of the Roots, which depends on the auscultation technique and clinical experience of doctors to a great extent, and has strong subjectivity. Compared with the traditional auscultation, the research of a non-invasive objective lung abnormal sound detection technology with high precision is particularly important.
At present, more and more researchers at home and abroad are dedicated to lung sound signal research, lung sounds are analyzed through a signal processing technology, physiological and pathological information in the lung sounds are extracted, and a classification system is established to accurately quantify normal and abnormal lung sounds, so that more bases are provided for lung state assessment and disease diagnosis. At present, the characteristic (power, fractal dimension, etc.) of the rale can be extracted through time-frequency analysis (short-time Fourier transform, wavelet decomposition, etc.) by adopting a signal processing technology to detect the rale, and a classifier is trained through a machine learning or neural network method, so that the rale is identified.
Wavelet analysis is widely applied to time-frequency analysis due to variable time-frequency resolution, wavelet transform (WT for short) is compared with Fourier transform, wavelet transform is local analysis of time (space) frequency, signals (functions) are gradually refined in a multi-scale mode through telescopic translation operation, and finally the requirements of time subdivision at high frequency and frequency subdivision at low frequency can be met automatically to adapt to time-frequency signal analysis, so that any details of signals can be focused, the problem of difficulty of Fourier transform is solved, but due to the particularity of a Roots waveform, a proper wavelet is not found at present to be applied to Roots detection, and therefore detection accuracy is influenced.
Disclosure of Invention
In order to solve the above problem, in a first aspect, an embodiment of the present invention provides a classifier training method for rhone detection, where the method includes: receiving a lung sound training signal, convolving the wavelet signal with the received lung sound training signal, and performing wavelet transformation to obtain a first-level decomposition signal; carrying out singular value decomposition on the first-stage decomposition signal to obtain a corresponding first-stage singular value; constructing a feature vector of the lung sound training signal at least according to the first-level singular value; and training a classifier through the feature vector of the lung sound training signal.
Optionally, the method further includes: performing n-time down-sampling on the first-level decomposition signal to obtain a first down-sampled signal, and performing wavelet transformation on a convolution of a wavelet signal and the first down-sampled signal to obtain a second-level decomposition signal; wherein n is a positive integer greater than 1, such as 2; performing n-time down-sampling on the second-level decomposition signal to obtain a second down-sampled signal, and performing wavelet transformation on the convolution of the wavelet signal and the second down-sampled signal to obtain a third-level decomposition signal; and respectively carrying out singular value decomposition on the second-stage decomposition signal and the third-stage decomposition signal to obtain a corresponding second-stage singular value and a corresponding third-stage singular value.
Optionally, in the above method, the wavelet signal is a rhone analog signal.
Optionally, in the above method, when the first-stage singular value is L singular values obtained by performing singular value decomposition on the first-stage decomposed signal, the step of constructing the feature vector of the lung sound training signal at least according to the first-stage singular value includes: adding the L singular values to obtain a singular value sum; comparing each singular value with the singular value sum to obtain L specific values, and forming the L specific values into a first-stage feature vector; and constructing a feature vector of the lung sound training signal at least according to the first-stage feature vector.
Optionally, the method further includes: performing singular value decomposition according to the second-stage decomposition signal to obtain corresponding L singular values serving as second-stage singular values, and adding the L singular values to obtain singular value sums; comparing each singular value with the singular value sum to obtain L specific values, and forming the L specific values into corresponding second-stage feature vectors; performing singular value decomposition according to the third-level decomposition signal to obtain corresponding L singular values serving as third-level singular values, and adding the L singular values to obtain singular value sums; and comparing each singular value with the singular value sum to obtain L ratios, and forming the L ratios into corresponding third-stage feature vectors.
Optionally, in the above method, the step of constructing a feature vector of the lung sound training signal according to at least the first-level singular value includes: and constructing a feature vector of the lung sound training signal according to the first-stage feature vector, or the first-stage feature vector and the second-stage feature vector, or the first-stage feature vector, the second-stage feature vector and the third-stage feature vector.
Optionally, in the above method, the classifier is a support vector machine classifier.
