WO2020034126A1

WO2020034126A1 - Sample training method, classification method, identification method, device, medium, and system

Info

Publication number: WO2020034126A1
Application number: PCT/CN2018/100711
Authority: WO
Inventors: 杜文静; 王磊; 李慧慧
Original assignee: 深圳先进技术研究院
Priority date: 2018-08-15
Filing date: 2018-08-15
Publication date: 2020-02-20

Abstract

Provided are a sample training method, a classification method, a lower back pain symptom identification method, a computer device, a computer-readable storage medium, and a lower back pain symptom identification system applicable in the technical field of computers. A first classifier not only takes into consideration the generality of sample features in a second sample set pertaining to a first category compared with changes in reference sample features of a second category, but also takes into consideration the relative scarcity of sample features in a third sample set pertaining to the first category compared with the changes in the reference sample features of the second category, thus allowing a quick and accurate classification; meanwhile, a second classifier focuses on taking into consideration the scarcity of the sample features in the third sample set compared with the changes in the reference sample features of the second category, thus allowing a corrective reclassification when the first classifier misclassifies, and effectively ensuring the accuracy of a test sample classification.

Description

Sample training method, classification method, recognition method, device, medium and system

Technical field

The invention belongs to the field of computer technology, and particularly relates to a sample training method, a classification method, a method for identifying lower back pain symptoms, a computing device, a computer-readable storage medium, and a system for identifying lower back pain symptoms.

Background technique

Due to the uneven distribution of information resources, for the information belonging to the same category, the amount of information in some categories is obviously scarce but it is significantly different from the information characteristics of another reference category, and the amount of information in some categories is significantly more abundant than that of the reference category. There is no significant difference in characteristics, which makes the scarce scarce category information quantitatively unable to be compared with the obviously abundant ordinary category information. In the existing classification process, the rare category information is not unique to the ordinary category information because of its special characteristics. Different treatment results in the relatively weak information of rare class information in the trained classifier, and the classification accuracy of the test samples cannot be effectively guaranteed.

Summary of the Invention

The purpose of the present invention is to provide a sample training method, a classification method, a method for identifying lower back pain symptoms, a computing device, a computer-readable storage medium, and a system for identifying lower back pain symptoms. The problem of sample classification accuracy.

In one aspect, the present invention provides a sample training method, which includes the following steps:

A first sample set composed of samples to be trained belonging to the first category is obtained, and the first sample set includes: compared to reference samples belonging to the second category, the samples to be trained do not have significant differences in feature changes A second sample set of, and a sample to be trained that has a significant difference between the feature change compared to the reference sample and the feature change of the sample to be trained in the second sample set compared to the reference sample The third sample set formed;

The features of the samples to be trained in the first sample set and the features of the samples to be trained in the third sample set are trained by a machine learning classification method, and a first classifier and a second classifier are obtained correspondingly.

In another aspect, the present invention provides a classification method, which includes the following steps:

Input the characteristics of the sample to be tested into the first classifier for the first judgment, and if the obtained first judgment result indicates that the sample to be tested belongs to the first category, use the first judgment result as the classification result,

If the first judgment result indicates that the sample to be tested does not belong to the first category, the characteristics of the sample to be tested are input to a second classifier for a second judgment; if the obtained second judgment result indicates the The sample to be tested belongs to the first category, and the second judgment result is used as a classification result,

The classified samples in the first sample set corresponding to the first classifier and the classified samples in the third sample set corresponding to the second classifier both belong to the first category, and the first sample The set includes a second sample set composed of classified samples that have no significant difference in feature changes compared to reference samples belonging to the second category, and a feature change compared to the reference sample, and the first The third sample set composed of the classified samples with a significant difference between the classified samples in the two sample set compared to the reference sample.

