CN111110268A - Human body muscle sound signal prediction method based on random vector function connection network technology - Google Patents
Human body muscle sound signal prediction method based on random vector function connection network technology Download PDFInfo
- Publication number
- CN111110268A CN111110268A CN201911170547.5A CN201911170547A CN111110268A CN 111110268 A CN111110268 A CN 111110268A CN 201911170547 A CN201911170547 A CN 201911170547A CN 111110268 A CN111110268 A CN 111110268A
- Authority
- CN
- China
- Prior art keywords
- muscle
- signal
- integrated
- training
- vector function
- 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.)
- Granted
Links
Images
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B7/00—Instruments for auscultation
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/103—Detecting, measuring or recording devices for testing the shape, pattern, colour, size or movement of the body or parts thereof, for diagnostic purposes
- A61B5/11—Measuring movement of the entire body or parts thereof, e.g. head or hand tremor, mobility of a limb
- A61B5/1118—Determining activity level
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/68—Arrangements of detecting, measuring or recording means, e.g. sensors, in relation to patient
- A61B5/6801—Arrangements of detecting, measuring or recording means, e.g. sensors, in relation to patient specially adapted to be attached to or worn on the body surface
-
- 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
- A61B5/7203—Signal processing specially adapted for physiological signals or for diagnostic purposes for noise prevention, reduction or removal
-
- 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
- A61B5/7225—Details of analog processing, e.g. isolation amplifier, gain or sensitivity adjustment, filtering, baseline or drift compensation
-
- 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
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Public Health (AREA)
- Veterinary Medicine (AREA)
- General Health & Medical Sciences (AREA)
- Animal Behavior & Ethology (AREA)
- Physics & Mathematics (AREA)
- Surgery (AREA)
- Molecular Biology (AREA)
- Medical Informatics (AREA)
- Biomedical Technology (AREA)
- Heart & Thoracic Surgery (AREA)
- Signal Processing (AREA)
- Biophysics (AREA)
- Pathology (AREA)
- Artificial Intelligence (AREA)
- Physiology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Psychiatry (AREA)
- Power Engineering (AREA)
- Evolutionary Computation (AREA)
- Fuzzy Systems (AREA)
- Mathematical Physics (AREA)
- Dentistry (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Rehabilitation Tools (AREA)
Abstract
The invention relates to a human body muscle sound signal prediction method based on a random vector function connection network technology, which overcomes the defects of incomplete muscle sound signal denoising treatment, low prediction algorithm precision, high complexity and the like compared with the prior art. The invention comprises the following steps: acquiring a human body muscle tone signal; integrating the muscle sound signals; denoising the integrated muscle sound signal; constructing and training a random vector function connection network model; integrating the prediction of the muscle tone signal. The wearable muscle sound signal sensor can be used for collecting the muscle sound signal of the elbow joint of the human body, so that the muscle sound signal can be rapidly obtained and accurately predicted.
Description
Technical Field
The invention relates to the technical field of signal processing, in particular to a human muscle tone signal prediction method based on a random vector function connection network technology.
Background
Human muscle tone (MMG), a biological signal used to study muscle function, is an acceleration signal generated by muscle contraction, when muscle contracts, the cross section of muscle fiber changes, generating transverse low-frequency vibration, which is transmitted to the skin surface to make sound, and the intensity of the signal is proportional to the muscle tension. A muscle tone signal is a time series signal with non-linear and non-stationary characteristics in which active muscle twitches can add up linearly or non-linearly. Muscle tone signals have been widely used in the fields of research of muscle fatigue muscle strength detection, human body rehabilitation engineering, pattern recognition, etc., and compared with the traditionally used Electromyographic (EMG), the muscle tone signals have the following advantages:
(1) due to the propagation characteristic, the sensor does not need to be placed on the surface of muscle skin like a myoelectric signal sensor, the signal is not greatly influenced even if the sensor moves when the sensor is used, and meanwhile, the requirement on the placement position of the sensor is low and the robustness is high;
(2) the muscle tone signal has immunity to the change of skin impedance and is not influenced by the change of the impedance;
(3) since the skin, fat and other tissues of the human body act as a low-pass filter, the energy of the muscle tone signal collected on the skin surface is mainly concentrated in the low frequency band (5-35Hz) to help improve the real-time processing efficiency of the MMG signal.
In recent years, the processing problem of the muscle tone signal is widely concerned in a plurality of fields, because the muscle tone signal is a time sequence with time variability, randomness and nonlinearity. At present, the processing method for such time series is mainly developed from three aspects of time domain, frequency domain and time-frequency domain.