In a second aspect, an embodiment of the present invention provides a method for detecting a rale, where the method includes: receiving a lung sound signal to be detected, convolving the wavelet signal with the lung sound signal to be detected, and performing wavelet transformation to obtain a first-level decomposition signal; carrying out singular value decomposition on the first-stage decomposition signal to obtain a corresponding first-stage singular value; constructing a feature vector of the lung sound signal to be detected at least according to the first-level singular value; and (5) sending the feature vector of the lung sound signal to be detected into a trained classifier for discrimination, and outputting a discrimination result.
Optionally, the method further includes: performing n-time down-sampling on the first-level decomposition signal to obtain a first down-sampled signal, and performing wavelet transformation on a convolution of a wavelet signal and the first down-sampled signal to obtain a second-level decomposition signal; performing n-time down-sampling on the second-level decomposition signal to obtain a second down-sampled signal, and performing wavelet transformation on the convolution of the wavelet signal and the second down-sampled signal to obtain a third-level decomposition signal; wherein n is a positive integer greater than 1; and respectively carrying out singular value decomposition on the second-stage decomposition signal and the third-stage decomposition signal to obtain a corresponding second-stage singular value and a corresponding third-stage singular value.
Optionally, in the above method, the wavelet signal is a rhone analog signal.
Optionally, in the above method, when the first-stage singular value is L singular values obtained by performing singular value decomposition on the first-stage decomposed signal, the step of constructing the feature vector of the lung sound training signal at least according to the first-stage singular value includes: adding the L singular values to obtain a singular value sum; comparing each singular value with the singular value sum to obtain L specific values, and forming the L specific values into a first-stage feature vector; and constructing a feature vector of the lung sound training signal at least according to the first-stage feature vector.
Optionally, the method further includes performing singular value decomposition on the second-stage decomposed signal to obtain corresponding L singular values as second-stage singular values, and adding the L singular values to obtain a singular value sum; comparing each singular value with the singular value sum to obtain L specific values, and forming the L specific values into corresponding second-stage feature vectors; performing singular value decomposition according to the third-level decomposition signal to obtain corresponding L singular values serving as third-level singular values, and adding the L singular values to obtain singular value sums; and comparing each singular value with the singular value sum to obtain L ratios, and forming the L ratios into corresponding third-stage feature vectors.
Optionally, in the above method, the step of constructing a feature vector of the lung sound signal to be detected at least according to the first-level singular value includes: and constructing a feature vector of the lung sound training signal according to the first-stage feature vector, or the first-stage feature vector and the second-stage feature vector, or the first-stage feature vector, the second-stage feature vector and the third-stage feature vector.
Optionally, in the above method, the classifier is a support vector machine classifier.
In a third aspect, an embodiment of the present invention provides a rhone detection apparatus, including: the wavelet transformation unit is used for receiving the lung sound signal to be detected, and performing wavelet transformation on the convolution of the wavelet signal and the lung sound signal to be detected to obtain at least one level of decomposition signal; the singular value decomposition unit is used for receiving at least one stage of decomposition signals and respectively carrying out singular value decomposition on the at least one stage of decomposition signals to obtain at least one stage of corresponding singular values; the characteristic vector construction unit is used for constructing a characteristic vector of the lung sound signal to be detected according to the at least one stage of singular value; the classification unit is used for sending the feature vector of the lung sound signal to be detected into the trained classifier for discrimination and outputting a discrimination result.
According to the method and the device for detecting the Roots tone provided by the embodiment of the invention, the Roots tone analog signal is used as a base wavelet, the wavelet decomposition is firstly carried out on the lung tone signal, the characteristic waveform of the Lung tone signal is extracted, then the singular value decomposition is carried out on the decomposed signal, the singular value is formed into the characteristic vector, the support vector machine is selected to train the classifier, and whether the Lung tone contains the Roots tone or not is detected. The method has high detection accuracy rate which can reach 100 percent, and has good application prospect.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained according to these drawings without creative efforts.
Fig. 1 is a classifier training method for rhone detection according to an embodiment of the present invention;
fig. 2 is a flowchart of a method for detecting a rhone according to an embodiment of the present invention;
fig. 3 is a Roots detection apparatus according to an embodiment of the present invention.
Detailed Description
The technical solution of the present invention is further described in detail by the accompanying drawings and embodiments.
The Roots have obvious characteristics in the time-frequency domain: in the time domain, the rale starts with a sudden sharp oscillation and then gradually expands, and the duration is generally less than 20 ms; in the frequency domain, the frequency is typically 150-1800Hz, which is greatly different from the frequency of normal lung sounds.