In another aspect, the present invention also provides a method for identifying symptoms of lower back pain, which includes the following steps:

Obtain local electromyographic signals of the waist of the test subject;

Pre-processing the local electromyographic signal of the waist muscle to obtain a sample to be tested;

Processing the sample to be tested to obtain characteristics of the sample to be tested;

Input the characteristics of the sample to be tested into the first classifier for the first judgment, and if the obtained first judgment result indicates that the sample to be tested belongs to the first category, use the first judgment result as the classification result, If the first judgment result indicates that the sample to be tested does not belong to the first category, the characteristics of the sample to be tested are input to a second classifier for a second judgment; if the obtained second judgment result indicates the The sample to be tested belongs to the first category, and the second judgment result is used as the classification result, wherein the classified samples in the first sample set corresponding to the first classifier and the second classifier correspond to The classified samples in the third sample set of the all belong to the first category, and the first sample set includes: the first sample set consisting of the classified samples that have no significant difference in feature changes compared to the reference samples belonging to the second category. A two-sample set, and a significant difference between a feature change compared to the reference sample and a feature change of a classified sample in the second sample set compared to the reference sample The third sample set composed of the classified samples, the first category is a category of lower back pain symptoms, and the second category is a category of no lower back pain symptoms.

In another aspect, the present invention also provides a computing device including a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor is implemented when the computer program is executed. As in the steps above.

In another aspect, the present invention also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in the method are implemented.

In another aspect, the present invention also provides a low back pain symptom recognition system, which includes:

An acquisition module, configured to obtain a local muscle electromyography signal of the waist of the subject;

A preprocessing module, configured to preprocess the myoelectric signals of the local muscles of the waist to obtain a sample to be tested;

A feature extraction module, configured to process the sample to be tested to obtain characteristics of the sample to be tested; and

A classification module, configured to input features of the sample to be tested into a first classifier for a first judgment, and if the obtained first judgment result indicates that the sample to be tested belongs to the first category, use the first judgment The result is a classification result. If the first judgment result indicates that the sample to be tested does not belong to the first category, the characteristics of the sample to be tested are input to a second classifier for a second judgment. The judgment result indicates that the sample to be tested belongs to the first category, and the second judgment result is used as a classification result, wherein the classified samples in the first sample set corresponding to the first classifier and the classification The classified samples in the third sample set corresponding to the second classifier all belong to the first category, and the first sample set includes: those that have no significant difference in feature changes compared to the reference samples belonging to the second category A second sample set composed of classified samples, and a feature change of the feature compared to the reference sample, and a feature change of the classified sample in the second sample set compared to the reference sample The third sample set composed of classified samples with significant differences among them, the first category is a category of lower back pain symptoms, and the second category is a category of no lower back pain symptoms.

In the present invention, the characteristics of the sample to be tested are input to the first classifier for the first judgment. If the obtained first judgment result indicates that the sample to be tested belongs to the first category, the first judgment result is used as the classification result. If not, the characteristics of the sample to be tested are input to a second classifier for a second judgment. If the obtained second judgment result indicates that the sample to be tested belongs to the first category, the second judgment result is used. As a classification result, wherein the classified samples in the first sample set corresponding to the first classifier and the classified samples in the third sample set corresponding to the second classifier both belong to the first category, the The first sample set includes: a second sample set consisting of a reference sample that is compared to a second category, a classified sample that does not have a significant difference in feature change, and a feature change that is compared to the reference sample, and The third sample set composed of the classified samples having a significant difference between the classified samples in the second sample set and the feature changes of the reference sample. In this way, the first classifier considers not only the generality of the changes in the sample features in the second sample set belonging to the first category compared to the reference sample features in the second category, but also the sample features in the third sample set belonging to the first category. Relative rarity of feature changes compared to the second category of reference samples, so that classification can be performed quickly and accurately, while the second classifier focuses on the features of the samples in the third sample set belonging to the first category compared to the second The rarity of the feature change of the category reference sample can be used to correct the reclassification when the classification of the first classifier is wrong, thereby effectively ensuring the accuracy of the classification of the sample to be tested.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is an implementation flowchart of a sample training method provided by Embodiment 1 of the present invention;

FIG. 2 is an implementation flowchart of a classification method provided by Embodiment 2 of the present invention; FIG.

3 is a flowchart of a method for identifying symptoms of lower back pain provided by Embodiment 3 of the present invention;

FIG. 4 is a schematic diagram of an experimental result of the classification recognition accuracy rate when the patient is in different motion states in Embodiment 3 of the present invention; FIG.

