WO2019014933A1 - 信号计算法中的参数的设定方法及装置 - Google Patents

信号计算法中的参数的设定方法及装置 Download PDF

Info

Publication number
WO2019014933A1
WO2019014933A1 PCT/CN2017/093909 CN2017093909W WO2019014933A1 WO 2019014933 A1 WO2019014933 A1 WO 2019014933A1 CN 2017093909 W CN2017093909 W CN 2017093909W WO 2019014933 A1 WO2019014933 A1 WO 2019014933A1
Authority
WO
WIPO (PCT)
Prior art keywords
parameter
parameters
signal
calculation method
metric
Prior art date
Application number
PCT/CN2017/093909
Other languages
English (en)
French (fr)
Inventor
李国梁
王鑫山
曾端
罗朝洪
杨柯
Original Assignee
深圳市汇顶科技股份有限公司
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 深圳市汇顶科技股份有限公司 filed Critical 深圳市汇顶科技股份有限公司
Priority to PCT/CN2017/093909 priority Critical patent/WO2019014933A1/zh
Priority to CN201780000780.1A priority patent/CN109791564B/zh
Publication of WO2019014933A1 publication Critical patent/WO2019014933A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions

Definitions

  • the present application relates to the field of signal processing technologies, and in particular, to a method and an apparatus for setting parameters in a signal calculation method.
  • Heart rate refers to the number of beats per minute of the human heart, which is a very important physiological indicator in clinical diagnosis.
  • Traditional medical devices require the user to be at rest while measuring heart rate, and are not convenient to carry; therefore, many manufacturers have already produced wearable devices that can perform heart rate measurement, so that users can measure heart rate in daily life.
  • the most commonly used method for heart rate measurement is the photoplethysmography (PPG) method, which uses LEDs to emit light of a specific wavelength and returns, diffuses, diffracts, and reflects through human tissue, and returns the returned optical signal into an electrical signal. Obtain the corresponding PPG signal.
  • the light beam is attenuated by the absorption of human tissue during the propagation of human tissue.
  • the absorption of static tissues such as skin, fat, muscle, etc. is a constant value, and the blood undergoes periodic volume changes due to the contraction and diastolic cycles of the heart.
  • the PPG signal generates a periodic waveform consistent with the heartbeat, so the heartbeat frequency can be measured by the PPG signal, and the photoelectric pulse volume method measures the heart rate as a non-invasive measurement method.
  • existing heart rate calculation methods are mainly sometimes Domain waveform method, spectrum analysis method, independent component analysis method, empirical mode decomposition method, adaptive filter method, etc.
  • parameter variables in these heart rate calculation methods, wherein the range of parameter variables can be based on mathematical principles It is estimated that the specific value of the parameter variable can be set according to experience or clinical physiological characteristics, but the setting of the parameter variable is subjective and uncertain, and the accuracy of the set parameter variable cannot be guaranteed.
  • the purpose of some embodiments of the present application is to provide a method and a device for setting parameters in a signal calculation method, and to evaluate the quality of parameters in a parameter space by using a metric model, and objectively setting a signal calculation method. Accurate parameters to avoid the subjectivity and uncertainty of artificially set parameters.
  • the embodiment of the present application provides a method for setting parameters in a signal calculation method, including: establishing a parameter space, where the parameter space includes at least one set of parameters whose values are within a parameter range of the signal calculation method; The measurement model of the parameters is evaluated; the signals in the preset signal data set and the parameters in the parameter space are substituted into the measurement model, and the parameter merits and demerits are calculated; and the parameters in the signal calculation method are set according to the parameter merits and measures.
  • the embodiment of the present application further provides a parameter setting device in a signal calculation method, including: a first establishing module, configured to establish a parameter space; and a parameter space including at least one set of values in which the value is located in the signal calculation method. a parameter in the range; a second establishing module, configured to establish a measurement model for evaluating the quality of the parameter; and a calculation module, configured to substitute the signal in the preset signal data set and the parameter in the parameter space into the measurement model, and calculate the parameter Good and bad metrics; a setting module for setting parameters in the signal calculation method according to the parameter merits and measures.
  • the embodiment of the present application establishes a parameter space and is established for evaluation. Estimating the quality of the parameters, so that the metrics can be used to evaluate the pros and cons of the parameters in the parameter space, and objectively set more accurate parameters for the signal calculation method, avoiding the subjectivity of the artificial parameters and not Certainty.
  • the parameter space includes multiple sets of parameters; the signal in the preset signal data set and the parameter in the parameter space are substituted into the metric model, and the metric value of the parameter is calculated, specifically: the signal and parameters of the preset signal data set.
  • the parameters in the space are substituted into the metric model, and the parameters of the multi-group parameters are calculated.
  • the parameters in the signal calculation method are set according to the metrics of the parameters, specifically: a set of parameters that minimize the metrics of the parameters. Determined as a parameter in the signal calculation method.
  • This embodiment provides a specific implementation manner of setting parameters in the signal calculation method.
  • the signal in the preset signal data set and the parameter in the parameter space are substituted into the metric model, and the metric value of the parameter is calculated, which is: the signal in the preset signal data set and the parameter in the parameter space are substituted into the metric model.
  • the gradient descent method is used to calculate the parameter merits and demerits that meet the preset conditions. Among them, any set of parameters in the parameter space is used as the initial value of the gradient descent method; according to the parameter merits and measures, the signal calculation method is used.
  • the parameter is specifically: setting a group of parameters corresponding to the parameter merits and values of the preset condition as the parameters in the signal calculation method. This embodiment provides another specific implementation manner of setting parameters in the signal calculation method.
  • the signal in the preset signal data set and the parameter in the parameter space are substituted into the metric model, and the parameter metrics of the plurality of sets of parameters are calculated, specifically: the signal in the preset signal data set and the parameter space
  • the parameters are substituted into the metric model, and the parameter traits of multiple sets of parameters are obtained in parallel.
  • the parameter merits and demerits of multiple sets of parameters are calculated in parallel, which improves the calculation speed, thereby reducing the calculation time.
  • a set of parameters that minimize the parameter merits are set as parameters in the signal calculation method. Before, it also includes: filtering and processing the parameter metrics of the plurality of parameters. In this embodiment, the parameter metrics of the plurality of sets of parameters are filtered, and the error of the parameter metrics caused by noise or abnormal data can be prevented.
  • the filtering processing method is mean smoothing filtering or median filtering. This embodiment provides a specific filtering processing manner.
  • the metric model is a cost function based on a minimum mean square error or a cost function based on a least squares. This embodiment provides a specific implementation manner of the metric model.
  • FIG. 1 is a specific flowchart of a method for setting parameters in a signal calculation method according to a first embodiment of the present application
  • FIG. 2 is a specific flowchart of a method for setting parameters in a signal calculation method according to a second embodiment of the present application
  • FIG. 3 is a specific flowchart of a method of setting parameters in a signal calculation method according to a third embodiment of the present application.
  • FIG. 4 is a specific flowchart of a method for setting parameters in a signal calculation method according to a fourth embodiment of the present application.
  • FIG. 5 is a block diagram showing a setting device of a parameter in a signal calculation method according to a fifth embodiment of the present application.
  • Fig. 6 is a block diagram showing a setting means of a parameter in a signal calculation method according to an eighth embodiment of the present application.
  • the first embodiment of the present application relates to a method for setting parameters in a signal calculation method.
  • the signal may be a biological signal such as a heart rate signal or a blood oxygen signal
