CN109934139A - A kind of muscle electrical signal paths combined optimization method based on swarm intelligence algorithm - Google Patents
A kind of muscle electrical signal paths combined optimization method based on swarm intelligence algorithm Download PDFInfo
- Publication number
- CN109934139A CN109934139A CN201910154588.9A CN201910154588A CN109934139A CN 109934139 A CN109934139 A CN 109934139A CN 201910154588 A CN201910154588 A CN 201910154588A CN 109934139 A CN109934139 A CN 109934139A
- Authority
- CN
- China
- Prior art keywords
- data
- ant
- road
- precision
- channel
- 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
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
Muscle electrical signal paths combined optimization method based on swarm intelligence algorithm, includes the following steps: step 1, acquires the data in each channel, and data are pre-processed, step 2 needs to choose number of channels and feature, step 3, precision establishment method is established, precision budget is carried out.Step 4, the precision that step 3 is obtained are calculated using ant group algorithm, obtain optimal precision.Invention more accurately solves complicated np problem by heuritic approach, and greatly reduces the training time.And after being optimized channel selection.
Description
Technical field
The present invention relates to a kind of muscle electrical signal paths combined optimization methods
Background technique
Surface electromyogram signal contains a large amount of characterization muscle biological informations, can differentiate some hands by these information
Gesture acts and produces mechanical prosthetic limb corresponding with people.But traditional mechanical prosthetic limb precision is not high, if desired improves essence
Degree must just increase channel amount, cause the high price of mechanical prosthetic limb, and the data similarity degree in channel is higher, have larger
Repeatability, but include certain independent data.Therefore, how in the case where reducing the quantity in channel, guarantee mechanical
It is a technology to be broken through that the error rate of arm, which does not increase,.
Summary of the invention
The present invention will overcome the above problem of the prior art, propose a kind of muscle electrical signal paths based on swarm intelligence algorithm
Combined optimization method, for solving the problems, such as CHANNEL OPTIMIZATION.
Present example provides the definition that optimization algorithm includes: objective function and constraint condition.To objective function, constraint item
Part is formed by component and is integrated, and establishes algorithm model.
Muscle electrical signal paths combined optimization method based on swarm intelligence algorithm, includes the following steps:
Step 1: acquiring the data in each channel, and data are pre-processed, the pretreatment to data specifically includes:
11. pair some abnormal datas carry out smoothing denoising, abnormal data includes: more than maximum, minimum measured value data
The data not changed after occurring with movement;
12. a pair invalid data arranges, wherein invalid data includes: replacing data value with data mean value is 0 Value Data;
Step 2: needing to choose number of channels and feature, specifically including to reduce number of channels:
21. setting mechanical prosthetic limb and sharing N number of channel is respectively S1、S2...SNSurface flesh is carried out in the case where the time is the period of T
The extraction of electric signal;
22. using having the surface electromyogram signal data progress feature choosing for extracting characterization method to obtaining in T time section
It takes;If extraction number of features is M, remember that M feature is C1、C2...CM;
Step 3: establishing precision establishment method, the calculating of precision in the case of a small amount of channel and feature is established;Wherein precision is true
It is vertical to include:
31. pair having N number of channel and M characteristic carrying out classification gesture information, algorithm classification is closed on using K, specifically
Include:
A, the measurement using Euclidean distance to two similarity degrees;
Euclidean distance refers to that n dimension Euclidean space is the set of a coordinate points, and each of it point X can be expressed as (x1、
x2…xn) wherein xi(i=12 ... n) is i-th of coordinate, two point A=(a that real number is known as X1、a2…an) and B=(b1、b2…
bn) the distance between d (AB) be defined as following formula
B, selecting K value, i.e. K most like samples, K value can use natural number in a sample space,
C, the formulation of categorised decision rule, using simple majority vote rule and apart from rule;Simple majority vote rule
Include: with apart from rule
C1. simple majority vote refers to that needing to obtain responses more than half just directly judges point classification situation;
C2. refer to that by distance, the longer weight of distance is lower as in weight input simple majority vote apart from rule;Distance is got over
Short weight is higher, then carries out judging point classification situation;
D, it establishes kd-tree and training data includes:
Input is target point X, kd-tree;Output is affiliated classification situation;
I, preparatory function A: when accessing each node, if simultaneously less than K data are added in the node by ambient data
In, if selecting K data being closer beyond k.
