Objective quantitative assessment method of brain cognitive load based on pulse wave morphological characteristics

Abstract

The present invention discloses an objective quantitative evaluation method of brain cognitive load based on pulse wave morphological characteristics, and the steps are as follows: collecting pulse wave signals during meditation and mental tasks; preprocessing the pulse wave signals; extracting the morphological characteristics of the preprocessed pulse wave signals cycle by cycle; extracting the time domain index of the morphological characteristics of the pulse wave signals, and performing de-differentiation processing on the time domain indexes; using a sequence backward selection algorithm to screen features, and training an SVM classifier through an optimal feature combination; constructing an SVM probability output model through a Sigmoid parameter model; and using the SVM probability output model to perform a quantitative evaluation of cognitive load on the pulse wave signals of the object to be tested. The present invention avoids the subjective limitations of traditional subjective scale detection, and objectively evaluates cognitive load; adopts a comprehensive evaluation of the optimal feature combination after de-differentiation processing to make the evaluation result more accurate; and further objectively quantifies the cognitive load level.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to an objective quantitative evaluation method for brain cognitive load based on pulse wave morphological characteristics. Background Art

[0002] Cognitive load refers to the total amount of psychological activity that a person applies to an individual's cognitive system during a specific working period. Cognitive load is determined by external factors (task requirements, environmental factors, etc.) and internal factors (operator's ability, effort, etc.). A task that may make an operator feel high-loaded may not necessarily cause the same high-load feeling to an operator with stronger ability or more experience. Therefore, it is necessary to timely and accurately determine whether the operator's cognitive load exceeds the load level they can bear. When the operator's cognition is overloaded, the task difficulty should be adjusted in a timely manner to ensure work efficiency.

[0003] At present, the methods for measuring cognitive load can be divided into three categories, namely subjective scale assessment, task performance assessment and physiological index assessment. Subjective scale assessment relies on subjective scales. It is necessary to assess cognitive load based on the operator's recollection and subjective feelings after the task is completed. It has time lag and subjective limitations. Although task performance can indirectly assess cognitive load, its assessment criteria need to be set according to specific tasks, which also has limitations. Physiological index assessment is an objective measurement method. Commonly used physiological indicators include electroencephalogram, electrocardiogram, blood pressure and skin electromyography activity, but these measurement methods have various limitations in sensitivity, reliability and availability. Studies have shown that pulse wave signals contain very rich cardiovascular information. Different load levels will cause changes in the morphological characteristics of pulse waves. Therefore, the activity of the autonomic nervous system can be indirectly analyzed by detecting the differences in the morphological characteristics of pulse wave signals. The acquisition of pulse wave signals is simple and convenient, and has high practical application value.

[0004] Most of the existing studies on the measurement of physiological indicators of cognitive load only distinguish the cognitive load levels under tasks of different difficulty levels, and do not further subdivide the cognitive load levels related to task performance, so it is impossible to accurately determine the level of cognitive overload. Summary of the invention

[0005] The purpose of the present invention is to solve the subjective limitations of existing cognitive load assessment methods and provide an objective quantitative assessment method for brain cognitive load based on the morphological characteristics of pulse wave signals. The present invention collects short-term pulse waves in the baseline state of sitting and the task state, divides the training data in combination with task performance, divides the cognitive load level during the task into cognitive load matching level and cognitive load overload level, and converts the hard discrimination of the classifier into a probability output. The output result can not only reflect the user's cognitive load level, but also reflect the user's relative load degree under this type of cognitive load level. The assessment result is objective and detailed, and the assessment method can be applied to different mental tasks, with a wide range of applications.

[0006] The purpose of the present invention can be achieved by adopting the following technical solutions:

[0007] An objective quantitative evaluation method for brain cognitive load based on pulse wave morphological characteristics, the evaluation method comprising the following steps:

[0008] S1. Collect the baseline pulse wave signal y in the sitting baseline state baseline and the task-state pulse wave signal y under the mental task state task ; Among them, 1 minute before the mental task begins is called the baseline state, and the pulse wave measured during this period is the baseline pulse wave signal y baseline ; The period from the beginning to the end of the mental task is called the mental task state, and the pulse wave signal collected is the task state pulse wave signal y task ;

[0009] S2, the collected pulse wave signal y baseline and task Perform preprocessing to obtain the preprocessed pulse wave signal y' baseline and y' task ;

[0010] Since there are various noise interferences in the collected pulse wave signal, including baseline drift, high-frequency interference, power frequency interference, etc., it is necessary to pre-process the pulse wave according to the characteristics of the pulse wave signal and common noises. The pre-processing includes removing high-frequency noise, removing baseline drift, period segmentation and screening.

[0011] S3, the pre-processed pulse wave signal y' baseline and y' taskPerform cycle-by-cycle pulse wave signal feature point identification, and on this basis extract the morphological features of each cycle of the pulse wave signal to obtain the morphological feature sequences of the baseline state and the task state; wherein the pulse wave signal feature points include the pulse starting point, the main wave peak point, and the dicrotic wave peak point of the pulse wave signal, and the morphological features include the main wave amplitude A1, the dicrotic wave amplitude A2, the descending branch time T1, the area under the waveform S, the ascending branch area S1, and the reflection index RI;

[0012] S4, calculate the time domain index for the baseline state morphological feature sequence and the task state morphological feature sequence obtained in step S3 respectively to obtain the time domain feature vector F baseline and F task , respectively called the baseline time domain feature vector and the task time domain feature vector, and the task time domain feature vector F of the task state task Subtract the baseline time domain feature vector F from the baseline state baseline , obtain the feature vector F after removing the baseline; wherein the time domain indicators include the feature sequence mean, the feature sequence standard deviation, the feature sequence first-order difference root mean square, and the feature sequence first-order difference standard deviation;