  • the signal calculation method may be an adaptive filtering algorithm for calculating a heart rate and calculating a frequency of a heart rate.
  • the domain heart rate algorithm, the blood oxygen algorithm for calculating blood oxygen, and the like, the present embodiment does not impose any limitation on the type of signal and the signal calculation method. Among them, the parameters involved in different signal calculation methods are different.
  • the parameters in the adaptive filtering algorithm for calculating the heart rate are the number of adaptive filter segments, the adaptive filter iteration step, etc.; in the frequency domain heart rate algorithm for calculating the heart rate
  • the parameters are the frequency search start frequency, the termination frequency, the frequency amplitude threshold, etc., and the specific parameter types need to be obtained according to the signal calculation method.
  • the range of values in the signal calculation method can be obtained according to mathematical principles, algorithmic experience, and signal characteristics.
  • a corresponding data set is established for the signal.
  • the signal as the heart rate signal as an example
  • the data set of the established heart rate signal needs to satisfy the following three conditions:
  • Each data in the data set has a unified data format, including unified data length, unified data gain and storage format, etc., to improve the standardization of heart rate algorithm design and apply to parallel processing platforms.
  • the data set needs to have enough data to prevent a small amount of data from causing over-fitting and generalization of the heart rate algorithm, while avoiding the accuracy of the heart rate algorithm in the training data calculation results, in the test data and Other calculations for label data are less accurate.
  • the data in the data set is complete, that is, the data set can cover the application scene data of a plurality of heart rate algorithms.
  • the data set includes data of different heart rate values, and is defined according to physiological characteristics of the human body or product requirements.
  • the heart rate measurement algorithm's measurement range [HeartRate low , HeartRate high ] the true heart rate of the data in the data set can cover every heart rate value in the measurement range [HeartRate low , HeartRate high ]; the second is that the data set includes different signal quality Data, different people's physiological characteristics are different, and the generated heart rate signal quality is different. Therefore, the data set needs to include both data with better signal quality and data with poor signal quality.
  • the heart rate algorithm can not only measure the heart rate at rest, It is also necessary to be able to measure the heart rate of the exercise state (walking, going upstairs, running, climbing, etc.). Therefore, the data set should include static state data, data of different motion states, data of different motion state interferences, data of different measurement parts, and the like.
  • step 101 a parameter space is established.
  • a parameter space S capable of covering the values of all the parameters is established, and the parameter space S includes at least one set of parameters whose values are within the parameter range of the signal calculation method.
  • step 102 a metric model for evaluating the merits of the parameters is established.
  • a metric model based on the accuracy of the signal calculation method is established to evaluate the parameters, that is, for each set of parameters, the metric model can calculate a metric to measure the merits of the set of parameters.
  • the metric model may be a cost function based on a minimum mean square error or a cost function based on a least squares, where the calculated value of the cost function is called a metric value, and the smaller the metric value, the more accurate the calculated signal is, ie, the parameter The higher the accuracy, the closer to the theoretical optimal value; otherwise, the less accurate the calculated signal is, that is, the lower the accuracy of the parameter, the greater the difference from the theoretical optimal value.
  • a cost function based on the minimum mean square error of the heart rate ie, the measurement model.
  • the specific expression of the function is as follows:
  • J(s) indicates that when the set parameter is a set of parameters in the parameter space S, the heart rate algorithm calculates the mean square error of the heart rate;
  • HR s (i, t) indicates when the set parameter is a set of parameters in the parameter space S
  • the i-th signal in the data set calculates the heart rate at time t;
  • HR ref (i, t) represents the reference heart rate of the i-th signal in the data set at time t;
  • M represents a total of M signals in the data set;
  • T represents the data set The length of each signal is T.
  • the process of calculating the metric value based on the cost function of the minimum mean square error of the heart rate can be divided into two steps: firstly, the heart rate algorithm is configured according to the parameter, and the heart rate signal in the data set is input to the heart rate algorithm to calculate the heart rate value; The calculated heart rate value and the reference heart rate value of the heart rate signal in the data set are substituted into the cost function, and the value of the calculated cost function is the metric value.
  • the cost function of the minimum mean square error is used as the metric model, but it is not limited thereto, and the cost function based on the least squares can also be used as the metric model, which is not limited in this embodiment.
  • Step 103 Substituting the signal in the preset signal data set and the parameter in the parameter space into the metric model, and calculating the parameter metric value.
  • the signal in the preset signal data set and the parameter in the parameter space are substituted into the metric model, and the metric value can be calculated and used as the parameter metric value.
  • the metric model is a cost function based on the minimum mean square error of the heart rate
  • the parameter traits of a set of parameters in the parameter space are: when the parameter is set, the heart rate algorithm calculates the heart rate. Mean square error.
  • Step 104 Set parameters in the signal calculation method according to the parameter merits and measures.
  • a better set of parameters is set as a parameter in the signal calculation method according to the parameter meritorious value, or a parameter corresponding to the parameter good and bad measure value that satisfies the preset condition is set as a signal calculation method.
  • the parameters are not limited in this embodiment.
  • step 101 To evaluate the metric model of the parameters, then perform step 101 to establish the parameter space.
  • the present embodiment establishes a parameter space and establishes a metric model for evaluating the merits and demerits of the parameters, so that the metric model can be used to evaluate the pros and cons of the parameters in the parameter space, objectively
  • the signal calculation method sets more accurate parameters to avoid the subjectivity and uncertainty of artificially setting parameters.
  • the second embodiment of the present application relates to a method for setting parameters in a signal calculation method.
  • This embodiment is a refinement of the first embodiment, and the main refinement is that a setting signal calculation method is provided.
  • step 201 a parameter space is established.
  • the parameter space includes a plurality of sets of parameters, and the parameter space includes a finite set of parameters, for example, there are N parameters [p 1 , p 2 , . . . , p i , . . . , p N ] in the signal calculation method,
  • the range of values of each parameter is known.
  • the value of the parameter p i is [p i 1 , p i 2 , ⁇ , p i K , ⁇ , p i Ki ], and a total of K i may be taken. value.
  • the parameter space S is defined as a matrix of K ⁇ N, where N represents the number of parameters, K i represents the number of possible values of the parameter p i .
  • N represents the number of parameters
  • K i represents the number of possible values of the parameter p i .
  • Each row of the parameter space S represents a combination of N parameters, and the parameter space S traverses any combination of N parameter ranges, and there are a total of K combinations, and the parameter space S can cover all parameter values.
  • S [s 1 ,s 2 , ⁇ , s i , ⁇ , s K ] T , [ ⁇ ] T represents matrix transposition
  • s K represents the combination of the Kth of the parameter, called the Kth group parameter
  • K is greater than 1.
  • there are three parameters in the parameter space S which are respectively p 1 , p 2 , and p 3 .
  • a discrete parameter value selection method is provided, that is, the parameter p i is taken in intervals within the range of values, and a discrete parameter value point is formed.
  • the parameter p i is equally spaced within the range of values, so that the values of the various regions in the range of values can be obtained, and the values of the parameters are better in a certain area without being The obtained situation; wherein, the smaller the interval is, the closer the value of the parameter is to the theoretical optimal value of the parameter, and even the parameter value is the theoretical optimal value of the parameter.