The method can be used for accurately predicting the change condition of the human muscle tone signals at the future time by predicting the muscle tone signals in a short term or a long term, thereby having important guiding significance for the evaluation of the functional condition of the human muscle and the rehabilitation evaluation of the joint of a patient. Meanwhile, the prediction result can be applied to the flexible control of the rehabilitation robot joint, so that the aim of man-machine fusion is fulfilled. At present, the research conditions for processing and predicting human body biological signals at home and abroad are as follows:
1. the 'human quadriceps femoris muscle strength estimation method based on muscle movement signals' published by Wangdangqing develops research work based on four aspects of acquisition, signal processing, feature extraction and regression model construction of surrounding muscle sound signals, proposes a muscle strength estimation regression model based on EEMD decomposition and support vector machine combination, estimates the magnitude of muscle strength, but as the muscle sound signals have the characteristics of strong interference noise, weak signal strength, high randomness, low frequency distribution and the like, other external crosstalk caused by respiratory interference in organisms, power frequency interference in organisms, poor grounding and the like can not be completely removed, and therefore the estimation accuracy of the algorithm on the aspect of signal processing needs to be further improved.
2. The Real-Time Myoprocessors for a Neural controlled powered Exoskeleton Arm, issued by Cavallaro et al, predicts human motor intent by designing a set of Hill muscle models. The myoelectric signal is collected by a 28-lead myoelectric signal collector, pattern recognition is carried out by using a genetic algorithm, and then the angle change of the upper limb is estimated, but the experiment needs more myoelectric sensors, the algorithm application complexity is higher, the estimation accuracy is not high, and the actual application is more complicated.
3. 'artificial limb hand action mode classification research based on muscle sound signals' published by the people of thankshood et al takes the muscle sound signals as the control source of the artificial limb hand, and realizes the effective control of the artificial limb by the transformation and the characteristic extraction of the muscle sound signals.
4. The study of the human body fall prediction method based on the surface muscle sound signals, published by the shin-Yang et al, provides and uses an SVM and a CNN algorithm to rapidly and accurately predict the human body fall through the collection, pretreatment and analysis of the muscle sound signals of the lower limbs during the human body fall and the daily activities.
However, the above preprocessing algorithm for human muscle sound signals still has certain defects, and fails to remove or completely eliminate noise and interference in the signals, and still has great influence on the use of subsequent algorithms, so that the accuracy of the algorithms is reduced. Meanwhile, how to reduce the time delay between the actual generation time and the acquisition time of the muscle sound signal and further improve the real-time performance and the accuracy of the acquired signal is a problem to be solved urgently in the field of medical engineering at present. Meanwhile, how to predict the muscle sound signals in advance timely and accurately is an effective way for improving the real-time performance of the acquired signals at present, and the prediction result can be used as an effective input signal of a rehabilitation medical equipment control system and a human-computer interface.
Therefore, how to accurately predict the human muscle sound signals in real time becomes a technical problem which needs to be solved urgently.
Disclosure of Invention
The invention aims to solve the defects of incomplete muscle sound signal denoising treatment, low precision of a prediction algorithm, high complexity and the like in the prior art, and provides a human body muscle sound signal prediction method based on a random vector function connection network technology to solve the problems.
In order to achieve the purpose, the technical scheme of the invention is as follows:
a human muscle sound signal prediction method based on a random vector function connection network technology comprises the following steps:
11) acquiring a human body muscle tone signal: the method comprises the following steps of collecting and storing acceleration signals of a wearable muscle sound signal sensor on human elbow joint muscle, obtaining acceleration information of the human elbow joint muscle from the 1 st moment to the L th moment in three directions of an x axis, a y axis and a z axis, and forming 3 groups of muscle sound signals with the lengths of L:
wherein L is more than or equal to 1000 and less than or equal to 10000;
12) integration of muscle tone signals: integrating the muscle tone signals collected by the sensors;
13) denoising integrated muscle sound signals: denoising the integrated muscle sound signals by using a lifting wavelet transform technology;
14) constructing and training a random vector function connection network model: constructing a random vector function connection network model, and training by using the denoised and updated integrated muscle sound signal;
15) prediction of integrated muscle tone signal: and predicting the integrated muscle sound signal by using the trained random vector function connection network model.
The muscle tone signal integration comprises the following steps:
in formula (1), xiRepresenting the integrated muscle sound signal value at the ith moment; acceleration components of the muscle sound signals collected by the sensor at the ith moment in an x axis, a y axis and a z axis are respectively represented, wherein i is 1, 2.
22) Taking the integrated signal as an integrated muscle sound signal, and recording as:
X(L)={x1,x2,...,xL};
wherein x is1Representing the integrated muscle sound signal value at the 1 st moment; x is the number of2Representing the integrated muscle tone signal value at the 2 nd time; x is the number ofLRepresents the integrated muscle tone signal value at the lth time instant.
