CN113705649B - EMD-SVD feature extraction-based hand tremor detection method and system - Google Patents
EMD-SVD feature extraction-based hand tremor detection method and system Download PDFInfo
- Publication number
- CN113705649B CN113705649B CN202110959839.8A CN202110959839A CN113705649B CN 113705649 B CN113705649 B CN 113705649B CN 202110959839 A CN202110959839 A CN 202110959839A CN 113705649 B CN113705649 B CN 113705649B
- Authority
- CN
- China
- Prior art keywords
- data
- fingertip
- acceleration
- time sequence
- classification
- 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.)
- Active
Links
- 206010044565 Tremor Diseases 0.000 title claims abstract description 109
- 238000001514 detection method Methods 0.000 title claims abstract description 51
- 238000000605 extraction Methods 0.000 title claims abstract description 28
- 230000006870 function Effects 0.000 claims abstract description 69
- 238000000034 method Methods 0.000 claims abstract description 52
- 239000011159 matrix material Substances 0.000 claims abstract description 30
- 238000012545 processing Methods 0.000 claims abstract description 18
- 230000001133 acceleration Effects 0.000 claims description 113
- 238000013145 classification model Methods 0.000 claims description 53
- 238000012549 training Methods 0.000 claims description 35
- 238000012360 testing method Methods 0.000 claims description 20
- 238000000354 decomposition reaction Methods 0.000 claims description 19
- 230000000694 effects Effects 0.000 claims description 15
- 230000004927 fusion Effects 0.000 claims description 7
- 238000007781 pre-processing Methods 0.000 claims description 7
- 230000008569 process Effects 0.000 claims description 6
- 230000009467 reduction Effects 0.000 claims description 6
- 238000012216 screening Methods 0.000 claims description 5
- 238000010801 machine learning Methods 0.000 abstract description 5
- 239000003814 drug Substances 0.000 abstract 1
- 239000013598 vector Substances 0.000 description 7
- 238000004458 analytical method Methods 0.000 description 4
- 210000004247 hand Anatomy 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 230000003247 decreasing effect Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000003672 processing method Methods 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 210000000707 wrist Anatomy 0.000 description 2
- 208000014094 Dystonic disease Diseases 0.000 description 1
- 208000002972 Hepatolenticular Degeneration Diseases 0.000 description 1
- 208000012902 Nervous system disease Diseases 0.000 description 1
- 208000025966 Neurological disease Diseases 0.000 description 1
- 208000018737 Parkinson disease Diseases 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000004888 barrier function Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000002490 cerebral effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000010224 classification analysis Methods 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 208000010118 dystonia Diseases 0.000 description 1
- 238000002567 electromyography Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 230000002981 neuropathic effect Effects 0.000 description 1
- 230000010355 oscillation Effects 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
- 238000000611 regression analysis Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000000638 solvent extraction Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/2431—Multiple classes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2132—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on discrimination criteria, e.g. discriminant analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/251—Fusion techniques of input or preprocessed data
Abstract
A hand tremor detection method and system based on EMD-SVD feature extraction relates to the technical field of machine learning and is used for solving the problem that an existing hand tremor detection system cannot be used for effectively classifying collected hand data rapidly and accurately. The technical key points of the invention include: the method comprises the steps of acquiring a time sequence of hand data by using a hand tremor detection device, performing signal processing based on an EMD method to obtain an eigenmode function in the time sequence, extracting singular values in an eigenmode function matrix by adopting an SVD method to obtain effective characteristics of the hand tremor data, and finally constructing a multi-classification strategy by using an SVM classifier to classify the characteristics so as to achieve the purpose of hand tremor detection. The invention can be applied to clinical medicine to judge whether tremors exist on the hands of patients.
Description
Technical Field
The invention relates to the technical field of machine learning, in particular to a hand tremor detection method and system based on EMD-SVD feature extraction.
Background
Hand tremors can be seen in a variety of neurological disorders such as primary tremors, parkinson's disease, hepatolenticular degeneration, dystonia tremors, peripheral neuropathic tremors, cerebellar tremors, and the like. The frequency and magnitude of tremors vary widely among different diseases, e.g., primary tremor frequencies are typically 8-10Hz, whereas Parkinson's tremor frequencies are typically 4-6Hz. Electromyography has been used in recent years to objectively evaluate tremor frequency and amplitude, but the examination has a certain traumatism, and the development of noninvasive tremor detection means is urgently needed.
The existing tremor detection methods are mainly completed based on special detection devices, such as an acceleration sensor is adopted to acquire tremor information of a tested person in the movement process, or surface electromechanical signals are collected for tremor analysis, and quantitative measurement of tremor signals is realized by the detection methods, but limitations still exist: different equipment or modules are required to be designed for detecting tremor of fingers, wrists and arms by a detection device based on equipment such as an acceleration sensor, a patient needs to wear the detection device to perform daily activities, and using barriers are large; the special instrument for detecting the physiological signals of the tremor patients is generally expensive, and the psychological burden of the patients in the detection process is large.