[0013] S5. Use a sequential backward selection algorithm to select the optimal feature combination for the feature vector F after removing the baseline, use the optimal feature combination of the training samples with known load levels to train the SVM classifier, and finally obtain a threshold-free output of the SVM classifier; wherein the sequential backward selection algorithm uses the classification accuracy of the SVM classifier as an evaluation criterion, and retains the first m features with the highest accuracy as the optimal feature combination;

[0014] Among them, the sequential backward selection algorithm uses the classification accuracy of the SVM classifier as the evaluation criterion and retains the top m features with the highest accuracy as the optimal feature combination;

[0015] S6, input the threshold-free output of the trained SVM classifier into the Sigmoid parameter model, and use it as the SVM probability output model to obtain the posterior probability estimate of the classification;

[0016] Using the Sigmoid parameter model, the hard discrimination of the SVM classifier is converted into a probability output, and the probability P of the cognitive load level belonging to the cognitive load matching level is obtained. low and the probability P of belonging to the cognitive load overload level high ;

[0017] S7, using the SVM probability output model to quantitatively evaluate the cognitive load level of the pulse wave signal of the subject to be tested;

[0018] The quantitative evaluation criteria are: high The value of reflects the cognitive load level of the subject to be tested. highWhen P < 0.5, the cognitive load level of the subject is at the cognitive load matching level, and the closer the value is to 0.5, the higher the relative load level; when P high When ≥0.5, the cognitive load level of the subject is at the cognitive load overload level, and the closer the value is to 0.5, the lower the relative load level is, and the closer it is to the load matching level; if the value is higher, the probability of meeting the cognitive load overload level is higher, and the task difficulty is very likely to exceed the load level that the subject can bear, and at this time the task difficulty needs to be reduced.

[0019] Furthermore, in step S1, the baseline pulse wave signal y is collected in the sitting baseline state. baseline and the task-state pulse wave signal y under the mental task state task ;

[0020] The pulse wave signal acquisition device is a reflective photoelectric sensor, which is fixed on the left index finger of the tester during acquisition. The pulse wave signal used for training the classifier is collected in a stable, quiet and comfortable indoor place. After the mental task is completed, the accuracy of the mental task is recorded and used as the task performance for the subsequent classification of cognitive load categories. A questionnaire survey is conducted to obtain the subjective cognitive load scale score, which is used as the subjective cognitive load score of the test subjects for the comparison and analysis of the subsequent quantitative output results.

[0021] Furthermore, the process of step S2 is as follows:

[0022] S201, collect the pulse wave signal y baseline and task Use FIR low-pass filter to filter and get the pulse wave signal y after removing high-frequency noise base_low and task_low ; This step is used to filter out high-frequency noise interference such as power frequency noise interference and myoelectric noise interference in the pulse wave signal;

[0023] S202, extract y base_low and task_low The pulse starting point;

[0024] S203, use cubic spline interpolation method to calculate y base_low and task_low The pulse starting point position is fitted to obtain the baseline drift curve y base_noise and task_noise , the pulse wave signal y after removing high-frequency noise base_low ,y task_low Subtract the baseline drift curve y base_noise ,y task_noise, the denoised pulse wave signal can be obtained; this step takes into account that the pulse wave signal acquisition process may be disturbed by factors such as human breathing, resulting in baseline drift, so the cubic spline interpolation method is used to fit the baseline drift curve to filter out the baseline drift noise interference in the pulse wave signal;

[0025] S204, performing period segmentation on the denoised pulse wave signal according to the pulse starting point to obtain Z pulse wave period signals Y;

[0026] S205, pulse wave period signal Y of adjacent periods i and Y i+1 Perform interpolation resampling to obtain two pulse wave cycles Y' with the same number of sampling points i , Y' i+1 , calculate Y' i With Y' i+1 Pearson correlation coefficient, set the threshold and filter out the continuous and stable pulse wave period signal y' according to certain screening rules baseline and y' task ; This step takes into account that the pulse wave signal acquisition process may be disturbed by motion artifacts, resulting in abnormal cycles in the pulse wave signal. Therefore, by calculating the Pearson correlation coefficient of adjacent cycles of the pulse wave signal, the abnormal cycles disturbed by motion artifacts are eliminated, and finally a continuous and stable pulse wave cycle signal is obtained;

[0027] Since the pulse wave signal cycle length is not fixed, the length of two adjacent pulse wave cycles Y i With Y i+1 The lengths of may not be consistent, so Y cannot be calculated directly i and Y i+1 Pearson coefficient, so we need to i and Y i+1 Perform interpolation resampling to obtain two pulse wave cycles Y' with the same number of sampling points m i , Y' i+1 , and then calculate Y' i and Y' i+1 The Pearson correlation coefficient P i ;

[0028] Among them, the Pearson correlation coefficient P i The calculation formula is:

[0029]

[0030] In the formula, Y' i With Y' i+1 are two sets of pulse wave data with length m, μ i , μ i+1 Y' i, Y' i+1 The mean of .

[0031] Furthermore, the process of step S3 is as follows:

[0032] S301, the processed pulse wave signal y' baseline and y' task Extract the characteristic points of the pulse wave signal cycle by cycle, the characteristic points of the pulse wave signal include the pulse starting point, the main wave peak point, and the dicrotic wave peak point; this step extracts the characteristic points of the pulse wave signal according to the waveform characteristics of the pulse wave signal, which are used for the subsequent extraction of the morphological characteristics of the pulse wave signal;

[0033] S302: According to the pulse wave signal feature points extracted in step S301, the pulse wave signal y' baseline and y' task Each morphological feature is extracted cycle by cycle, and the morphological features include the main wave amplitude A1, the dicrotic wave amplitude A2, the descending branch time T1, the area under the waveform S, the ascending branch area S1, and the reflection index RI; wherein the calculation formula of the reflection index RI is:

[0034]

[0035] In the formula, A2 is the amplitude of the dicrotic wave, and A1 is the amplitude of the main wave;

[0036] For the pulse wave signal y' baseline and y' task Extract the above morphological features period by period, denoted by y' baseline The period number is h baseline , y' task The period number is h task , and the length is h baseline The six baseline state morphological feature sequences g A1_base , g A2_base , g T1_base , g S_base , g S1_base , g RI_base , and length is h task The 6 task state morphological feature sequences g A1_task , g A2_task , g T1_task , g S_task , g S1_task , g RI_task ; This step extracts multiple morphological features of the pulse wave signal cycle by cycle based on the characteristic points of the pulse wave signal, so as to obtain multiple morphological feature sequences of the pulse wave signal.

