A bionic hand control system and method based on binary tree support vector machine

Abstract

The application discloses a kind of bionic hand control system and method based on binary tree support vector machine, comprising: action training process and real-time control process;Wherein, the action training process further includes multi-action offline training and single action online training, by surface electromyogram acquisition, pre-processing and training, obtain the parameter matrix of training completion;The real-time control process loads the parameter matrix, for realizing gesture action recognition and bionic hand real-time control.The application supports multi-action offline training mode, single action online training mode, multi-action real-time recognition control mode, and single action real-time recognition control mode, can support the offline training of pre-defined multiple gesture actions, classification identification and real-time control on the basis of supporting, support single gesture action of amputee user online training, classification identification and real-time control.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of intelligent prostheses, and particularly relates to a bionic hand control system and method based on a binary tree support vector machine. BACKGROUND

[0002] An intelligent bionic hand is an important research direction in the field of intelligent prostheses, and is realized by relying on machine learning technology to classify and recognize surface electromyographic signals of amputation stumps and to control the motion of the bionic hand. The classification and recognition capability of the surface electromyographic signals will directly affect the control level and wearing experience of the intelligent bionic hand. In the development process of the intelligent bionic hand, the classification and recognition accuracy of the biological surface electromyographic signals and the overall operation complexity of the related machine learning methods need to be considered in depth, which will directly affect the quality and cost of the intelligent bionic hand. The support vector machine (SVM) machine learning algorithm often has higher recognition accuracy than other algorithms, such as the Bayesian algorithm and linear discriminant analysis. The SVM is mainly aimed at binary classification problems. For the multi-classification problem of the surface electromyographic signals, the SVM often adopts a one-class vs. another-class classification method. However, with the increase in the number of classes, the support vectors to be trained will increase sharply. The calculation complexity of the one-class vs. the rest-class classification method of the SVM is significantly reduced compared with the former method, but there is a serious imbalance problem between the training sample quantities of the two classes in the training process, which affects the classification and recognition accuracy. Therefore, there is an urgent need for a surface electromyographic signal classification and recognition method that can reduce the calculation complexity while ensuring the recognition accuracy. SUMMARY

[0003] The application aims to provide a bionic hand control system and method based on a binary tree support vector machine to solve the problems existing in the prior art.

[0004] To achieve the above-mentioned purpose, the application provides a bionic hand control system and method based on a binary tree support vector machine, which comprises a motion training process and a real-time control process.

[0005] The motion training process comprises multi-motion offline training and single-motion online training, and a trained multi-dimensional parameter matrix is obtained through surface electromyographic signal acquisition, preprocessing and training.

[0006] The real-time control process loads the multi-dimensional parameter matrix to realize gesture motion recognition and real-time control of the bionic hand.

[0007] Optionally, the multi-motion offline training comprises:

[0008] According to a plurality of predefined gesture actions, the surface electromyogram signals of the related gesture actions are collected; the surface electromyogram signals are preprocessed offline on a computer, and the binary tree support vector machine model is trained; after the training is completed, the multi-dimensional parameter matrix of the binary tree support vector machine model is input into the embedded control system platform of the bionic hand.

[0009] Optionally, before the surface electromyogram signals are collected, pre-collection is required, and the pre-collection process is as follows: the surface electromyogram signal waveform is observed, and after each motion waveform is relatively stable, formal collection is entered.

[0010] Optionally, the surface electromyogram signals are collected based on sensors, the sampling frequencies of the sensors are all set to 1000 Hz, the sampling times are kept synchronous, and the sample data amounts obtained by each sensor are the same.

[0011] Optionally, the preprocessing process of the surface electromyogram signals is as follows: the surface electromyogram signals of the related gestures are intercepted, the sliding time window is set to 200 ms, the overlap parameter is 50%, the following four time domain features are extracted from the data collected by the eight surface electromyogram sensors: average absolute value, slope sign change, zero-crossing rate, and waveform length, and the feature data extracted by the eight sensors are combined into a 32-dimensional feature data matrix, which is used for subsequent training of the binary tree support vector machine model.