  • the theoretical optimal value of the parameter p i is 1.4.
  • the parameter p i of the final set signal calculation method is 1.5; when the value interval is 0.2, the value of p i is [1, 1.2, 1.4, 1.6, 1.8, 2 ], the parameter A in the final set signal calculation method is the theoretical optimal value of 1.4 (that is, the theoretical optimal value can be taken).
  • step 202 a metric model for evaluating the merits of the parameters is established.
  • this step is substantially the same as step 102 in the first embodiment, and details are not described herein again.
  • Step 203 Substituting the signal in the preset signal data set and the parameter in the parameter space into the metric model, and calculating the parameter traits of the plurality of parameters.
  • the metric model is the cost function based on the minimum mean square error of the heart rate in step 102 in the first embodiment
  • the parameter metric value is J ( s K ), indicating that the heart rate algorithm calculates the mean square error of the heart rate when the setting parameter is the kth group parameter s K .
  • Substituting the signal in the preset signal data set and the plurality of parameters s K in the parameter space S into the metric model specifically, taking the metric model as the cost function based on the mean square error as an example, using the metric value of the parameter to be calculated
  • the set of parameters configures the heart rate algorithm and then inputs the plurality of signals in the data set into the configured heart rate algorithm and calculates the heart rate value, and substitutes the calculated heart rate value and the known reference heart rate value into the cost function to obtain the parameters of the set of parameters.
  • the signal in the preset signal data set and the parameter in the parameter space are substituted into the metric model, and the parameter metrics of the plurality of parameters are calculated in parallel, which specifically includes two aspects: one is to calculate the parameters of each group of parameters. In the case of good and bad metrics, the cost function values corresponding to multiple signals in the data set are calculated in parallel, and then the parameter merits and demerits of the set of parameters are calculated. Second, when calculating the parameter merits and values of the plurality of sets of parameters, the parallel computing is more The parameter merits and metrics for each set of parameters in the group parameters. Among them, the parallel calculation results in the parameter merits of multiple sets of parameters, which improves the calculation speed and reduces the calculation time.
  • step 204 a set of parameters that minimize the parameter merits are set as parameters in the signal calculation method.
  • a threshold may be set for the parameter metric value, and a set of parameters is selected as a signal calculation method from all the plurality of parameters corresponding to the parameter metric value less than the threshold.
  • the parameters in the embodiment are not limited in this embodiment.
  • the present embodiment provides a specific implementation manner of setting parameters in the signal calculation method; that is, selecting an optimal set of parameters from a discrete parameter value as a signal.
  • the parameters in the calculation method are not limited to, selecting an optimal set of parameters from a discrete parameter value as a signal.
  • the third embodiment of the present application relates to a method for setting parameters in a signal calculation method.
  • This embodiment is a refinement of the first embodiment, and the main refinement is that another set signal calculation method is provided.
  • step 301 a parameter space is established.
  • the parameter in the parameter space S may take any value within its continuous value range, and the parameter space S may include an infinite group parameter.
  • there are three parameters in the parameter space S which are A, B, and C.
  • the value range of the parameter A is [a 1 , a 3 ]
  • the value range of the parameter B is [b 1 , b 3 ].
  • parameter A can take any value of [a 1 , a 3 ]
  • parameter B can take any value of [b 1 , b 3 ].
  • the parameter C can take any one of [c 1 , c 3 ].
  • a continuous parameter value selection method that is, the parameter p i continuously takes values within the range of values.
  • step 302 a metric model for evaluating the merits of the parameters is established.
  • this step is substantially the same as step 102 in the embodiment, and details are not described herein again.
  • Step 303 substituting the signal in the preset signal data set and the parameter in the parameter space into the metric
  • the model uses the gradient descent method to calculate the metrics of the parameters that meet the preset conditions.
  • the parameter metric value of the preset condition is minJ(s)
  • the specific calculation process is as follows:
  • Update S Where is the alpha update step size.
  • Step 304 Set a set of parameters corresponding to the parameter merits and metrics that meet the preset condition as parameters in the signal calculation method.
  • a set of parameters corresponding to the parameter merits and metrics that satisfy the preset condition calculated by the gradient descent method in step 303 is set as a parameter in the signal calculation method, that is, J(S') is corresponding.
  • the parameter S' is set to a parameter in the signal calculation method.
  • this embodiment provides another specific implementation manner of setting parameters in the signal calculation method, that is, setting a group of parameters corresponding to the parameter merits and demerits of the preset conditions as signals.
  • the parameters in the calculation method are set by setting a group of parameters corresponding to the parameter merits and demerits of the preset conditions as signals.
  • the fourth embodiment of the present application relates to a method for setting parameters in a signal calculation method, and the embodiment is an improvement on the basis of the second embodiment, and the main improvement is: in this embodiment, in the second implementation Before step 204 in the example, the parameter merits and metrics of the plurality of sets of parameters are filtered.
  • Step 401 to step 403 are substantially the same as step 201 to step 203.
  • Step 405 is substantially the same as step 204, and is not described here.
  • the main difference is that in this embodiment, step 404 is added, as follows:
  • Step 404 Filter the parameter metrics of the plurality of parameters.
  • the parameter metrics of the plurality of sets of parameters are filtered, and then proceeds to step 405 to perform filtering processing.
  • the filtering processing may be a mean smoothing filter or a median filtering.
  • this embodiment does not impose any limitation.
  • the present embodiment performs filtering processing on the parameter merits and metrics of multiple sets of parameters, which can prevent errors in the parameter metrics caused by noise or abnormal data.
  • the fifth embodiment of the present invention relates to a parameter setting device in a signal calculation method, wherein the signal may be a heart rate signal, a blood oxygen signal, or the like, and the signal calculation method may be an adaptive filtering algorithm for calculating a heart rate, and calculating a heart rate frequency.
  • the domain heart rate algorithm, the blood oxygen algorithm for calculating blood oxygen, and the like, the present embodiment does not impose any limitation on the type of signal and the signal calculation method. Among them, the parameters involved in different signal calculation methods are different.
  • the parameters in the adaptive filtering algorithm for calculating the heart rate are the number of adaptive filter segments, the adaptive filter iteration step, etc.; in the frequency domain heart rate algorithm for calculating the heart rate
  • the parameters are the frequency search start and stop frequency, the termination frequency, the frequency amplitude threshold, etc., and the specific parameter types need to be obtained according to the signal calculation method.
  • the range of values in the signal calculation method can be obtained according to mathematical principles, algorithmic experience, and signal characteristics.
  • the parameter setting device in the signal calculation method includes: The vertical module 1, the second establishing module 2, the calculating module 3, and the setting module 4.
  • the first establishing module 1 is used to establish a parameter space S; the parameter space S comprises at least one set of parameters whose values are within a parameter range of the signal calculation method.
  • the second establishing module 2 is configured to establish a metric model for evaluating the merits and demerits of the parameters, that is, for each set of parameters, the metric model can calculate a metric value to measure the quality of the set of parameters; the metric model can be based on The cost function of the least mean square error or the cost function based on the least squares, where the metric is the calculated value of the cost function.
  • the calculation module 3 is configured to substitute the signal in the preset signal data set and the parameter in the parameter space S into the metric model, and calculate the parameter metric value.
  • the setting module 4 is configured to set parameters in the signal calculation method according to the parameter merits and measures.