The detection process of the Roots comprises a training process and a recognition process. The training process comprises the steps of receiving a plurality of training lung sound signals of normal lung sound and Luo sound, extracting characteristic information of the training lung sound signals, constructing a characteristic vector of the training lung sound signals, and training a classifier by using the characteristic vector; the identification process comprises the steps of receiving the lung sounds to be detected, extracting the characteristic information of the lung sounds to be detected, constructing the characteristic vector of the lung sound signals to be detected, classifying and identifying the lung sounds to be detected according to the characteristic vector of the lung sounds to be detected by using the classifier obtained in the training process, and outputting the identification result.
Fig. 1 is a classifier training method for rhone detection according to an embodiment of the present invention, as shown in fig. 1, the method includes steps S101 to S104:
step S101, receiving a lung sound training signal, convolving a wavelet signal with the received lung sound training signal, and performing wavelet transformation to obtain a first-level decomposition signal;
optionally, "step S101" further includes: performing n-time down-sampling on the first-level decomposition signal to obtain a first down-sampled signal, and performing wavelet transformation on a convolution of a wavelet signal and the first down-sampled signal to obtain a second-level decomposition signal; performing n-time down-sampling on the second-level decomposition signal to obtain a second down-sampled signal, and performing wavelet transformation on the convolution of the wavelet signal and the second down-sampled signal to obtain a third-level decomposition signal; wherein n is a positive integer greater than 1, and in this embodiment n is 2.
Optionally, the lung sound training signal is wavelet-transformed by using a rale analog signal as a wavelet signal, and the rale waveform analog formula is as follows:
wherein, a1＝0.5，a2＝1.49，a3＝0.78，f0＝2.0。
The received lung sound training signal is accompanied by noise signals such as heart sounds, digestive tract sounds, and environmental noises. The basic wavelet signal is a Roots sound analog signal which approaches to a real Roots sound waveform to the maximum extent, so that the lung sound training signal is decomposed based on the Roots sound analog signal, and the decomposition effect of the basic wavelet signal is superior to that of general wavelet transformation. It should be noted that the rhone analog signal is not a wavelet in a strict sense, and it does not satisfy the normalization condition of the wavelet.
And S102, carrying out singular value decomposition on the first-stage decomposition signal to obtain a corresponding first-stage singular value.
Specifically, singular value decomposition is performed on the first-stage decomposed signal to obtain corresponding L singular values sigma1，σ2，…，σLIs a first order singular value, where σ1≥σ2≥…≥σr＞σr+1＝σr+2＝σL＝0(r<＝L)。
Optionally, "step S102" further includes: the specific manner of obtaining the second-level singular value and the third-level singular value is the same as that of obtaining the first-level singular value, and details thereof are omitted here.
Optionally, in the embodiment of the present invention, the length L of the singular value is 20. It should be noted that the specific length of the singular value may be determined according to the specific requirement of the detection accuracy.
Singular Value Decomposition (SVD) is an important matrix Decomposition in linear algebra, is a generalization of unitary diagonalization of a normal matrix in matrix analysis, and is one of the most basic and important tools in modern numerical analysis.
And S103, constructing a feature vector of the lung sound training signal at least according to the first-level singular value.
Specifically, adding the L first-level singular values to obtain a singular value sum; comparing each singular value with a singularityValue and proportion are calculated, a vector formed by L ratios after the proportion is used as a first-level feature vector, and for more accurately describing the working principle and process of the first-level feature vector, for example, the following description is given: calculating the sum of L first-level singular values: s ═ σ -1+σ2+…+σr(ii) a The finally obtained first-stage feature vector is:
optionally, "step S103" further includes: performing singular value decomposition according to the second-stage decomposition signal to obtain corresponding L singular values serving as second-stage singular values, and adding the L singular values to obtain singular value sums; comparing each singular value with the singular value sum to obtain L specific values, and forming the L specific values into corresponding second-stage feature vectors; performing singular value decomposition according to the third-level decomposition signal to obtain corresponding L singular values serving as third-level singular values, and adding the L singular values to obtain singular value sums; and comparing each singular value with the singular value sum to obtain L ratios, and forming the L ratios into corresponding third-stage feature vectors.