Development of machine learning techniques provides a new approach and method for detection of hand tremor. Empirical mode decomposition (Empirical Mode Decomposition, EMD), essentially a smooth processing method for nonlinear, non-smooth time series or signals, and a method capable of successively screening, processing or mining data; singular value decomposition (Singular Value Decomposition, SVD) is a feature extraction method widely used in the field of data mining and machine learning; the support vector machine (Support Vector Machine, SVM) is a supervised learning model and its associated learning algorithm that uses classification and regression analysis to analyze data. Therefore, how to combine the above methods to effectively utilize the methods, so that the detection of hand tremors is more objective, rapid and accurate, is a problem to be solved.
Disclosure of Invention
In view of the above problems, the invention provides a hand tremor detection method and system based on EMD-SVD feature extraction, which are used for solving the problem that the existing hand tremor detection system cannot be used for effectively classifying collected hand data rapidly and accurately.
According to an aspect of the present invention, there is provided a hand tremor detection method based on EMD-SVD feature extraction, the method comprising the steps of:
step one, collecting fingertip acceleration data and preprocessing the fingertip acceleration data to obtain an acceleration time sequence; the fingertip acceleration data comprise three acceleration data corresponding to tremors of hands, mild tremors and severe tremors;
decomposing the acceleration time sequence based on an EMD method to obtain a feature matrix composed of a plurality of eigen mode functions;
step three, performing dimension reduction on the feature matrix based on an SVD method, and extracting singular values of each eigenmode function as fingertip feature data;
dividing the fingertip characteristic data into training set data and test set data, and training a multi-classification model based on an SVM algorithm according to the training set data to obtain a trained multi-classification model;
step five, inputting the fingertip acceleration data to be classified into a trained multi-classification model to obtain a classification result, wherein the fingertip characteristic data to be classified is obtained after pretreatment, step two treatment and step three treatment; the classification results include three results of hand tremors, mild tremors and severe tremors.
In the fourth step, the test set data are input into the multi-classification model based on the SVM algorithm for classification, the classification effect of the multi-classification model is checked by using the test set, the influence of different amounts of fingertip characteristic data on the multi-classification model is compared, and the multi-classification model with the best classification effect and fingertip characteristic data are saved to be used as the final trained multi-classification model.
Further, the specific steps of the first step include: collecting fingertip acceleration data by using a sensor, wherein the triaxial acceleration of the sensor is a respectively when the sensor is stationary x0 ,a y0 ,a z0 The three-axis acceleration in each fingertip acceleration data is a respectively x ,a y ,a z Fusing the triaxial acceleration in each fingertip acceleration data, so as to obtain a fused acceleration time sequence which changes along with time; the fused formula is as follows:
in the formula, a represents the fingertip acceleration data after fusion.
Further, the specific steps of the second step include: firstly, finding all maximum value points and minimum value points of an acceleration time sequence; then, respectively fitting a maximum value and a minimum value by utilizing cubic spline interpolation to obtain an upper envelope curve and a lower envelope curve of the acceleration time sequence; calculating the average value of the upper envelope curve and the lower envelope curve; then, the acceleration time sequence is differenced from the average value to obtain a difference value; checking whether the difference value meets two basic conditions of the eigenmode function, and if so, determining the difference value as a first eigenmode function; if the difference is not satisfied, the difference is used as a new acceleration time sequence, and the steps are repeated until a first eigenmode function meeting basic conditions is screened out; after the first eigenmode function is obtained, the original acceleration time sequence is subjected to difference with the first eigenmode function to obtain a residual component; and taking the residual component as a new acceleration time sequence, repeating the steps to obtain a plurality of residual eigen-mode functions and monotonic residual components, wherein the eigen-mode functions form a feature matrix.
Further, in the fourth step, three two-term classifiers are established through one-vs-rest strategies in an SVM algorithm, and each two-term classifier classifies one category and the other two categories; in the sixth step, when the multi-classification model classifies the fingertip characteristic data to be classified, three two-term classifiers classify the fingertip characteristic data to be classified respectively, the classification result of each two-term classifier is a prediction probability, and the category with the highest prediction probability in the three two-term classifiers is the final classification result.
According to another aspect of the present invention, there is provided a hand tremor detection system based on EMD-SVD feature extraction, the system comprising:
the fingertip data acquisition module is used for acquiring fingertip acceleration data and preprocessing the fingertip acceleration data to obtain an acceleration time sequence; the fingertip acceleration data comprise three acceleration data corresponding to tremors of hands, mild tremors and severe tremors;
the signal decomposition module is used for carrying out decomposition processing on the acceleration time sequence based on an EMD method to obtain a feature matrix composed of a plurality of eigen mode functions;
the feature extraction module is used for reducing the dimension of the feature matrix based on the SVD method, and extracting the singular value of each eigenmode function as fingertip feature data;
the model training module is used for dividing the fingertip characteristic data into training set data and test set data, training a multi-classification model based on an SVM algorithm according to the training set data, inputting the test set data into the multi-classification model based on the SVM algorithm obtained by training for classification in the training process, checking the classification effect of the multi-classification model by using the test set, comparing the influence of different quantity of fingertip characteristic data on the multi-classification model, and storing the multi-classification model with the best classification effect and the fingertip characteristic data as a final trained multi-classification model;
the classifying detection module is used for preprocessing the fingertip acceleration data to be classified, inputting the fingertip characteristic data to be classified obtained after the processing of the signal decomposing module and the characteristic extracting module into a trained multi-classification model, and obtaining a classifying result; the classification results include three results of hand tremors, mild tremors and severe tremors.