  • the present embodiment can be implemented in cooperation with the first embodiment.
  • the related technical details mentioned in the first embodiment are still effective in this embodiment, and the technical effects that can be achieved in the first embodiment can also be implemented in the present embodiment. To reduce repetition, details are not described herein again. Accordingly, the related art details mentioned in the embodiment can also be applied to the first embodiment.
  • the present embodiment establishes a parameter space capable of covering all parameter values, and establishes a measurement model for evaluating the parameters, so that the parameters of the parameter space can be used by the measurement model.
  • the degree of evaluation objectively set more accurate parameters for the signal calculation method, to avoid the subjectivity and uncertainty of artificially set parameters.
  • the sixth embodiment of the present application relates to a parameter setting device in a signal calculation method.
  • This embodiment is a refinement of the fifth embodiment, and the main refinement is that a setting signal calculation method is provided.
  • the parameter space S established by the first establishing module 1 includes a plurality of sets of parameters, and the parameter space S includes a limited set of parameters.
  • N there are N parameters in the signal calculation method [p 1 , p 2 , . . . , p i , ⁇ , p N ], the range of values of each parameter is known, and the range of the parameter p i is [p i 1 , p i 2 , ⁇ , p i K , ⁇ , p i Ki ]
  • a total of K i may take values.
  • a discrete parameter value selection method is provided, that is, the parameter p i is taken in intervals within the range of values, and a discrete parameter value point is formed.
  • the parameter p i is equally spaced within the range of values, so that the values of the various regions in the range of values can be obtained, and the values of the parameters are better in a certain area without being In the case of the obtained value, the smaller the interval is, the closer the value of the parameter is to the theoretical optimal value of the parameter, and even the parameter value is the theoretical optimal value of the parameter.
  • the theoretical optimal value of the parameter p i is 1.4.
  • the value range of p i is [1, 2].
  • the parameter A value in the final set signal calculation method is 1.5; when the value interval is 0.2, the value of p i is [1, 1.2, 1.4, 1.6, 1.8, 2] Then, the parameter p i in the finally set signal calculation method is a theoretical optimal value of 1.4 (that is, a theoretical optimal value can be obtained).
  • the calculation module 3 is configured to substitute the signal in the preset signal data set and the parameter in the parameter space S into the metric model.
  • the metric model is a cost function based on the mean square error, and the metric of the parameter to be calculated is used.
  • the heart rate algorithm is configured by a set of parameters of the value, and then multiple signals in the data set are respectively input into the configured heart rate algorithm and the heart rate value is calculated, and the calculated heart rate value and the known reference heart rate value are substituted into the cost function to obtain the set of parameters.
  • the calculation module 3 is configured to set the signal of the preset signal data into the parameter space S.
  • the parameters are substituted into the metric model, and the parameter metrics of the plurality of sets of parameters are calculated in parallel; specifically, the calculating module 3 may include a plurality of calculating units, and when calculating the parameter metrics of each set of parameters, the plurality of calculating units Parallel computing the cost function values corresponding to the plurality of signals in the data set; when calculating the parameter merits and metrics of the plurality of sets of parameters, the plurality of sets of parameters are allocated to the plurality of computing units for parallel calculation.
  • the parallel calculation results in the parameter merits of multiple sets of parameters, which improves the calculation speed and reduces the calculation time.
  • the setting module 4 is used to set a set of parameters that minimize the parameter merits and values as parameters in the signal calculation method.
  • a threshold may be set for the parameter metric value, and a group of parameters may be set as a signal from all the plurality of parameters corresponding to the parameter metric value less than the threshold.
  • the parameters in the calculation method are not limited in this embodiment.
  • the present embodiment can be implemented in cooperation with the second embodiment.
  • the technical details mentioned in the second embodiment are still effective in this embodiment, and the technical effects that can be achieved in the second embodiment can also be implemented in the embodiment. To reduce the repetition, details are not described herein again. Accordingly, the related art details mentioned in the embodiment can also be applied to the second embodiment.
  • this embodiment provides a specific implementation manner of setting parameters in the signal calculation method; that is, selecting an optimal set of parameters from a discrete parameter value as a signal.
  • the parameters in the calculation method are not limited to, selecting an optimal set of parameters from a discrete parameter value as a signal.
  • the seventh embodiment of the present application relates to a parameter setting device in a signal calculation method, and this embodiment It is a refinement of the fifth embodiment, and the main refinement is that another specific implementation method of setting parameters in the signal calculation method is provided.
  • the parameter in the parameter space S may take any value within its continuous value range, and the parameter space S may include an infinite group parameter.
  • the parameter space S there are three parameters in the parameter space S, which are respectively p 1 , p 2 , and p 3 .
  • the value of the parameter p 1 is [a 1 , a 3 ], and the value of the parameter p 2 is [b 1 , b 3 ], the value of the parameter p 3 is [c 1 , c 3 ], then the parameter p 1 can take any one of [a 1 , a 3 ], and the parameter p 2 can take [b 1 , b Any one of the values of 3 ], the parameter p 3 can take any one of [c 1 , c 3 ].
  • a continuous parameter value selection method that is, the parameter p i continuously takes values within the range of values.
  • the calculation module 3 is configured to substitute the signal in the preset signal data set and the parameter in the parameter space S into the metric model, and use the gradient descent method to calculate the parameter metric value that satisfies the preset condition; wherein, in the parameter space S Any set of parameters is used as the initial value of the gradient descent method.
  • the setting module 4 is configured to set a set of parameters corresponding to the parameter merit metrics that satisfy the preset condition as parameters in the signal calculation method.
  • the present embodiment can be implemented in cooperation with the third embodiment.
  • the technical details mentioned in the third embodiment are still effective in this embodiment, and the technical effects that can be achieved in the third embodiment can also be implemented in this embodiment. To reduce repetition, details are not described herein again. Accordingly, the related art details mentioned in the embodiment can also be applied to the third embodiment.
  • This embodiment provides another setting signal calculation method with respect to the fifth embodiment.
  • the specific implementation mode of the parameter that is, a set of parameters corresponding to the parameter merits and demerits of the preset condition is set as a parameter in the signal calculation method.
  • the eighth embodiment of the present invention relates to a parameter setting device in a signal calculation method.
  • the present embodiment is an improvement on the basis of the fifth embodiment.
  • the main improvement is that, in this embodiment, please refer to FIG.
  • the setting means of the parameters in the signal calculation method further includes a filtering module 5.
  • the filtering module 5 is configured to filter the parameter metrics of the plurality of parameters.
  • the filter processing mode may be mean smoothing filtering or median filtering, but this embodiment does not impose any limitation.
  • the present embodiment can be implemented in cooperation with the fourth embodiment.
  • the technical details mentioned in the fourth embodiment are still effective in this embodiment.
  • the technical effects that can be achieved in the fourth embodiment can also be implemented in the embodiment. To reduce the repetition, details are not described herein again. Accordingly, the related art details mentioned in the embodiment can also be applied to the fourth embodiment.
  • the present embodiment performs filtering processing on the parameter merits and demerits of the plurality of sets of parameters, and can prevent the error of the parameter pros and cons of the noise or the abnormal data.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Measuring Pulse, Heart Rate, Blood Pressure Or Blood Flow (AREA)
  • Indication And Recording Devices For Special Purposes And Tariff Metering Devices (AREA)

