Classification system and method based on data inspection and controllable feature selection

Abstract

The invention belongs to the technical field of data representation, classification and electric signal processing, and relates to a classification system and method based on data inspection and controllable feature selection. The system comprises a data acquisition module, a feature extraction module, a feature crossing unit and a machine learning classification module, and further comprises a data preprocessing module and an I / O control module. And the I / O control module realizes controllable feature selection by controlling input and output. According to the system and the method, autocorrelation and active section secondary inspection are introduced by means of the data preprocessing module, so that the data quality is improved; multi-dimensional feature extraction is realized through a feature extraction module; controllable feature selection is introduced into the I / O control module, dynamic selection and optimization of features are achieved, and the accuracy and flexibility of diagnosis are improved; machine learning model classification is adopted, and automation and intelligentization of signal classification are achieved; the modular design enables the functional units to be closely connected, data flow is smooth, and stable operation, upgrading and maintenance of the system are facilitated.

Description

Technical Field

[0001] The present invention belongs to the technical field of data representation, classification and electrical signal processing, and relates to a classification system and method based on data verification and controllable feature selection. Background Art

[0002] With the rapid development of information technology and the acceleration of digital transformation, data is growing at an alarming rate. This data explosion not only brings rich information resources, but also poses unprecedented challenges to data processing, analysis and utilization. The complexity, diversity and massiveness of data make it difficult for traditional data processing methods to cope with it, and new technologies and methods are urgently needed to improve the efficiency and accuracy of data processing. Classification is a basic and important task in all walks of life. Whether it is finance, telecommunications, industrial production or biomedicine, data needs to be classified to extract valuable information or knowledge. Data quality is an important guarantee for the accuracy of classification results. However, in practical applications, data often has problems such as missing, anomalies, and noise, which will seriously affect the performance of classification models. Therefore, data verification before classification is an indispensable step. Therefore, a more efficient and accurate classification method is needed to meet these challenges. Through data verification, problems in the data can be discovered and handled in a timely manner to ensure that the input data of the classification model is accurate and reliable. In addition, in the classification task, through feature selection, redundant, irrelevant or noise features can be excluded to improve the performance and efficiency of the classification model. At the same time, feature selection can also reduce computing costs and storage requirements, making classification models easier to implement and deploy. Summary of the invention

[0003] The purpose of the present invention is to address the problems of insufficient data verification and incomplete feature extraction in existing classification methods, and to propose a classification system and method based on data verification and controllable feature selection.

[0004] In order to achieve the above object, the present invention adopts the following technical scheme:

[0005] As one aspect of the present invention, an I / O control module is proposed, which receives data preprocessed and selects output for feature extraction, including a mutually independent preamble and source feature selection unit, a high-order feature selection unit and a posture selection unit; the preamble and source feature selection unit selects the peak envelope or / and the bandpass filtered output data of the channel selected data for feature extraction; the high-order feature selection unit selects the low-pass filtered output data of the channel selected data or / and the output data of the channel selected data after bandpass filtering and autocorrelation test or / and the output data of the channel selected data after bandpass filtering, autocorrelation test and active segment detection or / and the output data of the channel selected data after low-pass filtering and static detection for feature extraction; the posture selection unit selects the output data of the channel selected data after low-pass filtering and / or static detection and / or normalization for feature extraction.

[0006] The preamble and source feature selection unit selects data after channel selection and analog-to-digital conversion or data after channel selection, analog-to-digital conversion and band-pass filtering as data for extracting preamble features; selects peak envelope as data for extracting source features; the peak envelope is obtained by autocorrelation test, active segment detection and peak extraction of the data after channel selection, analog-to-digital conversion and band-pass filtering.

[0007] The high-order feature selection unit selects the low-pass filtered output data of the data after channel selection, the output data of the data after band-pass filtering, autocorrelation test and active segment detection, the output data of the data after channel selection, analog-to-digital conversion, low-pass filtering and static detection, and the output data of the data after analog-to-digital conversion, band-pass filtering, autocorrelation test and active segment detection as data for extracting dynamic and fatigue features; and selects the output data of the data after band-pass filtering, autocorrelation test and active segment detection, the analytical envelope, and the output data of the data after analog-to-digital conversion, band-pass filtering, autocorrelation test and active segment detection as data for extracting high-order features.

[0008] The analytical envelope is obtained by channel selection, analog-to-digital conversion, bandpass filtering, autocorrelation test, active segment detection, and envelope analysis.

[0009] The input data of the static detection is the output data after the channel selection is processed through low-pass filtering; the normalized input data is the output data after the channel selection is processed through low-pass filtering and then processed through static detection.

[0010] As another aspect of the present invention, the present invention proposes a classification system based on data verification and controllable feature selection, including a data acquisition module, a data preprocessing module, a feature extraction module, a feature cross unit and a machine learning classification module, the output of the data acquisition module is given to the data preprocessing module for data preprocessing; the feature extraction module receives the data preprocessing and then selects it for the feature extraction module to extract surface muscle electricity, high-order classes and inertial features; the extracted surface muscle electricity, high-order classes and inertial features are input into the feature cross unit for feature crossover, and then output to the machine learning classification module for classification, characterized in that it also includes an I / O control module, the I / O control module is the I / O control module described in the first aspect, the output of the data preprocessing module is the input of the I / O control module, the I / O control module receives the output of the data preprocessing module and selects data that can extract surface muscle electricity, high-order classes and inertial features to output to the feature extraction module.

[0011] The system relies on the feature extraction module to realize multi-dimensional feature extraction, including surface electromyography features, pre-order features, source features, dynamic features, fatigue features, high-order features and posture angle features, which comprehensively describes the signal characteristics;

[0012] The data preprocessing module is used to perform preprocessing operations on the collected original signal, including a bandpass filtering unit, a low-pass filtering unit, a stationary detection unit, an envelope processing unit and a normalization unit, and is characterized in that it also includes an autocorrelation detection unit and an activity segment detection unit for realizing data verification;

[0013] The autocorrelation test unit removes data without obvious changes by calculating the autocorrelation of the surface electromyography data after bandpass filtering after channel selection; the activity segment detection unit divides the data after channel selection and autocorrelation test after bandpass filtering, and then retains the activity segment based on sample entropy and the number of slope changes; the division uses a sliding window method to segment the data, and the value range of the data segment is 40 to 160; the sliding step range is 10 to 60.