Further, the fingertip data acquisition module comprises a data acquisition unit and a data processing unit, wherein the data acquisition unit is an MPU6050 sensor, and the data processing unit is an STM32F103 microcontroller; the triaxial acceleration of the sensor is a when the sensor is stationary x0 ,a y0 ,a z0 The three-axis acceleration in each fingertip acceleration data is a respectively x ,a y ,a z Fusing the triaxial acceleration in each fingertip acceleration data, so as to obtain a fused acceleration time sequence which changes along with time; the fused formula is as follows:
in the formula, a represents the fingertip acceleration data after fusion.
Further, the specific steps of decomposing the acceleration time sequence based on the EMD method in the signal decomposition module to obtain the feature matrix composed of a plurality of eigen mode functions include: firstly, finding all maximum value points and minimum value points of an acceleration time sequence; then, respectively fitting a maximum value and a minimum value by utilizing cubic spline interpolation to obtain an upper envelope curve and a lower envelope curve of the acceleration time sequence; calculating the average value of the upper envelope curve and the lower envelope curve; then, the acceleration time sequence is differenced from the average value to obtain a difference value; checking whether the difference value meets two basic conditions of the eigenmode function, and if so, determining the difference value as a first eigenmode function; if the difference is not satisfied, the difference is used as a new acceleration time sequence, and the steps are repeated until a first eigenmode function meeting basic conditions is screened out; after the first eigenmode function is obtained, the original acceleration time sequence is subjected to difference with the first eigenmode function to obtain a residual component; and taking the residual component as a new acceleration time sequence, repeating the steps to obtain a plurality of residual eigen-mode functions and monotonic residual components, wherein the eigen-mode functions form a feature matrix.
Further, the method for constructing the multi-classification model in the model training module comprises the following steps: three two-term classifiers are established through one-vs-rest strategy in the SVM algorithm, and each two-term classifier classifies one category and the other two categories.
Further, when the multi-classification model in the classification detection module classifies the fingertip characteristic data to be classified, three two-term classifiers respectively classify the fingertip characteristic data to be classified, the classification result of each two-term classifier is a prediction probability, and the category with the highest prediction probability in the three two-term classifiers is a final classification result.
The beneficial technical effects of the invention are as follows:
the basis function of the EMD is obtained by decomposing time series data, and is based on the time scale local characteristic of the time series, so that the EMD method is posterior and visual and has self-adaptability; for unknown signals, the decomposition can be directly started without pre-analysis and research; the method can be automatically classified according to some fixed modes, and manual setting and intervention are not needed; SVD can simplify data set and data characteristics, and can eliminate data noise, thereby improving algorithm prediction results. The tremor data is extracted by the EMD-SVD method, so that the tremor data can be extracted to obtain more abundant features in the original signals than other methods.
The method integrates hand tremor data acquisition and analysis, analyzes the hand tremor data by using a machine learning related method, gives out a prediction result, eliminates subjective factors and enables a hand tremor assessment method to be objective and accurate; the hand tremor detection device is convenient to use, simple and easy to operate, can achieve rapid detection, is suitable for popularization among vast hand tremor patients, and provides a powerful auxiliary means for hand tremor detection.
Drawings
The invention may be better understood by reference to the following description taken in conjunction with the accompanying drawings, which are included to provide a further illustration of the preferred embodiments of the invention and to explain the principles and advantages of the invention, together with the detailed description below.
Fig. 1 is an overall flowchart of a hand tremor detection method based on EMD-SVD feature extraction in accordance with the present invention.
Fig. 2 is a diagram of an apparatus for collecting fingertip tremor data in an embodiment of the present invention.
Fig. 3 is an exploded flow chart of an EMD in an embodiment of the invention.
FIG. 4 is a schematic diagram of the SVD algorithm decomposition result in an embodiment of the invention; wherein panels (a), (b), (c) represent tremor-free, mild tremor and severe tremor data, respectively.
Fig. 5 is a graph showing the exploded results of the fingertip tremor information EMD of a patient in an embodiment of the present invention.
Fig. 6 is a graph of EMD singular value variation of three classes of data in an embodiment of the invention.
FIG. 7 is a flow chart of one-vs-rest strategy in an embodiment of the invention.
FIG. 8 is a classification result of increasing feature numbers in an embodiment of the invention.
FIG. 9 is a classification result of a single feature in an embodiment of the invention.
Fig. 10 is a schematic structural diagram of a hand tremor detection system based on EMD-SVD feature extraction according to the present invention.
Detailed Description
In order that those skilled in the art will better understand the present invention, exemplary embodiments or examples of the present invention will be described below with reference to the accompanying drawings. It is apparent that the described embodiments or examples are only implementations or examples of a part of the invention, not all. All other embodiments or examples, which may be made by one of ordinary skill in the art without undue burden, are intended to be within the scope of the present invention based on the embodiments or examples herein.