The denoising processing of the integrated muscle sound signal comprises the following steps:
31) for integrated muscle tone signal x (l) { x1,x2,...,xLPerforming splitting treatment, wherein the calculation formula of the splitting treatment is shown as formulas (2-1), (2-2) and (2-3):
in the formula (2-1), Split [ x (l) ] is a splitting operator, which indicates that the sequence x (l) is Split;
in the formulas (2-2) and (2-3), the value range of n is more than or equal to 500 and less than or equal to 5000;
splitting the integrated muscle tone signal x (l) into odd and even subsequences, denoted as:
wherein the content of the first and second substances,representing the odd sequence, x, obtained after splitting1Representing the 1 st signal value, x, of the corresponding integrated muscle tone signal3Representing the 3 rd signal value, x, of the corresponding integrated muscle tone signal2n-1Representing the 2n-1 signal value corresponding to the integrated muscle sound signal;
wherein the content of the first and second substances,denotes the even sequence obtained after splitting, x2Representing the 2 nd signal value, x, of the corresponding integrated muscle signal4Representing the 4 th signal value, x, of the corresponding integrated muscle tone signal2nRepresenting a2 nth signal value corresponding to the integrated muscle sound signal;
32) based on even sequences obtained after splittingBy means of predictorsP, predicting the corresponding odd sequence to obtain the predicted value of the odd sequence, and recording the predicted value asThe prediction calculation formula is shown as formula (3):
33) using pairs of update operatorsUpdating to obtain an updated integrated muscle sound signal, wherein a calculation formula is shown as formula (5):
in formula (5), X1(L) represents the low-frequency component obtained after the integrated muscle sound signal is updated once,representing a primary high-frequency component, U1The operator is updated once, and the calculation is shown as the formula (6):
34) repeating steps 31) to 33) for the primary low-frequency component X obtained in step 33)1(L) performing denoising updating for M times to obtain
wherein, the value range of M is as follows: m is more than or equal to 1 and less than or equal to 10, X2(L) represents the updated 2-times low-frequency component, X3(L) represents the 3-time low-frequency component obtained after updating, XM(L) represents the M-th low-frequency component obtained after updating,it is shown that the high-frequency component is divided into 2,which represents the high-frequency components of 3 times,representing the high frequency components M times.
The construction and training of the random vector function connection network model comprises the following steps:
41) updating the denoised Mth low-frequency component XMAnd (L) taking the first K data as a training sequence of a random vector function connection network, and recording as:
wherein K is more than or equal to 5000 and less than or equal to 8000, and a sequence consisting of the rest L-K data is taken as a prediction sequence and is recorded as:
42) setting parameters of each layer of the random vector function connection network model structure:
wherein, the input layer is 1 layer, the number of nodes of the input layer is m, the value range of m is 1-50, and the weight omega from the input layer to the hidden layern,lIn [0, 1 ]]Produce randomlyRaw;
the hidden layer is 1 layer, the number of nodes of the hidden layer is n, the numerical range of n is 1-20, the hidden layer selects a sigmoid function as an activation function, and the form of the sigmoid function is shown as a formula (7):
in the formula (7), ωn,lB is the weight value from the input layer to the hidden layer and the bias from the input layer to the hidden layer respectively;
the network hidden layer kernel function mapping matrix H is as shown in equation (8):
in the formula (8), L represents the number of hidden layer nodes;
the output layer is 1 layer, the number of output layer nodes is 1, and the weight β of the output layernThe method can be estimated and solved by a standardized regular least square method, and the calculation formula is shown as formula (9):
βn=(HTH+λI)-1HTY(9)
in equation (9), Y is the output column vector corresponding to the input vector in the training sample space; h is a hidden layer kernel function mapping matrix, lambda is a constant, and I represents an identity matrix;
43) the training sequence is segmented in sequence by adopting a sliding window model,
431) mixing XM(K) And taking the continuous q data as an input subsequence of the random vector function connection network model training stage, and recording as:
wherein q is the size of a sliding window, and the value is consistent with the number of nodes of an input layer of a random vector function connection network, namely q is m;
taking the corresponding training input subsequence and the training output data as a group, and forming K-q groups of random vector function connection network training subsequence and training output data together;
44) inputting the training subsequence to the random vector function connection network model for training, and obtaining the prediction output of the random vector function connection network after the training is finishedCalculating as shown in equation (10):
in formula (10), βlThe weight from the 1 st node of the hidden layer to the output layer; h (x, ω)l,bl) The activation function value corresponding to the 1 st node of the hidden layer.
The predicting the integrated muscle sound signal comprises the following steps:
51) sequence composed of L-K dataAs a prediction sequence, and adopting a sliding window model to segment the prediction sequence in sequence,
is recorded as:wherein i is more than or equal to 1 and less than or equal to L-K-q + 1; the size of the sliding window is q, and the sliding step length is 1;
52) will predict the sequenceInputting the input into the trained random vector function connection network, and calculating a predicted value according to the formula (10) to obtain a first predicted value of the muscle tone signal
53) And (L-K-q +1) group sequences with the length of q are continuously and sequentially input into the trained random vector function connection network, and the predicted value of the muscle tone signal in a future continuous (L-K-q +1) period can be predicted.
