CN106901751A - A kind of recognition methods of the speed movement status based on brain hemoglobin information - Google Patents

Abstract

The invention discloses a method for identifying the state of motion speed based on brain hemoglobin information, the steps of which include: (1), the subject autonomously performs riding exercise in three different speed states of low, medium and high; (2), for the motion For the cortical hemoglobin concentration information recorded at the initial moment, the difference between oxyhemoglobin and deoxygenated hemoglobin is used as the analysis parameter, and based on the average value of the change rate of the corresponding difference, the parameter characteristics of the key channels are considered in four frequency bands; (3 ), identify three different speed states: directly apply the average value of the difference change rate of the oxygenated and deoxygenated hemoglobin in the key channels under the four frequency bands as the feature vector, and use the extreme learning machine ELM algorithm to identify the speed state level.

Description

A recognition method of motion speed state based on brain hemoglobin information

technical field

The invention belongs to the technology of intelligent walking aid and rehabilitation training, and in particular relates to a realization method for identifying the movement speed state of lower limbs based on cerebral cortex hemoglobin information.

Background technique

According to statistics from the Disabled Persons' Federation, the number of disabled people in my country is about 85 million, of which the number of physically disabled people accounts for 29.08%. Among them, there are more and more lower limb walking obstacles caused by stroke and traumatic brain injury. Among them, only stroke is newly diagnosed every year. The number of patients has reached about 2 million, and about 70% to 80% of patients cannot live independently due to disabilities. Their physical impairments have brought a great burden to the family and society. Therefore, the prognosis and rehabilitation of these patients with physical impairments is very important. Due to the late awareness of rehabilitation prognosis training in my country, and most of the non-intelligent passive training equipment currently on the market, the effect of rehabilitation training for patients is not good, and providing a rehabilitation training method with active awareness of patients will be beneficial to patients. Rehabilitation plays a very positive role in prognosis and provides a great possibility for them to live independently again and integrate into society.

In order to improve the intelligence of rehabilitation training equipment and the effect of rehabilitation training, many research institutions are committed to developing new rehabilitation training products based on brain-computer interface technology. However, the current brain-computer interface technology still has the following major problems:

1. Implantable or semi-implantable brain-computer interface technology has made breakthroughs, but micro-electrodes need to be implanted in the gray matter of the experimenter's brain or on the cerebral cortex under the dura mater, which may trigger immune reactions and Callus, but also psychological and ethical issues after implantation, is not yet suitable for widespread use.

2. Non-invasive brain information testing techniques include electroencephalography (EEG), magnetoencephalography (MEG), functional magnetic resonance imaging (fMRI), positron emission tomography (PET) and near-infrared spectral brain functional imaging (NIRS) and other technologies, among which the spatial resolution of fMRI and PET technology is high, but the time resolution is low, and the body is often confined to a static state during the test, which is very restrictive; the application of MEG requires the external magnetic field to be controlled. Fully shielded, so currently EEG and NIRS technologies are mainly used in the research and development of products to help the elderly and the disabled. However, in the study of brain-computer interface systems based on EEG signals, the commonly used methods based on visual evoked potential (VEP) and event-related potential (P300) require additional stimulation devices to provide stimulation to generate evoked potentials, and rely on human A certain sense (such as vision) of the experimenter is forced to synchronize with external stimuli. Since long-term operation is likely to cause visual fatigue or reduce the significance of P300 potential, the corresponding brain-computer interface operation time should not be too long. The spontaneous EEG depends on the user's spontaneous mental activity. Only special thinking processes can produce detectable brain activity. It requires a lot of training by the experimenter to generate a specific pattern of EEG, which is greatly affected by subjective factors. Therefore, most experiments are completed under specific conditions, requiring the experimenter to concentrate, the movements realized are simple and limited, lack of naturalness and flexibility, and are not very practical.

Relatively speaking, NIRS technology is non-invasive, has few restrictions on the test environment and subjects, supports long-term measurement in the natural context of cognitive activities, does not require extensive training, and has ideal spatial and temporal resolution functions It has great advantages in the field of brain-computer interface applications.

Contents of the invention

Purpose of the invention: To propose a method for identifying the state of movement speed based on brain hemoglobin information, and to use non-invasive NIRS technology to record the information of cerebral cortex hemoglobin in the process of human movement, so that autonomous control of movement does not require external stimulation and pre-training, in natural situations Realize the tracking and measurement of brain biological information and identify the state of motion speed in real time; and further integrate the identified motion patterns into motion control to improve the intelligence of helping the elderly and the disabled, and lay an important theory for intelligent control of walking/rehabilitation training equipment Base.