Optionally, the feature vector of the lung sound training signal may be constructed according to a first-level feature vector, or the first-level feature vector and a second-level feature vector, or the first-level feature vector and the second-level feature vector and a third-level feature vector. It should be noted that, the specific length of the feature vector may be determined according to the specific requirement of the detection accuracy.
And step S104, training a classifier according to the feature vector of the lung sound training signal.
Optionally, the classifier is a Support Vector Machine (SVM) classifier. The support vector machine is based on the structure risk minimization and VC Dimension (Vapnik-Chervonenkis Dimension) theory, and selects a proper function subset and a decision function, so that the actual risk of machine learning is minimized, the support vector machine has better performance, and the problems existing in small sample, non-linear and high-dimensional pattern recognition can be better solved. In the embodiment, a linear function is selected as a kernel function of the support vector machine, and the classifier is trained.
It should be noted that, in a specific training process, the adaptive wavelet decomposition times can be selected according to the specific requirement of the detection accuracy to obtain at least one level of decomposition signals, such as: performing wavelet transformation on the convolution of the Roots simulation signal and the lung sound training signal to obtain a first-level decomposition signal, performing singular value decomposition on the first-level decomposition signal to obtain a first-level singular value, obtaining a first-level feature vector according to the first-level singular value, constructing a feature vector of the lung sound signal according to the first-level feature vector, and training a classifier according to the feature vector.
After the classifier is trained by the above method, the trained classifier can be used to perform the speech detection and recognition, and the following embodiment describes a speech detection method.
Fig. 2 is a flowchart of a method for detecting a rhone according to an embodiment of the present invention. As shown in fig. 2, the method includes the following steps S201 to S204:
step S201, receiving the lung sound signal to be detected, and performing wavelet transformation by convolving the wavelet signal with the lung sound signal to be detected to obtain a first-level decomposition signal.
Optionally, "step S201" further includes: carrying out n-time down-sampling on the first-level decomposition signal to obtain a first down-sampled signal, and carrying out wavelet transformation on the Roots analog signal and the convolution of the first down-sampled signal to obtain a second-level decomposition signal; performing n-time down-sampling on the second-level decomposition signal to obtain a second down-sampled signal, and performing wavelet transformation on the Roots analog signal and the convolution of the second down-sampled signal to obtain a third-level decomposition signal; wherein n is a positive integer greater than 1, and in this embodiment n is 2.
Optionally, the lung sound training signal is wavelet-transformed by using a rale analog signal as a wavelet signal, and the rale waveform analog formula is as follows:
wherein, a1＝0.5，a2＝1.49，a3＝0.78，f0＝2.0。
The received lung sound training signal is accompanied by noise signals such as heart sounds, digestive tract sounds, and environmental noises. The basic wavelet signal is a Roots sound analog signal which approaches to a real Roots sound waveform to the maximum extent, so that the lung sound training signal is decomposed based on the Roots sound analog signal, and the decomposition effect of the basic wavelet signal is superior to that of general wavelet transformation. It should be noted that the rhone analog signal is not a wavelet in a strict sense, and it does not satisfy the normalization condition of the wavelet.
Step S202, singular value decomposition is carried out on the first-stage decomposition signal to obtain a corresponding first-stage singular value.
Specifically, singular value decomposition is performed on the first-stage decomposed signal to obtain corresponding L singular values sigma1，σ2，…，σLIs a first order singular value, where σ1≥σ2≥…≥σr＞σr+1＝σr+2＝σL＝0(r<＝L)。
Optionally, "step S102" further includes: the specific manner of obtaining the second-level singular value and the third-level singular value is the same as that of obtaining the first-level singular value, and details thereof are omitted here. Optionally, in the embodiment of the present invention, the length L of the singular value is 20. It should be noted that the specific length of the singular value may be determined according to the specific requirement of the detection accuracy.
Singular Value Decomposition (SVD) is an important matrix Decomposition in linear algebra, is a generalization of unitary diagonalization of a normal matrix in matrix analysis, and is one of the most basic and important tools in modern numerical analysis.
And step S203, constructing a feature vector of the lung sound signal to be detected at least according to the first-level singular value.