[0014] The data acquisition module is used to collect surface electromyography and attitude angle analog signals, including multi-channel surface muscle electrodes, a channel selection unit, a nine-axis inertial measurement unit and an analog-to-digital conversion unit, and converts the analog signals into digital signals after channel and axis number selection; the channel selection unit is respectively connected to the multi-channel surface muscle electrodes and the nine-axis inertial measurement unit; is used to select the surface electromyography corresponding channel and the nine-axis inertial measurement unit and the axis number and output the surface electromyography and attitude angle analog signals; the analog-to-digital conversion unit is connected to the bandpass filtering unit and the low-pass filtering unit, and is used to convert the surface electromyography and attitude angle analog signals into digital signals; the envelope processing unit performs analytical envelope and peak envelope extraction;

[0015] The feature extraction module includes a surface electromyography feature extraction unit, a fatigue feature extraction unit, a high-order class feature extraction unit and an inertia feature extraction unit, which are used to extract multi-dimensional features; the surface electromyography feature extraction unit is used for pre-order feature extraction and source feature extraction; the inertia feature extraction unit is used for posture feature extraction; the pre-order feature extraction and source feature extraction are connected to the pre-order and source feature selection unit in the I / O control module; the dynamic feature extraction unit and the high-order feature extraction unit are connected to the fatigue feature extraction unit at the same time; the inertia feature extraction unit is connected to the posture feature selection unit in the I / O control module; the feature cross unit is used to screen and combine features; the outputs of all feature extraction modules are connected to the feature cross unit; the machine learning classification module uses the cross-screened features to perform signal intelligent classification.

[0016] As a third aspect of the present invention, a classification method based on data verification and controllable feature selection is proposed, and the classification system based on data verification and controllable feature selection provided in the second aspect is applied, including the following steps:

[0017] S1, collecting the surface electromyography signal and posture angle signal of the subject, and obtaining the surface electromyography and posture data through analog-to-digital conversion; in specific implementation, S1 is realized by a data acquisition module;

[0018] S2, filtering, autocorrelation test, activity segment detection, static detection and envelope processing are performed on the data acquired by S1 to obtain preprocessed data; in specific implementation, S2 is realized by a data preprocessing module;

[0019] S3, select the feature type to be extracted according to the preset conditions; in specific implementation, S3 is realized through the I / O control module;

[0020] S4, extracting multi-dimensional features from the pre-processed data according to the preset conditions of S3; in specific implementation, S4 is implemented through a feature extraction module;

[0021] S5, screening and combination optimization of the extracted multi-dimensional features to obtain data to be classified; in specific implementation, S5 is implemented through a feature cross unit;

[0022] S6. Classify the data to be classified.

[0023] In specific implementation, S6 is realized through a machine learning classification module;

[0024] S2 The filtering of data is specifically: band-pass filtering and low-pass filtering are performed on the muscle electrical data and posture data after channel selection and analog-to-digital conversion respectively; the autocorrelation test is specifically: autocorrelation test is performed on the muscle electrical data after channel selection and analog-to-digital conversion, and the data output by the autocorrelation test is used for active segment detection; the active segment detection is specifically: active segment detection is performed on the data after autocorrelation test on the muscle electrical data after channel selection and analog-to-digital conversion, and the data output by the active segment detection is used to extract envelopes and extract high-order class features; the stillness detection is specifically: stillness detection is performed on the data after low-pass filtering on the posture data after channel selection and analog-to-digital conversion, and the data output by the stillness detection is used for high-order class feature extraction or normalized and then used for posture feature selection; the envelope processing is specifically: extracting analytical envelopes and peak envelopes from the data obtained by active segment detection after autocorrelation test on the muscle electrical data after channel selection and analog-to-digital conversion, and then using them for high-order class feature selection and pre-order feature selection respectively.

[0025] S3 describes selecting the feature type to be extracted according to preset conditions, specifically: selecting the low-pass filtered output data of the data after channel selection, the output data of the data after band-pass filtering, autocorrelation test and active segment detection, the output data of the data after channel selection, analog-to-digital conversion, low-pass filtering and static detection, and the output data of the data after analog-to-digital conversion, band-pass filtering, autocorrelation test and active segment detection as data for extracting dynamic and fatigue features; selecting the output data of the data after band-pass filtering, autocorrelation test and active segment detection, the analytical envelope, and the output data of the data after analog-to-digital conversion, band-pass filtering, autocorrelation test and active segment detection as data for extracting high-order features.

[0026] The S1 collects the surface electromyography signal and posture angle signal of the subject, and obtains the surface electromyography and posture data through analog-to-digital conversion; in specific implementation, it is realized through a data acquisition module, specifically: the surface electromyography signal and posture angle signal of the subject are collected respectively through a multi-channel surface muscle electrode or a nine-axis inertial sensor unit (IMU);

[0027] The surface electromyography signal is an electromyographic signal of the forearm muscle of the subject collected by a multi-channel surface muscle electrode; the posture angle signal is a posture angle signal of the subject collected by a nine-axis IMU;

[0028] The multi-channel surface muscle electrodes include but are not limited to 4-channel and 8-channel electrodes;

[0029] The attitude angle signal includes, but is not limited to, attitude angle signals of the forearm node and the thigh node; the attitude angle signal is a three-axis, six-axis or nine-axis output signal collected by the nine-axis IMU;

[0030] The method and the system it relies on use multi-channel surface muscle electrodes and a nine-axis inertial sensing unit (IMU), which enriches the types and dimensions of data collection and provides a more comprehensive data basis for subsequent feature extraction.