[0037] Furthermore, the process of step S4 is as follows:

[0038] S401, extracting the corresponding four time domain indicators from the six morphological feature sequences of the baseline state and the task state obtained in step S3, and obtaining the time domain indicator vector AVG of the baseline state baseline 、STD baseline 、dRMS baseline ,dSTD baseline , the time domain indicator vector AVG of the task status task 、STD task 、dRMS task ,dSTD task , according to the time domain index vector, the baseline time domain feature vector F is obtained baseline And the task time domain feature vector F task ; Among them, the time domain indicator vector in the baseline time domain feature vector is:

[0039] AVG baseline =[avg(g A1_base ),avg(g A2_base ),avg(g T1_base ),avg(g S_base ),avg(g S1_base ),avg(g RI_base )]

[0040] STD baseline =[std(g A1_base ),std(g A2_base ),std(g T1_base ),std(g S_base ),std(g S1_base ),std(g RI_base )]

[0041] dRMS baseline =[drms(g A1_base ),drms(g A2_base ),drms(g T1_base ),drms(g S_base ),drms(g S1_base ),drms(g RI_base )]

[0042] dSTD baseline =[dstd(g A1_base ),dstd(g A2_base ),dstd(g T1_base ),drms(g S_base ),dstd(g S1_base ),dstd(g RI_base )]

[0043] The time domain indicator vector in the task time domain feature vector is:

[0044] AVG task =[avg(g A1_task ),avg(g A2_task ),avg(g T1_task ),avg(g S_task ),avg(g S1_task ),avg(g RI_task )]

[0045] STD task =[std(g A1_task ),std(g A2_task ),std(g T1_task ),std(g S_task ),std(g S_task ),std(g RI_task )]

[0046] dRMS task =[drms(g A1_task ),drms(g A2_task ),drms(g T1_task ),drms(g S_task ),drms(g S1_task ),drms(g RI_task )]

[0047] dSTD task =[dstd(g A1_task ),dstd(g A2_task ),dstd(g T1_task ),dstd(g S_task ),dstd(g S1_task ),dstd(g RI_task )]

[0048] The above avg means to calculate the average value; std means to calculate the standard deviation; drms means to calculate the first-order difference root mean square; dstd means to calculate the first-order difference standard deviation;

[0049] Among them, the baseline time domain feature vector F baseline And the task time domain feature vector F task for:

[0050] F baseline =[AVG baseline ,STD baseline ,dRMS baseline ,dSTD baseline ],

[0051] F task =[AVG task,STD task ,dRMS task ,dSTD task ]

[0052] In this step, four time-domain indicators commonly used in cognitive load assessment based on physiological signals are extracted according to the morphological feature sequence of the pulse wave signal, and the time-domain feature vector of the pulse wave signal is obtained;

[0053] S402, perform baseline removal on the time domain feature vector of the task state, and convert the task time domain feature vector F task Subtract the baseline time domain feature vector F from the baseline state baseline , and obtain the feature vector F after removing the baseline; this step takes into account the individual differences in the extracted time domain feature vectors. If the individual differences are not removed, the classification accuracy of the classifier will decrease when the time domain feature vectors with individual differences are used to train the classifier; considering that the subjects remain relaxed during the sitting state, the cognitive load is the lowest at this time, so the sitting state is regarded as the baseline state, and the feature sequence of the task state is subjected to baseline removal to reduce the impact of individual differences and improve the classification accuracy.

[0054] Furthermore, the process of step S5 is as follows:

[0055] S501, use the sequence backward selection algorithm to screen the features in the feature vector F, use the classification accuracy of the SVM classifier as the evaluation standard, retain the first m features with the highest accuracy in F, and obtain the feature vector with the optimal feature combination for subsequent classifier training; since the feature vector F obtained in step S4 is calculated by extracting the corresponding 4 time domain indicators from the 6 morphological feature sequences, there are a total of 24 features in the feature vector F. This step takes into account that there must be information redundancy in these 24 features. If all features are used for classifier training, the complexity of classification will increase and the generalization ability of the classifier will decrease. Therefore, the sequence backward selection algorithm is used to screen the features, and the classification accuracy of the SVM classifier is used as the evaluation standard to screen out the optimal feature combination, so as to ensure that the SVM classifier obtained by subsequent training has high generalization performance;

[0056] S502, extracting the feature vectors of all training samples to form a feature vector set x, and performing standardization on x to obtain a standardized feature vector set x'; the standardization formula is:

[0057]

[0058] Where n∈[1,N], N is the total number of training samples, x' n is the feature vector of the nth sample after standardization, x nis the feature vector of the nth sample, μ is the feature mean vector of all samples, and σ is the feature standard deviation vector of all samples; this step takes into account that the extreme maximum and minimum values ​​in the feature vector set will prolong the training time of the classifier and reduce the performance of the classifier, so the feature vector set is standardized to reduce the impact of the extreme maximum and minimum values ​​on the classifier;