The invention provides a hand tremor detection method and system based on EMD-SVD feature extraction and SVM classification, which are characterized in that a time sequence of hand tremor data is acquired by using a hand tremor detection device, signal processing is performed based on an EMD method to obtain an intrinsic mode function in a complex sequence, singular values in an intrinsic mode function matrix are extracted by adopting the SVD method to obtain effective features of the hand tremor data, and finally a multi-classification strategy is constructed by using an SVM classifier to classify the features, so that the purpose of hand tremor detection is achieved.
The flow chart of the method of the invention is shown in fig. 1, and mainly comprises the following steps:
step one: data is collected by a hand tremor detection device.
According to the embodiment of the invention, the hand tremor detection device mainly comprises a data acquisition unit and a data processing unit. The data acquisition unit selects MPU6050 as a sensor, the data processing unit selects STM32F103 as a microcontroller, and the serial port standard is RS-232.
In this example, the tremor data at the fingertips of the testers are collected, and the device is shown in fig. 2. The testers wear the fingerstall on the index finger of the left hand or the right hand, the main body part of the hand tremor detection device is bound at the corresponding wrist, and the data of fingertip tremor are acquired according to different motion states. The collected fingertip tremor data was manually classified into three types of tremor-free, mild tremor and severe tremor data to be used as training and test sets, and the corresponding tester names were saved.
The data processing unit adopts the following processing method for the collected fingertip acceleration data: the triaxial acceleration of the sensor is respectively a when the sensor is static x0 ,a y0 ,a z0 The three-axis acceleration in the data are fused, and the three-axis acceleration of each fingertip data in the training data set is respectively a x ,a y ,a z The fusion formula is:
wherein a is the acceleration after fusion; the fused acceleration sequence over time is denoted as X (t).
Step two: EMD-based hand tremor signal processing.
EMD, i.e. empirical mode decomposition, is capable of adaptively decomposing hand tremor signals into mutually different but all have eigenmode function forms with narrow-band characteristics. The EMD method needs to obtain all local maximum points and local minimum points in an original time sequence, firstly, fitting all local maximum points by using a cubic spline interpolation function to form an upper envelope curve of the original time sequence X (t), then, fitting all local minimum points by using the cubic spline interpolation function to form a lower envelope curve of the original time sequence X (t), determining an instantaneous balance position by using an average value m (t) of each corresponding point in the upper envelope curve and the lower envelope curve, and then, making a difference between the original time sequence and the envelope average value, namely, h (t) =X (t) -m (t), continuously removing gentle trend components, and further extracting an intrinsic mode function of a signal. The continuous sequence screening process is realized by an EMD method, the components under different frequencies are effectively extracted sequentially from high frequency to low frequency according to the sequence, and the decomposition flow chart of EMD on fingertip data is shown in figure 3. The specific decomposition steps can be described as:
(1) Firstly, finding all maximum value points and minimum value points of a time sequence X (t);
(2) Fitting maximum and minimum values respectively by using cubic spline interpolation, and solving by adopting natural boundary conditions to obtain an upper envelope X of a time sequence max (t) and lower envelope X min (t)：
S i (x)＝a i +b i x+c i x 2 +d i x 3
Wherein S is i (x) Is between each segmented cell [ x ] i ,x i+1 ]Fitting equation above, a i ,b i ,c i ,d i Is the coefficient to be solved;
(3) Calculating an average value m (t) of the upper envelope and the lower envelope:
(4) Taking the difference between the original time series X (t) and the average value m (t), the result is h (t):
h(t)＝X(t)-m(t)
(5) Checking whether the difference h (t) satisfies two basic conditions of the eigenmode function, if h (t) satisfies the conditions, c 1 (t) =h (t), i.e. the difference h (t) is the first selected eigenmode function; if h (t) does not meet the two basic conditions, repeating (1) - (4) with h (t) as a new time sequence until screening out an eigenmode function c meeting the basic conditions 1 (t)；
(6) Obtaining a first eigenmode function c 1 After (t), the original time series X (t) is combined with the first eigenmode function c 1 (t) taking the difference to obtain the residual component r 1 (t)：
r 1 (t)＝X(t)-c 1 (t)
(7) Residual component r 1 (t) repeating steps (1) - (6) as new original signals to obtain a plurality of residual eigenmode functions c 2 (t),c 3 (t),...,c n (t) and monotonic residual component r n (t), the original time series X (t) can be expressed as:
since too many screens are prone to error, deviate from the original signal, so that the termination condition cannot trigger a trap into dead-loop, an additional termination criterion SD is introduced:
the acceleration time series X (t) is decomposed by the above EMD method, SD is set to 0.28 in the termination condition, and the test shows that all the eigenmode functions of the data can be screened out without causing the program to enter a dead cycle on the premise of maintaining the linearity and stability of the eigenmode functions, and the EMD decomposition result is shown in fig. 4.
Six eigen-mode functions and a residual error are obtained after three kinds of fingertip tremor data are decomposed by an EMD method. From the first eigenmode to the sixth eigenmode c 1 (t),c 2 (t),...,c 6 (t) the original signal can be approximated as c 1 (t),c 2 (t),...,c 6 The superposition of (t) and the oscillation times of the decomposed eigenmode functions gradually decrease, the time scale gradually increases, the information of the original signals contained by each eigenmode is gradually decreased, and the relevant eigenvalues are further analyzed and extracted, and the eigenvalue matrix is expressed as follows:
[c 1 (t),c 2 (t),...,c 6 (t)] T
step three: SVD-based feature extraction method.