Advantageous effects
Compared with the prior art, the human muscle tone signal prediction method based on the random vector function connection network technology can acquire the muscle tone signal of the elbow joint of the human body through the wearable muscle tone signal sensor, and quickly acquire and accurately predict the muscle tone signal. Compared with the traditional neural network algorithm, the algorithm in the invention has the advantages of high learning speed, high precision, strong network generalization performance and the like. In the field of human joint rehabilitation, the activity state of human muscles at the future moment is judged by predicting the muscle tone signals of the joints of the disabled patients, the activity form of the elbow joints at the future moment is further deduced, information can be input into a rehabilitation machine control system in advance, the flexible control of a machine and the human joints is realized, and the effect of man-machine fusion is achieved. In the field of artificial limb control, the recognition and control of the human body limb actions can be realized by extracting the characteristics of human body muscle sound signals in the field of limb action recognition.
The invention effectively solves the defects of weak signal intensity, complex acquisition process, low anti-interference performance and the like of other biological signals such as myoelectric signals and the like by adopting the myoelectric signals as the sequences to be predicted; the muscle sound signals are subjected to an integration processing algorithm, so that the amplitude of the muscle sound signals is increased, the cost caused by expensive signal amplification equipment is saved, meanwhile, the sensitivity of the single-component muscle sound signals to the angle change of the sensor is avoided, and the stability of the signals is improved; the signal denoising algorithm provided by the invention can realize accurate analysis of the myotone signals with any length, has the advantages of simple and flexible structure, low calculation complexity and the like compared with the traditional signal denoising algorithm, and is suitable for occasions with limited performance of a signal processor.
The muscle sound signals are processed by utilizing the lifting wavelet transform algorithm, so that the problems of large calculation amount, high complexity and large storage space required by the traditional wavelet processing algorithm are solved; the random vector function is used for connecting a network algorithm to realize random generation of network parameters, and compared with the traditional neural network learning technology, the method has the characteristics of short time consumption and high precision. Meanwhile, the direct connection design in the connection network based on the random vector function is also more favorable for mastering hidden features inside the muscle sound signal sequence. The method solves the disadvantage that the initial weight of the artificial neural network needs to be determined by artificial experience, has higher convergence speed, avoids the network from falling into local optimization, and has the characteristics of good generalization performance, high precision and the like.
Drawings
FIG. 1 is a sequence diagram of the method of the present invention;
FIG. 2 is a schematic diagram of a random vector function connection network model structure according to the present invention;
FIG. 3 is a graph of acceleration signals and integrated signals in the X-axis, Y-axis, and Z-axis directions;
FIG. 4 is a graph illustrating the denoising effect of a muscle sound signal;
FIG. 5 is a graph comparing the effects of different prediction algorithms.
Detailed Description
So that the manner in which the above recited features of the present invention can be understood and readily understood, a more particular description of the invention, briefly summarized above, may be had by reference to embodiments, some of which are illustrated in the appended drawings, wherein:
as shown in fig. 1, the method for predicting human muscle tone signals based on random vector function connection network technology according to the present invention includes the following steps:
firstly, acquiring a human muscle tone signal.
The method comprises the following steps of collecting and storing acceleration signals of a wearable muscle sound signal sensor on human elbow joint muscle, obtaining acceleration information of the human elbow joint muscle in three directions of an x axis, a y axis and a z axis from the 1 st moment to the 3000 th moment, and forming 3 groups of muscle sound signals with the lengths of 3000, and recording the signals as:
secondly, integrating the muscle tone signals: and integrating the muscle tone signals collected by the sensors. The method comprises the following specific steps:
in formula (1), xiRepresenting the integrated muscle sound signal value at the ith moment; the components of the muscle tone signal acquired by the sensor at the i-th time point in the x-axis, the y-axis and the z-axis are respectively represented, wherein i is 1, 2.
(2) Taking the integrated signal as an integrated muscle sound signal, and recording as:
X(3000)={x1,x2,...,x3000};
wherein x is1Representing the integrated muscle sound signal value at the 1 st moment; x is the number of2Representing the integrated muscle tone signal value at the 2 nd time; x is the number of3000Indicating the integrated muscle tone signal value at the 3000 th instant.
Thirdly, integrating denoising treatment of the muscle sound signals: and denoising the integrated muscle sound signals by using a lifting wavelet transform technology. Considering that the integrated muscle sound signal still contains a large number of high-frequency signals, the characteristics that the implementation efficiency of the wavelet transformation algorithm is high, the Fourier transformation is not depended on, and signals with any length can be analyzed are fully considered, and meanwhile, the advantages of the wavelet transformation algorithm in the aspects of structural design and self-adaptive construction are combined, and the integrated muscle sound signal is subjected to denoising processing. The method comprises the following specific steps:
(1) for integrated muscle tone signal X (3000) { X }1,x2,...,x3000The splitting treatment is carried out, and then,the calculation formula of the splitting treatment is shown in formulas (2-1), (2-2) and (2-3):
in the formula (2-1), Split [ x (l) ] is a splitting operator, which indicates that the sequence x (l) is Split;
in the formulae (2-2) and (2-3), n is 1500; the value of L is 3000;
the integrated muscle tone signal X (3000) is split into odd and even subsequences, denoted as:
wherein the content of the first and second substances,representing the odd sequence, x, obtained after splitting1Representing the 1 st signal value, x, of the corresponding integrated muscle tone signal3Representing the 3 rd signal value, x, of the corresponding integrated muscle tone signal2999A 2999 th signal value representing a corresponding integrated muscle tone signal;
wherein the content of the first and second substances,denotes the even sequence obtained after splitting, x2Representing the 2 nd signal value, x, of the corresponding integrated muscle signal4Representing the 4 th signal value, x, of the corresponding integrated muscle tone signal3000Indicating corresponding integrated muscle voice messageSignal value number 3000.