[0012] Optionally, the binary tree support vector machine model creation process is as follows:

[0013] According to expert prior knowledge, the difficulty of recognizing the gesture actions based on the surface electromyogram signals is determined, and based on the traversal idea of the binary tree, the gesture actions are combined and grouped according to the gesture recognition difficulty, so that each time the related gesture actions are divided into two groups, and the gesture actions are divided into different leaf nodes in stages until each type of gesture action is divided into different leaf nodes, wherein the most easily recognizable group is located at the root node of the binary tree; the division result is used for training of the binary tree support vector machine model.

[0014] Optionally, the binary tree support vector machine model training process is as follows: the offline training of multiple actions is performed in the binary tree support vector machine model, the offline training of multiple actions is performed; the multi-dimensional feature data matrix of the related gesture actions is used to train the two groups of gesture actions under each root node in turn, the corresponding support vector machine model parameters are obtained, and the support vector machine model parameters obtained by training each root node are saved; after each root node is trained, the support vector machine model parameters corresponding to each node are combined into a multi-dimensional parameter matrix; the multi-dimensional parameter matrix is input into the embedded control system platform of the bionic hand, which is used for subsequent gesture action recognition and real-time control of the bionic hand.

[0015] Optionally, the single action online training includes:

[0016] Carry out the collection of the surface myoelectric signal of the self-defined gesture action, the embedded control platform completes the pre-processing of the surface myoelectric signal and the training of the binary tree support vector machine model online, and the parameter matrix of the model is stored to the embedded control system platform of the bionic hand after the training is completed;

[0017] Wherein, the surface myoelectric signal collection and pre-processing work in the single action online training is same as the multi-action offline training related work.

[0018] Optionally, the real-time control process specifically includes:

[0019] The parameter matrix of the SVM model trained and the driving parameter of the bionic hand finger steering engine of the related gesture action are carried out the real-time collection and processing of the arm surface myoelectric signal, the recognition of the gesture action and the real-time control of the bionic hand.

[0020] The application also provides a bionic hand control system based on binary tree support vector machine, the control system includes support vector machine model training module and bionic hand real-time control module, the support vector machine model training module and the bionic hand real-time control module are connected through model parameter matrix;

[0021] The support vector machine model training module is used for supporting multi-action offline training mode and single action online training mode through the collection of the arm surface myoelectric signal, the pre-processing of the surface myoelectric signal, the creation of the binary tree support vector machine model, the training of the binary tree support vector machine model and the storage of the model parameter matrix.

[0022] The bionic hand real-time control module is used for supporting multi-action real-time recognition control mode and single action real-time recognition control mode through the loading of the SVM model parameter, the real-time collection and processing of the surface myoelectric signal, the recognition of the gesture action and the real-time control of the bionic hand.

[0023] The technical effect of the application is:

[0024] The application supports multi-action offline training mode, single action online training mode, multi-action real-time recognition control mode and single action real-time recognition control mode, can support the online training, classification identification and real-time control of the single gesture action defined by the amputated user on the basis of supporting the offline training, classification identification and real-time control of the multiple gesture actions defined in advance. BRIEF DESCRIPTION OF DRAWINGS

[0025] The drawings that constitute a part of this application are used to provide further understanding of the application, the illustrative embodiment of the application and its description are used to explain the application, and do not constitute improper limitation to the application.

[0026] Figure 1This is a schematic diagram of the architecture of a bionic hand control system based on a binary tree support vector machine in an embodiment of the present invention;

[0027] Figure 2 This is a basic flowchart of a bionic hand control method based on a binary tree support vector machine in an embodiment of the present invention;

[0028] Figure 3 This is a schematic diagram of a binary tree support vector machine model for a bionic hand control method based on a binary tree support vector machine in an embodiment of the present invention. Detailed Implementation

[0029] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0030] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0031] Example 1

[0032] like Figure 1 As shown, this embodiment provides a bionic hand control system based on binary tree support vector machine, including: an SVM model training module and a bionic hand real-time control module, wherein the SVM model training module supports multi-action offline training mode and single-action online training mode, and the bionic hand real-time control module supports multi-action real-time recognition control mode and single-action real-time recognition control mode.