Specifically, adding the L first-level singular values to obtain a singular value sum; comparing each singular value with the singular value sum, and taking a vector formed by the L compared ratios as a first-stage feature vector, for example, to describe the working principle and process of the first-stage feature vector more accurately, such as: calculating the sum of L first-level singular values: s ═ σ -1+σ2+…+σr(ii) a The finally obtained first-stage feature vector is:
optionally, "step S103" further includes: performing singular value decomposition according to the second-stage decomposition signal to obtain corresponding L singular values serving as second-stage singular values, and adding the L singular values to obtain singular value sums; comparing each singular value with the singular value sum to obtain L specific values, and forming the L specific values into corresponding second-stage feature vectors; performing singular value decomposition according to the third-level decomposition signal to obtain corresponding L singular values serving as third-level singular values, and adding the L singular values to obtain singular value sums; and comparing each singular value with the singular value sum to obtain L ratios, and forming the L ratios into corresponding third-stage feature vectors.
Specifically, the feature vector of the lung sound signal to be detected may be constructed according to a first-level feature vector, or the first-level feature vector and a second-level feature vector, or the first-level feature vector, the second-level feature vector and a third-level feature vector. It should be noted that, the specific length of the feature vector may be determined according to the specific requirement of the detection accuracy.
And step S204, sending the characteristic vector of the lung sound signal to be detected into a trained classifier for classification and identification, and outputting an identification result which indicates whether the lung sound signal to be detected has a Roots sound.
Optionally, the classifier is a support vector machine classifier. The support vector machine is based on the structure risk minimization and VC (vitamin C) dimension theory, and selects a proper function subset and a decision function, so that the actual risk of machine learning is minimized, the support vector machine has better performance, and the problems existing in small sample, non-linear and high-dimensional pattern recognition can be better solved. In the present embodiment, a linear function is selected as the kernel function of the support vector machine.
It should be noted that the above embodiment is only a specific implementation manner of the technical solution of the present invention, and does not limit the present invention, and in the specific identification process, an appropriate wavelet function and wavelet decomposition frequency may be selected according to the specific requirement of the detection accuracy to obtain at least one level of decomposition signal, such as: performing wavelet transformation on the convolution of the Roots simulation signal and the lung sound training signal to obtain a first-level decomposition signal, performing singular value decomposition on the first-level decomposition signal to obtain a first-level singular value, obtaining a first-level feature vector according to the first-level singular value, constructing a feature vector of the lung sound signal to be detected according to the first-level feature vector, sending the feature vector into a trained classifier for classification and identification, and outputting an identification result. And the length of the feature vector of the lung sound signal to be detected is consistent with the length of the feature vector of the lung sound training signal.
Correspondingly, the embodiment of the invention provides a detection device corresponding to the detection method in the embodiment.
Fig. 3 is a Roots detection apparatus according to an embodiment of the present invention, and as shown in fig. 3, the apparatus 3 includes: a wavelet transformation unit 31, a singular value decomposition unit 32, a feature vector construction unit 33, and a classification unit 34; wherein,
the wavelet transformation unit 31 is configured to receive a lung sound signal to be detected, perform wavelet transformation on the convolution of the wavelet signal and the lung sound signal to be detected, and obtain a one-level or multi-level decomposition signal.
Specifically, the wavelet transform unit 31 determines the specific number of levels of the decomposed signal to be acquired according to the specific detection precision requirement;
if only the first-level decomposition signal is obtained, the wavelet transform unit 31 performs wavelet transform on the Roots analog signal and the lung sound signal to be detected by convolution after receiving the lung sound signal to be detected, so as to obtain a first-level decomposition signal;
if a multi-level decomposition signal is obtained, such as three levels (a first-level decomposition signal, a second-level decomposition signal, and a third-level decomposition signal), the wavelet transform unit 31 performs wavelet transform by convolving the wavelet signal with the lung sound signal to be detected after receiving the lung sound signal to be detected, so as to obtain a first-level decomposition signal; performing n-time down-sampling on the first-level decomposition signal to obtain a first down-sampled signal, and performing wavelet transformation on a convolution of a wavelet signal and the first down-sampled signal to obtain a second-level decomposition signal; performing n-time down-sampling on the second-level decomposition signal to obtain a second down-sampled signal, and performing wavelet transformation on the convolution of the wavelet signal and the second down-sampled signal to obtain a third-level decomposition signal; wherein n is a positive integer greater than 1, and in this embodiment n is 2.