Ii, from root node, recursively access kd-tree downwards, if the coordinate of target X leading dimension is less than cut-off
Coordinate, then be moved to left child node, is otherwise moved to right child node, until node is leaf node;Execute preparatory function A;
Iii, recursive upward rollback, perform the following operation in each node: 1, executing preparatory function A, check child node
Whether capacity is less than, or the data more closer than top heap data, if so, executing i operation;
Iv, when retracting root node, training terminates;
32. pair a small amount of channel and feature establish classification information, wherein a small amount of channel and feature include: the candidate solution of problem:
S2、S3...SN-1,C2、C3...CM-1;Classification method include in step 31 K close on algorithm classification;
33. the determination of precision, specifically includes:
(1) scramble data matrix is established, matrix content includes
Equipped with W label, respectively Z1、Z2、...ZW;If training tag along sort n and testing classification label m, (m, n < W),
Matrix label fmnAdding 1, F is confusion matrix.All classification data are written in confusion matrix,
(2) ∑ f is calculatedijThe value of (i, j < w), as a result X, calculates ∑ fiiThe value of (i < w), accuracy value are
Step 4: carrying out optimal solution search to the obtained precision of step 3, is searched for, is specifically included using ant group algorithm:
41. initialization with Ant colony parameter, by the road for being selected as ant selection in channel, by the reciprocal as correspondence of precision
The length of road;Assuming that n is road amount, m is ant number, and L is corresponding link length, then m ant is put on n road,
Ant k carries out Model choices according to the pheromones length on each road, does not stop to change pheromone concentration on road;Ant choosing
The publicity for selecting road is
Tabu is taboo list in formula, the road that record ant has been passed by, with the target for forbidding ant to select to have been selected.
τijFor residual risk on starting point i to terminal j road, α τijRelative importance, ηijFor heuristic greedy method, β ηijPhase
To significance level,Indicate the desired value of ant k selection road,For the total amount of pheromones on each road.
42. the update of pheromones, m ant carries out the update of pheromones before choosing quality preferably;Assuming that changing in kth time
After generation terminates, ant is to pheromones τijIt is updated, publicity is as follows:
τij=(1-p) × τij(t)+Δτij(t)
Wherein Q is pheromones intensity;
The optimal value Y of precision is obtained by ant search algorithmMAX。
The invention has the advantages that more accurately being solved by heuritic approach to complicated np problem, and big
The training time is reduced greatly.And after being optimized channel selection.
Detailed description of the invention
Fig. 1 is flow chart of the method for the present invention.
Fig. 2 is electromyography signal datagram.
Fig. 3 is the flow chart of ant group algorithm of the present invention.
Fig. 4 is experimental result picture of the present invention.
Specific embodiment
Technical solution of the present invention is further illustrated with reference to the accompanying drawing.
One mechanical arm will identify n movement, and corresponding channel amount is m, at this stage, need to identify in mechanical arm dynamic
It is effective to reduce corresponding number of channels in the case where not influencing.Minor issue n=5, m=16 are set.Particular problem form is such as
Shown in following table:
Movement | Existing port number | Port number after optimization |
Movement 1 | 16 | ? |
Movement 2 | 16 | ? |
Movement 3 | 16 | ? |
Movement 4 | 16 | ? |
Movement 5 | 16 | ? |
Total movement | 16 | ? |
This problem is a kind of relatively conventional optimization problem, and optimization problem is exactly to construct a mapping function, by asking
The extreme value of the problem obtains required solution.There is now product is to be made that requirement one by one to 16 channels, and by 16
The data of the electric potential signal in channel are trained, thus achieve the purpose that classification, but channel manufacturing process is complex, and
Cost of manufacture is higher, therefore the quantity for reducing channel is imperative.The input and output of example
The data of input are as follows:
(1) potential value corresponding to each channel in movement 1.
(2) potential value corresponding to each channel in movement 2.
(3) potential value corresponding to each channel in movement 3.