[0033] The SVM model training module supports two modes: offline training mode for multiple actions and online training mode for single actions. It mainly performs the acquisition of surface electromyography (EMG) signals of the arm, preprocessing of surface EMG signals, creation of binary tree support vector machine (SVM) models, training of binary tree SVM models, and storage of model parameter matrices.

[0034] The bionic hand real-time control module supports two modes: multi-action real-time recognition and control mode and single-action real-time recognition and control mode. It mainly performs SVM model parameter loading, real-time acquisition and processing of surface electromyography signals, gesture recognition, and real-time control of the bionic hand.

[0035] Example 2

[0036] Based on the above embodiments, the present invention also discloses a bionic hand control method based on a binary tree support vector machine, such as... Figure 2 As shown, it includes:

[0037] Step 1: Multi-action offline training. Based on a variety of predefined hand gestures, collect surface electromyography (EMG) signals related to the gestures, and perform offline preprocessing of the EMG signals and training of the binary tree support vector machine model on the computer. After training, download the multidimensional parameter matrix of the model to the bionic hand embedded control system platform.

[0038] Step 2: Single-action online training. Amputees can activate the single-action training mode of the embedded control platform according to their actual needs and collect surface electromyography (EMG) signals for customized gestures. The embedded control platform completes the preprocessing of surface EMG signals and the training of the SVM model online. After training, the parameter matrix of the model is stored in the bionic hand embedded control system platform.

[0039] Step 3: Real-time control. Based on the working mode set by the amputee user, the trained SVM model parameter matrix and the bionic hand finger servo drive parameters for relevant gestures are loaded. Real-time acquisition and processing of electromyographic signals on the arm surface, recognition of gestures, and real-time control of the bionic hand are then performed.

[0040] Further, step 1 includes:

[0041] Step 1.1: Surface Electromyography (EMG) Signal Acquisition. Based on the physiological distribution of arm muscles, eight EMG sensors are evenly arranged along the circumference of the arm to collect EMG signals from the amputee user under different hand gestures. The sampling frequency of all sensors is set to 1000Hz, and the sampling time is synchronized, ensuring that each sensor obtains the same amount of sample data. The collected gestures mainly include the following five active gestures: wrist adduction, wrist abduction, OK gesture, grasping a cylinder, three-finger pinch, and one relaxed gesture. Before formally collecting EMG signal data, pre-collection is performed to observe the waveform of the EMG signal. Formal collection only begins after the waveform of each gesture has stabilized. During the formal EMG signal collection, the amputee user imagines that the amputated area is intact and imitates the hand gestures shown on the screen according to the prompts. Each gesture lasts for 2 seconds and is repeated 5 times. The system sequentially records the sample data of different gestures and labels the data.

[0042] Step 1.2: Preprocessing of surface electromyography (SEMG) signals. SEMG signals related to the gestures were extracted, with a sliding time window of 200ms and an overlap parameter of 50%. The following four temporal features were extracted from the data collected by the eight SEMG sensors: Mean Absolute Value (MAV), Slope Sign Change (SSC), Zero Crossing Rate (ZC), and Waveform Length (WL). The feature data extracted from the eight sensors were then merged into a 32-dimensional feature data matrix for subsequent SVM model training.

[0043] Step 1.3: Binary Tree Support Vector Machine Model Creation. Based on expert prior knowledge, the difficulty of recognizing the hand gestures based on surface electromyography (EMG) signals is assessed. Using the binary tree traversal approach, gestures are grouped according to recognition difficulty, with each gesture divided into two groups, progressively increasing until each category is assigned to a different leaf node. The easiest group to recognize is located at the root node of the binary tree. Based on the division results, a binary tree support vector machine model is created for subsequent training. The binary tree support vector machine model is arranged according to a binary tree model, with each root node having two child nodes. Each child node corresponds to a category, and each root node corresponds to a set of support vectors determining the category of its child nodes. The binary tree support vector machine model is as follows: Figure 3 As shown;

[0044] Step 1.4: Binary Tree Support Vector Machine Model Training. On the computer, perform offline training for multiple actions based on the created binary tree support vector machine model. Using the 32-dimensional feature data matrix of relevant gesture actions, train the two sets of gesture actions under each root node sequentially to obtain the corresponding SVM model parameters, and save the SVM model parameters obtained from training each root node. After all root nodes have been trained, merge the SVM model parameters corresponding to each node into a multi-dimensional parameter matrix. Download the multi-dimensional parameter matrix to the bionic hand embedded control system platform for subsequent gesture recognition and real-time control of the bionic hand.