[0059] S503. The task accuracy rate recorded after the subject's mental task is completed is used as task performance to divide the cognitive load level: cognitive load overload level and cognitive load matching level; samples with lower task performance are divided into cognitive load overload level, and samples with higher task performance are divided into cognitive load matching level, and the cognitive load level of the sample and the standardized feature vector set x' are input into the SVM classifier to train the SVM classifier; this step divides the cognitive load level of the training samples according to the task performance, and establishes a mapping relationship between the feature vector and the cognitive load level through the training of the SVM classifier, and obtains the threshold-free output of the SVM classifier;

[0060] Furthermore, in step S6, the threshold-free output of the trained SVM classifier is input into the Sigmoid parameter model, and this is used as the SVM probability output model to obtain the posterior probability estimate of the classification; wherein the decision function f(x' n )for:

[0061] f(x' n )=sign(h(x' n ))=sign(ω T ·x' n +b) Formula (4)

[0062] In the formula, h(x' n ) is the threshold-free output of the SVM classifier, ω T , b is a parameter related to the decision boundary, x' n is the feature vector of the nth sample, sign(j) is a sign function that returns 1 when j>0 and returns -1 when j<0;

[0063] The unthresholded output h(x' n ) is input into the Sigmoid parameter model to obtain the posterior probability estimate; the posterior probability P(y=1|h(x' n )) is calculated as:

[0064]

[0065] In the formula, α and β are the parameters to be fitted, and the maximum likelihood estimation is used to fit the two parameter values;

[0066] By transforming formula (5), h(x' n ) is mapped to a probability value, thereby constructing an SVM probability output model; this step takes into account that the SVM threshold-free output can only obtain the category to which the sample's cognitive load level belongs, which is a hard discrimination output, but the cognitive load itself should be a continuous quantity. If the cognitive load level is clearly attributed to a certain category, it may cause sample discrimination errors near the classification boundary, and it is impossible to compare the relative cognitive load levels of samples with the same cognitive load level, resulting in the loss of effective information. Therefore, through the Sigmoid parameter model, the threshold-free output of the SVM classifier is converted into the probability output of the sample belonging to each category through the Sigmoid parameter model, thereby achieving the purpose of converting the hard discrimination output of the classifier into a probability output; through this conversion, the probability P of the sample's cognitive load level belonging to the cognitive load matching level can be obtained. low and the probability P of belonging to the cognitive load overload level high .

[0067] Furthermore, in step S7, the SVM probability output model is used to quantitatively evaluate the cognitive load level of the pulse wave signal of the subject to be measured;

[0068] The quantitative evaluation criteria are: high The value of reflects the cognitive load level of the subject to be tested. high When P < 0.5, the cognitive load level of the subject is at the cognitive load matching level, and the closer the value is to 0.5, the higher the relative load level; when P high When ≥0.5, the cognitive load level of the subject is at the cognitive load overload level, and the closer the value is to 0.5, the lower the relative load level is, and the closer it is to the load matching level; if the value is higher, the probability of meeting the cognitive load overload level is higher, and the task difficulty is very likely to exceed the load level that the subject can bear, and at this time it is necessary to reduce the task difficulty; this step quantitatively evaluates the cognitive load level according to the SVM probability output model, and realizes a detailed description of the cognitive load level.

[0069] Compared with the prior art, the present invention has the following advantages and effects:

[0070] 1) The present invention objectively evaluates the cognitive load level based on the morphological characteristics of the pulse wave signal. It is an objective and non-invasive cognitive load assessment method. The pulse wave signal is easy to collect and can realize daily monitoring of cognitive load. It avoids the subjective limitations of traditional cognitive load scale assessment methods and improves the signal acquisition difficulties of other commonly used cognitive load physiological signal assessment methods such as electrocardiogram signals and electroencephalogram signals.

[0071] 2) The present invention extracts multiple morphological features of the pulse wave signal and performs individual difference processing on the morphological features, thereby reducing the influence of individual differences in the features on the classification accuracy of the SVM classifier and improving the accuracy of the assessment of the cognitive load level. In addition, before training the SVM classifier, the present invention uses a sequential backward selection algorithm to screen out the optimal feature combination, and trains the SVM classifier through the optimal feature combination, thereby reducing the influence of redundant features and ensuring the generalization performance of the SVM classifier.

[0072] 3) The present invention uses the SVM probability output model to objectively and quantitatively evaluate the cognitive load level, thereby achieving a further detailed description of the cognitive load level. The quantitative evaluation results can not only reflect the category to which the user's cognitive load level belongs, but also reflect the relative load degree of the user under this type of cognitive load level. The evaluation results are objective and detailed, and the evaluation method can be applied to different mental tasks and is universal. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0074] Figure 1 is a flow chart of a method for objectively quantifying brain cognitive load evaluation based on pulse wave morphological characteristics disclosed in Example 1 of the present invention;

[0075] Figure 2 is a flow chart of pulse wave signal preprocessing in Example 1 of the present invention;

[0076] Figure 3 is a flow chart of pulse wave signal cycle screening in Example 1 of the present invention;

[0077] Figure 4 is a schematic diagram of extracting pulse wave morphological features in Example 1 of the present invention;

[0078] Figure 5 is a flow chart of training an SVM classifier based on pulse wave morphology features in Example 1 of the present invention;

[0079] Figure 6 Schematic diagram of extracting characteristic points of the pulse wave signal of the object to be measured in Example 2 of the present invention. DETAILED DESCRIPTION

[0080] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0081] Example 1

[0082] Figure 1 This is a flow chart of a method for objectively quantifying the brain cognitive load assessment based on pulse wave morphology characteristics provided by an embodiment of the present invention. This embodiment can be applied to the assessment of cognitive load levels during any mental task.

[0083] S1. Collect the baseline pulse wave signal y of the sitting baseline state baseline and the task-state pulse wave signal y of the mental task state task .