[0031] The preprocessing described in S2, when implemented specifically, includes the following sub-steps:

[0032] S21, performing bandpass filtering on the surface electromyography signal collected in S1 to obtain surface electromyography data after bandpass filtering;

[0033] The filtering range of the bandpass filter is 20-500 Hz;

[0034] S22, performing a 10 Hz low-pass filter on the collected attitude angle signal to obtain attitude data after low-pass filtering;

[0035] S23, calculating the autocorrelation of the surface electromyography data after bandpass filtering and screening, removing the data without obvious changes, and obtaining the data after autocorrelation test;

[0036] The autocorrelation is calculated using a lagged autocorrelation function and a discrete cross-correlation function;

[0037] S24, dividing the data after the autocorrelation test, retaining the active segment based on the sample entropy and the number of slope changes, and obtaining the data after the active segment detection;

[0038] The partitioning uses a sliding window method to segment the data, the window value range is 0.4-0.8s, and the step value range is 0.1-0.3s;

[0039] S25, performing analytical envelope extraction and peak envelope extraction on the data after the activity segment detection, respectively, to obtain an analytical envelope and a peak envelope;

[0040] The analytical envelope extraction includes but is not limited to extracting envelopes based on Hilbert transform and short-time energy method; the peak envelope extraction uses a processing window with a value range of 2-10 for extraction;

[0041] S26, excluding the static segment from the posture data after the low-pass filtering in S22, to obtain data after excluding the static segment;

[0042] The static segment exclusion is based on the sliding window method to divide the data, including but not limited to excluding the static data segment based on the threshold method, the standard deviation method and the energy integration method; the window value range is 0.4-0.8s, and the step value range is 0.1-0.3s;

[0043] S27. Normalize the data after excluding the static segment to obtain normalized data.

[0044] In specific implementation, S3 consists of three parts:

[0045] (31), the preamble and source feature selection unit controls whether the signal after bandpass filtering and peak envelope processing is used for surface electromyography feature extraction;

[0046] (32) The high-order feature selection unit controls whether the signal after low-pass filtering, autocorrelation test, static detection, active segment detection and analytical envelope processing is used for dynamic features, high-order features and fatigue extraction;

[0047] (33), the attitude feature selection unit controls whether the normalized signal is used for inertial feature extraction;

[0048] S3, select the feature type to be extracted according to the preset conditions; in specific implementation, S3 is realized through the I / O control module;

[0049] S4, according to the preset conditions of S3, extract multi-dimensional features from the pre-processed data; in specific implementation, it is realized by a feature extraction module; the multi-dimensional features are extracted from surface electromyography features, high-order class features, and inertial features, specifically:

[0050] S41, extracting the pre-sequence features and source features of the input signal; the pre-sequence features are signal features when the muscle begins to contract and the action has not yet started; the source features include but are not limited to root mean square, average amplitude and average power frequency; when S41 is specifically implemented, it is realized by a surface electromyography feature extraction unit;

[0051] S42, extracting dynamic features, fatigue features and high-order features of the input signal;

[0052] The dynamic features include but are not limited to activity level, momentum and relative rate of change; the high-order features include but are not limited to skewness, kurtosis, entropy and fractal dimension; the fatigue features include but are not limited to dynamic features, high-order features, median frequency, average absolute value and average amplitude; S42 is implemented by a high-order feature extraction unit;

[0053] S43, extracting posture features of the input signal; the posture features include but are not limited to acceleration, angular velocity, relative position and features in S42; when S43 is specifically implemented, it is realized by an inertial feature extraction unit;

[0054] S5. Screen and optimize the extracted multi-dimensional features to obtain data to be classified. In the specific implementation, this is achieved through a feature cross unit, which specifically includes the following sub-steps:

[0055] S51, the feature cross unit cross-combines the screened features to generate new composite features;

[0056] S52, using the feature selection method again to perform a second screening on the feature set containing the composite features to obtain a final optimized feature set;

[0057] The cross combination includes, but is not limited to, feature crossover based on factorization machine, Kronecker product, and attention decomposition machine;

[0058] The feature selection method includes but is not limited to the NCFS method, the maximum correlation minimum redundancy method, the Relief algorithm, and the random forest algorithm.

[0059] S6. Classify the data to be classified, which specifically includes the following sub-steps:

[0060] S61, inputting the optimized features into a pre-trained machine learning model;

[0061] S62, the machine learning model outputs classification results, including classification categories and confidence levels;

[0062] S63. If the confidence level is lower than the preset threshold, the system will mark the sample as "pending confirmation";

[0063] The machine learning model includes but is not limited to SVM, decision tree, multi-layer perceptron model, logistic regression and neural network.

[0064] Beneficial Effects

[0065] Compared with the existing classification systems, the classification system based on data verification and controllable feature selection described in the present invention has the following beneficial effects:

[0066] 1. The system uses multi-channel surface muscle electrodes and a nine-axis inertial measurement unit, which enriches the types and dimensions of data collection and provides a more comprehensive data basis for subsequent feature extraction;

[0067] 2. The system relies on the feature extraction module to realize multi-dimensional feature extraction, including surface electromyography features, pre-order features, source features, dynamic features, fatigue features, high-order features and posture angle features, which comprehensively describes the signal characteristics;

[0068] 3. The system introduces a controllable feature selection mechanism based on the I / O control module. Through the I / O control module and the feature crossover and selection module, the dynamic selection and optimization combination of features are realized, thereby improving the accuracy and flexibility of subsequent diagnosis;

[0069] 4. The machine learning classification module of the system adopts machine learning method to classify signals, realizing the automation and intelligence of signal classification process;

[0070] 5. The modular design of the system enables close connection between various functional units and smooth data flow, which is conducive to the stable operation of the system and subsequent upgrade and maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 A classification system based on data verification and controllable feature selection in an embodiment of the present invention;

[0072] Figure 2 Schematic diagram of the I / O control interface of the classification system based on data verification and controllable feature selection in an embodiment of the present invention. DETAILED DESCRIPTION

[0073] The classification system based on data verification and controllable feature selection described in the present invention is further illustrated and described in detail below in conjunction with the accompanying drawings and embodiments.