[0045] Further, step 2 includes:

[0046] Step 2.1: Acquisition of surface electromyography (EMG) signals. The preparatory work is the same as in Step 1.1. After deploying and setting up the 8 surface EMG sensors, the user starts the single-action online training mode. Following the prompts on the embedded control system's LCD interface, the user acquires surface EMG signals for a single action. The amputee user imagines the amputated limb is intact and performs the desired customized gesture for 2 seconds, then relaxes for 2 seconds. This action is repeated 5 times. The acquired data is directly stored in the embedded control system platform, with the activated and relaxed data labeled "1" and "-1" respectively, for subsequent online training of the SVM model.

[0047] Step 2.2: Online preprocessing of surface electromyography (SEMG) signals. The online preprocessing is basically the same as in Step 1.2, and the relevant work is completed on the embedded control system platform. The main tasks are to extract the SEMG signals of the customized gesture to be recognized, set a sliding time window, extract the temporal features of the data collected by the 8 SEMG sensors, and construct a 32-dimensional feature data matrix of the customized gesture for subsequent online training of the SVM model.

[0048] Step 2.3: Online training of the SVM model. On the embedded control system platform, the SVM model is trained online using a 32-dimensional feature data matrix of customized hand gestures, and the corresponding SVM model parameter matrix is ​​saved for subsequent recognition of single actions and real-time control of the bionic hand;

[0049] Furthermore, step 3 includes:

[0050] Step 3.1: Load Model Parameters. Based on the working mode set by the user on the embedded control system platform, load the corresponding SVM model parameters, and at the same time load the bionic hand finger servo drive parameters for the relevant gesture actions; if set to multi-action mode, load the multi-dimensional parameter matrix trained in step 1.4; if set to single-action mode, load the parameter matrix trained in step 2.3.

[0051] Step 3.2: Real-time acquisition and processing of surface electromyography (EMG) signals. After the eight surface EMG sensors are deployed in the forearm, they sample at a frequency of 1000Hz. The embedded control system extracts four temporal features from the eight sensors in real time using a 200ms sliding time window with 50% overlap, and obtains a 32-dimensional feature data matrix for subsequent gesture recognition and real-time control of the bionic hand.

[0052] Step 3.3: Gesture recognition and real-time control of the bionic hand.

[0053] The 32-dimensional feature data matrix obtained in real time is input into the trained SVM model to calculate and identify the gesture action category label; the motion mode of the bionic hand servo is determined based on the label to realize real-time control of the bionic hand.

[0054] In summary, this invention addresses the problem that existing SVM-based one-class vs. another classification methods suffer from excessive computational complexity, making them difficult to apply to embedded intelligent prosthetic limb control systems. This invention proposes a bionic hand control system and method based on a binary tree support vector machine. By adding expert prior knowledge, the difficulty of recognizing the hand gesture based on surface electromyography (EMG) signals is assessed, and a binary tree-based SVM model is constructed. This significantly reduces computational complexity while maintaining the accuracy of EMG signal classification.