[0084] The mental task refers to the classic 2-back task used in psychology to stimulate cognitive load and psychological stress. Each experiment has a 10-minute preparation process, allowing the subjects to familiarize themselves with the environment, sit quietly and relax. The 1 minute before the mental task begins is called the baseline state. The pulse wave measured during this period is the baseline pulse wave signal. baseline The period from the start of a mental task to the end is called the mental task state, and the pulse wave signal collected is the task state pulse wave signal y task .

[0085] The pulse wave signal acquisition device is a reflective photoelectric sensor, which is fixed on the left index finger of the tester during acquisition. The pulse wave signal used for training the classifier is collected in a stable, quiet and comfortable indoor place. After the mental task is completed, the task accuracy is recorded as the task performance for the subsequent classification of cognitive load categories, and a questionnaire survey is conducted to obtain the subjective cognitive load scale score, which is used as the subjective cognitive load score of the test subjects for the comparison and analysis of the subsequent quantitative output results.

[0086] S2, the collected pulse wave signal y baseline and task Perform preprocessing to obtain the preprocessed pulse wave signal y' baseline and y' task .

[0087] The pulse wave signal collected in this example has various noise interferences, including baseline drift noise interference, myoelectric noise interference, power frequency noise interference, etc. The pulse wave signal preprocessing flow chart is as follows: Figure 2As shown in the figure, the signal preprocessing in this example includes removing high-frequency noise, extracting the pulse starting point, removing baseline drift, cycle segmentation and cycle screening. The process is as follows:

[0088] S201, collect the pulse wave signal y baseline and task A high-order FIR low-pass filter with a passband cutoff frequency of 10 Hz, a stopband cutoff frequency of 15 Hz, and a Hanning window as the window function is used to filter out the high-frequency noise above 10 Hz in the original pulse wave signal, and the pulse wave signal y after high-frequency noise removal is obtained. base_low and task_low .

[0089] S202, extract y base_low and task_low The pulse starting point.

[0090] S203, use cubic spline interpolation method to calculate y base_low and task_low The pulse starting point position is fitted to obtain the baseline drift curve y base_noise and task_noise , the pulse wave signal y after removing high-frequency noise base_low ,y task_low Subtract the baseline drift curve y base_noise ,y task_noise , the denoised pulse wave signal can be obtained.

[0091] Since the test part may shake during the signal acquisition process, motion artifacts may exist in the collected pulse wave signal, causing abnormal cycles in the signal and interfering with subsequent feature extraction. Since motion artifacts cannot be removed by filtering, it is necessary to perform period screening on the denoised pulse wave signal to eliminate the periods with motion artifacts. The pulse wave signal period screening flow chart is shown below: Figure 3 The specific screening process is as follows:

[0092] S204 , periodically dividing the denoised pulse wave signal according to the pulse starting point to obtain Z pulse wave periodic signals Y.

[0093] S205, pulse wave period signal Y of adjacent periods i and Y i+1 Perform interpolation resampling to obtain two pulse wave cycles Y' with the same number of sampling points i , Y' i+1 , calculate Y' i With Y' i+1 Pearson correlation coefficient, set the threshold and filter out the continuous and stable pulse wave period signal y' according to certain screening rules baseline and y' task .

[0094] Since the pulse wave signal cycle length is not fixed, the length of two adjacent pulse wave cycles Y i With Y i+1 The lengths of may not be consistent, so Y cannot be calculated directly i and Y i+1 Pearson coefficient, so we need to i and Y i+1 Perform interpolation resampling to obtain two pulse wave cycles Y' with the same number of sampling points m i , Y' i+1 , and then calculate Y' according to formula (1) i and Y' i+1 The Pearson correlation coefficient P i .

[0095] Set the threshold σ = 0.95 and the number of continuous bands K = 6 to filter out the continuous and stable pulse wave period signals that meet the threshold. The period screening rule is: starting from i = 1, P i and P i+1 The pulse wave signal Y is greater than σ i+1 is regarded as a qualified cycle, and the number of consecutive qualified cycles s is recorded. When s = K, Y i+1 and Y i+1 The pulse wave signal of the previous s-1 segment is retained and continues to be counted. i and P i+1 When all meet the threshold, retain Y i+1 When s>K but the correlation coefficient P i or P i+1 When it does not meet the threshold, s is cleared until a period that meets the threshold appears, and then the counting starts again. Repeat the above steps to filter out the continuous and stable pulse wave period signal y' that meets the threshold baseline and y' task .

[0096] In this example, by performing period segmentation and screening on the pulse wave signal, continuous periods whose correlation coefficients meet the threshold requirements are selected, thereby eliminating abnormal periods caused by motion artifacts in the original pulse wave signal, and facilitating accurate extraction of feature sequences in subsequent processing.

[0097] S3, the pre-processed pulse wave signal y' baseline and y' task The pulse wave signal feature points are identified cycle by cycle, and on this basis, the morphological features of each cycle of the pulse wave signal are extracted to obtain the morphological feature sequences of the baseline state and the task state.

[0098] This example extracts the morphological feature sequence of the pulse wave signal based on the morphological feature of the pulse wave signal, and extracts the time domain index of the morphological feature by referring to the commonly used time domain index of the feature sequence variability. Figure 4 As shown in Figure 2, the time domain feature extraction process is as follows:

[0099] S301, the processed pulse wave signal y' baseline and y' task The pulse wave signal feature points are extracted cycle by cycle, and the pulse wave signal feature points include the pulse starting point, the main wave peak point, and the dicrotic wave peak point.

[0100] S302: According to the pulse wave signal feature points extracted in step S301, the pulse wave signal y' baseline and y' task Each morphological feature is extracted cycle by cycle, and the morphological features include the main wave amplitude A1, the dicrotic wave amplitude A2, the descending branch time T1, the area under the waveform S, the ascending branch area S1, and the reflection index RI.