5 is a schematic structural diagram of a computing device according to a fourth embodiment of the present invention;

FIG. 6 is a schematic structural diagram of a lower back pain symptom recognition system according to a sixth embodiment of the present invention.

detailed description

In order to make the objectives, technical solutions, and advantages of the present invention clearer, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

The following describes the specific implementation of the present invention in detail with reference to specific embodiments:

Embodiment one:

FIG. 1 shows the implementation process of the sample training method provided in the first embodiment of the present invention. For convenience of explanation, only the parts related to the embodiment of the present invention are shown, and the details are as follows:

In step S101, a first sample set consisting of to-be-trained samples belonging to a first category is obtained, and the first sample set includes: A second sample set composed of training samples, and a to-be-trained train that has a significant difference between the feature change compared to the reference sample and the feature change of the to-be-trained sample in the second sample set compared to the reference sample The sample constitutes a third sample set.

In the embodiment of the present invention, the to-be-trained samples in the first sample set and the to-be-trained samples in the third sample set have the same number of index parameters, and the difference between the sample features can be indicated by the calculated distance. The following uses several application examples to explain.

For example, the samples to be trained belong to the category of lower back pain symptoms, and relatively there are no categories of lower back pain symptoms (or health categories). These samples to be trained constitute a first sample set, and the first sample set contains There are a second set of samples and a third set of samples, where the samples to be trained in the second set of samples are similar: that is, the characteristics change compared to the reference sample with no back pain symptoms (no back pain symptoms or healthy samples) There is no significant difference. It may be that the average electromyography (AEMG) obtained by processing the local electromyographic signals of the lower back muscles of the patients with lower back pain corresponding to the second sample set is lower than that of those without normal back pain or healthy. The AEMG and other changes in the same direction, while the characteristics of the samples to be trained in the third sample set are changed compared to the characteristics of the above reference samples, and the characteristics of the samples to be trained in the second sample set are compared to the characteristics of the above reference samples. There is a significant difference between them. It may be that the AEMG of the patients with lower back pain corresponding to the second sample set is the same as that of the normal people who have no symptoms of lower back pain or healthy people. Changes in the AEMG and other aspects of patients with lower back pain corresponding to the third sample set compared with those of normal persons who have no symptoms of low back pain or healthy people, and the AEMG and other phases of patients with lower back pain corresponding to the second sample set Compared with normal people who have no symptoms of lower back pain or healthy people, such as AEMG, the direction of change is opposite. Whether there is a significant difference in feature changes may also be reflected in the change in the relevant index parameters, and the number of samples to be trained in the second sample set is the majority. It belongs to the ordinary category, and the samples to be trained in the third sample set are few and rare, but both the samples to be trained in the second sample set and the samples to be trained in the third sample set are symptoms of lower back pain category.

For another example: the collected samples to be trained belong to the low-emotion category, and there are relatively non-emotional categories. These to-be-trained samples constitute a first sample set, and the first sample set includes a second sample set and a third sample Set, where the samples to be trained in the second sample set are similar: that is, compared with the non-emotional low-level reference samples (emotion calm samples or emotional agitation samples), the feature changes do not have significant differences, which may be corresponding to the second sample set The eye shape indications and other values obtained from the facial image processing of the depressed person are changed in the same direction compared to the eye shape indications and the like of the person with a peaceful mood, and the samples to be trained in the third sample set are compared with the reference samples. There is a significant difference between the change in the feature value of the second sample set and the feature change of the sample to be trained in the second sample set compared to the reference sample, which may be the eye shape indicator value of the depressed person corresponding to the second sample set. The eye shape of the person with a peaceful mood changes in the same direction, and the eye shape of the person with low mood corresponding to the third sample set Compared with the direction of changes in the eye shape of the person who is at peace, the instruction value and other changes are compared with the values of the eye shape of the person at low mood corresponding to the second sample set. The direction is opposite, while the samples to be trained in the second sample set are majority and belong to the ordinary category, the samples to be trained in the third sample set are few and belong to the rare category, but whether it is the samples to be trained in the second sample set or the third The samples to be trained in the sample set all belong to the depressed mood category.