(2) Based on even sequences obtained after splittingPredicting the corresponding odd sequence through the predictor P to obtain the predicted value of the odd sequence, and recording the predicted value as the odd sequenceThe prediction calculation formula is shown as formula (3):
wherein L is 1, 2, 3.., 3000, n is 1, 2, 3.., 1500;
wherein L is 1, 2, 3.
(3) Using pairs of update operatorsUpdating to obtain an updated integrated muscle sound signal, wherein a calculation formula is shown as formula (5):
wherein L ═ 1, 2, 3.., 3000
In formula (5), X1(L) represents the low-frequency component obtained after the integrated muscle sound signal is updated once,representing a primary high-frequency component, U1The operator is updated once, and the calculation is shown as the formula (6):
(4) repeating the steps (1) to (3), and carrying out the primary low-frequency component X obtained in the step (3)1(3000) Performing denoising updating for 5 times to obtain
wherein, X2(3000) Representing the updated 2-times low-frequency component, X3(3000) Represents the updated 3 times low frequency component, X5(3000) Representing the updated 5 th low frequency component,it is shown that the high-frequency component is divided into 2,which represents the high-frequency components of 3 times,representing the 5 high frequency components.
Fourthly, constructing and training a random vector function connection network model: and constructing a random vector function connection network model, and training by using the denoised and updated integrated muscle sound signal. The traditional neural network algorithm has the defects of low convergence rate, long learning time, poor network generalization performance and the like. The random vector function connection network has certain advantages in these aspects, the learning speed of the network is improved by training the output weight of the network by adopting a least square method, and meanwhile, the generalization capability of the network is further improved by utilizing the enhanced nonlinear core of the hidden layer. The method comprises the following specific steps:
(1) updating the denoised 5 th low-frequency component X5(3000) The first 2408 data in the middle are taken as training sequences of random vector function connection networks, and are recorded as:
the sequence of the remaining 592 data was taken as the predicted sequence and noted as:
(2) the random vector function is set to connect the parameters of each layer of the network model structure, as shown in fig. 2.
Wherein, the input layer is 1 layer, the number of nodes of the input layer is 8, and the weight omega from the input layer to the hidden layern,lIn [0, 1 ]]Randomly generating;
the hidden layer is 1 layer, the number of nodes of the hidden layer is 10, the hidden layer selects a sigmoid function as an activation function, and the form of the sigmoid function is shown in formula (7):
in the formula (7), ωn,lB is the weight value from the input layer to the hidden layer and the bias from the input layer to the hidden layer respectively;
the network hidden layer kernel function mapping matrix H is as shown in equation (8):
the output layer is 1 layer, the number of output layer nodes is 1, and the weight β of the output layernThe method can be estimated and solved by a standardized regular least square method, and the calculation formula is shown as formula (9):
βn=(HTH+λI)-1HTY(9)
in equation (9), Y is the output column vector corresponding to the input vector in the training sample space; h is a hidden layer kernel function mapping matrix, lambda is a constant, and I represents an identity matrix.
(3) The training sequence is segmented in sequence by adopting a sliding window model,
A1) mixing X5(2408) And taking 8 continuous data as an input subsequence of a random vector function connection network model training stage, and recording as:wherein, i is 1, 2.., 2401;
and taking the corresponding training subsequence and the training output data as a group, and forming a K-q 2400 group random vector function connection network training subsequence and training output data together.
(4) Inputting the training subsequence to the random vector function connection network model for training, and obtaining the prediction output of the random vector function connection network after the training is finishedCalculating as shown in equation (10):
in formula (10), βlThe weight from the 1 st node of the hidden layer to the output layer; h (x, ω)l,bl) The activation function value corresponding to the 1 st node of the hidden layer.
And fifthly, integrating the prediction of the muscle sound signal: and predicting the integrated muscle sound signal by using the trained random vector function connection network model. The method comprises the following specific steps:
(1) sequence of 592 dataColumn(s) ofAs a prediction sequence, and adopting a sliding window model to segment the prediction sequence in sequence,
is recorded as:wherein i is more than or equal to 1 and less than or equal to 585; the size of the sliding window is 8 and the sliding step is 1.
(2) Will predict the sequenceInputting the input into the trained random vector function connection network, and calculating a predicted value according to the formula (10) to obtain a first predicted value of the muscle tone signal
(3) And (L-K-q +1 ═ 585) sequences with the group length of 8 are sequentially input into the trained random vector function connection network, and the predicted value of the muscle tone signal in the future continuous (L-K-q +1 ═ 585) period is predicted.