Optionally, the lung sound training signal is wavelet-transformed by using a rale analog signal as a wavelet signal, and the rale waveform analog formula is as follows:
wherein, a1＝0.5，a2＝1.49，a3＝0.78，f0＝2.0。
The received lung sound training signal is accompanied by noise signals such as heart sounds, digestive tract sounds, and environmental noises. The basic wavelet signal is a Roots sound analog signal which approaches to a real Roots sound waveform to the maximum extent, so that the lung sound training signal is decomposed based on the Roots sound analog signal, and the decomposition effect of the basic wavelet signal is superior to that of general wavelet transformation. It should be noted that the rhone analog signal is not a wavelet in a strict sense, and it does not satisfy the normalization condition of the wavelet.
A singular value decomposition unit 32, configured to receive the one-stage or multi-stage decomposed signals, and perform singular value decomposition on the one-stage or multi-stage decomposed signals respectively to obtain corresponding one-stage or multi-stage singular values; and if the first-stage decomposition signal, the second-stage decomposition signal and the third-stage decomposition signal are received, performing singular value decomposition on the first-stage decomposition signal to obtain a corresponding first-stage singular value, performing singular value decomposition on the second-stage decomposition signal to obtain a second-stage singular value, and performing singular value decomposition on the third-stage decomposition signal to obtain a third-stage singular value.
Specifically, singular value decomposition is performed on the first, second or third-level decomposition signal to obtain corresponding L singular values sigma1，σ2，…，σLIs a first, second or third order singular value, where σ1≥σ2≥…≥σr＞σr+1＝σr+2＝σL＝0(r<L). Optionally, in the embodiment of the present invention, the length L of the singular value is 20. It should be noted that the specific length of the singular value may be determined according to the specific requirement of the detection accuracy.
The singular value decomposition is an important matrix decomposition in linear algebra, is a popularization of unitary diagonalization of a normal matrix in matrix analysis, and is one of the most basic and important tools of modern numerical analysis.
The feature vector constructing unit 33 is configured to construct a feature vector of the lung sound signal to be detected according to the received at least one level of singular value.
Specifically, the L first-stage singular values σ are set1，σ2，…，σL(σ1≥σ2≥…≥σr＞σr+1＝σr+2＝σL＝0，r<L) to obtain a singular value sum; and (3) comparing each singular value with the singular value sum to obtain a vector formed by L ratios after comparison as a first-stage feature vector, wherein the first-stage feature vector comprises the following steps: calculating the sum of L first-level singular values: s ═ σ -1+σ2+…+σr(ii) a The finally obtained first-stage feature vector is:
optionally, the second-level feature vector is obtained from the second-level singular value, the third-level feature vector is obtained from the third-level singular value, and the specific manner of obtaining the second-level feature vector and the third-level feature vector is the same as that of obtaining the first-level feature vector, which is not described herein again.
Specifically, the one-dimensional feature vector of the lung sound training signal may be constructed according to a first-level feature vector, or the first-level feature vector and a second-level feature vector, or the first-level feature vector, the second-level feature vector and a third-level feature vector. It should be noted that, the specific length of the feature vector may be determined according to the specific requirement of the detection accuracy.
And the classification unit 34 is used for receiving the feature vector, sending the feature vector of the lung sound signal to be detected into a trained classifier for judgment, and outputting a judgment result, wherein the judgment result indicates whether the lung sound signal to be detected has the Roots or not. And the length of the feature vector of the lung sound signal to be detected is consistent with the length of the feature vector of the lung sound training signal.
Optionally, the classifier is a support vector machine classifier. The support vector machine is based on the structure risk minimization and VC (vitamin C) dimension theory, and selects a proper function subset and a decision function, so that the actual risk of machine learning is minimized, the support vector machine has better performance, and the problems existing in small sample, non-linear and high-dimensional pattern recognition can be better solved. In the present embodiment, a linear function is selected as the kernel function of the support vector machine.
The above-mentioned embodiments are intended to illustrate the objects, technical solutions and advantages of the present invention in further detail, and it should be understood that the above-mentioned embodiments are merely exemplary embodiments of the present invention, and are not intended to limit the scope of the present invention, and any modifications, equivalent substitutions, improvements and the like made within the spirit and principle of the present invention should be included in the scope of the present invention.