SVD (singular value decomposition) can remove some characteristics with small duty ratio, reduce redundancy of data, improve generalization capability of a prediction model to a certain extent, reduce calculation amount in model training, and improve training and prediction speed of the model.
For matrix A, A εR n×n Lambda is the characteristic value of A, w 1 ,w 2 …w n For a linear independent feature vector, a can be decomposed into:
A＝WΣW -1
wherein w= (W 1 ,w 2 ,…w n ) Is a feature vector matrix; Σ∈r n×n Is a singular value matrix.
For non-square matrix A, A ε R m×n A may decompose into:
A＝UΣV T
wherein U is E R m×m Is AA T Is a feature vector matrix of (a); v epsilon R n×n Is A T A feature vector matrix of A; Σ∈r m×n Is a singular value matrix.
In the singular value matrix, elements on a diagonal are singular values of A, right singular value vectors corresponding to the first k singular values compress data from the column, only the first k features are selected from a plurality of features representing data on the data, k bases with the largest singular values are extracted as new coordinates, and the features are screened.
Simplifying data set and data characteristics by SVD method, eliminating noise of data, and matching characteristic matrix [ c ] 1 (t),c 2 (t),...,c 6 (t)] T SVD dimension reduction is carried out to extract singular values of each mode, the left singular vector, the right singular vector and the singular values of the fingertip data after dimension reduction are shown in a shadow part of fig. 5, SVD dimension reduction is carried out on the feature matrix to extract the singular values of each mode, and the change condition of the singular values of three types of data eigen mode functions is shown in fig. 6.
In the embodiment, six fingertip tremor features are extracted by using the above EMD-SVD method, and the first five of these six features show high correlation, all of which are above 0.7, and the last correlation is not high, only 0.3, and also show no significant difference, so they are not considered in feature selection.
Step four: and constructing a multi-classification model based on the SVM.
And the SVM adopts a kernel function to map to a high-dimensional space for nonlinear samples, so that the SVM is linearly separable, and has good popularization capability. And constructing a strategy by utilizing a multi-classification model of the SVM, and constructing a multi-classification model to realize the detection of hand tremors. SVM is a classification model, given a d= { (x) for a classification problem 1 ,y 1 ),(x 2 ,y 2 ),…,(x m ,y m ) -wherein y i E { -1, +1} is the training set of samples, hope to find a hyperplane in the sample space, can divide the positive and negative samples in the training set, and achieve the aim of classifying the data. For better partitioning of the data set, a hyperplane with the best robustness should be preferentially selected in the sample space, so that classification errors of the classification model due to local disturbance or noise factors of the sample are avoided, and meanwhile generalization capability of the model is improved.
In the embodiment, three levels of detection of fingertip tremors are realized, namely three classification problems are realized, and the detection is realized through a one-vs-rest strategy in an SVM. The strategy is as follows: establishing 3 two-term classifiers, wherein each two-term classifier classifies one class and the remaining two classes, namely a target class and other classes; when prediction is carried out, the three two classifiers respectively classify the classes to be detected, the classification result of the classifiers is the prediction probability, and the class with the highest probability in the three classifiers is the final result of the three classifications. A one-vs-rest strategy flow chart is shown in fig. 7.
Step five: training and testing of the model.
And training the multi-classification model by using different amounts of characteristic data in the training set, checking the classification effect of the classification model by using the data in the test set, comparing the influence of different amounts of characteristics on the classification model, and storing the model with the best classification effect to obtain the fingertip tremor detection system.
In the embodiment, fingertip tremors are detected by using EMD-SVD feature extraction and SVM classification methods, analysis is performed by using a 5-fold cross validation method, and finally, the average value of the five results is used as an evaluation index. Through experiments, the classification result of the feature quantity is shown in fig. 8. As can be seen from FIG. 8, there is a c 1 In the case of features, increasing or decreasing the number of features is not a major factor affecting the classification result, and thus for c 1 ～c 5 Each feature is analyzed, the classification result is shown in fig. 9, the classification effect of the first feature classifier is good, and the performance of the classifier trained by other features is gradually reduced.
Another embodiment of the present invention provides a hand tremor detection system based on EMD-SVD feature extraction, as shown in fig. 10, the system comprising:
the fingertip data acquisition module 110 is used for acquiring fingertip acceleration data and preprocessing the fingertip acceleration data to obtain an acceleration time sequence; the fingertip acceleration data comprise three acceleration data corresponding to tremors of hands, mild tremors and severe tremors; the signal decomposition module 120 is configured to decompose the acceleration time sequence based on an EMD method, and obtain a feature matrix composed of a plurality of eigen-mode functions; the feature extraction module 130 is configured to perform dimension reduction on the feature matrix based on an SVD method, and extract a singular value of each eigenmode function as fingertip feature data; the model training module 140 is configured to divide the fingertip feature data into training set data and test set data, train the multi-classification model based on the SVM algorithm according to the training set data, input the test set data into the multi-classification model based on the SVM algorithm obtained by training to classify, test the classification effect of the multi-classification model by using the test set, compare the influence of different amounts of fingertip feature data on the multi-classification model, and store the multi-classification model with the best classification effect and the fingertip feature data as the final trained multi-classification model; the classification detection module 150 is configured to pre-process the fingertip acceleration data to be classified, and input the fingertip feature data to be classified obtained after processing by the signal decomposition module 120 and the feature extraction module 130 into a trained multi-classification model, so as to obtain a classification result; the classification results included three results of tremor-free, mild tremor and severe tremor in the hands.