(4) potential value corresponding to each channel in movement 4.
(5) potential value corresponding to each channel in movement 5.
The data of output are as follows:
The necessary number of channels left and channel in (1) five movement.
(2) precision of corresponding channel after optimizing.
Example exports result: the optimization precision of number of channels
After the input of each action input electric potential signal, obtained result is as shown in figure 4, abscissa is selected channel
Quantity, ordinate are the size of precision, and blue column diagram is the number of channels, corresponding optimal precision number, yellow histogram
It is multipair to answer mean accuracy for the number of channels.
As shown in figure 4, corresponding adaptive optimal control degree (precision) has just been higher than 90% since choosing three channels, from
From the point of view of economic situation, needed for three channels have substantially met optimization, in 5-10, corresponding precision is for it is as shown in the table port number
Reach 98%-99%, precision size is substantially uniform.And channel number be 9 when, corresponding precision highest corresponds to accuracy value and is
0.997, when channel number is 5, it is 0.9836 that corresponding precision highest, which corresponds to accuracy value, but due to channel comparison fiber crops
It is tired, and cost of manufacture is higher, therefore it is more good selection that selector channel number, which is 5,.Therefore this algorithm good can obtain
Complete Optimization Work.
Corresponding channel number | Optimal precision |
1 | 0.4772 |
2 | 0.8331 |
3 | 0.9569 |
4 | 0.9459 |
5 | 0.9736 |
6 | 0.9889 |
7 | 0.9974 |
8 | 0.9897 |
9 | 0.9997 |
10 | 0.9994 |
It repeats above-mentioned work and successively CHANNEL OPTIMIZATION is carried out to five movements, obtain following data:
Movement | Existing port number | Port number after optimization |
Movement 1 | 16 | 5 |
Movement 2 | 16 | 4 |
Movement 3 | 16 | 4 |
Movement 4 | 16 | 6 |
Movement 5 | 16 | 5 |
Total movement | 16 | 8 |
Work is optimized by the potential data acted to five, the optimal channel number always acted is 8, because
This, this example final optimization pass port number is 8.
Content described in this specification embodiment is only enumerating to the way of realization of inventive concept, protection of the invention
Range should not be construed as being limited to the specific forms stated in the embodiments, and protection scope of the present invention is also and in art technology
Personnel conceive according to the present invention it is conceivable that equivalent technologies mean.
Claims (1)
1. the muscle electrical signal paths combined optimization method based on swarm intelligence algorithm, includes the following steps:
Step 1: acquiring the data in each channel, and data are pre-processed, the pretreatment to data specifically includes:
11. pair some abnormal datas carry out smoothing denoising, abnormal data includes: more than maximum, minimum measured value data and to move
Make the data not changed after occurring;
12. a pair invalid data arranges, wherein invalid data includes: replacing data value with data mean value is 0 Value Data;
Step 2: needing to choose number of channels and feature, specifically including to reduce number of channels:
21. setting mechanical prosthetic limb and sharing N number of channel is respectively S1、S2...SNSurface myoelectric letter is carried out in the case where the time is the period of T
Number extraction;
22. extracting characterization method to the surface electromyogram signal data progress Feature Selection obtained in T time section using having;If
Extraction number of features is M, remembers that M feature is C1、C2...CM;
Step 3: establishing precision establishment method, the calculating of precision in the case of a small amount of channel and feature is established;Wherein precision establishes packet
It includes:
31. pair having N number of channel and M characteristic carrying out classification gesture information, algorithm classification is closed on using K, is specifically included:
A, the measurement using Euclidean distance to two similarity degrees;
Euclidean distance refers to that n dimension Euclidean space is the set of a coordinate points, and each of it point X can be expressed as (x1、x2…xn)
Wherein xi(i=12 ... n) is i-th of coordinate, two point A=(a that real number is known as X1、a2…an) and B=(b1、b2…bn) between
Distance d (AB) be defined as following formula
D (AB)=√ { ∑ [(ai-bi)2] (i=1,2 ... n) (1)
B, selecting K value, i.e. K most like samples, K value can use natural number in a sample space,
C, the formulation of categorised decision rule, using simple majority vote rule and apart from rule;Simple majority vote rule and away from
Include: from rule
C1. simple majority vote refers to that needing to obtain responses more than half just directly judges point classification situation;
C2. refer to that by distance, the longer weight of distance is lower as in weight input simple majority vote apart from rule;The shorter power of distance
It is again higher, then carry out judging point classification situation;
D, it establishes kd-tree and training data includes:
Input is target point X, kd-tree;Output is affiliated classification situation;
I, preparatory function A: when accessing each node, if the node is added in data simultaneously less than K for ambient data, if
Beyond k, then K data being closer are selected.