In step S102, the features of the samples to be trained in the first sample set and the features of the samples to be trained in the third sample set are trained by a machine learning classification method, and a first classifier and a second classifier are obtained correspondingly. In the embodiment of the present invention, when performing sample training, not only the samples to be trained, but also reference samples are needed, so as to train to obtain a classifier. Due to the different sets of samples targeted for training, the first and second classifiers obtained are also different.

In the embodiment of the present invention, because the first classifier considers both the generality of the changes in the sample features in the second sample set belonging to the first category compared to the reference sample features in the second category, and the first The rarity of the sample features in the three-sample set compared to the reference samples in the second category, which can provide fast and accurate classification. The second classifier focuses on the third sample set that belongs to the first category. Compared with the rarity of the characteristic changes of the reference sample of the second category, it can further provide a guarantee for the correct re-classification of the first classifier when it is incorrectly classified, thereby effectively ensuring the accuracy of the sample to be classified.

Embodiment two:

FIG. 2 shows the implementation flow of the classification method provided in Embodiment 2 of the present invention. The classification method is based on the first classifier and the second classifier implemented in the first embodiment. The first classifier and the second classifier can perform stages. To obtain a Support Vector Machine (SVM) classifier. For ease of description, only the parts related to the embodiment of the present invention are shown, and the details are as follows:

In step S201, a sample to be tested is obtained.

In step S202, the sample to be tested is processed to obtain the characteristics of the sample to be tested.

In step S203, the characteristics of the sample to be tested are input to the first classifier for the first judgment. If the obtained first judgment result indicates that the sample to be tested belongs to the first category, step S204 is performed; otherwise, step S205 is performed.

In step S204, the first determination result is used as the classification result.

In step S205, the characteristics of the sample to be tested are input to the second classifier for a second judgment. If the obtained second judgment result indicates that the sample to be tested belongs to the first category, step S206 is performed.

In step S206, the second judgment result is used as the classification result.

In the embodiment of the present invention, because the first classifier considers both the generality of the changes in the sample features in the second sample set belonging to the first category compared to the reference sample features in the second category, and the first The relative rarity of the characteristics of the samples in the three sample set compared to the characteristics of the reference sample in the second category, so that the classification can be performed quickly and accurately. The second classifier focuses on the samples in the third sample set that belong to the first category. The rarity of feature changes compared to the reference samples of the second category, so that the first classifier can be reclassified in a correct manner when the classification is incorrect, thereby effectively ensuring the accuracy of the sample to be tested.

In addition, if a test sample that is known to belong to the first category is input in the above process, if the first judgment result obtained by the first classifier indicates that the test sample does not belong to the first category, the first classifier judges incorrectly, indicating that The test sample may belong to the category corresponding to the third sample set, and then the second classifier needs to be used to judge the test sample. If the second judgment result obtained by the second classifier indicates that the test sample belongs to the first category, the first The two classifiers make correct judgments, otherwise they make incorrect judgments. If the number of test samples targeted by cumulative judgment errors is large, these test samples need to be used as training samples in the third sample set, and retrained to update to obtain a new A classifier and a second classifier.

Embodiment three:

FIG. 3 shows the implementation flow of the method for identifying lower back pain symptoms provided by the third embodiment of the present invention. For ease of description, only parts related to the embodiment of the present invention are shown, and the details are as follows:

In step S301, a waist local muscle electromyogram signal of the subject is obtained.

In the embodiment of the present invention, an electrode sheet may be pasted on the surface of the waist muscle of the subject, and the electrode sheet records the bioelectrical signal released during the neuromuscular activity, that is, the above-mentioned local muscle electromyographic signal of the waist.

In step S302, pre-processing is performed on the lumbar local muscle EMG signal to obtain a test sample.

In the embodiment of the present invention, the preprocessing involves filtering, denoising, and normalizing the myoelectric signals of local waist muscles. The effective frequency of the EMG signal is 10-500 Hz. Therefore, the original signal collected must be processed by a 10-500 Hz band-pass filter. The 50 Hz power frequency interference generated by the acquisition equipment and China's voltage of 220 volts usually causes interference to the EMG signal. Therefore, the power frequency denoising of the signal needs to be performed at 50 Hz. Due to the differences between different individuals, in order to eliminate this difference and make individuals judge at a consistent standard level, the filtered and denoised signals need to be standardized, and the maximum normalization normalization algorithm is used to obtain each time The EMG signals are standardized and finally the test sample is obtained.