The fingertip data acquisition module 110 comprises a data acquisition unit 1110 and a data processing unit 1120, wherein the data acquisition unit 1110 is an MPU6050 sensor, and the data processing unit 1120 is an STM32F103 microcontroller; the triaxial acceleration of the sensor is a when the sensor is stationary x0 ,a y0 ,a z0 The three-axis acceleration in each fingertip acceleration data is a respectively x ,a y ,a z Fusing the triaxial acceleration in each fingertip acceleration data, so as to obtain a fused acceleration time sequence which changes along with time; the fused formula is as follows:
in the formula, a represents the fingertip acceleration data after fusion.
The specific steps of the signal decomposition module 120 for decomposing the acceleration time sequence based on the EMD method to obtain the feature matrix composed of a plurality of eigen-mode functions include: firstly, finding all maximum value points and minimum value points of an acceleration time sequence; then, respectively fitting a maximum value and a minimum value by utilizing cubic spline interpolation to obtain an upper envelope curve and a lower envelope curve of the acceleration time sequence; calculating the average value of the upper envelope curve and the lower envelope curve; then, the acceleration time sequence is differenced from the average value to obtain a difference value; checking whether the difference value meets two basic conditions of the eigenmode function, and if so, determining the difference value as a first eigenmode function; if the difference is not satisfied, the difference is used as a new acceleration time sequence, and the steps are repeated until a first eigenmode function meeting basic conditions is screened out; after the first eigenmode function is obtained, the original acceleration time sequence is subjected to difference with the first eigenmode function to obtain a residual component; and taking the residual component as a new acceleration time sequence, repeating the steps to obtain a plurality of residual eigen-mode functions and monotonic residual components, wherein the eigen-mode functions form a feature matrix.
The method for constructing the multi-classification model in the model training module 140 is as follows: three two-term classifiers are established through one-vs-rest strategy in the SVM algorithm, and each two-term classifier classifies one category and the other two categories. When the multi-classification model in the classification detection module 150 classifies the fingertip characteristic data to be classified, three two-term classifiers respectively classify the fingertip characteristic data to be classified, the classification result of each two-term classifier is a prediction probability, and the category with the highest prediction probability in the three two-term classifiers is a final classification result.
The function of the hand tremor detection system based on EMD-SVD feature extraction in this embodiment may be described by the hand tremor detection method based on EMD-SVD feature extraction, so that details of this embodiment are not described, and reference may be made to the above method embodiments, which are not described herein.
While the invention has been described with respect to a limited number of embodiments, those skilled in the art, having benefit of the above description, will appreciate that other embodiments are contemplated within the scope of the invention as described herein. The disclosure of the present invention is intended to be illustrative, but not limiting, of the scope of the invention, which is defined by the appended claims.
Claims (8)
1. The hand tremor detection method based on EMD-SVD feature extraction is characterized by comprising the following steps of:
step one, collecting fingertip acceleration data and preprocessing the fingertip acceleration data to obtain an acceleration time sequence; the fingertip acceleration data comprise three acceleration data corresponding to tremors of hands, mild tremors and severe tremors;
decomposing the acceleration time sequence based on an EMD method to obtain a feature matrix composed of a plurality of eigen mode functions; the method comprises the following specific steps:
finding all maximum value points and minimum value points of the acceleration time sequence;
respectively fitting a maximum value and a minimum value by utilizing cubic spline interpolation to obtain an upper envelope curve and a lower envelope curve of the acceleration time sequence;
calculating the average value of the upper envelope curve and the lower envelope curve;
the acceleration time sequence is differenced from the average value to obtain a difference value;
checking whether the difference value meets two basic conditions of the eigenmode function, and if so, determining the difference value as a first eigenmode function; if the difference is not satisfied, the difference is used as a new acceleration time sequence, and the steps are repeated until a first eigenmode function meeting basic conditions is screened out;
after the first eigenmode function is obtained, the original acceleration time sequence is subjected to difference with the first eigenmode function to obtain a residual component;
taking the residual component as a new acceleration time sequence, repeating the steps to obtain a plurality of residual eigen-mode functions and monotonic residual components, wherein the eigen-mode functions form a feature matrix;
wherein, screening termination conditions are:
step three, performing dimension reduction on the feature matrix based on an SVD method, and extracting singular values of each eigenmode function as fingertip feature data;
dividing the fingertip characteristic data into training set data and test set data, and training a multi-classification model based on an SVM algorithm according to the training set data to obtain a trained multi-classification model;
step five, inputting the fingertip acceleration data to be classified into a trained multi-classification model to obtain a classification result, wherein the fingertip characteristic data to be classified is obtained after pretreatment, step two treatment and step three treatment; the classification results include three results of hand tremors, mild tremors and severe tremors.