Ii, from root node, recursively downwards access kd-tree, if the coordinate of target X leading dimension be less than cut-off coordinate,
It is then moved to left child node, is otherwise moved to right child node, until node is leaf node;Execute preparatory function A;
Iii, recursive upward rollback, perform the following operation in each node: 1, executing preparatory function A, whether check child node
Capacity is less than, or the data more closer than top heap data, if so, executing i operation;
Iv, when retracting root node, training terminates;
32. pair a small amount of channel and feature establish classification information, wherein a small amount of channel and feature include: the candidate solution of problem: S2、
S3...SN-1,C2、C3...CM-1;Classification method include in step 31 K close on algorithm classification;
33. the determination of precision, specifically includes:
(1) scramble data matrix is established, matrix content includes
Equipped with W label, respectively Z1、Z2、...ZW;If training tag along sort n and testing classification label m, (m, n < W), the square
Battle array label fmnAdding 1, F is confusion matrix.All classification data are written in confusion matrix,
(2) ∑ f is calculatedijThe value of (i, j < w), as a result X, calculates ∑ fiiThe value of (i < w), accuracy value are
Step 4: carrying out optimal solution search to the obtained precision of step 3, is searched for, is specifically included using ant group algorithm:
41. initialization with Ant colony parameter, by the road for being selected as ant selection in channel, by the reciprocal as corresponding road of precision
Length;Assuming that n is road amount, m is ant number, and L is corresponding link length, then m ant is put on n road, ant k
Model choices are carried out according to the pheromones length on each road, do not stop to change pheromone concentration on road;Ant selects road
The publicity on road is
Tabu is taboo list in formula, the road that record ant has been passed by, with the target for forbidding ant to select to have been selected.τijFor
Starting point i residual risk on j road to terminal, α τijRelative importance, ηijFor heuristic greedy method, β ηijIt is relatively heavy
Degree is wanted,Indicate the desired value of ant k selection road,For the total amount of pheromones on each road.
42. the update of pheromones, m ant carries out the update of pheromones before choosing quality preferably;Assuming that in kth time iteration knot
Shu Yihou, ant is to pheromones τijIt is updated, publicity is as follows:
τij=(1-p) × τij(t)+Δτij(t)
Wherein Q is pheromones intensity;
The optimal value Y of precision is obtained by ant search algorithmMAX。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910154588.9A CN109934139B (en) | 2019-03-01 | 2019-03-01 | Muscle electric signal channel combination optimization method based on group intelligent algorithm |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910154588.9A CN109934139B (en) | 2019-03-01 | 2019-03-01 | Muscle electric signal channel combination optimization method based on group intelligent algorithm |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109934139A true CN109934139A (en) | 2019-06-25 |
CN109934139B CN109934139B (en) | 2021-05-11 |
Family
ID=66986259
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910154588.9A Active CN109934139B (en) | 2019-03-01 | 2019-03-01 | Muscle electric signal channel combination optimization method based on group intelligent algorithm |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109934139B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116308798A (en) * | 2023-03-22 | 2023-06-23 | 广东博众智能科技投资有限公司 | Stock index drawing method, device, equipment and storage medium |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
TW201827008A (en) * | 2017-01-25 | 2018-08-01 | 國立清華大學 | Portable multi-channel electromyography system |
CN108388855A (en) * | 2018-02-11 | 2018-08-10 | 宁波工程学院 | A kind of pin electrode EMG Signal Decomposition Based method |
CN108805067A (en) * | 2018-05-31 | 2018-11-13 | 陕西师范大学 | Surface electromyogram signal gesture identification method |
CN108985157A (en) * | 2018-06-07 | 2018-12-11 | 北京邮电大学 | A kind of gesture identification method and device |