As shown in fig. 3, in this experiment, the acceleration signals of 3000 sets of sensors in the X-axis, Y-axis and Z-axis directions are respectively intercepted, and the signals are integrated by using an integration formula, and the integrated signals contain the acceleration information of the sensors in three directions, so that the multidimensional characteristics of the muscle sound signals can be more comprehensively expressed.
As shown in fig. 4, it is a result of denoising the integrated muscle sound signal by the lifting wavelet transform algorithm proposed by the present invention, and it can be seen that the processed signal greatly reduces the high frequency noise contained in the original signal, and improves the signal-to-noise ratio of the muscle sound signal.
As shown in fig. 5, which is a comparison graph of the effect of the muscle tone signal under different prediction algorithms, wherein the solid line is the predicted value of the muscle tone signal of the present invention, and the black dots are the measured values of the muscle tone signal, it can be seen that the prediction accuracy of the method of the present invention is the highest, while the prediction effect based on the other two algorithms is inferior to that of the method of the present invention, and the prediction algorithm of the present invention is the most efficient.
The foregoing shows and describes the general principles, essential features, and advantages of the invention. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, which are merely illustrative of the principles of the invention, but that various changes and modifications may be made without departing from the spirit and scope of the invention, which fall within the scope of the invention as claimed. The scope of the invention is defined by the appended claims and equivalents thereof.
Claims (5)
1. A human muscle sound signal prediction method based on a random vector function connection network technology is characterized by comprising the following steps:
11) acquiring a human body muscle tone signal: the method comprises the following steps of collecting and storing acceleration signals of a wearable muscle sound signal sensor on human elbow joint muscle, obtaining acceleration information of the human elbow joint muscle from the 1 st moment to the L th moment in three directions of an x axis, a y axis and a z axis, and forming 3 groups of muscle sound signals with the lengths of L:
wherein L is more than or equal to 1000 and less than or equal to 10000;
12) integration of muscle tone signals: integrating the muscle tone signals collected by the sensors;
13) denoising integrated muscle sound signals: denoising the integrated muscle sound signals by using a lifting wavelet transform technology;
14) constructing and training a random vector function connection network model: constructing a random vector function connection network model, and training by using the denoised and updated integrated muscle sound signal;
15) prediction of integrated muscle tone signal: and predicting the integrated muscle sound signal by using the trained random vector function connection network model.
2. The method for predicting human muscle tone signals based on stochastic vector function connectivity network technology as claimed in claim 1, wherein the integrating of the muscle tone signals comprises the steps of:
in formula (1), xiRepresenting the integrated muscle sound signal value at the ith moment;acceleration components of the muscle sound signals collected by the sensor at the ith moment in an x axis, a y axis and a z axis are respectively represented, wherein i is 1, 2.
22) Taking the integrated signal as an integrated muscle sound signal, and recording as:
X(L)={x1,x2,…,xL};
wherein x is1Representing the integrated muscle sound signal value at the 1 st moment; x is the number of2Representing the integrated muscle tone signal value at the 2 nd time; x is the number ofLRepresents the integrated muscle tone signal value at the lth time instant.
3. The method for predicting the human muscle sound signal based on the stochastic vector function connection network technology as claimed in claim 1, wherein the denoising process of the integrated muscle sound signal comprises the following steps:
31) for integrated muscle tone signal x (l) { x1,x2,...,xLPerforming splitting treatment, wherein the calculation formula of the splitting treatment is shown as formulas (2-1), (2-2) and (2-3):
in the formula (2-1), Split [ x (l) ] is a splitting operator, which indicates that the sequence x (l) is Split;
in the formulas (2-2) and (2-3), the value range of n is more than or equal to 500 and less than or equal to 5000;
splitting the integrated muscle tone signal x (l) into odd and even subsequences, denoted as:
wherein the content of the first and second substances,representing the odd sequence, x, obtained after splitting1Representing the 1 st signal value, x, of the corresponding integrated muscle tone signal3Representing the 3 rd signal value, x, of the corresponding integrated muscle tone signal2n-1Representing the 2n-1 signal value corresponding to the integrated muscle sound signal;
wherein the content of the first and second substances,denotes the even sequence obtained after splitting, x2Representing the 2 nd signal value, x, of the corresponding integrated muscle signal4Representing the 4 th signal value, x, of the corresponding integrated muscle tone signal2nRepresenting a2 nth signal value corresponding to the integrated muscle sound signal;
32) based on even sequences obtained after splittingPredicting the corresponding odd sequence through the predictor P to obtain the predicted value of the odd sequence, and recording the predicted value as the odd sequenceThe prediction calculation formula is shown as formula (3):
33) using pairs of update operatorsUpdating to obtain an updated integrated muscle sound signal, wherein a calculation formula is shown as formula (5):
in formula (5), X1(L) represents the low-frequency component obtained after the integrated muscle sound signal is updated once,representing a primary high-frequency component, U1The operator is updated once, and the calculation is shown as the formula (6):
34) repeating steps 31) to 33) for the primary low-frequency component X obtained in step 33)1(L) performing denoising updating for M times to obtain
wherein, the value range of M is as follows: m is more than or equal to 1 and less than or equal to 10, X2(L) represents the updated 2-times low-frequency component, X3(L) represents the 3-time low-frequency component obtained after updating, XM(L) represents the M-th low-frequency component obtained after updating,it is shown that the high-frequency component is divided into 2,which represents the high-frequency components of 3 times,representing the high frequency components M times.