In step S303, the sample to be tested is processed to obtain the characteristics of the sample to be tested.

In the embodiment of the present invention, time-domain and frequency-domain index parameters representing muscle function status can be screened for the sample to be measured, and time-domain index parameters are obtained: AEMG, Root Mean Square (RMS), Co-contraction ratio (CCR), sample entropy (SamEn), and frequency-domain index parameters: Mean Power Frequency (Mean), Frequency (MPF), and Median Frequency (MDF). Among them, AEMG largely represents the dominating output of the surface electromyography of the selected muscle under a given task or given action; RMS is directly related to the energy of the electromyographic signal and is often used to reflect the energy that generates myoelectricity; CCR reflects The coordination ability of each muscle in a specific task; SamEn reflects the complexity of muscle movement patterns in a specific task; MPF represents the center of gravity frequency of the EMG signal spectrum; MDF represents its total power less than the MPF part and greater than the MPF part The total power is equal, these two indicators reflect the degree of muscle fatigue.

The parameters of each indicator are calculated as follows:

In formula (1-1), N represents the number of sample points of the EMG signal, and Data [i] represents the original EMG signal of a period of time, specifically a time series signal of a period of time. One time point corresponds to a voltage value, such as If i is 10, it refers to the continuous voltage value of 10 points. In the formula (1-2), the molecular part represents the average EMG value of the antagonist muscle muscle signal, and the denominator part represents the average EMG value of all the antagonist muscles and active muscles tested. The mean EMG value is also the instantaneous voltage value obtained by summing and averaging the voltage values at all time points.

SamEn is a set of data consisting of K points of the original data x (1), x (2), x (3), ..., x (K), where K is the total Data length, m is the vector dimension, that is, the total length of the data is converted into an m-dimensional vector as shown in the following formula (1-4) by an algorithm. The value of m is: m <K.

X (i) = [x (i), x (i + 1), ..., x (i + m-1)]

i = 1, 2, ..., K-m + 1 ... (1-4)

Define the distance between vector X (i) and vector X (j) as:

d [X (i), X (j)] = max | x (i + k) -x (j-k) | ...... (1-5)

Among them, k = 0,1,2, ..., m-1,1≤i, j≤K-m + 1, given the similarity tolerance r, calculate d [X ( i), X (j)] <r The ratio of the number of r to the total number of vectors Km-1, the ratio is shown in formula (1-6), where d [X (i), X (j)] <r The number refers to the sum of the number of all sample points whose distance between the vectors x (i) and x (j) is less than r.

All averages for i are:

Adding 1 to the dimension gives:

When K is finite, the sample entropy SamEn of the sequence is:

In formula (1-9), m represents the maximum template length, r represents the matching tolerance, and K is the total data length. Generally, the value of m is 1 or 2, and the range of r is [0.1SD, 0.25SD]. SD is the standard deviation of the time series (Standard Deviation). In the specific application example, m can take the value of 2, and r can take the value. Is 0.15, and the entire data length is 10,000 sample points.

In the above formula, PSD is Power Spectral Density.

Six muscles of each tester were obtained from 172 testers: left and right lateral oblique muscles, external oblique muscles, erector spinae / multifidus muscles. After feature extraction, 31 muscle function parameters were obtained. Characteristically, they are left internal oblique AEMG, right internal oblique AEMG, left external oblique AEMG, right external oblique AEMG, left erector spinae / multisplit AEMG, and right vertical Spinal / multifidus AEMG, Six-muscle overall coordination parameter CCR, Left internal oblique RMS, Right internal oblique RMS, Left external oblique RMS, Right external oblique RMS, Left Side erector spinae / polysacral RMS, right erector spine / polysacral RMS, left internal oblique SamEn, right internal oblique SamEn, left external oblique SamEn, right external oblique SamEn, left erector spinae / multifidus SamEn, right erector spinae / multifidus SamEn, left inner oblique muscle MPF, right inner oblique muscle MPF, left outer oblique muscle MPF, right abdomen Outer oblique muscle MPF, left erector spinae / multifidus MPF, right erector spinae / multifidus MPF, left internal oblique MDF, right internal oblique MDF, left external oblique MDF, Right external oblique MDF, left vertical spine / Multifidus muscle MDF, right erector spinae / multifidus muscle MDF, so that these features can be used for training samples and the subsequent classification.