[0074] Example 1

[0075] This embodiment describes the specific implementation of the classification system based on data verification and controllable feature selection for clinical attention deficit hyperactivity disorder (ADHD) auxiliary diagnosis and detection. ADHD is a common neurodevelopmental disorder with symptoms of difficulty concentrating, overactivity and increased impulsive behavior. The initial diagnosis of ADHD often begins in childhood, but has far-reaching effects, often continuing into adulthood and hindering daily life and social interaction. At present, the diagnosis process of ADHD is highly dependent on detailed clinical evaluation, which is not only time-consuming, but also faces the challenge of scarce professional diagnostic resources worldwide, often leading to delays in diagnosis. In view of the above dilemma, in recent years, innovations in the fields of machine learning and artificial intelligence have opened up new ways for the rapid and accurate diagnosis of ADHD. However, existing auxiliary diagnosis systems mostly rely on high-end medical equipment, such as magnetic resonance imaging and electroencephalograms, which are not only expensive but also require patients to remain still for a long time, increasing the complexity of diagnosis and the difficulty of patient cooperation. Although they have shown certain potential in improving diagnostic accuracy, their high cost, complex operation and limited popularity cannot be ignored. In contrast, the combination of surface electromyography and accelerometer nodes is relatively simple to collect signals. It can not only fully monitor the movement state, but also has the advantages of ease of use and low cost. The introduction of a controllable feature selection mechanism allows the feature set to be flexibly adjusted according to specific needs, making feature selection more flexible and diverse. Combined with machine learning, it can achieve efficient and accurate diagnosis of ADHD, overcoming the shortcomings of existing methods.

[0076] This embodiment describes the application of the system of the present invention to the diagnosis of adult ADHD. Figure 1The classification system based on data inspection and controllable feature selection shown in the figure includes a data acquisition module, a data preprocessing module, an I / O control module, a feature extraction module, a feature cross unit and a machine learning classification module. The system introduces a controllable feature selection mechanism based on the I / O control module. Through the I / O control module and the feature cross and selection module, the dynamic selection and optimization combination of features are realized, and the accuracy and flexibility of subsequent diagnosis are improved; the system adopts a machine learning classification module for signal classification, realizing the automation and intelligence of the signal classification process; the modular design of the system makes the various functional units closely connected and the data flow smooth, which is conducive to the stable operation of the system and subsequent upgrade and maintenance.

[0077] The I / O control module receives data preprocessed and selects output for feature extraction, including a mutually independent preamble and source feature selection unit, a high-order feature selection unit and a posture selection unit; the preamble and source feature selection unit selects the peak envelope or / and the bandpass filtered output data of the channel selected data for feature extraction; the high-order feature selection unit selects the low-pass filtered output data of the channel selected data or / and the output data of the channel selected data after bandpass filtering and autocorrelation inspection or / and the output data of the channel selected data after bandpass filtering, autocorrelation inspection and active segment detection or / and the output data of the channel selected data after low-pass filtering and static detection for feature extraction; the posture selection unit selects the output data of the channel selected data after low-pass filtering and / or static detection and / or normalization for feature extraction.

[0078] The preamble and source feature selection unit selects data after channel selection and analog-to-digital conversion or data after channel selection, analog-to-digital conversion and band-pass filtering as data for extracting preamble features; selects peak envelope as data for extracting source features; the peak envelope is obtained by autocorrelation test, active segment detection and peak extraction of the data after channel selection, analog-to-digital conversion and band-pass filtering.

[0079] The high-order feature selection unit selects the low-pass filtered output data of the data after channel selection, the output data of the data after band-pass filtering, autocorrelation test and active segment detection, the output data of the data after channel selection, analog-to-digital conversion, low-pass filtering and static detection, and the output data of the data after analog-to-digital conversion, band-pass filtering, autocorrelation test and active segment detection as data for extracting dynamic and fatigue features; and selects the output data of the data after band-pass filtering, autocorrelation test and active segment detection, the analytical envelope, and the output data of the data after analog-to-digital conversion, band-pass filtering, autocorrelation test and active segment detection as data for extracting high-order features.

[0080] The analytical envelope is obtained by channel selection, analog-to-digital conversion, bandpass filtering, autocorrelation test, active segment detection, and envelope analysis.

[0081] The input data of the static detection is the output data after the channel selection is processed through low-pass filtering; the normalized input data is the output data after the channel selection is processed through low-pass filtering and then processed through static detection.

[0082] On the other hand, a classification system based on data verification and controllable feature selection is proposed, including a data acquisition module, a data preprocessing module, a feature extraction module, a feature cross unit and a machine learning classification module, the output of the data acquisition module is given to the data preprocessing module for data preprocessing; the feature extraction module receives the data preprocessing and then selects it for the feature extraction module to extract surface muscle electricity, high-order classes and inertial features; the extracted surface muscle electricity, high-order classes and inertial features are input into the feature cross unit for feature crossover, and then output to the machine learning classification module for classification, characterized in that it also includes an I / O control module, the I / O control module is the I / O control module described in the first aspect, the output of the data preprocessing module is the input of the I / O control module, the I / O control module receives the output of the data preprocessing module and selects data that can extract surface muscle electricity, high-order classes and inertial features to output to the feature extraction module.