In order to achieve the above object, the present invention provides the following technical solutions:

A method for identifying a movement speed state based on brain hemoglobin information, the steps of which include:

1. Subjects autonomously perform riding exercise at three different speed states of low, medium and high;

2. For the cerebral cortex hemoglobin concentration information recorded at the beginning of the exercise, the difference between oxyhemoglobin and deoxygenated hemoglobin is used as the analysis parameter, starting from the fifth sampling point and combining with the previous fourth sampling point. Numerically calculate the rate of change of the difference for the corresponding 5 sampling periods, and consider the parameter characteristics of the key channels in four frequency bands;

The specific analysis method is as follows:

(1) Time-domain perspective, using statistical analysis methods to determine the key test channels under the three speed states;

(2) From the perspective of frequency domain, according to the power spectral density distribution of each test channel under the three speed states, focus on observing four frequency bands (the first frequency band: 0.01～0.03Hz, the second frequency band: 0.03～0.06Hz, the third frequency band Frequency band: 0.06～0.09Hz, the fourth frequency band: 0.09～0.12Hz) The average value of the change rate of the difference between oxygenated hemoglobin and deoxygenated hemoglobin in each key test channel;

3. Identify three different speed states: directly use the average value of the difference change rate of oxygenated and deoxygenated hemoglobin in the key channels under the four frequency bands as the feature vector, and use the extreme learning machine ELM algorithm to identify the speed state level.

Beneficial effect:

1. The application of non-invasive NIRS brain information acquisition technology solves the psychological and ethical problems after the invasion, and the test is carried out during the exercise, which ensures an application premise of using the motion pattern recognition results in the control of walking aid equipment; exercise autonomy Control enables acquisition of cortical biological information in the natural context of cognitive activity, increasing the practical value of motor velocity states.

2. Based on the rate of change of cerebral cortical hemoglobin concentration to identify the state of movement speed, and using the relative change (difference) between oxyhemoglobin and deoxygenated hemoglobin as the main index, it can improve the recognition rate and reduce the lag of cerebral blood oxygen information in cognition The negative impact of activities is conducive to the rapid identification of patterns, laying an important preliminary foundation for providing control information to mobility aids in a timely manner.

3. Combining time domain and frequency domain information is conducive to more comprehensive extraction of typical features and improved recognition rate.

Description of drawings

Fig. 1 is the sequence diagram of motion of the experimental process of the present invention;

Fig. 2 is a distribution diagram of cerebral cortex motion-associated regions and test channels in the present invention;

Fig. 3 is the schematic diagram of the relative changes of oxyhemoglobin and deoxyhemoglobin in each test channel before and after exercise under three different riding speed states in the present invention (T1, T2 are two time periods before exercise, T3 is the time after exercise segment, each time segment is 1.04s);

Fig. 4 is the corresponding frequency distribution figure of each test channel main power density under three kinds of different riding speed states in the present invention;

Fig. 5 is the frequency band characteristics of three riding speed states in the present invention (I, II, III, IV respectively represent four frequency bands of 0.01-0.03Hz, 0.03-0.06Hz, 0.06-0.09Hz and 0.09-0.12Hz).

detailed description

Example:

1. Experimental design: Subjects autonomously perform cycling at different speeds (such as low speed: 30rpm, medium speed: 60rpm, high speed: 90rpm); explain the whole process and precautions of the experiment to the subjects. In the natural state, cycling was completed at different speeds. During the whole experiment, the near-infrared optical brain imaging equipment FORIE-3000 was used to collect the cortical hemoglobin information of the subjects, and each sampling period was 0.13 seconds.

The specific process of the experiment: before the start of the task, the subjects kept resting for about 2 minutes, and then started the riding task, and the task segment and the rest segment were alternated; the order of riding speed was low speed, medium speed and high speed; After the mission, the riding mission at three speeds was repeated.

The start and end of the task are completely controlled by the subjects themselves, in a state of spontaneity, and the rest time is also controlled by the subjects. Before the experiment, the subjects were told to rest enough time, at least 25 seconds (but not by counting). control).

During the experiment, the experimenter marked the start and end of the subject's task with markers.

In the process of collecting cerebral hemoglobin information, the headgear with optical fiber was fixed on the top of the subject's head. During the process, the head should not shake too much. Task 1, task 2 and task 3 represent the low-speed riding section and the medium-speed riding section respectively. Travel section and high-speed riding section (as shown in Figure 1).