In step S304, the characteristics of the sample to be tested are input to the first classifier for the first judgment. If the obtained first judgment result indicates that the sample to be tested belongs to the first category, the first judgment result is used as the classification result. If the judgment result indicates that the sample to be tested does not belong to the first category, the characteristics of the sample to be tested are input to the second classifier for a second judgment. If the obtained second judgment result indicates that the sample to be tested belongs to the first category, the second judgment is used. The results are used as classification results.

In the implementation of the present invention, the classified samples in the first sample set corresponding to the first classifier and the classified samples in the third sample set corresponding to the second classifier both belong to the first category, and the first sample set includes : A second sample set composed of classified samples that do not have significant differences in feature changes compared to a reference sample belonging to the second category, and a classification of the second sample set that consists of feature changes compared to reference samples A third sample set consisting of classified samples that have significant differences between the feature changes of the reference sample compared to the reference sample, the first category is the category of lower back pain symptoms, and the second category is the category without lower back pain symptoms or health category.

Because the first classifier considers both the generality of the sample features in the second sample set belonging to the first category compared to the reference sample features in the second category, and also considers the sample feature phases in the third sample set belonging to the first category Relative rarity of feature changes compared to the second category of reference samples, so that classification can be performed quickly and accurately, while the second classifier focuses on the features of the samples in the third sample set belonging to the first category compared to the second category The rarity of the change in the reference sample characteristics can be used to correct the reclassification of the first classifier when it is incorrectly classified, thereby effectively ensuring the accuracy of classification and recognition of lower back pain symptoms; in addition, patients based on lower back pain symptoms The 31 index parameters in the time and frequency domains can more fully reflect the patient's disease characterization, and the method can achieve low-cost, non-invasive, non-radiation effects when applied.

Preferably, the above-mentioned method for identifying low back pain is used to classify and identify 89 of 172 patients with normal low back pain and 83 patients of low back pain. The recognition result is shown in FIG. 4, and the results show that: There is a certain difference in the accuracy of recognition in the state of motion. The specific performance is as follows: in the forward lean state, the accuracy of identifying low back pain is 96.08%; in the backward state, the accuracy of identifying low back pain is 89.13%; In the left-leaning state, the accuracy rate of identifying low back pain is 88.89%; in the right-leaning state, the accuracy rate of identifying low back pain is 90.38%, so the four types of exercise can better distinguish between painless people and lower back. Painful crowd. It can be seen from the recognition effect of the four motion states that when identifying the symptoms of lower back pain, an appropriate human motion state should be selected to improve the accuracy of classification and recognition, such as forward leaning motion mode or right leaning motion mode. Furthermore, the classification and recognition effect of the second-level classifier is significantly better than that of the first-level classifier.

Embodiment 4:

FIG. 5 shows a structure of a computing device provided in Embodiment 4 of the present invention. For convenience of explanation, only parts related to the embodiment of the present invention are shown.

The computing device according to the embodiment of the present invention includes a processor 501, a memory 502, and a computer program 503 stored in the memory 502 and executable on the processor 501. When the processor 501 executes the computer program 503, the steps in the foregoing method embodiments are implemented, for example, steps S101 to S102 shown in FIG. 1 and the like.

The computing device in the embodiment of the present invention may be a personal computer, a smart phone, a tablet computer, or the like. For the steps implemented when the processor 501 executes the computer program 503 in the computing device to implement the foregoing method, reference may be made to the description of the foregoing method embodiment, and details are not described herein again.

Embodiment 5:

In the embodiment of the present invention, a computer-readable storage medium is provided. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in the foregoing method embodiment are implemented, for example, as shown in FIG. 1. The steps S101 to S102 are shown.