CN108983973A (en) * | 2018-07-03 | 2018-12-11 | 东南大学 | A kind of humanoid dexterous myoelectric prosthetic hand control method based on gesture identification |
-
2019
- 2019-03-01 CN CN201910154588.9A patent/CN109934139B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
TW201827008A (en) * | 2017-01-25 | 2018-08-01 | 國立清華大學 | Portable multi-channel electromyography system |
CN108388855A (en) * | 2018-02-11 | 2018-08-10 | 宁波工程学院 | A kind of pin electrode EMG Signal Decomposition Based method |
CN108805067A (en) * | 2018-05-31 | 2018-11-13 | 陕西师范大学 | Surface electromyogram signal gesture identification method |
CN108985157A (en) * | 2018-06-07 | 2018-12-11 | 北京邮电大学 | A kind of gesture identification method and device |
CN108983973A (en) * | 2018-07-03 | 2018-12-11 | 东南大学 | A kind of humanoid dexterous myoelectric prosthetic hand control method based on gesture identification |
Non-Patent Citations (3)
Title |
---|
HU HUANG等: "Ant colony optimization-based feature selection method for surface electromyography signals classification", 《COMPUTERS IN BIOLOGY AND MEDICINE》 * |
王永宗: "面向情绪识别的脑电特征组合及通道优化选择研究", 《中国优秀硕士学位论文全文数据库》 * |
黄虎等: "基于蚁群算法的表面肌电信号特征选择", 《北京生物医学工程》 * |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116308798A (en) * | 2023-03-22 | 2023-06-23 | 广东博众智能科技投资有限公司 | Stock index drawing method, device, equipment and storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN109934139B (en) | 2021-05-11 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108491765B (en) | Vegetable image classification and identification method and system | |
CN110413924A (en) | A kind of Web page classification method of semi-supervised multiple view study | |
CN109086799A (en) | A kind of crop leaf disease recognition method based on improvement convolutional neural networks model AlexNet | |
CN108696331B (en) | Signal reconstruction method based on generation countermeasure network | |
CN109992670A (en) | A kind of map completion method of knowledge based map neighbour structure | |
CN111000553B (en) | Intelligent classification method for electrocardiogram data based on voting ensemble learning | |
CN109977234A (en) | A kind of knowledge mapping complementing method based on subject key words filtering | |
CN104035996B (en) | Field concept abstracting method based on Deep Learning | |
CN106095872A (en) | Answer sort method and device for Intelligent Answer System | |
CN107392919B (en) | Adaptive genetic algorithm-based gray threshold acquisition method and image segmentation method | |
CN109145965A (en) | Cell recognition method and device based on random forest disaggregated model | |
CN105760821A (en) | Classification and aggregation sparse representation face identification method based on nuclear space | |
CN111079784B (en) | Flue-cured tobacco baking stage identification method in baking process based on convolutional neural network | |
CN109993057A (en) | Method for recognizing semantics, device, equipment and computer readable storage medium | |
CN106503458A (en) | A kind of surface air temperature data quality control method | |
CN110363229A (en) | A kind of characteristics of human body's parameter selection method combined based on improvement RReliefF and mRMR | |
CN110399846A (en) | A kind of gesture identification method based on multichannel electromyography signal correlation | |
CN109325513B (en) | Image classification network training method based on massive single-class images | |
CN109800795A (en) | A kind of fruit and vegetable recognition method and system | |
CN113569742A (en) | Broadband electromagnetic interference source identification method based on convolutional neural network | |
CN111553186A (en) | Electromagnetic signal identification method based on depth long-time and short-time memory network | |
CN113887517A (en) | Crop remote sensing image semantic segmentation method based on parallel attention mechanism | |
CN116524960A (en) | Speech emotion recognition system based on mixed entropy downsampling and integrated classifier | |
CN109934139A (en) | A kind of muscle electrical signal paths combined optimization method based on swarm intelligence algorithm | |
CN109886097A (en) | A kind of method for detecting fatigue driving based on artificial fish school optimization H-ELM |
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 |