[0101] The reflectance index RI is calculated according to formula (2).

[0102] For the pulse wave signal y' baseline and y' task Extract the above morphological features period by period, denoted by y' baseline The period number is h baseline , y' task The period number is h task , and the length is h baseline The six baseline state morphological feature sequences g A1_base , g A2_base , g T1_base , g S_base , g S1_base , g RI_base , and length is h task The 6 task state morphological feature sequences g A1_task , g A2_task , g T1_task , g S_task , g S1_task , g RI_task .

[0103] S4, calculate the time domain index for the baseline state morphological feature sequence and the task state morphological feature sequence obtained in step S3 respectively to obtain the time domain feature vector F baseline and F task , respectively called the baseline time domain feature vector and the task time domain feature vector, and the task time domain feature vector F of the task state task Subtract the baseline time domain feature vector F from the baseline state baseline, and obtain the feature vector F after removing the baseline.

[0104] The specific process is as follows:

[0105] S401, extracting the corresponding four time domain indicators from the six morphological feature sequences of the baseline state and the task state obtained in step S3, and obtaining the time domain indicator vector AVG of the baseline state baseline 、STD baseline 、dRMS baseline ,dSTD baseline , the time domain indicator vector AVG of the task status task 、STD task 、dRMS task ,dSTD task , according to the time domain index vector, the baseline time domain feature vector F is obtained baseline And the task time domain feature vector F task ; Among them, the time domain indicator vector in the baseline time domain feature vector is:

[0106] AVG baseline =[avg(g A1_base ),avg(g A2_base ),avg(g T1_base ),avg(g S_base ),avg(g S1_base ),avg(g RI_base )]

[0107] STD baseline =[std(g A1_base ),std(g A2_base ),std(g T1_base ),std(g S_base ),std(g S1_base ),std(g RI_base )]

[0108] dRMS baseline =[drms(g A1_base ),drms(g A2_base ),drms(g T1_base ),drms(g S_base ),drms(g S1_base ),drms(g RI_base )]

[0109] dSTD baseline =[dstd(g A1_base ),dstd(g A2_base ),dstd(g T1_base ),drms(g S_base ),dstd(gS1_base ),dstd(g RI_base )]

[0110] The time domain indicator vector in the task time domain feature vector is:

[0111] AVG task =[avg(g A1_task ),avg(g A2_task ),avg(g T1_task ),avg(g S_task ),avg(g S1_task ),avg(g RI_task )]

[0112] STD task =[std(g A1_task ),std(g A2_task ),std(g T1_task ),std(g S_task ),std(g S_task ),std(g RI_task )]

[0113] dRMS task =[drms(g A1_task ),drms(g A2_task ),drms(g T1_task ),drms(g S_task ),drms(g S1_task ),drms(g RI_task )]

[0114] dSTD task =[dstd(g A1_task ),dstd(g A2_task ),dstd(g T1_task ),dstd(g S_task ),dstd(g S1_task ),dstd(g RI_task )]

[0115] The above avg means the average value calculation; std means the standard deviation calculation; drms means the first-order difference root mean square calculation; dstd means the first-order difference standard deviation calculation.

[0116] Among them, the baseline time domain feature vector F baseline And the task time domain feature vector F task for:

[0117] F baseline =[AVG baseline ,STD baseline ,dRMSbaseline ,dSTD baseline ],

[0118] F task =[AVG task ,STD task ,dRMS task ,dSTD task ]

[0119] S402, perform baseline removal on the time domain feature vector of the task state, and convert the task time domain feature vector F task Subtract the baseline time domain feature vector F from the baseline state baseline , and get the feature vector F after removing the baseline.

[0120] There are individual differences in the time domain feature vectors extracted in this example. If the individual differences are not removed and the classifier is trained using the feature vectors with individual differences, the classification accuracy of the classifier will decrease. Considering that the subjects are relaxed in the sitting state, the cognitive load is the lowest at this time. Therefore, the sitting state is regarded as the baseline state, and the time domain feature vectors of the task state are de-baselined to reduce the impact of individual differences and improve the classification accuracy.

[0121] S5. The optimal feature combination is selected by using the sequential backward selection algorithm for the feature vector F after removing the baseline, and the SVM classifier is trained by using the optimal feature combination of the training samples with known load levels, and finally the threshold-free output of the SVM classifier is obtained.

[0122] The flowchart of training SVM classifier based on pulse wave morphology features is as follows Figure 5 shown.

[0123] S501. Use the sequential backward selection algorithm to screen the features in the feature vector F, use the classification accuracy of the SVM classifier as the evaluation criterion, retain the first m features with the highest accuracy in F, and obtain the feature vector with the optimal feature combination for subsequent classifier training.

[0124] Since the feature vector F obtained in step S4 is calculated by extracting the corresponding 4 time domain indicators from the 6 morphological feature sequences, there are a total of 24 features in the feature vector F, but these features cannot all be used for feature classification because there is information redundancy in the 24 features, and too many data dimensions will increase the complexity of classifier training and reduce the generalization ability of the classifier. Therefore, the sequence backward selection algorithm is used to screen the features, and the classification accuracy of the SVM classifier is used as the evaluation standard. The top m features with the highest accuracy in F are retained to form the optimal feature combination. In this example, m=10, that is, the top 10 features with the highest accuracy in F are retained.

[0125] S502: extract the feature vectors of all training samples to form a feature vector set x, and perform standardization on x according to formula (3) to obtain a standardized feature vector set x'.

[0126] S503. The task accuracy recorded after the subject completes the mental task is used as task performance to divide the cognitive load level into: cognitive load overload level and cognitive load matching level; samples with lower task performance are classified as cognitive load overload level, and samples with higher task performance are classified as cognitive load matching level. The cognitive load level of the sample and the standardized feature vector set x' are input into the SVM classifier to train the SVM classifier.