Abstract

本申请部分实施例提供了一种信号计算法中的参数的设定方法及装置。信号计算法中的参数的设定方法包括：建立参数空间，参数空间包括至少一组取值位于信号计算法中的参数取值范围内的参数；建立用于评估参数优劣的度量模型；将预设的信号数据集中的信号、参数空间中的参数代入度量模型，计算得到参数优劣度量值；根据参数优劣度量值设定信号计算法中的参数。采用本申请的实施例，通过度量模型对参数空间中的参数的优劣程度进行评估，客观的为信号计算法设定了较为准确的参数，避免人为设定参数的主观性以及不确定性。

Description

信号计算法中的参数的设定方法及装置 技术领域
本申请涉及信号处理技术领域，特别涉及一种信号计算法中的参数的设定方法及装置。
背景技术
随着生活水平的提高，人们越来越重视生活的健康水平。心率是指人体心脏每分钟跳动的次数，在临床诊断上是一项非常重要的生理指标。传统医疗设备在测量心率时要求使用者处于静止状态，同时不方便携带；因此，很多厂商已经生产出可以进行心率测量的穿戴设备，以便于使用者可以在日常生活状态下进行心率的测量。
现有的最常用的心率测量方法是光电脉搏容积(PPG)法，利用LED发出特定波长的光并经人体组织传播、散射、衍射和反射后返回，将返回的光信号转换为电信号，从而获取相应的PPG信号。光束在人体组织传播过程中，由于人体组织的吸收作用而衰减，其中静态组织如皮肤、脂肪、肌肉等的吸收是恒定值，而血液由于心脏的收缩和舒张周期而产生周期性容积变化，因而PPG信号中产生与心跳一致的周期性波形，所以通过PPG信号可以测量出心跳频率，且光电脉搏容积法测量心率是一种无创无害的测量方法。
发明人发现现有技术至少存在以下问题：现有的心率计算方法主要有时 域波形法、频谱分析法、独立成分分析方法、经验模态分解方法、自适应滤波器方法等，在这些心率计算方法中均有较多的参数变量，其中，参数变量的范围可以根据数学原理估计，参数变量的具体数值可以根据经验或者临床生理特征设定，但是参数变量的设定具有主观性和不确定性，并且无法保证设定的参数变量的准确度。
发明内容
本申请部分实施例的目的在于提供一种信号计算法中的参数的设定方法及装置，通过度量模型对参数空间中的参数的优劣程度进行评估，客观的为信号计算法设定了较为准确的参数，避免人为设定参数的主观性以及不确定性。
本申请实施例提供了一种信号计算法中的参数的设定方法，包括：建立参数空间，参数空间包括至少一组取值位于信号计算法中的参数取值范围内的参数；建立用于评估参数优劣的度量模型；将预设的信号数据集中的信号、参数空间中的参数代入度量模型，计算得到参数优劣度量值；根据参数优劣度量值设定信号计算法中的参数。
本申请实施例还提供了一种信号计算法中的参数的设定装置，包括：第一建立模块，用于建立参数空间；参数空间包括至少一组取值位于信号计算法中的参数取值范围内的参数；第二建立模块，用于建立用于评估参数优劣的度量模型；计算模块，用于将预设的信号数据集中的信号、参数空间中的参数代入度量模型，计算得到参数优劣度量值；设定模块，用于根据参数优劣度量值设定信号计算法中的参数。
本申请实施例相对于现有技术而言，建立了参数空间，并建立了用于评 估参数优劣的度量模型，从而可以通过度量模型对参数空间中的参数的优劣程度进行评估，客观的为信号计算法设定了较为准确的参数，避免人为设定参数的主观性以及不确定性。
另外，参数空间包括多组参数；将预设的信号数据集中的信号、参数空间中的参数代入度量模型，计算得到参数优劣度量值，具体为：将预设的信号数据集中的信号、参数空间中的参数代入度量模型，计算得到多组参数的参数优劣度量值；根据参数优劣度量值设定信号计算法中的参数，具体为：将参数优劣度量值最小的一组参数设定为信号计算法中的参数。本实施例提供了一种设定信号计算法中的参数的具体实现方式。
另外，将预设的信号数据集中的信号、参数空间中的参数代入度量模型，计算得到参数优劣度量值，具体为：将预设的信号数据集中的信号、参数空间中的参数代入度量模型，采用梯度下降法计算得到满足预设条件的参数优劣度量值；其中，以参数空间中的任意一组参数作为梯度下降法的初值；根据参数优劣度量值设定信号计算法中的参数，具体为：将满足预设条件的参数优劣度量值对应的一组参数设定为信号计算法中的参数。本实施例提供了另一种设定信号计算法中的参数的具体实现方式。
另外，将预设的信号数据集中的信号、参数空间中的参数代入度量模型，计算得到多组参数的参数优劣度量值，具体为：将预设的信号数据集中的信号、参数空间中的参数代入度量模型，并行计算得到多组参数的参数优劣度量值。本实施例中，并行计算多组参数的参数优劣度量值，提高了计算速度，从而减少了计算时间。
另外，将参数优劣度量值最小的一组参数设定为信号计算法中的参数之 前，还包括：对多组参数的参数优劣度量值进行滤波处理。本实施例中对多组参数的参数优劣度量值进行滤波处理，能够防止噪声或异常数据导致的参数优劣度量值的误差。
另外，滤波处理方式为均值平滑滤波或中值滤波。本实施例提供了具体的滤波处理方式。
另外，度量模型为基于最小均方误差的代价函数或基于最小二乘的代价函数。本实施例提供了度量模型的具体实现方式。
附图说明
一个或多个实施例通过与之对应的附图中的图片进行示例性说明，这些示例性说明并不构成对实施例的限定，附图中具有相同参考数字标号的元件表示为类似的元件，除非有特别申明，附图中的图不构成比例限制。
图1是根据本申请第一实施例中的信号计算法中的参数的设定方法的具体流程图；
图2是根据本申请第二实施例中的信号计算法中的参数的设定方法的具体流程图；
图3是根据本申请第三实施例中的信号计算法中的参数的设定方法的具体流程图；
图4是根据本申请第四实施例中的信号计算法中的参数的设定方法的具体流程图；
图5是根据本申请第五实施例中的信号计算法中的参数的设定装置的方框示意图；
图6是根据本申请第八实施例中的信号计算法中的参数的设定装置的方框示意图。
具体实施方式
为了使本申请的目的、技术方案及优点更加清楚明白，以下结合附图及实施例，对本申请部分实施例进行进一步详细说明。应当理解，此处所描述的具体实施例仅仅用以解释本申请，并不用于限定本申请。
本申请第一实施例涉及一种信号计算法中的参数的设定方法，信号可以为心率信号、血氧信号等生物信号，信号计算法可以为计算心率的自适应滤波算法、计算心率的频域心率算法、计算血氧的血氧算法等，本实施例对信号的种类以及信号计算法均不作任何限制。其中，不同信号计算法中涉及的参数是不同，例如，计算心率的自适应滤波算法中的参数为自适应滤波器节数、自适应滤波器迭代步长等；计算心率的频域心率算法中的参数为频率搜索起始频率、终止频率、频率幅度阈值等，需要根据信号计算法来获取具体参数种类。信号计算法中的参数的取值范围可以根据数学原理、算法经验以及信号特征来获取。
本实施例中在确定信号的种类后，会为信号建立相应的数据集，以信号为心率信号为例，建立的心率信号的数据集需满足以下三个条件：
1、数据集中每个数据具有统一的数据格式，包括统一的数据长度、统一的数据增益和存储格式等，以提高心率算法设计的标准化和适用于并行处理平台。
2、数据集需要有足够的数据量，以防止少量数据导致心率算法的过拟合、泛化能力差，同时避免心率算法在训练数据计算结果准确度高，在测试数据和 其他为标注数据计算准确度低。
3、数据集中的数据是完备的，即，数据集能够覆盖多种心率算法的应用场景数据，具体来说，一是该数据集包括不同心率值的数据，根据人体生理特征或产品需求，定义了心率测量算法的测量范围[HeartRatelow，HeartRatehigh]，该数据集中的数据的真实心率能够覆盖测量范围[HeartRatelow，HeartRatehigh]内的每一个心率值；二是数据集包括不同信号质量的数据，不同人的生理特征不同，产生的心率信号质量不同，因此该数据集需既包含信号质量较好的数据，也包含信号质量较差的数据；三是心率算法不仅能测量静止状态心率，还要能够测量运动状态(走路、上下楼、跑步、爬山等)心率，因此，该数据集要包括静止状态数据、不同运动状态的数据、不同运动状态干扰的数据、不同测量部位的数据等。
本实施例的信号计算法中的参数的设定方法的具体流程如图1所示。
步骤101，建立参数空间。
具体而言，建立一个能够覆盖所有参数取值的参数空间S，参数空间S包括至少一组取值位于信号计算法中的参数取值范围内的参数。
步骤102，建立用于评估参数优劣的度量模型。
具体而言，建立一种基于信号计算法准确度的度量模型来评估参数优劣，即，对于每一组参数，该度量模型均能计算出一个度量值来衡量该组参数的优劣。其中，度量模型可以为基于最小均方误差的代价函数或基于最小二乘的代价函数，这里代价函数的计算值称为度量值，度量值越小，则说明计算得到的信号越准确，即参数的准确度越高，越接近理论最优值；反之，则说明计算得到的信号越不准确，即参数的准确度越低，与理论最优值相差越大。
下面以信号为心率信号为例，建立一种基于心率最小均方误差的代价函数(即度量模型)，函数具体表达式如下：
Figure PCTCN2017093909-appb-000001