2. For the cerebral cortex hemoglobin concentration information recorded at the beginning of the exercise, the difference between oxyhemoglobin and deoxygenated hemoglobin is used as the analysis parameter, starting from the fifth sampling point and combining with the previous fourth sampling point. Numerically calculate the rate of change of the difference for the corresponding 5 sampling periods, and consider the parameter characteristics of the key channels in four frequency bands;

①Time domain Step1: For each test channel, calculate the difference (CZ) between oxyhemoglobin and deoxyhemoglobin based on each sampling point as the basic characterization index;

②Time domain Step2: For the difference (CZ) corresponding to each test channel, start from the 5th sampling point and combine the value of the previous 4th sampling point to calculate the change rate of the difference for the corresponding 5 sampling periods (CZ_K), that is, to smooth the data within 0.65(0.13*5) seconds;

③Time domain Step3: Take the starting point of the movement (the position of the second mark point in Figure 1) as the turning point, take two time periods T1 and T2 before the movement, and take a time period T3 after the movement, and each time period is separated by 8 samples Period, 9 sampling points (8 sampling periods total 0.13*8=1.04 seconds);

④Time domain Step4: By analyzing the statistical variance (ANOVA1), if there is no significant difference in the average value (CZ_K) of the difference change rate (CZ_K) measured in T1 and T2 of a certain test channel, and T3 and T1 and T2 respectively If there is a significant difference in the difference change rate mean value (CZ_K), then it is determined to be selected as the key test channel; the key test channels corresponding to each motion speed are shown in Table 1 and Figure 3;

⑤Frequency domain Step1: Perform power spectral density analysis on the difference (CZ) of each test channel, confirm the main power density in the frequency range ≥0.01Hz, exclude the influence of DC components, and record the corresponding frequency values of the main power density;

⑥Frequency domain Step2: According to the distribution range of frequency values corresponding to the main power density of the three kinds of motion speeds (Figure 4), intercept four frequency band information (filtering: 0.01-0.03Hz, 0.03-0.06Hz, 0.06-0.09Hz, 0.09- 0.12Hz, because the frequency values corresponding to the main power spectral density of some channels of individual subjects are in the range of 0.09-0.12Hz, so this frequency band is reserved), and for the data in the T3 time period after exercise, analyze the statistical variance of the four frequency bands ( ANOVA1), it is further concluded that the size relationship and statistical difference characteristics of the energy of each frequency band are significantly different under different speed states (as shown in Figure 5); low speed state: the average value of data in the frequency band 0.09-0.12Hz is significantly greater than the other three frequency bands , at the same time, the value in the frequency band 0.01-0.03Hz is significantly greater than the value in the frequency band 0.06-0.09Hz; the medium-speed state is also satisfied, and the average value of the data in the frequency band 0.09-0.12Hz is obviously greater than the other three frequency bands; high-speed state: frequency band 0.01-0.03 The average value of the data in Hz is significantly smaller than that of the other three frequency bands.

3. Identify three different speed states:

①Pattern recognition (training): Use the CZ_K average value of 9 key channels in four frequency bands as the feature vector (4*9=36), randomly select 7 people’s test data for a total of 42 groups (7 people*3 state/person* 2 repetitions) for training;

②Pattern recognition (discrimination): Use the ELM pattern recognition method to discriminate 12 states of the other two people (2 persons*3 states/person*2 repetitions), distinguish according to the eigenvectors in each state, and compare with the actual results Calculate the recognition rate;

③ Calculate the average recognition rate: repeat steps ①② for more than 10 times, each time randomly select 7 people for training, and the other 2 people for verification, calculate the average recognition rate based on more than 10 times of recognition results: low, medium and high speed three motion states The average recognition rates are 72.2%, 66.7%, 83.3%, and the total average recognition rate can reach 74.1%.

Table 1 Time-domain characteristics of low, middle and high riding speed states

Claims (5)

1. a kind of recognition methods of the speed movement status based on brain hemoglobin information, it is characterised in that its step includes：

(1), subject is autonomous under low middle three kinds of friction speed states high performs motion of riding；

(2), the cortex HC information recorded for motion initial time, to close oxygen hemoglobin and deoxidation blood The difference of Lactoferrin starts from each sampled point and combines above the 4th number of sampled point as analytical parameters, from the 5th sampled point Value calculates the difference rate of change in corresponding 5 sampling periods, divides four frequency ranges to consider the parameter attribute of emphasis passage respectively；

(3) three kinds of friction speed states, are recognized：Directly apply the oxygen-containing and deoxyhemoglobin of emphasis passage under four frequency ranges Difference rate of change average value as characteristic vector, using extreme learning machine ELM algorithm recognition speed state grades.