The computer-readable storage medium of the embodiment of the present invention may include any entity or device capable of carrying computer program code, a recording medium, for example, a memory such as a ROM / RAM, a magnetic disk, an optical disk, a flash memory, or the like.

Embodiment 6:

FIG. 6 shows a structure of a lower back pain symptom recognition system provided in Embodiment 3 of the present invention. For ease of description, only parts related to the embodiment of the present invention are shown, including:

An acquisition module 601, configured to obtain a local muscle electromyography signal of a waist of a person to be measured;

A preprocessing module 602, configured to preprocess the myoelectric signals of the local muscles of the waist to obtain a test sample;

A feature extraction module 603, configured to process the sample to be tested to obtain features as the sample to be tested; and

The classification module 604 is configured to input the characteristics of the sample to be tested into the first classifier for the first judgment. If the obtained first judgment result indicates that the sample to be tested belongs to the first category, the first judgment result is used as the classification result. If the judgment result indicates that the sample to be tested does not belong to the first category, the characteristics of the sample to be tested are input to the second classifier for a second judgment. If the obtained second judgment result indicates that the sample to be tested belongs to the first category, the second judgment is used. The result is the classification result, where the classified samples in the first sample set corresponding to the first classifier and the classified samples in the third sample set corresponding to the second classifier both belong to the first category, and the first sample set Contains: a second sample set composed of classified samples that have no significant difference in feature changes compared to a reference sample belonging to the second category, and a feature change compared to the reference sample that has been in the second sample set. The classified sample is a third sample set composed of classified samples that have significant differences between the feature changes of the reference sample, and the first category is lower back pain Symptoms category, while the second category is no lower back pain symptoms category or health category.

In this embodiment, for the steps implemented when the lower back pain symptom recognition system implements the foregoing method, reference may be made to the description of the foregoing method embodiment, and details are not described herein again.

In this embodiment, each module of the low back pain symptom recognition system may be implemented by corresponding hardware or software units, and each unit may be an independent software and hardware unit, or may be integrated into one software and hardware unit, which need not be limited here. this invention.

The above description is only the preferred embodiments of the present invention, and is not intended to limit the present invention. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present invention shall be included in the protection of the present invention. Within range.

Claims

A sample training method, characterized in that the method includes the following steps:

A first sample set composed of samples to be trained belonging to the first category is obtained, and the first sample set includes: compared to reference samples belonging to the second category, the samples to be trained do not have significant differences in feature changes A second sample set of, and a sample to be trained that has a significant difference between the feature change compared to the reference sample and the feature change of the sample to be trained in the second sample set compared to the reference sample The third sample set formed;

The features of the samples to be trained in the first sample set and the features of the samples to be trained in the third sample set are trained by a machine learning classification method, and a first classifier and a second classifier are obtained correspondingly.
A classification method, wherein the classification method includes the following steps:

Input the characteristics of the sample to be tested into the first classifier for the first judgment, and if the obtained first judgment result indicates that the sample to be tested belongs to the first category, use the first judgment result as the classification result,

If the first judgment result indicates that the sample to be tested does not belong to the first category, the characteristics of the sample to be tested are input to a second classifier for a second judgment; if the obtained second judgment result indicates the The sample to be tested belongs to the first category, and the second judgment result is used as a classification result,

The classified samples in the first sample set corresponding to the first classifier and the classified samples in the third sample set corresponding to the second classifier both belong to the first category, and the first sample The set includes a second sample set composed of classified samples that have no significant difference in feature changes compared to reference samples belonging to the second category, and a feature change compared to the reference sample, and the first The third sample set composed of the classified samples with a significant difference between the classified samples in the two sample set compared to the reference sample.
A method for identifying symptoms of lower back pain, wherein the method for identifying symptoms of lower back pain includes the following steps:

Obtain local electromyographic signals of the waist of the test subject;

Pre-processing the local electromyographic signal of the waist muscle to obtain a sample to be tested;

Processing the sample to be tested to obtain characteristics of the sample to be tested;