[0083] The system relies on the feature extraction module to realize multi-dimensional feature extraction, including surface electromyography features, pre-order features, source features, dynamic features, fatigue features, high-order features and posture angle features, which comprehensively describes the signal characteristics;

[0084] The data preprocessing module is used to perform preprocessing operations on the collected original signal, including a bandpass filtering unit, a low-pass filtering unit, a stationary detection unit, an envelope processing unit and a normalization unit, and is characterized in that it also includes an autocorrelation detection unit and an activity segment detection unit for realizing data verification;

[0085] The autocorrelation test unit removes data without obvious changes by calculating the autocorrelation of the surface electromyography data after the channel selection and bandpass filtering; the activity segment detection unit divides the data after the channel selection and bandpass filtering and autocorrelation test, and then retains the activity segment based on the sample entropy and the number of slope changes; the division uses the sliding window method to segment the data, and the value range of the data segment is 40 to 160; the sliding step range is 10 to 60; specifically in this embodiment: the value of the data segment is 40, 100 or 160; the sliding step value is 10, 30 or 60;

[0086] The data acquisition module is used to collect surface electromyography and attitude angle analog signals, including multi-channel surface muscle electrodes, a channel selection unit, a nine-axis inertial measurement unit and an analog-to-digital conversion unit, and converts the analog signals into digital signals after channel and axis number selection; the channel selection unit is respectively connected to the multi-channel surface muscle electrodes and the nine-axis inertial measurement unit; is used to select the surface electromyography corresponding channel and the nine-axis inertial measurement unit and the axis number and output the surface electromyography and attitude angle analog signals; the analog-to-digital conversion unit is connected to the bandpass filtering unit and the low-pass filtering unit, and is used to convert the surface electromyography and attitude angle analog signals into digital signals; the envelope processing unit performs analytical envelope and peak envelope extraction;

[0087] The feature extraction module includes a surface electromyography feature extraction unit, a fatigue feature extraction unit, a high-order class feature extraction unit and an inertia feature extraction unit, which are used to extract multi-dimensional features; the surface electromyography feature extraction unit is used for pre-order feature extraction and source feature extraction; the inertia feature extraction unit is used for posture feature extraction; the pre-order feature extraction and source feature extraction are connected to the pre-order and source feature selection unit in the I / O control module; the dynamic feature extraction unit and the high-order feature extraction unit are connected to the fatigue feature extraction unit at the same time; the inertia feature extraction unit is connected to the posture feature selection unit in the I / O control module; the feature cross unit is used to screen and combine features; the outputs of all feature extraction modules are connected to the feature cross unit; the machine learning classification module uses the cross-screened features to perform signal intelligent classification.

[0088] As a third aspect, a classification method based on data verification and controllable feature selection is proposed, and the classification system based on data verification and controllable feature selection provided in the second aspect is applied, including the following steps:

[0089] S1, collecting the surface electromyography signal and posture angle signal of the subject, and obtaining the surface electromyography and posture data through analog-to-digital conversion; in specific implementation, S1 is realized by a data acquisition module;

[0090] S2, filtering, autocorrelation test, activity segment detection, static detection and envelope processing are performed on the data acquired by S1 to obtain preprocessed data; in specific implementation, S2 is realized by a data preprocessing module;

[0091] S3, select the feature type to be extracted according to the preset conditions; in specific implementation, S3 is realized through the I / O control module;

[0092] S4, extracting multi-dimensional features from the pre-processed data according to the preset conditions of S3; in specific implementation, S4 is implemented through a feature extraction module;

[0093] S5, screening and combination optimization of the extracted multi-dimensional features to obtain data to be classified; in specific implementation, S5 is implemented through a feature cross unit;

[0094] S6. Classify the data to be classified.

[0095] In specific implementation, S6 is realized through a machine learning classification module;

[0096] S2 The filtering of data is specifically: band-pass filtering and low-pass filtering are performed on the muscle electrical data and posture data after channel selection and analog-to-digital conversion respectively; the autocorrelation test is specifically: autocorrelation test is performed on the muscle electrical data after channel selection and analog-to-digital conversion, and the data output by the autocorrelation test is used for active segment detection; the active segment detection is specifically: active segment detection is performed on the data after autocorrelation test on the muscle electrical data after channel selection and analog-to-digital conversion, and the data output by the active segment detection is used to extract envelopes and extract high-order class features; the stillness detection is specifically: stillness detection is performed on the data after low-pass filtering on the posture data after channel selection and analog-to-digital conversion, and the data output by the stillness detection is used for high-order class feature extraction or normalized and then used for posture feature selection; the envelope processing is specifically: extracting analytical envelopes and peak envelopes from the data obtained by active segment detection after autocorrelation test on the muscle electrical data after channel selection and analog-to-digital conversion, and then using them for high-order class feature selection and pre-order feature selection respectively.

[0097] S3 describes selecting the feature type to be extracted according to preset conditions, specifically: selecting the low-pass filtered output data of the data after channel selection, the output data of the data after band-pass filtering, autocorrelation test and active segment detection, the output data of the data after channel selection, analog-to-digital conversion, low-pass filtering and static detection, and the output data of the data after analog-to-digital conversion, band-pass filtering, autocorrelation test and active segment detection as data for extracting dynamic and fatigue features; selecting the output data of the data after band-pass filtering, autocorrelation test and active segment detection, the analytical envelope, and the output data of the data after analog-to-digital conversion, band-pass filtering, autocorrelation test and active segment detection as data for extracting high-order features.

[0098] The S1 collects the surface electromyography signal and posture angle signal of the subject, and obtains the surface electromyography and posture data through analog-to-digital conversion; in specific implementation, it is realized through a data acquisition module, specifically: the surface electromyography signal and posture angle signal of the subject are collected respectively through a multi-channel surface muscle electrode or a nine-axis inertial measurement unit (IMU);

[0099] The surface electromyography signal is an electromyographic signal of the forearm muscle of the subject collected by a multi-channel surface muscle electrode; the posture angle signal is a posture angle signal of the subject collected by a nine-axis IMU;

[0100] The multi-channel surface muscle electrode includes, but is not limited to, 4-channel and 8-channel electrodes; specifically in this embodiment, the number of channels of the surface muscle electrode is 4, 6 or 8 channels.

[0101] The attitude angle signal includes but is not limited to the attitude angle signals of the forearm node and the thigh node; the attitude angle signal is a three-axis, six-axis or nine-axis output signal collected by the nine-axis IMU; specifically in this embodiment, the IMU is a nine-axis output.

[0102] The method and the system it relies on use multi-channel surface muscle electrodes and a nine-axis inertial sensing unit (IMU), which enriches the types and dimensions of data collection and provides a more comprehensive data basis for subsequent feature extraction.