Claims (15)
1. A classifier training method for Luo-phone detection, the method comprising:
receiving a lung sound training signal, convolving a wavelet signal with the lung sound training signal, and performing wavelet transformation to obtain a first-level decomposition signal;
respectively carrying out singular value decomposition on the first-stage decomposition signals to obtain corresponding first-stage singular values;
constructing a feature vector of the lung sound training signal at least according to the first-level singular value;
and training a classifier according to the feature vector of the lung sound training signal.
2. The method of claim 1, further comprising:
performing n-time down-sampling on the first-level decomposition signal to obtain a first down-sampled signal, and performing wavelet transformation on a convolution of a wavelet signal and the first down-sampled signal to obtain a second-level decomposition signal;
performing n-time down-sampling on the second-level decomposition signal to obtain a second down-sampled signal, and performing wavelet transformation on a convolution of a wavelet signal and the second down-sampled signal to obtain a third-level decomposition signal; wherein n is a positive integer greater than 1;
and respectively carrying out singular value decomposition on the second-stage decomposition signal and the third-stage decomposition signal to obtain a corresponding second-stage singular value and a corresponding third-stage singular value.
3. A method according to claim 1 or 2, wherein the wavelet signal is a rhone analog signal.
4. The method of claim 1, wherein when the first-level singular values are L singular values obtained by performing singular value decomposition on the first-level decomposed signal, the constructing the feature vector of the lung sound training signal according to at least the first-level singular values comprises:
adding the L singular values to obtain a singular value sum;
comparing each singular value with the singular value sum to obtain L specific values, and forming a first-stage feature vector by the L specific values;
and constructing a feature vector of the lung sound training signal at least according to the first-stage feature vector.
5. The method according to claim 1 or 2, characterized in that the method further comprises:
performing singular value decomposition on the second-stage decomposition signal to obtain corresponding L singular values serving as the second-stage singular values, and adding the L singular values to obtain a singular value sum; comparing each singular value with the singular value sum to obtain L specific values, and forming the L specific values into corresponding second-stage feature vectors;
performing singular value decomposition on the third-stage decomposition signal to obtain corresponding L singular values serving as the third-stage singular values, and adding the L singular values to obtain a singular value sum; and comparing each singular value with the singular value sum to obtain L ratios, and forming the L ratios into corresponding third-stage feature vectors.
6. The method according to claim 1, 4 or 5, wherein said constructing a feature vector of said lung sound training signal at least according to said first-stage singular values comprises:
and constructing the feature vector of the lung sound training signal according to the first-stage feature vector, or the first-stage feature vector and the second-stage feature vector, or the first-stage feature vector, the second-stage feature vector and the third-stage feature vector.
7. The method of claim 1, wherein the classifier is a support vector machine classifier.
8. A method for detecting a rale, the method comprising:
receiving a lung sound signal to be detected, and performing wavelet transformation on a wavelet signal and the convolution of the lung sound signal to be detected to obtain a first-level decomposition signal;
carrying out singular value decomposition on the first-stage decomposition signal to obtain a corresponding first-stage singular value;
constructing a feature vector of the lung sound signal to be detected at least according to the first-level singular value;
and sending the feature vector of the lung sound signal to be detected into a trained classifier for discrimination, and outputting a discrimination result.
9. The method of claim 8, further comprising:
performing n-time down-sampling on the first-level decomposition signal to obtain a first down-sampled signal, and performing wavelet transformation on a convolution of a wavelet signal and the first down-sampled signal to obtain a second-level decomposition signal;
performing n-time down-sampling on the second-level decomposition signal to obtain a second down-sampled signal, and performing wavelet transformation on a convolution of a wavelet signal and the second down-sampled signal to obtain a third-level decomposition signal; wherein n is a positive integer greater than 1;
and respectively carrying out singular value decomposition on the second-stage decomposition signal and the third-stage decomposition signal to obtain a corresponding second-stage singular value and a corresponding third-stage singular value.
10. A method according to claim 8 or 9, wherein the wavelet signal is a rhone analog signal.
11. The method of claim 8, wherein when the first-level singular values are L singular values obtained by performing singular value decomposition on the first-level decomposed signal, the constructing the feature vector of the lung sound training signal according to at least the first-level singular values comprises:
adding the L singular values to obtain a singular value sum;
comparing each singular value with the singular value sum to obtain L specific values, and forming a first-stage feature vector by the L specific values;
and constructing a feature vector of the lung sound training signal at least according to the first-stage feature vector.