[0055] The above description is merely a preferred embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A bionic hand control method based on binary tree support vector machine, characterized in that, This includes the motion training process and the real-time control process; The exercise training process includes offline training of multiple exercises and online training of a single exercise. By collecting, preprocessing and training surface electromyography signals, a multidimensional parameter matrix of the completed training is obtained. The real-time control process loads the multi-dimensional parameter matrix to realize gesture recognition and real-time control of the bionic hand; The training process of the binary tree support vector machine model is as follows: Offline training with multiple actions is performed within the binary tree support vector machine model; using the multi-dimensional feature data matrix of relevant gesture actions, the two sets of gesture actions under each root node are trained sequentially to obtain the corresponding support vector machine model parameters, and the support vector machine model parameters obtained from the training of each root node are saved; after all root nodes have been trained, the support vector machine model parameters corresponding to each node are merged into a multi-dimensional parameter matrix; the multi-dimensional parameter matrix is ​​input into the bionic hand embedded control system platform for subsequent gesture recognition and real-time control of the bionic hand; The single-action online training includes: acquiring surface electromyography (EMG) signals for customized gestures; preprocessing the EMG signals online using an embedded control platform; training a binary tree support vector machine model; and storing the model's parameter matrix in the bionic hand embedded control system platform after training. The signal acquisition and preprocessing in the single-action online training are the same as those in the multi-action offline training.

2. The bionic hand control method based on binary tree support vector machine according to claim 1, characterized in that, The multi-action offline training includes: Based on a variety of predefined hand gestures, the surface electromyography (EMG) signals of the relevant hand gestures are collected. The EMG signals are preprocessed offline on the computer, and the binary tree support vector machine model is trained. After training, the multidimensional parameter matrix of the binary tree support vector machine model is input into the bionic hand embedded control system platform.

3. The bionic hand control method based on binary tree support vector machine according to claim 2, characterized in that, Before acquiring the surface electromyography (EMG) signal, a pre-acquisition process is required. The pre-acquisition process involves observing the surface EMG signal waveform and waiting for the waveform of each movement to stabilize before proceeding with the formal acquisition.

4. The bionic hand control method based on binary tree support vector machine according to claim 2, characterized in that, The surface electromyography (EMG) signals are acquired based on sensors. The sampling frequency of each sensor is set to 1000Hz, the sampling time is kept synchronized, and the amount of sample data obtained by each sensor is the same.

5. The bionic hand control method based on binary tree support vector machine according to claim 2, characterized in that, The preprocessing of the surface electromyography (SEMG) signal is as follows: SEMG signals of relevant gestures are extracted, the sliding time window is set to 200ms, the overlap parameter is 50%, and the following four time-domain features are extracted from the data collected by the eight SEMG sensors: mean absolute value, slope sign change, zero-crossing rate, and waveform length. The feature data extracted from the eight sensors are then merged into a 32-dimensional feature data matrix for subsequent training of the binary tree support vector machine model.

6. The bionic hand control method based on binary tree support vector machine according to claim 2, characterized in that, The process of creating the binary tree support vector machine model is as follows: Based on expert prior knowledge, the difficulty of recognizing the hand gestures based on surface electromyography signals is determined. Based on the idea of ​​binary tree traversal, the hand gestures are grouped according to their recognition difficulty. The relevant hand gestures are divided into two groups each time, and the division is carried out step by step until each type of hand gesture is assigned to a different leaf node. The group that is easiest to recognize is located at the root node of the binary tree. The division results are used to train the binary tree support vector machine model.

7. The bionic hand control method based on binary tree support vector machine according to claim 1, characterized in that, The real-time control process specifically includes: The trained SVM model parameter matrix and related hand servo drive parameters are obtained, and real-time acquisition and processing of electromyographic signals on the arm surface, hand gesture recognition, and real-time control of the bionic hand are performed.

8. A bionic hand control system based on a binary tree support vector machine, characterized in that, For implementing the method as described in claim 1, the control system includes a support vector machine model training module and a bionic hand real-time control module, wherein the support vector machine model training module and the bionic hand real-time control module are connected via a model parameter matrix. The support vector machine model training module supports multi-action offline training mode and single-action online training mode through the acquisition of surface electromyography signals of the arm, the preprocessing of surface electromyography signals, the construction of binary tree support vector machine model, the training of binary tree support vector machine model, and the storage of model parameter matrix. The bionic hand real-time control module supports multi-action real-time recognition and control modes and single-action real-time recognition and control modes by loading binary tree support vector machine model parameters, real-time acquisition and processing of surface electromyography signals, recognition of gestures, and real-time control of the bionic hand.

Images