[0103] The preprocessing described in S2, when implemented specifically, includes the following sub-steps:

[0104] S21, performing bandpass filtering on the surface electromyography signal collected in S1 to obtain surface electromyography data after bandpass filtering;

[0105] The filtering range of the bandpass filter is 20-500 Hz;

[0106] S22, performing a 10 Hz low-pass filter on the collected attitude angle signal to obtain attitude data after low-pass filtering;

[0107] S23, calculating the autocorrelation of the surface electromyography data after bandpass filtering and screening, removing the data without obvious changes, and obtaining the data after autocorrelation test;

[0108] The autocorrelation is calculated using a lagged autocorrelation function and a discrete cross-correlation function;

[0109] S24, dividing the data after the autocorrelation test, retaining the active segment based on the sample entropy and the number of slope changes, and obtaining the data after the active segment detection;

[0110] The partitioning uses a sliding window method to segment the data, the window value range is 0.4-0.8s, and the step value range is 0.1-0.3s; in specific implementation, the window value is 0.8s, and the step value is 0.1 or 0.2;

[0111] S25, performing analytical envelope extraction and peak envelope extraction on the data after the activity segment detection, respectively, to obtain an analytical envelope and a peak envelope;

[0112] The analytical envelope extraction includes but is not limited to extracting envelopes based on Hilbert transform and short-time energy method; the peak envelope extraction uses a processing window with a value range of 2-10 for extraction; in specific implementation, the size of the processing window used is 2, 6 or 10;

[0113] S26, excluding the static segment from the posture data after the low-pass filtering in S22, to obtain data after excluding the static segment;

[0114] The static segment exclusion is based on the sliding window method to divide the data, including but not limited to excluding the static data segment based on the threshold method, the standard deviation method and the energy integration method; the window value range is 0.4-0.8s, and the step value range is 0.1-0.3s; in specific implementation, the window value is 0.4 or 0.6s, and the step value is 0.3s;

[0115] S27. Normalize the data after excluding the static segment to obtain normalized data.

[0116] In specific implementation, S3 consists of three parts:

[0117] (31), the preamble and source feature selection unit controls whether the signal after bandpass filtering and peak envelope processing is used for surface electromyography feature extraction;

[0118] (32) The high-order feature selection unit controls whether the signal after low-pass filtering, autocorrelation test, static detection, active segment detection and analytical envelope processing is used for dynamic features, high-order features and fatigue extraction;

[0119] (33), the attitude feature selection unit controls whether the normalized signal is used for inertial feature extraction;

[0120] S3, select the feature type to be extracted according to the preset conditions; in specific implementation, S3 is realized through the I / O control module;

[0121] S4, according to the preset conditions of S3, extract multi-dimensional features from the pre-processed data; in specific implementation, it is realized by a feature extraction module; the multi-dimensional features are extracted from surface electromyography features, high-order class features, and inertial features, specifically:

[0122] S41, extracting the pre-sequence features and source features of the input signal; the pre-sequence features are signal features when the muscle begins to contract and the action has not yet started; the source features include but are not limited to root mean square, average amplitude and average power frequency; when S41 is specifically implemented, it is realized by a surface electromyography feature extraction unit;

[0123] S42, extracting dynamic features, fatigue features and high-order features of the input signal;

[0124] The dynamic features include but are not limited to activity level, momentum and relative rate of change; the high-order features include but are not limited to skewness, kurtosis, entropy and fractal dimension; the fatigue features include but are not limited to dynamic features, high-order features, median frequency, average absolute value and average amplitude; S42 is implemented by a high-order feature extraction unit;

[0125] S43, extracting posture features of the input signal; the posture features include but are not limited to acceleration, angular velocity, relative position and features in S42; when S43 is specifically implemented, it is realized by an inertial feature extraction unit;

[0126] S5. Screen and optimize the extracted multi-dimensional features to obtain data to be classified. In the specific implementation, this is achieved through a feature cross unit, which specifically includes the following sub-steps:

[0127] S51, the feature cross unit cross-combines the screened features to generate new composite features;

[0128] S52, using feature selection again to perform secondary screening on the feature set containing the composite feature to obtain a final optimized feature set;

[0129] The cross combination includes, but is not limited to, feature crossover based on factor decomposition machine, Kronecker product, and attention decomposition machine; in specific implementation: feature crossover is performed using Kronecker product;

[0130] The feature selection includes but is not limited to the NCFS method, the maximum correlation minimum redundancy method, the Relief algorithm, and the random forest algorithm. In specific implementation: the NCFS method is used for feature selection;

[0131] S6. Classify the data to be classified, which specifically includes the following sub-steps:

[0132] S61, inputting the optimized features into a pre-trained machine learning model;

[0133] S62, the machine learning model outputs classification results, including classification categories and confidence levels;

[0134] S63. If the confidence level is lower than the preset threshold, the system will mark the sample as "pending confirmation";

[0135] The machine learning model includes but is not limited to SVM, multi-layer perceptron model, decision tree, logistic regression and neural network; in specific implementation, SVM is used for classification.

[0136] In the specific implementation, the multi-channel surface muscle electrodes in the data acquisition module use Delsys Trigno wireless surface electromyography sensors, and the acquisition frequency is set to 600Hz; the nine-axis inertial measurement unit uses Xsens MTw Awinda inertial measurement unit, and the sampling frequency is set to 100Hz. The channel selection unit uses NI USB-6343 multi-function data acquisition card for signal selection and acquisition.

[0137] In the data preprocessing module, the bandpass filter unit adopts a 4th-order Butterworth filter with a frequency range of 20-500 Hz; the low-pass filter unit adopts a 4th-order Butterworth low-pass filter with a cutoff frequency of 20 Hz.

[0138] The autocorrelation check unit calculates the discrete cross-correlation of each channel of the signal; the calculation result is a symmetric sequence containing the correlation of positive delay and negative delay, the first 1 / 2 of the symmetric sequence is extracted to obtain the extracted sequence and the mean of the extracted sequence is calculated; it is determined whether the mean is less than 0.66 and greater than 0.54, if so, the channel is valid, otherwise the channel is deleted, that is, the channel data is not subjected to subsequent active segment detection.