Input the characteristics of the sample to be tested into the first classifier for the first judgment, and if the obtained first judgment result indicates that the sample to be tested belongs to the first category, use the first judgment result as the classification result, If the first judgment result indicates that the sample to be tested does not belong to the first category, the characteristics of the sample to be tested are input to a second classifier for a second judgment; if the obtained second judgment result indicates the The sample to be tested belongs to the first category, and the second judgment result is used as the classification result, wherein the classified samples in the first sample set corresponding to the first classifier and the second classifier correspond to The classified samples in the third sample set of the all belong to the first category, and the first sample set includes: the first sample set consisting of the classified samples that have no significant difference in feature changes compared to the reference samples belonging to the second category. A two-sample set, and a significant difference between a feature change compared to the reference sample and a feature change of a classified sample in the second sample set compared to the reference sample The classification of samples constituting a third sample set, the first symptoms of low back pain is a category category, the second category is a category no lower back pain symptoms.
A computing device, characterized in that the computing device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the processor A computer program implements the steps in the method according to any one of claims 1 to 3.
A computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the steps in the method according to any one of claims 1 to 3 are implemented.
A low back pain symptom recognition system, characterized in that the low back pain symptom recognition system includes:

An acquisition module, configured to obtain a local muscle electromyography signal of the waist of the subject;

A preprocessing module, configured to preprocess the myoelectric signals of the local muscles of the waist to obtain a sample to be tested;

A feature extraction module, configured to process the sample to be tested to obtain characteristics of the sample to be tested; and

A classification module, configured to input features of the sample to be tested into a first classifier for a first judgment, and if the obtained first judgment result indicates that the sample to be tested belongs to the first category, use the first judgment The result is a classification result. If the first judgment result indicates that the sample to be tested does not belong to the first category, the characteristics of the sample to be tested are input to a second classifier for a second judgment. The judgment result indicates that the sample to be tested belongs to the first category, and the second judgment result is used as a classification result, wherein the classified samples in the first sample set corresponding to the first classifier and the classification The classified samples in the third sample set corresponding to the second classifier all belong to the first category, and the first sample set includes: those that have no significant difference in feature changes compared to the reference samples belonging to the second category A second sample set composed of classified samples, and a feature change of the feature compared to the reference sample, and a feature change of the classified sample in the second sample set compared to the reference sample The classified sample configuration of a third set of samples having a significant difference, the first symptoms of low back pain is a category category, the second category is a category no lower back pain symptoms.
The low back pain symptom recognition system according to claim 6, characterized in that the feature relates to one or more of the following index parameters: the left intraabdominal oblique muscle average electromyography value AEMG, and the right intraabdominal oblique Muscles AEMG, left external oblique AEMG, right external oblique AEMG, left erector spinae / multisplitter AEMG, right erector spinae / multisplitter AEMG, overall coordination parameters of the six muscles Contraction rate CCR, left internal oblique muscle EMG RMS, right internal oblique RMS, left external oblique RMS, right external oblique RMS, left erector spinae / multifidus RMS , Right erector spinae / multifidus RMS, Left internal oblique sample entropy SamEn, Right internal oblique Sam Sam, left external oblique SamEn, right external oblique SamEn, left right spine / Multifidus SamEn, right erector spinae / multifidus SamEn, left internal oblique muscle average power frequency MPF, right internal oblique muscle MPF, left external oblique muscle MPF, right external oblique muscle MPF , Left erector spinae / multifidus MPF, Right erector spinae / multifidus MPF, Left medial oblique median frequency MDF, Right medial oblique MDF, Left medial oblique MDF, Right Lateral external oblique MDF, left Erector spinae / multifidus muscle MDF, on the right erector spinae / multifidus muscle MDF.
The low back pain symptom recognition system according to claim 6, wherein the acquisition module is specifically configured to:

A local electromyographic signal of the waist of a subject under test in a predetermined motion state is obtained, and the motion state is a forward leaning state, a backward leaning state, a left leaning state, or a right leaning state.
The lower back pain symptom recognition system according to claim 6, wherein the preprocessing module is specifically configured to:

Filtering, denoising, and normalizing the myoelectric signals of the lumbar local muscles.
The low back pain symptom recognition system according to claim 9, wherein the preprocessing module comprises:

10-500 Hz band-pass filter, 50 Hz power frequency denoiser and maximum normalization normalization unit.