[0127] S6. Input the threshold-free output of the trained SVM classifier into the Sigmoid parameter model and use it as the SVM probability output model to obtain the posterior probability estimate of the classification.

[0128] According to formula (4), the threshold-free output h(x' n ), through the transformation of formula (5), h(x' n ) is mapped to probability values, thereby constructing the SVM probability output model.

[0129] In this example, the threshold-free output of the trained SVM classifier is mapped to a probability value through the Sigmoid function to construct an SVM probability output model. The probability output of the model is the probability of the two categories to which the subject belongs. The probability of belonging to the cognitive load matching level is P low , the probability of belonging to the cognitive load overload level is P high . Use P high As the quantitative score of the cognitive load of the subject, the obtained quantitative score of cognitive load can not only reflect the load category to which the subject belongs, but also reflect the relative load level of the subject in the category. high >0.5, it belongs to the cognitive load overload level, and P high The closer it is to 0.5, the higher the tendency of the subject's characteristics to deviate from the cognitive load overload level is, and the lower the relative level of cognitive load in the cognitive load overload category is.

[0130] In order to verify the reliability of the probability output model, this example performs a Pearson correlation analysis on the quantitative scores of the trained probability output model and the subjective scale scores of the subjects used for classifier training. The analysis results show that the quantitative scores of the probability output model and the subjective scale scores have a significant and highly positive correlation. The quantitative scores are reliable and can be used as a quantitative reference for cognitive load levels.

[0131] S7. Use the SVM probability output model to quantitatively evaluate the cognitive load level of the pulse wave signal of the subject.

[0132] The quantitative evaluation criteria are: high The value of reflects the cognitive load level of the subject to be tested. high When P < 0.5, the cognitive load level of the subject is at the cognitive load matching level, and the closer the value is to 0.5, the higher the relative load level. high When ≥0.5, the cognitive load level of the subject is at the cognitive load overload level, and the closer the value is to 0.5, the lower the relative load level is, and the closer it is to the load matching level; if the value is higher, the probability of meeting the cognitive load overload level is higher, and the task difficulty is very likely to exceed the load level that the subject can bear, and at this time the task difficulty needs to be reduced.

[0133] Example 2

[0134] This embodiment further discloses a method for objectively quantifying brain cognitive load assessment based on pulse wave morphological characteristics, comprising the following steps:

[0135] S1. Referring to the corresponding steps in Example 1, a baseline pulse wave signal y of a subject to be tested is collected. baseline and task state pulse wave signal y task .

[0136] S2. Refer to the corresponding steps in Example 1 to obtain the pre-processed pulse wave signal y' baseline and y' task .

[0137] S3, referring to the corresponding steps in Example 1, the pre-processed pulse wave signal y' of the object to be measured baseline and y' task The pulse wave signal feature point recognition is performed cycle by cycle. The pulse wave signal feature point extraction diagram of the object to be measured is as follows: Figure 6 As shown, on this basis, the morphological features of each cycle of the pulse wave signal of the object to be measured are extracted to obtain the morphological feature sequences of the baseline state and the task state of the object to be measured.

[0138] S4. Refer to the corresponding steps in Example 1 and will not be repeated here.

[0139] S5. Based on Example 1, the optimal feature combination of the subject to be tested is input into the trained SVM probability output model to obtain a quantitative evaluation result of the cognitive load of the subject to be tested.

[0140] The optimal features and the corresponding feature values ​​of the objects to be tested are shown in Table 1 below:

[0141] Table 1 The optimal features and corresponding feature values ​​of the objects to be tested in Example 2

[0142] Optimal Features <![CDATA[T1_STD]]> <![CDATA[T1_dRMS]]> <![CDATA[T1_dSTD]]> <![CDATA[S1_STD]]> <![CDATA[A1_STD]]> Eigenvalue -4.368 -0.9868 -0.989 -5.564 -0.469 Optimal Features <![CDATA[A1_STD]]> <![CDATA[A2_dSTD]]> S_dSTD RI_dRMS RI_dSTD Eigenvalue -0.108 -0.047 -0.0064 -4.649 -4.652

[0143] Among them, P_STD represents the standard deviation of sequence P, P_dRMS represents the first-order difference root mean square of sequence P, and P_dSTM represents the first-order difference standard deviation of sequence P.

[0144] The optimal features of the test object in the above table are input into the SVM probability output model, and the cognitive load quantitative score of the test object is obtained as 0.79.

[0145] Since 0.79>0.5, it is judged that the subject is at the cognitive overload level, and since 0.79 is closer to 1, it is judged that the probability that the subject meets the cognitive overload level is high. The difficulty of the mental task is very likely to exceed the load level that the subject can bear, and the task difficulty needs to be reduced.

[0146] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the protection scope of the present invention.