[0139] The activity segment detection unit first uses the sliding window method to segment the data, with the window size set to 0.6s and the step size set to 0.2s; the slope change times (SSC) of each channel of the segmented data is calculated. If the SSC is greater than 2 and the sample entropy value is greater than 0.7, it indicates muscle activity, otherwise the channel is deleted.

[0140] In the envelope processing unit, the analytical envelope adopts the Hilbert transform method, the peak envelope adopts the sliding window method, and the window length is set to 100ms.

[0141] The static detection unit first uses the sliding window method to segment the data, which is consistent with the active segment detection unit. The window size is set to 0.6s and the step size is 0.2s. The acceleration mean of each channel of the segmented data is calculated. If the mean of each channel is less than 10.8g / ㎡ and greater than 8.8g / ㎡, this segment of data is marked as a static segment.

[0142] The normalization unit calculates the data by subtracting the mean of each channel and dividing by the standard deviation to convert the original data into a distribution with a mean of 0 and a standard deviation of 1.

[0143] The I / O control module is implemented using LabVIEW software programming, and dynamic control of feature selection is achieved through front panel controls. Figure 2As shown. In the specific implementation, click the panel front-order feature button, the software divides the data into equal lengths of 0.6s, calculates the root mean square, average absolute value, and variance of each segment of data after division and the data of the window size of 0.2s before 0.2s, and adds the result to the feature list; click the panel high-order feature button, the software calculates the cross entropy and classification dimension of the data and adds them to the feature list; click the panel fatigue feature button, the software calculates the kurtosis and median frequency of the data and adds them to the feature list; click feature extraction to complete.

[0144] In the feature crossover and selection module, the feature crossover unit uses the attention decomposition machine method to perform feature crossover, and the feature selection unit uses the random forest algorithm in the scikit-learn library to obtain the feature importance ranking, and takes the top N features as the features output to the classifier; the value range of N is 10 to 200; in specific implementation, N is 30.

[0145] The machine learning ADHD diagnosis unit uses the TensorFlow framework to build a multi-layer perceptron model. The hidden layer adopts a 3-layer structure, with 64, 32, and 16 neurons in each layer respectively. The activation function uses ReLU, and the output layer uses the Softmax function.

[0146] When the system is running, the surface electromyography signals and posture angle signals of the subjects are first collected when they perform standardized computer tasks. The data collection time is 15 minutes. After the collected raw data is preprocessed, the I / O control module selects the type of features to be extracted. The feature extraction module extracts the selected multi-dimensional features, which are optimized by the feature intersection and selection module and input into the machine learning ADHD diagnosis unit. The diagnosis unit outputs the diagnosis results and confidence of ADHD or non-ADHD, and the confidence threshold is set to 0.85.

[0147] In this embodiment, the system has a diagnostic accuracy of 88.0%, a sensitivity of 89.5%, and a specificity of 91.2% on a sample of 100 adult volunteers (50 ADHD patients and 50 healthy controls), and can provide objective auxiliary diagnosis results in a short time.

[0148] Table 1 shows the influence of the number of features selected by the feature cross unit on the classification performance.

[0149] Table 1 The influence of the number of selected features on the classification performance of feature cross unit

[0150]

[0151] It can be seen from Table 1 that the best effect is achieved when the number of features selected by the feature cross unit is 30.

[0152] Table 2 is a comparison of the effects of characteristic cross units.

[0153] Table 2 Comparison of the effects of characteristic cross units

[0154]

[0155] As shown in Table 2, after the feature crossover unit crosses the features and further selects the features, the auxiliary diagnosis effect of ADHD is significantly improved, with the accuracy increased by 3.5%, the sensitivity increased by 3.2%, and the specificity increased by 1.7%.

[0156] The above is only a preferred embodiment of the present invention, and the present invention should not be limited to the contents disclosed in the embodiment and the drawings. Any equivalent or modification completed without departing from the spirit disclosed in the present invention shall fall within the scope of protection of the present invention.

Claims

1. An I / O control module, characterized in that: The data preprocessing is received and the output is selected for feature extraction, including a mutually independent preamble and source feature selection unit, a high-order class feature selection unit and a posture selection unit; the preamble and source feature selection unit selects the peak envelope or / and the bandpass filter output data of the channel selected data for feature extraction; the high-order class feature selection unit selects the low-pass filter output data of the channel selected data or / and the output data of the channel selected data after bandpass filtering and autocorrelation test or / and the output data of the channel selected data after bandpass filtering, autocorrelation test and active segment detection or / and the output data of the channel selected data after low-pass filtering and static detection for feature extraction; the posture selection unit selects the output data of the channel selected data after low-pass filtering and / or static detection and / or normalization for feature extraction.

2. An I / O control module according to claim 1, characterized in that: The preamble and source feature selection unit selects data after channel selection and analog-to-digital conversion or data after channel selection, analog-to-digital conversion and bandpass filtering as data for extracting preamble features; and selects peak envelope as data for extracting source features; The peak envelope is obtained by channel selection, analog-to-digital conversion, bandpass filtering, autocorrelation inspection, active segment detection, and peak extraction.

3. An I / O control module according to claim 1, characterized in that: The high-order feature selection unit selects the low-pass filtered output data of the data after channel selection, the output data of the data after band-pass filtering, autocorrelation test and active segment detection, the output data of the data after channel selection, analog-to-digital conversion, low-pass filtering and static detection, and the output data of the data after analog-to-digital conversion, band-pass filtering, autocorrelation test and active segment detection as data for extracting dynamic and fatigue features; and selects the output data of the data after band-pass filtering, autocorrelation test and active segment detection, the analytical envelope, and the output data of the data after analog-to-digital conversion, band-pass filtering, autocorrelation test and active segment detection as data for extracting high-order features.