2. The hand tremor detection method based on EMD-SVD feature extraction according to claim 1, wherein in the fourth step, test set data are input into a multi-classification model based on SVM algorithm obtained through training to classify, the classification effect of the multi-classification model is checked by using the test set, influences of different amounts of fingertip feature data on the multi-classification model are compared, and the multi-classification model with the best classification effect and fingertip feature data are stored to be used as a final trained multi-classification model.
3. The method for detecting hand tremor based on EMD-SVD feature extraction of claim 2, wherein the specific steps of the first step include: collecting fingertip acceleration data by using a sensor, wherein the triaxial acceleration of the sensor is a respectively when the sensor is stationary x 0 ,a y 0 ,a z 0 The three-axis acceleration in each fingertip acceleration data is a respectively x ,a y ,a z Fusing the triaxial acceleration in each fingertip acceleration data, so as to obtain a fused acceleration time sequence which changes along with time; the fused formula is as follows:
in the formula, a represents the fingertip acceleration data after fusion.
4. The hand tremor detection method based on EMD-SVD feature extraction according to claim 3, wherein in the fourth step, three two-term classifiers are established through one-vs-rest strategy in an SVM algorithm, and each two-term classifier classifies one category and the other two categories; in the sixth step, when the multi-classification model classifies the fingertip characteristic data to be classified, three two-term classifiers classify the fingertip characteristic data to be classified respectively, the classification result of each two-term classifier is a prediction probability, and the category with the highest prediction probability in the three two-term classifiers is the final classification result.
5. A hand tremor detection system based on EMD-SVD feature extraction, comprising:
the fingertip data acquisition module is used for acquiring fingertip acceleration data and preprocessing the fingertip acceleration data to obtain an acceleration time sequence; the fingertip acceleration data comprise three acceleration data corresponding to tremors of hands, mild tremors and severe tremors;
the signal decomposition module is used for carrying out decomposition processing on the acceleration time sequence based on an EMD method to obtain a feature matrix composed of a plurality of eigen mode functions; the method comprises the following specific steps: finding all maximum value points and minimum value points of the acceleration time sequence; respectively fitting a maximum value and a minimum value by utilizing cubic spline interpolation to obtain an upper envelope curve and a lower envelope curve of the acceleration time sequence; calculating the average value of the upper envelope curve and the lower envelope curve; the acceleration time sequence is differenced from the average value to obtain a difference value; checking whether the difference value meets two basic conditions of the eigenmode function, and if so, determining the difference value as a first eigenmode function; if the difference is not satisfied, the difference is used as a new acceleration time sequence, and the steps are repeated until a first eigenmode function meeting basic conditions is screened out; after the first eigenmode function is obtained, the original acceleration time sequence is subjected to difference with the first eigenmode function to obtain a residual component; taking the residual component as a new acceleration time sequence, repeating the steps to obtain a plurality of residual eigen-mode functions and monotonic residual components, wherein the eigen-mode functions form a feature matrix; wherein, screening termination conditions are:
the feature extraction module is used for reducing the dimension of the feature matrix based on the SVD method, and extracting the singular value of each eigenmode function as fingertip feature data;
the model training module is used for dividing the fingertip characteristic data into training set data and test set data, training a multi-classification model based on an SVM algorithm according to the training set data, inputting the test set data into the multi-classification model based on the SVM algorithm obtained by training for classification in the training process, checking the classification effect of the multi-classification model by using the test set, comparing the influence of different quantity of fingertip characteristic data on the multi-classification model, and storing the multi-classification model with the best classification effect and the fingertip characteristic data as a final trained multi-classification model;
the classifying detection module is used for preprocessing the fingertip acceleration data to be classified, inputting the fingertip characteristic data to be classified obtained after the processing of the signal decomposing module and the characteristic extracting module into a trained multi-classification model, and obtaining a classifying result; the classification results include three results of hand tremors, mild tremors and severe tremors.
6. The hand tremor detection system based on EMD-SVD feature extraction of claim 5, wherein the fingertip data acquisition module comprises a data acquisition unit and a data processing unit, wherein the data acquisition unit is an MPU6050 sensor, and the data processing unit is an STM32F103 microcontroller; the triaxial acceleration of the sensor is a when the sensor is stationary x 0 ,a y 0 ,a z 0 The three-axis acceleration in each fingertip acceleration data is a respectively x ,a y ,a z Each is provided withThe three-axis acceleration in the fingertip acceleration data is fused, so that a fused acceleration time sequence which changes along with time is obtained; the fused formula is as follows:
in the formula, a represents the fingertip acceleration data after fusion.
7. The hand tremor detection system based on EMD-SVD feature extraction of claim 6, wherein the method for constructing the multi-classification model based on SVM algorithm in the model training module is as follows: three two-term classifiers are established through one-vs-rest strategy in the SVM algorithm, and each two-term classifier classifies one category and the other two categories.