Claims

1. An objective quantitative evaluation method for brain cognitive load based on pulse wave morphology characteristics, characterized in that: The evaluation method comprises the following steps: S1. Collect the baseline pulse wave signal in the sitting baseline state and task-state pulse wave signals under mental task conditions ; Among them, 1 minute before the start of the mental task is called the baseline state, and the pulse wave measured during this period is the baseline pulse wave signal The period from the beginning to the end of the mental task is called the mental task state, and the pulse wave signal collected is the task state pulse wave signal. ; S2. Collected pulse wave signal and Perform preprocessing to obtain the preprocessed pulse wave signal and ; S3. Preprocessed pulse wave signal and Perform cycle-by-cycle pulse wave signal feature point identification, and on this basis extract the morphological features of each cycle of the pulse wave signal to obtain the morphological feature sequences of the baseline state and the task state; wherein the pulse wave signal feature points include the pulse starting point, the main wave peak point, and the dicrotic wave peak point of the pulse wave signal, and the morphological features include the main wave amplitude , Dicrotic wave amplitude , descending branch time , Area under the waveform , ascending branch area , reflectance index RI; S4, calculate the time domain index for the baseline state morphological feature sequence and the task state morphological feature sequence obtained in step S3 respectively to obtain the time domain feature vector and , respectively called the baseline time domain feature vector and the task time domain feature vector, and the task time domain feature vector of the task state Subtract the baseline time domain feature vector of the baseline state , get the feature vector after removing the baseline ; Wherein, the time domain indicators include the feature sequence mean, the feature sequence standard deviation, the feature sequence first-order difference root mean square, and the feature sequence first-order difference standard deviation; S5. Feature vector after removing the baseline The optimal feature combination is screened out using a sequential backward selection algorithm, and the optimal feature combination of training samples with known load levels is used to train an SVM classifier, and finally a threshold-free output of the SVM classifier is obtained; wherein the sequential backward selection algorithm uses the classification accuracy of the SVM classifier as an evaluation criterion, and retains the top m features with the highest accuracy as the optimal feature combination; S6, input the threshold-free output of the trained SVM classifier into the Sigmoid parameter model, and use it as the SVM probability output model to obtain the posterior probability estimate of the classification; S7. Use the SVM probability output model to quantitatively evaluate the cognitive load level of the pulse wave signal of the subject.

2. The objective quantitative evaluation method for brain cognitive load based on pulse wave morphological characteristics according to claim 1 is characterized in that: The process of step S2 is as follows: S201, collect the pulse wave signal and Use FIR low-pass filter to filter and get the pulse wave signal after removing high-frequency noise and ; S202, Extraction and The pulse starting point; S203, using cubic spline interpolation method and The pulse starting point position is fitted to obtain the baseline drift curve and , the pulse wave signal after removing high-frequency noise , Subtract the baseline drift curve , , the denoised pulse wave signal can be obtained; S204, periodically segment the denoised pulse wave signal according to the pulse starting point to obtain Pulse wave cycle signal ; S205: pulse wave periodic signals of adjacent periods and Perform interpolation resampling to obtain two pulse wave cycles with the same number of sampling points , ,calculate and Pearson correlation coefficient, set the threshold and filter out continuous and stable pulse wave cycle signals according to certain screening rules and .

3. The objective quantitative evaluation method for brain cognitive load based on pulse wave morphological characteristics according to claim 1, characterized in that: The process of step S3 is as follows: S301: Processing the pulse wave signal and Extract characteristic points of pulse wave signal cycle by cycle; S302: according to the pulse wave signal feature points extracted in step S301, and Extract each morphological feature period by period and record The number of cycles is , The number of cycles is , and the length is Six baseline state morphological feature sequences , , , , , , and the length is The 6 task state morphological feature sequences , , , , , .

4. The objective quantitative evaluation method for brain cognitive load based on pulse wave morphological characteristics according to claim 3 is characterized in that: The process of step S4 is as follows: S401, extracting the corresponding four time domain indicators from the six morphological feature sequences of the baseline state and the task state obtained in step S3, and obtaining the time domain indicator vector of the baseline state , , , , the time domain indicator vector of the task status , , , , the baseline time domain feature vector is obtained according to the time domain indicator vector And the task time domain feature vector ; Among them, the time domain indicator vector in the baseline time domain feature vector is: The time domain indicator vector in the task time domain feature vector is: The above avg means to calculate the average value; std means to calculate the standard deviation; drms means to calculate the first-order difference root mean square; dstd means to calculate the first-order difference standard deviation; Among them, the baseline time domain feature vector And the task time domain feature vector for: , S402, removing the baseline of the time domain feature vector of the task state, and converting the task time domain feature vector of the task state Subtract the baseline time domain feature vector of the baseline state , get the feature vector after removing the baseline .

5. The objective quantitative evaluation method for brain cognitive load based on pulse wave morphological characteristics according to claim 1, characterized in that: The process of step S5 is as follows: S501, use the sequence backward selection algorithm to select the feature vector The features in are screened, and the classification accuracy of the SVM classifier is used as the evaluation criterion. The first m features with the highest accuracy are selected to obtain the feature vector with the optimal feature combination for subsequent classifier training; S502: Extract feature vectors of all training samples to form a feature vector set , and Perform standardization to obtain the standardized feature vector set ; S503. The task accuracy recorded after the subject's mental task is completed is used as task performance to divide the cognitive load level into: cognitive load overload level and cognitive load matching level; samples with lower task performance are classified as cognitive load overload level, and samples with higher task performance are classified as cognitive load matching level. The cognitive load level of the sample and the standardized feature vector set Input into the SVM classifier and train the SVM classifier.

6. The objective quantitative evaluation method for brain cognitive load based on pulse wave morphological characteristics according to claim 1, characterized in that: In step S6, the hard discriminant output of the SVM classifier is converted into a probability output through the Sigmoid parameter model to obtain the probability that the cognitive load level belongs to the cognitive load matching level. and the probability of being at the cognitive overload level .

7. The objective quantitative evaluation method for brain cognitive load based on pulse wave morphology characteristics according to claim 1, characterized in that: In step S7, the SVM probability output model is used to quantitatively evaluate the cognitive load level of the pulse wave signal of the subject to be measured, wherein the quantitative evaluation criterion is: The value of reflects the cognitive load level of the subject to be tested. When , the cognitive load level of the subject is at the cognitive load matching level, and the closer the value is to 0.5, the higher the relative load level; when When , the cognitive load level of the subject is at the cognitive load overload level, and the closer the value is to 0.5, the lower the relative load level is, and the closer it is to the load matching level; if the value is higher, the probability of meeting the cognitive load overload level is higher, and the task difficulty is very likely to exceed the load level that the subject can bear, and at this time it is necessary to reduce the task difficulty.

Images