4. The method for predicting human muscle tone signals based on stochastic vector function connected network technology as claimed in claim 1, wherein the constructing and training of the stochastic vector function connected network model comprises the following steps:
41) updating the denoised Mth low-frequency component XMAnd (L) taking the first K data as a training sequence of a random vector function connection network, and recording as:
wherein K is more than or equal to 5000 and less than or equal to 8000, and a sequence consisting of the rest L-K data is taken as a prediction sequence and is recorded as:
42) setting parameters of each layer of the random vector function connection network model structure:
wherein, the input layer is 1 layer, the number of nodes of the input layer is m, the value range of m is 1-50, and the weight omega from the input layer to the hidden layern,lIn [0, 1 ]]Randomly generating;
the hidden layer is 1 layer, the number of nodes of the hidden layer is n, the numerical range of n is 1-20, the hidden layer selects a sigmoid function as an activation function, and the form of the sigmoid function is shown as a formula (7):
in the formula (7), ωn,lB is the weight value from the input layer to the hidden layer and the bias from the input layer to the hidden layer respectively;
the network hidden layer kernel function mapping matrix H is as shown in equation (8):
in the formula (8), L represents the number of hidden layer nodes;
the output layer is 1 layer, the number of output layer nodes is 1, and the weight β of the output layernThe method is estimated and solved by a standardized regular least square method, and the calculation formula is shown as formula (9):
βn=(HTH+λI)-1HTY(9)
in equation (9), Y is the output column vector corresponding to the input vector in the training sample space; h is a hidden layer kernel function mapping matrix, lambda is a constant, and I represents an identity matrix;
43) the training sequence is segmented in sequence by adopting a sliding window model,
431) mixing XM(K) And taking the continuous q data as an input subsequence of the random vector function connection network model training stage, and recording as:
wherein q is the size of a sliding window, and the value is consistent with the number of nodes of an input layer of a random vector function connection network, namely q is m;
taking the corresponding training input subsequence and the training output data as a group, and forming K-q groups of random vector function connection network training subsequence and training output data together;
44) inputting the training subsequence to the random vector function connection network model for training, and obtaining the prediction output of the random vector function connection network after the training is finishedCalculating as shown in equation (10):
in formula (10), βlThe weight from the 1 st node of the hidden layer to the output layer; h (x, ω)l,bl) The activation function value corresponding to the 1 st node of the hidden layer.
5. The method for predicting a human muscle tone signal based on stochastic vector function connectivity network as claimed in claim 1, wherein the predicting the integrated muscle tone signal comprises the steps of:
51) sequence composed of L-K dataAs a prediction sequence, and adopting a sliding window model to segment the prediction sequence in sequence,
is recorded as:wherein i is more than or equal to 1 and less than or equal to L-K-q + 1; the size of the sliding window is q, and the sliding step length is 1;
52) will predict the sequenceInputting the input into the trained random vector function connection network, and calculating a predicted value according to the formula (10) to obtain a first predicted value of the muscle tone signal
53) And (L-K-q +1) groups of sequences with the length of q are continuously and sequentially input into the trained random vector function connection network, and the predicted value of the muscle tone signal in a future continuous (L-K-q +1) period is predicted.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911170547.5A CN111110268B (en) | 2019-11-26 | 2019-11-26 | Human body muscle sound signal prediction method based on random vector function connection network technology |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911170547.5A CN111110268B (en) | 2019-11-26 | 2019-11-26 | Human body muscle sound signal prediction method based on random vector function connection network technology |
Publications (2)
Publication Number | Publication Date |
---|---|
CN111110268A true CN111110268A (en) | 2020-05-08 |
CN111110268B CN111110268B (en) | 2022-03-08 |
Family
ID=70496662
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201911170547.5A Active CN111110268B (en) | 2019-11-26 | 2019-11-26 | Human body muscle sound signal prediction method based on random vector function connection network technology |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111110268B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111772669A (en) * | 2020-08-18 | 2020-10-16 | 中国科学院合肥物质科学研究院 | Elbow joint contraction muscle force estimation method based on adaptive long-time and short-time memory network |