其中，J(s)表示设置参数为参数空间S中的一组参数时，心率算法计算心率的均方误差；HRs(i,t)表示在设置参数为参数空间S中的一组参数时，数据集中第i个信号计算得到时刻t时的心率；HRref(i,t)表示数据集中第i个信号在时刻t时的参考心率；M表示数据集中共有M个信号；T表示数据集中每个信号的时间长度为T。
在基于心率最小均方误差的代价函数(即度量模型)计算度量值的过程可以分为两个步骤：首先根据参数配置心率算法，把数据集中的心率信号输入到心率算法计算得到心率值；然后把计算得到的心率值和数据集中的心率信号已知的参考心率值代入代价函数，计算得到的代价函数的值即为度量值。
需要说明的是，本实施例中以最小均方误差的代价函数作为度量模型，然不限于此，还可以以基于最小二乘的代价函数作为度量模型，本实施例对此不作任何限制。
步骤103，将预设的信号数据集中的信号、参数空间中的参数代入度量模型，计算得到参数优劣度量值。
具体而言，根据步骤102中建立的度量模型，将预设的信号数据集中的信号、参数空间中的参数代入度量模型，可以计算得到度量值并作为参数优劣度量值。例如，度量模型为基于心率最小均方误差的代价函数，则参数空间中的一组参数的参数优劣度量值为：设置参数该组参数时，心率算法计算心率的 均方误差。
步骤104，根据参数优劣度量值设定信号计算法中的参数。
具体而言，根据参数优劣度量值将较优的一组参数设定为信号计算法中的参数，或将满足预设条件的参数优劣度量值对应的参数设定为信号计算法中的参数，本实施例对此不作任何限制。
需要说明的是，图1中只是示意性给出步骤的执行顺序，然本实施例对步骤101与步骤102的实际执行顺序不作任何限制，即，本实施例中也可以先执行步骤102建立用于评估参数优劣的度量模型，然后在执行步骤101建立参数空间。
本实施例相对于现有技术而言，建立了参数空间，并建立了用于评估参数优劣的度量模型，从而可以通过度量模型对参数空间中的参数的优劣程度进行评估，客观的为信号计算法设定了较为准确的参数，避免人为设定参数的主观性以及不确定性。
本申请第二实施例涉及一种信号计算法中的参数的设定方法，本实施例是对第一实施例的细化，主要细化之处在于：提供了一种设定信号计算法中的参数的具体实现方式。
本实施例的信号计算法中的参数的设定方法的具体流程如图2所示。
步骤201，建立参数空间。
具体而言，参数空间包括多组参数，参数空间中包含有限组参数，例如信号计算法中有N个参数[p1，p2，···，pi，···，pN]，每个参数的取值范围已知，参数pi的取值范围为[pi 1，pi 2，···，pi K，···，pi Ki]，共Ki个可能取值。
进一步的，定义参数空间S为一个K×N的矩阵，其中N表示参数的个 数，
Figure PCTCN2017093909-appb-000002
Ki表示参数pi的可能取值个数。参数空间S的每一行表示N个参数的某一种组合，参数空间S遍历N个参数范围的任意组合，共有K种组合，参数空间S能够覆盖所有参数取值。S＝[s1，s2，···，si，···，sK]T，[·]T表示矩阵转置，sK表示参数的第K中组合，称为第K组参数，K大于1。例如，参数空间S中的参数有3个，分别为p1、p2、p3，参数p1的取值范围为[a1，a2，a3]，参数p2的取值范围为[b1，b2，b3]，参数p3的取值范围为[c1，c2，c3]，则参数空间S包括27种组合，S＝[a1b1c1，a1b1c2，a1b1c3，a1b2c1，a1b2c2，a1b2c3，a1b3c1，a1b3c2，a1b3c3，a2b1c1，a2b1c2，a2b1c3，a2b2c1，a2b2c2，a2b2c3，a2b3c1，a2b3c2，a2b3c3，a3b1c1，a3b1c2，a3b1c3，a3b2c1，a3b2c2，a3b2c3，a3b3c1，a3b3c2，a3b3c3]。
本实施例中，提供了一种离散型的参数取值方式，即参数pi在其取值范围内间隔取值，形成一个个离散的参数取值点。
较佳的，参数pi在其取值范围内等间隔取值，从而能够保证取值范围内各个区域的值都能被取到，避免参数较优的取值在某个区域内而没有被取到的情况；其中，取值间隔越小，参数的取值越接近参数的理论最优值，甚至参数取值为参数的理论最优值。例如，参数pi的理论最优值为1.4，在建立参数空间S时，pi的取值范围为[1，2]，当取值间隔为0.5时，pi的取值为[1，1.5，2]，此时，最终设定的信号计算法中的参数pi值为1.5；当取值间隔为0.2时，pi的取值为[1，1.2，1.4，1.6，1.8，2]，则最终设定的信号计算法中的参数A值为理论最优值1.4(即可以取到理论最优值)。
步骤202，建立用于评估参数优劣的度量模型。
具体而言，本步骤与第一实施例中的步骤102大致相同，在此不再赘述。
步骤203，将预设的信号数据集中的信号、参数空间中的参数代入度量模型，计算得到多组参数的参数优劣度量值。
具体而言，由于参数空间S中包含多组参数sK，例如，度量模型为第一实施例中的步骤102中的基于心率最小均方误差的代价函数，则参数优劣度量值为J(sK)，表示在设置参数为第k组参数sK时，心率算法计算心率的均方误差。将预设的信号数据集中的信号、参数空间S中的多组参数sK代入度量模型，具体来说，以度量模型为基于均方误差的代价函数为例，用待计算参数优劣度量值的一组参数配置心率算法然后将数据集中的多个信号分别输入已配置心率算法并计算得到心率值，将计算得到的心率值和已知参考心率值代入该代价函数求得该组参数的参数优劣度量值；对于每组参数，均采用上述计算方法，即可以分别计算得到多组参数sK的参数优劣度量值。
较佳的，将预设的信号数据集中的信号、参数空间中的参数代入度量模型，并行计算得到多组参数的参数优劣度量值，具体包括两个方面：一是计算每组参数的参数优劣度量值时，并行计算数据集中的多个信号对应的代价函数值，继而计算出该组参数的参数优劣度量值；二是计算多组参数的参数优劣度量值时，并行计算多组参数中的每组参数的参数优劣度量值。其中，并行计算得到多组参数的参数优劣度量值，提高了计算速度，从而减少了计算时间。
步骤204，将参数优劣度量值最小的一组参数设定为信号计算法中的参数。
具体而言，参数优劣度量值越小表示参数取值越接近最优值，因此，将参数优劣度量值最小的一组参数设定为信号计算法中的参数，可以保证参数取值的准确度；其中，若参数优劣度量值最小的参数不止一组，则可以从其中任 选一组设定为信号计算法中的参数，本实施例对此不作任何限制。
值得一提的是，本实施例中还可以为参数优劣度量值设定一阈值，从所有小于该阈值的参数优劣度量值对应的多组参数中选择一组参数设定为信号计算法中的参数，然本实施例对此不作任何限制。
本实施例相对于第一实施例而言，提供了一种设定信号计算法中的参数的具体实现方式；即从离散型的参数取值中，选取最优的一组参数设定为信号计算法中的参数。
本申请第三实施例涉及一种信号计算法中的参数的设定方法，本实施例是对第一实施例的细化，主要细化之处在于：提供了另一种设定信号计算法中的参数的具体实现方式。
本实施例的信号计算法中的参数的设定方法的具体流程如图3所示。
步骤301，建立参数空间。
具体而言，参数空间S中的参数可在其连续的取值范围内取任意值，参数空间S中可以包括无限组参数。例如，参数空间S中的参数有3个，分别为A、B、C，参数A的取值范围为[a1，a3]，参数B的取值范围为[b1，b3]，参数C的取值范围为[c1，c3]，则参数A可以取[a1，a3]中的任意一个值，参数B可以取[b1，b3]中的任意一个值，参数C可以取[c1，c3]中的任意一个值。
本实施例中，提供了一种连续型的参数取值方式，即参数pi在其取值范围内连续取值。
步骤302，建立用于评估参数优劣的度量模型。
具体而言，本步骤与实施例中的步骤102大致相同，在此不再赘述。
步骤303，将预设的信号数据集中的信号、参数空间中的参数代入度量 模型，采用梯度下降法计算得到满足预设条件的参数优劣度量值。
具体而言，度量模型J(s)，则满足预设条件的参数优劣度量值为minJ(s)，具体计算过程如下：
1、以步骤301中建立的参数空间中的任意一组参数sK作为梯度下降法的初值。
2、求J(s)关于S的梯度，即，
Figure PCTCN2017093909-appb-000003
3、更新S，
Figure PCTCN2017093909-appb-000004
其中为α更新步长。
4、将S′代入J(s)，计算得到J(S′)，如果|J(S′)-J(sK)|小于预设的阈值，则说明参数优劣度量值J(S′)满足预设条件，停止迭代；否则，令S＝S′，重复上述步骤2到步骤4，直至获取满足预设条件的参数优劣度量值。其中，设置预设的阈值越小，获取的满足预设条件的参数优劣度量值对应的参数越接近理论最优值，本实施例对此不作任何限制。
步骤304，将满足预设条件的参数优劣度量值对应的一组参数设定为信号计算法中的参数。
具体而言，将步骤303中采用梯度下降法计算得到的满足预设条件的参数优劣度量值对应的一组参数设定为信号计算法中的参数，即，将J(S′)对应的参数S′设定为信号计算法中的参数。
本实施例相对于第一实施例而言，提供了另一种设定信号计算法中的参数的具体实现方式，即将满足预设条件的参数优劣度量值对应的一组参数设定为信号计算法中的参数。