12. The method according to claim 8 or 9, characterized in that the method further comprises:
performing singular value decomposition on the second-stage decomposition signal to obtain corresponding L singular values serving as the second-stage singular values, and adding the L singular values to obtain a singular value sum; comparing each singular value with the singular value sum to obtain L specific values, and forming the L specific values into corresponding second-stage feature vectors;
performing singular value decomposition on the third-stage decomposition signal to obtain corresponding L singular values serving as the third-stage singular values, and adding the L singular values to obtain a singular value sum; and comparing each singular value with the singular value sum to obtain L ratios, and forming the L ratios into corresponding third-stage feature vectors.
13. The method according to claim 8, 11 or 12, wherein the constructing the feature vector of the lung sound signal to be tested according to at least the first-stage singular values comprises:
and constructing the feature vector of the lung sound training signal according to the first-stage feature vector, or the first-stage feature vector and the second-stage feature vector, or the first-stage feature vector, the second-stage feature vector and the third-stage feature vector.
14. The method of claim 8, wherein the classifier is a support vector machine classifier.
15. A rhone detection apparatus, the apparatus comprising:
the wavelet transformation unit is used for receiving the lung sound signal to be detected, carrying out wavelet transformation on the convolution of the wavelet signal and the lung sound signal to be detected, and obtaining at least one-level decomposition signal;
the singular value decomposition unit is used for receiving the at least one stage of decomposition signal and respectively carrying out singular value decomposition on the at least one stage of decomposition signal to obtain at least one stage of corresponding singular value;
the feature vector construction unit is used for constructing a feature vector of the lung sound signal to be detected according to the at least one stage of singular value;
and the classification unit is used for sending the feature vector of the lung sound signal to be detected into a trained classifier for discrimination and outputting a discrimination 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 |
|---|---|---|
| Mouawad et al. | Robust detection of COVID-19 in cough sounds: using recurrence dynamics and variable Markov model | |
| CN107657964B (en) | Depression auxiliary detection method and classifier based on acoustic features and sparse mathematics | |
| Rubin et al. | Recognizing abnormal heart sounds using deep learning | |
| Dibazar et al. | Feature analysis for automatic detection of pathological speech | |
| Tsanas et al. | Novel speech signal processing algorithms for high-accuracy classification of Parkinson's disease | |
| Tsanas et al. | Objective automatic assessment of rehabilitative speech treatment in Parkinson's disease | |
| CN110786850B (en) | Electrocardiosignal identity recognition method and system based on multi-feature sparse representation | |
| CN111444747B (en) | Epileptic state identification method based on migration learning and cavity convolution | |
| CN112137620B (en) | Ultra-wideband radar-based human body weak respiration signal detection method | |
| Ulukaya et al. | Overcomplete discrete wavelet transform based respiratory sound discrimination with feature and decision level fusion | |
| Qian et al. | Automatic detection, segmentation and classification of snore related signals from overnight audio recording | |
| Lederman et al. | Alternating diffusion for common manifold learning with application to sleep stage assessment | |
| Reddy et al. | The automatic detection of heart failure using speech signals | |
| Fontes et al. | Classification system of pathological voices using correntropy | |
| Haider et al. | Computerized lung sound based classification of asthma and chronic obstructive pulmonary disease (COPD) | |
| CN103456302A (en) | Emotion speaker recognition method based on emotion GMM model weight synthesis | |
| Majidnezhad et al. | A HMM-based method for vocal fold pathology diagnosis | |
| Huang et al. | Classification of cough sounds using spectrogram methods and a parallel-stream one-dimensional deep convolutional neural network | |
| See et al. | Heart sound classification based on nonlinear time-frequency features | |
| CN106073709A (en) | A kind of method and apparatus of rale detection | |
| CN112052880A (en) | Underwater sound target identification method based on weight updating support vector machine | |
| CN108461092B (en) | Method for analyzing Parkinson's disease voice | |
| He et al. | Automatic evaluation of hypernasality and speech intelligibility for children with cleft palate | |
| Uğuz et al. | Classification of internal carotid artery Doppler signals using fuzzy discrete hidden Markov model | |
| Xu et al. | Voiceprint recognition of Parkinson patients based on deep learning |
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 |