WO2022195657A1 (en) * | 2021-03-15 | 2022-09-22 | 日本電信電話株式会社 | Muscle sound extraction method, muscle sound extraction device, and program |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107255647A (en) * | 2017-05-10 | 2017-10-17 | 中国科学院合肥物质科学研究院 | Microelement contents of soil analyzing and predicting method based on X-ray fluorescence spectra and depth confrontation study |
CN107765259A (en) * | 2017-09-18 | 2018-03-06 | 国家电网公司 | A kind of transmission line of electricity laser ranging Signal denoising algorithm that threshold value is improved based on Lifting Wavelet |
CN109171124A (en) * | 2018-09-11 | 2019-01-11 | 华东理工大学 | A kind of muscle signals wireless collection bracelet for Sign Language Recognition |
US20190201691A1 (en) * | 2017-12-31 | 2019-07-04 | Neuroenhancement Lab, LLC | Method and apparatus for neuroenhancement to enhance emotional response |
US20190228579A1 (en) * | 2018-01-25 | 2019-07-25 | Ctrl-Labs Corporation | Techniques for anonymizing neuromuscular signal data |
-
2019
- 2019-11-26 CN CN201911170547.5A patent/CN111110268B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107255647A (en) * | 2017-05-10 | 2017-10-17 | 中国科学院合肥物质科学研究院 | Microelement contents of soil analyzing and predicting method based on X-ray fluorescence spectra and depth confrontation study |
CN107765259A (en) * | 2017-09-18 | 2018-03-06 | 国家电网公司 | A kind of transmission line of electricity laser ranging Signal denoising algorithm that threshold value is improved based on Lifting Wavelet |
US20190201691A1 (en) * | 2017-12-31 | 2019-07-04 | Neuroenhancement Lab, LLC | Method and apparatus for neuroenhancement to enhance emotional response |
US20190228579A1 (en) * | 2018-01-25 | 2019-07-25 | Ctrl-Labs Corporation | Techniques for anonymizing neuromuscular signal data |
CN109171124A (en) * | 2018-09-11 | 2019-01-11 | 华东理工大学 | A kind of muscle signals wireless collection bracelet for Sign Language Recognition |
Non-Patent Citations (1)
Title |
---|
王大庆 等: "一种基于肌动信号的股四头肌收缩力量估计方法研究", 《传感技术学报》 * |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111772669A (en) * | 2020-08-18 | 2020-10-16 | 中国科学院合肥物质科学研究院 | Elbow joint contraction muscle force estimation method based on adaptive long-time and short-time memory network |
CN111772669B (en) * | 2020-08-18 | 2022-08-19 | 中国科学院合肥物质科学研究院 | Elbow joint contraction muscle force estimation method based on adaptive long-time and short-time memory network |
WO2022195657A1 (en) * | 2021-03-15 | 2022-09-22 | 日本電信電話株式会社 | Muscle sound extraction method, muscle sound extraction device, and program |
Also Published As
Publication number | Publication date |
---|---|
CN111110268B (en) | 2022-03-08 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Bao et al. | A CNN-LSTM hybrid model for wrist kinematics estimation using surface electromyography | |
CN103989485A (en) | Human body fatigue evaluation method based on brain waves | |
CN111110268B (en) | Human body muscle sound signal prediction method based on random vector function connection network technology | |
CN111317600A (en) | Artificial limb control method, device, system, equipment and storage medium | |
Peng et al. | Feasibility of NeuCube spiking neural network architecture for EMG pattern recognition | |
CN110710984B (en) | Ankle moment prediction method of recursion cerebellum model based on surface electromyogram signal | |
CN109674445B (en) | Inter-muscle coupling analysis method combining non-negative matrix factorization and complex network | |
Xue et al. | SEMG-based human in-hand motion recognition using nonlinear time series analysis and random forest | |
CN111643321A (en) | Exoskeleton joint angle prediction method and system based on sEMG signals | |
Chen et al. | Towards robot-assisted post-stroke hand rehabilitation: Fugl-Meyer gesture recognition using sEMG | |
Tang et al. | Continuous estimation of human upper limb joint angles by using PSO-LSTM model | |
CN105534517A (en) | Method for removing vehicle motion noise in three-lead electrocardiosignal | |
Gao et al. | Real-time evaluation of the signal processing of sEMG used in limb exoskeleton rehabilitation system | |
Wang et al. | Prediction of sEMG-based tremor joint angle using the RBF neural network | |
CN111772669B (en) | Elbow joint contraction muscle force estimation method based on adaptive long-time and short-time memory network | |
Wang et al. | Deep convolutional neural network for decoding EMG for human computer interaction | |
Wang et al. | Lower limb motion recognition based on surface electromyography signals and its experimental verification on a novel multi-posture lower limb rehabilitation robots☆ | |
KR100994408B1 (en) | Method and device for deducting pinch force, method and device for discriminating muscle to deduct pinch force | |
CN110738093B (en) | Classification method based on improved small world echo state network electromyography | |
Veer | A flexible approach for segregating physiological signals | |
CN116831874A (en) | Lower limb rehabilitation device control method based on electromyographic signals | |
Sun et al. | Classification of sEMG signals using integrated neural network with small sized training data | |
Song et al. | Motion recognition of the bilateral upper-limb rehabilitation using sEMG based on ensemble EMD | |
Tang et al. | sEMG-based estimation of knee joint angles and motion intention recognition | |
Guo et al. | Long short term memory model based continuous estimation of human finger joint angles |
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 |