本申请第四实施例涉及一种信号计算法中的参数的设定方法，本实施例是在第二实施例基础上的改进，主要改进之处在于：本实施例中，在第二实施 例中的步骤204之前，对多组参数的参数优劣度量值进行滤波处理。
本实施例的信号计算法中的参数的设定方法的具体流程如图4所示。
其中，步骤401至步骤403与步骤201至步骤203大致相同，步骤405与步骤204大致相同，在此不再赘述，主要不同之处在于，本实施例中，增加了步骤404，具体如下：
步骤404，对多组参数的参数优劣度量值进行滤波处理。
具体而言，在步骤203(本实施例中步骤403)计算得到多组参数的参数优劣度量值后，对多组参数的参数优劣度量值进行滤波处理，继而进入步骤405，在滤波处理后的多组参数的参数优劣度量值中，将参数优劣度量值最小的一组参数设定为信号计算法中的参数。其中，滤波处理的方式可以是均值平滑滤波或中值滤波，然本实施例对此不作任何限制。
本实施例相对于第二实施例而言，对多组参数的参数优劣度量值进行滤波处理，能够防止噪声或异常数据导致的参数优劣度量值的误差。
本申请第五实施例涉及一种信号计算法中的参数的设定装置，信号可以为心率信号、血氧信号等生物信号，信号计算法可以为计算心率的自适应滤波算法、计算心率的频域心率算法、计算血氧的血氧算法等，本实施例对信号的种类以及信号计算法均不作任何限制。其中，不同信号计算法中涉及的参数是不同，例如，计算心率的自适应滤波算法中的参数为自适应滤波器节数、自适应滤波器迭代步长等；计算心率的频域心率算法中的参数为频率搜索起止频率、终止频率、频率幅度阈值等，需要根据信号计算法来获取具体参数种类。信号计算法中的参数的取值范围可以根据数学原理、算法经验以及信号特征来获取。
本实例中，请参考图5，信号计算法中的参数的设定装置包括：第一建 立模块1、第二建立模块2、计算模块3以及设定模块4。
第一建立模块1用于建立参数空间S；参数空间S包括至少一组取值位于信号计算法中的参数取值范围内的参数。
第二建立模块2用于建立用于评估参数优劣的度量模型，即，对于每一组参数，该度量模型均能计算出一个度量值来衡量该组参数的优劣；度量模型可以为基于最小均方误差的代价函数或基于最小二乘的代价函数，这里的度量值即为代价函数的计算值。
计算模块3用于将预设的信号数据集中的信号、参数空间S中的参数代入度量模型，计算得到参数优劣度量值。
设定模块4用于根据参数优劣度量值设定信号计算法中的参数。
由于第一实施例与本实施例相互对应，因此本实施例可与第一实施例互相配合实施。第一实施例中提到的相关技术细节在本实施例中依然有效，在第一实施例中所能达到的技术效果在本实施例中也同样可以实现，为了减少重复，这里不再赘述。相应地，本实施例中提到的相关技术细节也可应用在第一实施例中。
本实施例相对于现有技术而言，建立了能够覆盖所有参数取值的参数空间，并建立了用于评估参数优劣的度量模型，从而可以通过度量模型对参数空间中的参数的优劣程度进行评估，客观的为信号计算法设定了较为准确的参数，避免人为设定参数的主观性以及不确定性。
本申请第六实施例涉及一种信号计算法中的参数的设定装置，本实施例是对第五实施例的细化，主要细化之处在于:提供了一种设定信号计算法中的参数的具体实现方式。
本实施例中的信号计算法中的参数的设定装置的方框示意图请参考图5。其中，第一建立模块1建立的参数空间S包括多组参数，参数空间S中包含有限组参数，例如信号计算法中有N个参数[p1，p2，···，pi，···，pN]，每个参数的取值范围已知，参数pi的取值范围为[pi 1，pi 2，···，pi K，···，pi Ki]，共Ki个可能取值。
本实施例中，提供了一种离散型的参数取值方式，即参数pi在其取值范围内间隔取值，形成一个个离散的参数取值点。
较佳的，参数pi在其取值范围内等间隔取值，从而能够保证取值范围内各个区域的值都能被取到，避免参数较优的取值在某个区域内而没有被取到的情况，取值间隔越小，参数的取值越接近参数的理论最优值，甚至参数取值为参数的理论最优值。例如，参数pi的理论最优值为1.4，在建立参数空间S时，pi的取值范围为[1，2]，当取值间隔为0.5时，pi的取值为[1，1.5，2]，此时，最终设定的信号计算法中的参数A值为1.5；当取值间隔为0.2时，pi的取值为[1，1.2，1.4，1.6，1.8，2]，则最终设定的信号计算法中的参数pi值为理论最优值1.4(即可以取到理论最优值)。
计算模块3用于将预设的信号数据集中的信号、参数空间S中的参数代入度量模型，具体来说，以度量模型为基于均方误差的代价函数为例，用待计算参数优劣度量值的一组参数配置心率算法，然后将数据集中的多个信号分别输入已配置心率算法并计算得到心率值，将计算得到的心率值和已知参考心率值代入该代价函数求得该组参数的参数优劣度量值；对于每组参数，均采用上述计算方法，即可以分别计算得到多组参数sK的参数优劣度量值。
较佳的，计算模块3用于将预设的信号数据集中的信号、参数空间S中 的参数代入度量模型，并行计算得到多组参数的参数优劣度量值；具体而言，计算模块3可以包括多个计算单元，在计算每组参数的参数优劣度量值时，多个计算单元并行计算数据集中的多个信号对应的代价函数值；计算多组参数的参数优劣度量值时，将多组参数分配到多个计算单元并行计算。其中，并行计算得到多组参数的参数优劣度量值，提高了计算速度，从而减少了计算时间。
设定模块4用于将参数优劣度量值最小的一组参数设定为信号计算法中的参数。参数优劣度量值越小表示参数取值的越接近最优值，因此，将参数优劣度量值最小的一组参数设定为信号计算法中的参数，可以保证参数取值的准确度；其中，若参数优劣度量值最小的参数不止一组，则可以从其中任选一组设定为信号计算法中的参数，本实施例对此不作任何限制。
值得一提的是，本实施例中还可以为参数优劣度量值设定一阈值，从所有小于该阈值的参数优劣度量值对应的多组参数中选择一组参数都可以设定为信号计算法中的参数，然本实施例对此不作任何限制。
由于第二实施例与本实施例相互对应，因此本实施例可与第二实施例互相配合实施。第二实施例中提到的相关技术细节在本实施例中依然有效，在第二实施例中所能达到的技术效果在本实施例中也同样可以实现，为了减少重复，这里不再赘述。相应地，本实施例中提到的相关技术细节也可应用在第二实施例中。
本实施例相对于第五实施例而言，提供了一种设定信号计算法中的参数的具体实现方式；即从离散型的参数取值中，选取最优的一组参数设定为信号计算法中的参数。
本申请第七实施例涉及一种信号计算法中的参数的设定装置，本实施例 是对第五实施例的细化，主要细化之处在于:提供了另一种设定信号计算法中的参数的具体实现方式。
本实施例中的信号计算法中的参数的设定装置的方框示意图请参考图5。其中，第一建立模块1建立的参数空间S时，参数空间S中的参数可在其连续的取值范围内取任意值，参数空间S中可以包括无限组参数。例如，参数空间S中的参数有3个，分别为p1、p2、p3，参数p1的取值范围为[a1，a3]，参数p2的取值范围为[b1，b3]，参数p3的取值范围为[c1，c3]，则参数p1可以取[a1，a3]中的任意一个值，参数p2可以取[b1，b3]中的任意一个值，参数p3可以取[c1，c3]中的任意一个值。
本实施例中，提供了一种连续型的参数取值方式，即参数pi在其取值范围内连续取值。
计算模块3用于将预设的信号数据集中的信号、参数空间S中的参数代入度量模型，采用梯度下降法计算得到满足预设条件的参数优劣度量值；其中，以参数空间S中的任意一组参数作为梯度下降法的初值。
设定模块4用于将满足预设条件的参数优劣度量值对应的一组参数设定为信号计算法中的参数。
由于第三实施例与本实施例相互对应，因此本实施例可与第三实施例互相配合实施。第三实施例中提到的相关技术细节在本实施例中依然有效，在第三实施例中所能达到的技术效果在本实施例中也同样可以实现，为了减少重复，这里不再赘述。相应地，本实施例中提到的相关技术细节也可应用在第三实施例中。
本实施例相对于第五实施例而言，提供了另一种设定信号计算法中的 参数的具体实现方式，即将满足预设条件的参数优劣度量值对应的一组参数设定为信号计算法中的参数。
本申请第八实施例涉及一种信号计算法中的参数的设定装置，本实施例是在第五实施例基础上的改进，主要改进之处在于：本实施例中，请参考图6，信号计算法中的参数的设定装置还包括滤波模块5。
滤波模块5用于对多组参数的参数优劣度量值进行滤波处理。滤波处理方式可以为均值平滑滤波或中值滤波，然本实施例对此不作任何限制。
由于第四实施例与本实施例相互对应，因此本实施例可与第四实施例互相配合实施。第四实施例中提到的相关技术细节在本实施例中依然有效，在第四实施例中所能达到的技术效果在本实施例中也同样可以实现，为了减少重复，这里不再赘述。相应地，本实施例中提到的相关技术细节也可应用在第四实施例中。
本实施例相对于第六实施例而言，对多组参数的参数优劣度量值进行滤波处理，能够防止噪声或异常数据导致的参数优劣度量值的误差。
本领域的普通技术人员可以理解，上述各实施例是实现本申请的具体实施例，而在实际应用中，可以在形式上和细节上对其作各种改变，而不偏离本申请的精神和范围。