2. the recognition methods of the speed movement status based on brain hemoglobin information according to claim 1, its feature It is that in step (2), specific analytical method is as follows：

(1) time domain angle, applied statistics analysis method determines the stress test passage under three kinds of speed states；

(2) frequency domain angle, for each oxygen-containing hemoglobin of TCH test channel under three kinds of speed states and the difference of deoxyhemoglobin Rate of change carries out the main power density respective frequencies of each TCH test channel in power spectral-density analysis, four frequency ranges of selective analysis Distributional difference.

3. the recognition methods of the speed movement status based on brain hemoglobin information according to claim 1, its feature It is, in step (1)：

Subject is in low speed:30rpm, middling speed:60rpm, at a high speed:It is autonomous under 90rpm friction speed states to perform motion of riding；It is whole In individual experimentation, tested cortex hemoglobin information is gathered using near infrared light Brian Imaging equipment FORIE-3000, often One sampling period is 0.13 second；

Experiment idiographic flow：Before task starts, it is tested and keeps quiescent condition 2 minutes or so, the task of riding, task is started afterwards Section and rest section are alternately；The speed order for riding is successively respectively low speed, middling speed and high speed；After three tasks terminate, three The planting speed of the task of riding is repeated；The beginning and end of task is controlled by being tested oneself completely, in spontaneous state, And the time of having a rest is also, by tested control, the tested rest enough time to be informed before experiment, at least more than 25 seconds；It is real Test the beginning and end that operator marks tested task in experimentation with mark point.

4. the recognition methods of the speed movement status based on brain hemoglobin information according to claim 2, its feature It is that specific steps include：

1. time domain Step1：For each TCH test channel, calculated based on each sampled point and close oxygen hemoglobin and the blood red egg of deoxidation White difference, based on characteristic index；

2. time domain Step2：For the corresponding difference of each TCH test channel, each sampled point knot is started from from the 5th sampled point The difference rate of change in above numerical computations corresponding 5 sampling periods of the 4th sampled point is closed, so that data are carried out with smooth place Reason；

3. time domain Step3：To move starting point as turnover, two time periods T1 and T2 are taken before motion, a time is taken after motion Section T3, each time period is spaced 8 sampling periods, 9 sampled points；

4. time domain Step4：By analytic statistics variance (ANOVA1), if measured in the T1 and T2 of some TCH test channel Difference rate of change average value does not have significant difference, and difference rate of change average values of the T3 respectively between T1 and T2 has aobvious Write difference, it is determined that it is emphasis TCH test channel to select；

5. frequency domain Step1：Difference for each TCH test channel carries out power spectral-density analysis, in the band limits of >=0.01Hz Confirm main power density, exclude the influence of flip-flop, main power density respective frequencies value is recorded respectively；

6. frequency domain Step2：The distribution of the main power density respective frequencies value according to three kinds of motions speed, intercepts four frequency ranges Information, and for data in the T3 time periods after motion, four statistical variances of frequency range are analyzed, from which further follow that conclusion：It is not synchronized The magnitude relationship of each band energy and statistical discrepancy characteristic are significantly different under degree state；Lower-speed state：Frequency range 0.09-0.12Hz Interior statistical average is significantly greater than other three frequency ranges, meanwhile, numerical value is significantly greater than frequency range in frequency range 0.01-0.03Hz Numerical value in 0.06-0.09Hz；Middling speed state equally meets, and the statistical average in frequency range 0.09-0.12Hz is significantly greater than other Three frequency ranges；Fast state：Statistical average in frequency range 0.01-0.03Hz is significantly less than other three frequency ranges.

5. the recognition methods of the speed movement status based on brain hemoglobin information according to claim 1, its feature It is that in step (3), specific steps include：

1. train：It is random to select using 9 emphasis passages in four difference rate of change average values of frequency range as characteristic vector 7 test datas of people are trained for 42 groups totally；

2. differentiate：12 kinds of states of other two people are differentiated using ELM mode identification methods, according to the characteristic vector under every kind of state Differentiated, and gone out discrimination with actual result contrast conting；

3. average recognition rate is calculated：Repeat 1. 2. more than step 10 time, 7 personal datas to be randomly selected every time and is trained, in addition 2 people Verified, average recognition rate is calculated based on the recognition result of more than 10 times：The average knowledge of basic, normal, high fast three kinds of motions state Rate is not respectively 72.2%, 66.7%, 83.3%, and overall average discrimination is up to 74.1%.