4. An I / O control module according to claim 3, characterized in that: The analytical envelope is obtained by channel selection, analog-to-digital conversion, bandpass filtering, autocorrelation test, active segment detection, and envelope analysis.

5. The I / O control module according to claim 1, characterized in that: The input data of the static detection is the output data after the channel selection is processed through low-pass filtering; the normalized input data is the output data after the channel selection is processed through low-pass filtering and then processed through static detection.

6. A classification system based on data verification and controllable feature selection, comprising a data acquisition module, a data preprocessing module, a feature extraction module, a feature cross unit and a machine learning classification module, wherein the output of the data acquisition module is sent to the data preprocessing module for data preprocessing; the feature extraction module receives the data preprocessing and then selects the feature extraction module to extract surface muscle electricity, high-order classes and inertial features; the extracted surface muscle electricity, high-order classes and inertial features are input into the feature cross unit for feature crossover and then output to the machine learning classification module for classification, characterized in that: It also includes an I / O control module, which is the I / O control module described in any one of claims 1 to 5, and the output of the data preprocessing module is the input of the I / O control module. The I / O control module receives the output of the data preprocessing module and selects data that can extract surface muscle electricity, high-order classes and inertial features to output to the feature extraction module.

7. The data preprocessing module in the classification system based on data inspection and controllable feature selection according to claim 6 is used to perform preprocessing operations on the collected original signal, including a bandpass filtering unit, a low-pass filtering unit, a stationary detection unit, an envelope processing unit and a normalization unit, characterized in that: It also includes an autocorrelation checking unit and an activity segment detecting unit for realizing data checking; The autocorrelation test unit removes data without obvious changes by calculating the autocorrelation of the surface electromyography data after the channel selection and bandpass filtering; the activity segment detection unit divides the data after the channel selection and bandpass filtering and autocorrelation test, and then retains the activity segment based on the sample entropy and the number of slope changes; The partitioning uses a sliding window method to segment the data, and the value range of the data segment is 40 to 160; the sliding step range is 10 to 60. The data acquisition module is used to collect surface electromyography and attitude angle analog signals, including multi-channel surface muscle electrodes, a channel selection unit, a nine-axis inertial measurement unit and an analog-to-digital conversion unit, and converts the analog signals into digital signals after channel and axis number selection; the channel selection unit is respectively connected to the multi-channel surface muscle electrodes and the nine-axis inertial measurement unit; is used to select the surface electromyography corresponding channel and the nine-axis inertial measurement unit and the axis number and output the surface electromyography and attitude angle analog signals; the analog-to-digital conversion unit is connected to the bandpass filtering unit and the low-pass filtering unit, and is used to convert the surface electromyography and attitude angle analog signals into digital signals; the envelope processing unit performs analytical envelope and peak envelope extraction; The feature extraction module includes a surface electromyography feature extraction unit, a fatigue feature extraction unit, a high-order class feature extraction unit and an inertia feature extraction unit, which are used to extract multi-dimensional features; the surface electromyography feature extraction unit is used for pre-order feature extraction and source feature extraction; the inertia feature extraction unit is used for posture feature extraction; the pre-order feature extraction and source feature extraction are connected to the pre-order and source feature selection unit in the I / O control module; the dynamic feature extraction unit and the high-order feature extraction unit are connected to the fatigue feature extraction unit at the same time; the inertia feature extraction unit is connected to the posture feature selection unit in the I / O control module; the feature cross unit is used to screen and combine features; the outputs of all feature extraction modules are connected to the feature cross unit; the machine learning classification module uses the cross-screened features to perform signal intelligent classification.

8. A classification method based on data verification and controllable feature selection, using the classification system based on data verification and controllable feature selection according to claim 6 or 7, characterized in that: The steps include: S1, collecting the surface electromyography signal and posture angle signal of the subject, and obtaining the surface electromyography and posture data through analog-to-digital conversion; S2, filtering, autocorrelation test, activity segment detection, static detection and envelope processing are performed on the data acquired in S1 to obtain preprocessed data; S3, selecting the feature type to be extracted according to the preset conditions; S4, extracting multi-dimensional features from the preprocessed data according to the preset conditions of S3; S5. Screen and optimize the extracted multi-dimensional features to obtain data to be classified; S6. Classify the data to be classified.

9. A classification method based on data verification and controllable feature selection according to claim 8, characterized in that: S2 The filtering of data is specifically: band-pass filtering and low-pass filtering are performed on the muscle electrical data and posture data after channel selection and analog-to-digital conversion respectively; the autocorrelation test is specifically: autocorrelation test is performed on the muscle electrical data after channel selection and analog-to-digital conversion, and the data output by the autocorrelation test is used for active segment detection; the active segment detection is specifically: active segment detection is performed on the data after autocorrelation test on the muscle electrical data after channel selection and analog-to-digital conversion, and the data output by the active segment detection is used to extract envelopes and extract high-order class features; the stillness detection is specifically: stillness detection is performed on the data after low-pass filtering on the posture data after channel selection and analog-to-digital conversion, and the data output by the stillness detection is used for high-order class feature extraction or normalized and then used for posture feature selection; the envelope processing is specifically: extracting analytical envelopes and peak envelopes from the data obtained by active segment detection after autocorrelation test on the muscle electrical data after channel selection and analog-to-digital conversion, and then using them for high-order class feature selection and pre-order feature selection respectively.

10. A classification method based on data verification and controllable feature selection according to claim 8, characterized in that: S3 describes selecting the feature type to be extracted according to preset conditions, specifically: selecting the low-pass filtered output data of the data after channel selection, the output data of the data after band-pass filtering, autocorrelation test and active segment detection, the output data of the data after channel selection, analog-to-digital conversion, low-pass filtering and static detection, and the output data of the data after analog-to-digital conversion, band-pass filtering, autocorrelation test and active segment detection as data for extracting dynamic and fatigue features; selecting the output data of the data after band-pass filtering, autocorrelation test and active segment detection, the analytical envelope, and the output data of the data after analog-to-digital conversion, band-pass filtering, autocorrelation test and active segment detection as data for extracting high-order features.

Images