Claims (14)

  1. 一种信号计算法中的参数的设定方法，其特征在于，包括：
    建立参数空间，所述参数空间包括至少一组取值位于所述信号计算法中的参数取值范围内的参数；
    建立用于评估参数优劣的度量模型；
    将预设的信号数据集中的信号、所述参数空间中的参数代入所述度量模型，计算得到参数优劣度量值；
    根据所述参数优劣度量值设定所述信号计算法中的参数。
  2. 如权利要求1所述的方法，其特征在于，所述参数空间包括多组参数；
    所述将预设的信号数据集中的信号、所述参数空间中的参数代入所述度量模型，计算得到参数优劣度量值，具体为：
    将预设的信号数据集中的信号、所述参数空间中的参数代入所述度量模型，计算得到多组参数的参数优劣度量值；
    所述根据所述参数优劣度量值设定所述信号计算法中的参数，具体为：
    将参数优劣度量值最小的一组参数设定为所述信号计算法中的参数。
  3. 如权利要求1所述的方法，其特征在于，所述将预设的信号数据集中的信号、所述参数空间中的参数代入所述度量模型，计算得到参数优劣度量值，具体为：
    将预设的信号数据集中的信号、所述参数空间中的参数代入所述度量模型，采用梯度下降法计算得到满足预设条件的参数优劣度量值；其中，以所述参数空间中的任意一组参数作为梯度下降法的初值；
    所述根据所述参数优劣度量值设定所述信号计算法中的参数，具体为：
    将满足预设条件的所述参数优劣度量值对应的一组参数设定为所述信号计算法中的参数。
  4. 如权利要求2所述的方法，其特征在于，所述将预设的信号数据集中的信号、所述参数空间中的参数代入所述度量模型，计算得到多组参数的参数优劣度量值，具体为：
    将预设的信号数据集中的信号、所述参数空间中的参数代入所述度量模型，并行计算得到多组参数的参数优劣度量值。
  5. 如权利要求2所述的方法，其特征在于，所述将参数优劣度量值最小的一组参数设定为所述信号计算法中的参数之前，还包括：
    对所述多组参数的参数优劣度量值进行滤波处理。
  6. 如权利要求5所述的方法，其特征在于，所述滤波处理方式为均值平滑滤波或中值滤波。
  7. 如权利要求1-6任一项所述的方法，其特征在于，所述度量模型为基于最小均方误差的代价函数或基于最小二乘的代价函数。
  8. 一种信号计算法中的参数的设定装置，其特征在于，包括：
    第一建立模块，用于建立参数空间；所述参数空间包括至少一组取值位于所述信号计算法中的参数取值范围内的参数；
    第二建立模块，用于建立用于评估参数优劣的度量模型；
    计算模块，用于将预设的信号数据集中的信号、所述参数空间中的参数代入所述度量模型，计算得到参数优劣度量值；
    设定模块，用于根据所述参数优劣度量值设定所述信号计算法中的参数。
  9. 如权利要求8所述的装置，其特征在于，所述参数空间包括多组参数；
    所述计算模块用于将预设的信号数据集中的信号、所述参数空间中的参数代入所述度量模型，计算得到多组参数的参数优劣度量值；
    所述设定模块用于将参数优劣度量值最小的一组参数设定为所述信号计算法中的参数。
  10. 如权利要求8所述的装置，其特征在于，所述计算模块用于将预设的信号数据集中的信号、所述参数空间中的参数代入所述度量模型，采用梯度下降法计算得到满足预设条件的参数优劣度量值；其中，以所述参数空间中的任意一组参数作为梯度下降法的初值；
    所述设定模块用于将满足预设条件的所述参数优劣度量值对应的一组参数设定为所述信号计算法中的参数。
  11. 如权利要求9所述的装置，其特征在于，所述计算模块用于将预设的信号数据集中的信号、所述参数空间中的参数代入所述度量模型，并行计算得到多组参数的参数优劣度量值。
  12. 如权利要求9所述的装置，其特征在于，所述装置还包括滤波模块；所述滤波模块用于对所述多组参数的参数优劣度量值进行滤波处理。
  13. 如权利要求12所述的装置，其特征在于，所述滤波处理方式为均值平滑滤波或中值滤波。
  14. 如权利要求8-13任一项所述的装置，其特征在于，所述度量模型为基于最小均方误差的代价函数或基于最小二乘的代价函数。
PCT/CN2017/093909 2017-07-21 2017-07-21 信号计算法中的参数的设定方法及装置 WO2019014933A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
PCT/CN2017/093909 WO2019014933A1 (zh) 2017-07-21 2017-07-21 信号计算法中的参数的设定方法及装置
CN201780000780.1A CN109791564B (zh) 2017-07-21 2017-07-21 信号计算法中的参数的设定方法及装置

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2017/093909 WO2019014933A1 (zh) 2017-07-21 2017-07-21 信号计算法中的参数的设定方法及装置

Publications (1)

Publication Number Publication Date
WO2019014933A1 true WO2019014933A1 (zh) 2019-01-24

Family

ID=65014839

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/093909 WO2019014933A1 (zh) 2017-07-21 2017-07-21 信号计算法中的参数的设定方法及装置

Country Status (2)

Country Link
CN (1) CN109791564B (zh)
WO (1) WO2019014933A1 (zh)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1430832A (zh) * 2000-04-26 2003-07-16 记忆链公司 利用定期测试的分集接收
CN102722103A (zh) * 2012-05-18 2012-10-10 镇江雅迅软件有限责任公司 基于模糊分析和可视化技术的多目标参数寻优方法
WO2014168886A1 (en) * 2013-04-08 2014-10-16 Qualcomm Incorporated Low complexity blind detection of transmission parameters of interferers
CN104200087A (zh) * 2014-06-05 2014-12-10 清华大学 用于机器学习的参数寻优及特征调优的方法及系统

Family Cites Families (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090171226A1 (en) * 2007-12-31 2009-07-02 Nellcor Puritan Bennett Llc System and method for evaluating variation in the timing of physiological events
JP5493319B2 (ja) * 2008-09-11 2014-05-14 富士通株式会社 モデルパラメータ抽出プログラム
CN102663417B (zh) * 2012-03-19 2015-02-25 河南工业大学 一种小样本数据模式识别的特征选择方法
US10769320B2 (en) * 2012-12-18 2020-09-08 Kla-Tencor Corporation Integrated use of model-based metrology and a process model
CN105147248B (zh) * 2015-07-30 2019-02-05 华南理工大学 基于生理信息的抑郁症评估系统及其评估方法
CN105335619B (zh) * 2015-11-10 2019-02-01 广西大学 一种岩爆过程数值计算模型参数反分析的协同优化法
CN106599230A (zh) * 2016-12-19 2017-04-26 北京天元创新科技有限公司 一种分布式数据挖掘模型评估的方法与系统

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1430832A (zh) * 2000-04-26 2003-07-16 记忆链公司 利用定期测试的分集接收
CN102722103A (zh) * 2012-05-18 2012-10-10 镇江雅迅软件有限责任公司 基于模糊分析和可视化技术的多目标参数寻优方法
WO2014168886A1 (en) * 2013-04-08 2014-10-16 Qualcomm Incorporated Low complexity blind detection of transmission parameters of interferers
CN104200087A (zh) * 2014-06-05 2014-12-10 清华大学 用于机器学习的参数寻优及特征调优的方法及系统

Also Published As

Publication number Publication date
CN109791564B (zh) 2023-06-16
CN109791564A (zh) 2019-05-21

Similar Documents

Publication Publication Date Title
US10952682B2 (en) System and method of a biosensor for detection of health parameters
KR102622403B1 (ko) 생물학적 데이터 처리
US9936886B2 (en) Method for the estimation of the heart-rate and corresponding system
CN108478206B (zh) 运动状态下基于脉搏波的心率监测方法
US20190295729A1 (en) Universal non-invasive blood glucose estimation method based on time series analysis
CN102429649B (zh) 连续血压测量装置
CN102488503B (zh) 连续血压测量装置
CN114867410A (zh) 使用人工智能进行连续无创血压监测的校准
US20120136261A1 (en) Systems and methods for calibrating physiological signals with multiple techniques
Yang et al. Estimation and validation of arterial blood pressure using photoplethysmogram morphology features in conjunction with pulse arrival time in large open databases
JP2001517128A (ja) 血液成分値の最良な評価を得るために調停し，高調波を除去するための方法および装置
JP2001517991A (ja) データ信号適応平均化方法及び装置
Lang Beyond Fitbit: a critical appraisal of optical heart rate monitoring wearables and apps, their current limitations and legal implications
RU2010146145A (ru) Неинвазивный способ и устройство для определения стадий поверхностного и глубокого сна
CN109497977A (zh) 人体心率和血氧饱和度检测方法和装置
US10470717B2 (en) Pulse validation
US20190060568A1 (en) System and method of a biosensor for detection of vasodilation
Liu et al. Cuffless blood pressure measurement using smartwatches: a large-scale validation study
CN111466899A (zh) 一种基于mpga-miv-bp模型簇的连续血压无创监测方法
US11219376B2 (en) Identifying anomalous autoregulation state values
US20200065649A1 (en) Method for robust and noise-tolerant SpO2 determination
CN109643579B (zh) 质量评估方法及装置、模型建立方法及模块、穿戴设备
WO2019014933A1 (zh) 信号计算法中的参数的设定方法及装置
US20210401312A1 (en) Viscoelasticity characteristics acquisition device, viscoelasticity characteristics acquisition method, viscoelasticity characteristics acquisition program, and recording medium recording said program
TWI733378B (zh) 建立血壓模型的方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 17918028

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 17918028

Country of ref document: EP

Kind code of ref document: A1