8. The hand tremor detection system based on EMD-SVD feature extraction of claim 7, wherein when the multi-classification model based on SVM algorithm in the classification detection module classifies fingertip feature data to be classified, three two-term classifiers respectively classify the fingertip feature data to be classified, the classification result of each two-term classifier is prediction probability, and the classification with the highest prediction probability in the three two-term classifiers is the final classification result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110959839.8A CN113705649B (en) | 2021-08-20 | 2021-08-20 | EMD-SVD feature extraction-based hand tremor detection method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110959839.8A CN113705649B (en) | 2021-08-20 | 2021-08-20 | EMD-SVD feature extraction-based hand tremor detection method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113705649A CN113705649A (en) | 2021-11-26 |
| CN113705649B true CN113705649B (en) | 2024-01-12 |
Family
ID=78654109
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110959839.8A Active CN113705649B (en) | 2021-08-20 | 2021-08-20 | EMD-SVD feature extraction-based hand tremor detection method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113705649B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104776908A (en) * | 2015-04-17 | 2015-07-15 | 南京理工大学 | EMD generalized energy-based wheeltrack vibration signal fault feature extraction method |
| WO2019036749A1 (en) * | 2017-08-24 | 2019-02-28 | Adams Warwick Russell | A system for detecting early parkinson's disease and other neurological diseases and movement disorders |
| CN112075940A (en) * | 2020-09-21 | 2020-12-15 | 哈尔滨工业大学 | Tremor detection system based on bidirectional long-time and short-time memory neural network |
| CN112580588A (en) * | 2020-12-29 | 2021-03-30 | 西北工业大学 | Intelligent flutter signal identification method based on empirical mode decomposition |
| CN113100756A (en) * | 2021-04-15 | 2021-07-13 | 重庆邮电大学 | Stacking-based Parkinson tremor detection method |
-
2021
- 2021-08-20 CN CN202110959839.8A patent/CN113705649B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104776908A (en) * | 2015-04-17 | 2015-07-15 | 南京理工大学 | EMD generalized energy-based wheeltrack vibration signal fault feature extraction method |
| WO2019036749A1 (en) * | 2017-08-24 | 2019-02-28 | Adams Warwick Russell | A system for detecting early parkinson's disease and other neurological diseases and movement disorders |
| CN112075940A (en) * | 2020-09-21 | 2020-12-15 | 哈尔滨工业大学 | Tremor detection system based on bidirectional long-time and short-time memory neural network |
| CN112580588A (en) * | 2020-12-29 | 2021-03-30 | 西北工业大学 | Intelligent flutter signal identification method based on empirical mode decomposition |
| CN113100756A (en) * | 2021-04-15 | 2021-07-13 | 重庆邮电大学 | Stacking-based Parkinson tremor detection method |
Non-Patent Citations (2)
| Title |
|---|
| 一种求解信号包络曲线端点值的新方法;马小刚;王永平;杜历;;噪声与振动控制(04);第20-23页 * |
| 局部波动特征分解及其在滚动轴承故障诊断中的应用研究;张亢;石阳春;唐明珠;吴家腾;;振动与冲击(01);第97-103页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113705649A (en) | 2021-11-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Rubin et al. | Recognizing abnormal heart sounds using deep learning | |
| CN110188836B (en) | Brain function network classification method based on variational self-encoder | |
| CN113052113B (en) | Depression identification method and system based on compact convolutional neural network | |
| Sahin et al. | Pattern recognition with surface EMG signal based wavelet transformation | |
| CN111476158B (en) | Multi-channel physiological signal somatosensory gesture recognition method based on PSO-PCA-SVM | |
| CN111134664B (en) | Epileptic discharge identification method and system based on capsule network and storage medium | |
| Patil et al. | A Novel Approach for ABO Blood Group Prediction using Fingerprint through Optimized Convolutional Neural Network | |
| Cadieu et al. | The neural representation benchmark and its evaluation on brain and machine | |
| Dinning et al. | Real-time classification of multiunit neural signals using reduced feature sets | |
| CN108538388B (en) | Knee joint dyskinesia function judging method | |
| Mamun et al. | Vocal feature guided detection of parkinson’s disease using machine learning algorithms | |
| CN110289097A (en) | A kind of Pattern Recognition Diagnosis system stacking model based on Xgboost neural network | |
| CN113662560A (en) | Method for detecting seizure-like discharge between attacks, storage medium and device | |
| CN107045624B (en) | Electroencephalogram signal preprocessing and classifying method based on maximum weighted cluster | |
| CN115563484A (en) | Street greening quality detection method based on physiological awakening identification | |
| CN113988135A (en) | Electromyographic signal gesture recognition method based on double-branch multi-stream network | |
| CN108962379B (en) | Mobile phone auxiliary detection system for cranial nerve system diseases | |
| CN113705649B (en) | EMD-SVD feature extraction-based hand tremor detection method and system | |
| CN112580486A (en) | Human behavior classification method based on radar micro-Doppler signal separation | |
| CN117193537A (en) | Double-branch convolutional neural network motor imagery intention decoding method based on self-adaptive transfer learning | |
| CN113128585B (en) | Deep neural network based multi-size convolution kernel method for realizing electrocardiographic abnormality detection and classification | |
| CN115736920A (en) | Depression state identification method and system based on bimodal fusion | |
| CN113143275B (en) | Electroencephalogram fatigue detection method for quantitative evaluation of sample and characteristic quality in combined manner | |
| CN114947850A (en) | Mental load grade objective detection method based on pulse Bouss model characteristics | |
| CN113545789A (en) | Electroencephalogram analysis model construction method based on CSP algorithm and PSD algorithm, electroencephalogram analysis method and system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |