An active somatosensory brain-computer interface system based on tactile mapping to extend the input dimension of a single limb.

Abstract

This invention discloses an active somatosensory brain-computer interface system that expands the input dimension of a single limb based on tactile mapping. The system includes a tactile stimulation module, an EEG signal acquisition module, and a feature decoding module. During the stimulation phase, the tactile stimulation module provides localized tactile vibration stimulation to designated areas of the fingers and wrist of the subject's single hand, establishing a mapping relationship between tactile location sequences and discrete control commands. During the acquisition phase, the EEG signal acquisition module records and preprocesses the single-trial EEG signals induced when the designated areas receive tactile vibration stimulation. During the decoding phase, the feature decoding module identifies the preprocessed EEG signals, extracts and classifies the spatiotemporal features to decode the tactile response of specific areas. This invention can significantly increase the number of commands for a single limb, enabling the generation of multiple input states for a single limb, and breaking through the bottleneck of traditional brain-computer interfaces limited to passive somatosensory or large joint control.

Description

Technical Field

[0001] This invention relates to the fields of brain-computer interface and neural engineering technology, and in particular to an active somatosensory brain-computer interface system based on tactile mapping to extend the input dimension of a single limb. Background Technology

[0002] In current human-computer interaction and brain-computer interface (BCI) systems, establishing efficient and high-dimensional control command channels has always been a research focus. However, traditional control systems mostly have limitations: for example, systems that rely on large joint movements or fine finger movements directly occupy the human body's motor channels; while brain-computer interfaces that rely on steady-state visual evoked potentials (SSVEP) severely occupy the visual channels and are prone to causing visual fatigue; and paradigms that rely on motor imagery often suffer from high latency and high cognitive load.

[0003] In many practical applications (such as multitasking, driving, or rehabilitation scenarios involving limited function on one side of the body), the user's visual and motor channels are often occupied or restricted. Therefore, how to effectively increase the input dimension and control bandwidth of one side of the body without occupying the motor and visual channels has become a pressing technical challenge.

[0004] Current technologies lack an effective method for converting non-invasive EEG signals into active, high-dimensional control commands through non-intrusive sensory channels. To address this, researchers have widely adopted tactile stimulation as an auxiliary method. Tactile stimulation not only enhances the distinguishability of neural patterns but also effectively alleviates the "BCI illiteracy" phenomenon in some users, improving system calibration efficiency. In current closed-loop somatosensory brain-computer interface research, combining motion decoding with tactile feedback has become crucial for improving the naturalness and precision of prosthetic control. Current mainstream research typically uses external vibration motors to induce somatosensory evoked potentials, which are then used as control commands or feedback signals. Examples include the literature "Cortical responses totactile imagery: a high-density EEG study of the μ-rhythm event-related desynchronization and somatosensory evoked potentials" and "SensoryStimulation Training for BCI System Based on Somatosensory AttentionalOrientation".

[0005] However, existing research on tactile brain-computer interfaces mainly focuses on coarse somatosensory decoding or large joint movement recognition, such as distinguishing activation patterns at the wrist or whole hand level. Due to the low spatial resolution of electroencephalography (EEG), the high similarity of neural responses between different fingers, and its susceptibility to environmental noise, it remains unclear whether non-invasive EEG signals contain sufficient discriminative information to reliably decode single-finger-level tactile responses. This lack of research on finger-level tactile decoding limits the further application of somatosensory information in the expansion of refined commands and high-resolution brain-computer interface systems. Summary of the Invention

[0006] This invention provides an active somatosensory brain-computer interface system based on tactile mapping to expand the input dimension of a single limb. It does not occupy the motor and visual channels and, without increasing the natural freedom of movement, enables the system to have decodeable, discriminative, and stable EEG spatial features. This significantly increases the number of single-limb commands and enables the generation of multiple input states for a single limb, breaking through the bottleneck of traditional brain-computer interfaces that are limited to passive somatosensory or large joint control.

[0007] An active somatosensory brain-computer interface system based on tactile mapping to extend the input dimension of a single limb includes a tactile stimulation module, an EEG signal acquisition module, and a feature decoding module. During the stimulation phase, the tactile stimulation module provides localized tactile vibration stimulation to designated areas of the fingers and wrist on one side of the subject's hand without occupying the visual and motor channels. During the stimulation process, the subject needs to focus their attention on the vibration stimulation and identify the location of the tactile vibration stimulation in their mind, thereby establishing a mapping relationship between the tactile location sequence and discrete control commands. During the acquisition phase, the EEG signal acquisition module is used to record and preprocess the EEG signals induced by tactile vibration stimulation at designated points on one side of the fingers and wrist in real time. In the decoding stage, the feature decoding module is used to identify the preprocessed EEG signal, extract and classify the spatiotemporal features in the EEG signal, so as to decode the tactile response of a specific part.

[0008] Furthermore, the tactile stimulation module delivers stimulation via a tactile glove equipped with a linear resonant actuator motor.

[0009] Furthermore, the specific finger and wrist points on one side of the hand are the tips of the thumb, index finger, middle finger, ring finger, little finger, and wrist.

[0010] Furthermore, in the stimulation phase, the procedure for each trial includes: first, providing visual cues to indicate the start of tactile stimulation, followed by applying vibrational stimulation for 4 seconds at a single designated site; during these 4 seconds of vibrational stimulation, the subject must focus their attention on the vibrational stimulation and identify the location of the tactile vibrational stimulation in their mind; finally, there is a 1-second relaxation period and random 0 to 1 second intervals between trials.

[0011] Furthermore, the EEG signal acquisition module is used to acquire 64-channel EEG data of the subject under tactile stimulation, with the sampling rate set at 512Hz and AFz grounding and binaural average reference.

[0012] Furthermore, the preprocessing of EEG signals includes: A 48-52Hz notch filter was applied to suppress power frequency noise, and a 0.1-60Hz bandpass filter was applied. Independent component analysis and ICLabel were used to automatically remove electrooculography and electromyography artifacts, and signal segments were extracted from 0 to 4 seconds relative to the start of tactile stimulation.

[0013] Preferably, the feature decoding module includes a filter bank co-space pattern combined with a support vector machine FBCSP-SVM, as well as deep learning models EEGNet and IFNet; the three models are used to learn stable spatiotemporal features of sensorimotor related intentions and support six-class active tactile EEG signal decoding.

[0014] The operation and intervention process of this system strictly follows a scientific experimental paradigm, and is divided into a preparation phase, a tactile stimulation induction phase, and a data processing and evaluation phase, as detailed below: During the preparation phase, participants (such as BCI beginners) sit comfortably with their right arm placed horizontally on the support, and the system performs individualized calibration of the tactile gloves. At the start of each trial, a fixed cross is displayed on the screen, accompanied by an auditory cue at 500Hz for 200ms, guiding the participant into the preparation state.

[0015] During the stimulation phase, the control module selects one of six sites (thumb-D1, index finger-D2, middle finger-D3, ring finger-D4, little finger-D5, and wrist W1) in a pseudo-random order for stimulation. Visual cues are synchronized with a 4-second vibrational stimulus. After stimulation, a 1-second relaxation period and a random interval of 0-1 seconds are provided to reduce sensory fatigue and adaptation.

[0016] During the acquisition phase, signal preprocessing was first performed, applying a 48-52Hz notch filter to suppress power frequency noise and a 0.1-60Hz bandpass filter. Independent component analysis (ICA) and ICLabel were used to automatically remove electrooculography (EOG) and electromyography (EMG) artifacts. Then, feature extraction was performed, and event-related spectral perturbation (ERSP) was calculated to quantify changes in non-phase-locked oscillatory activity. Analysis revealed that fingertip stimulation induced significant and persistent event-related desynchronization (ERD) modes in the Alpha (8-13Hz) and Beta (13-30Hz) frequency bands.

[0017] During the decoding phase, the system provides three representative decoding models, including the traditional FBCSP-SVM and the deep learning models EEGNet and IFNet. Among them, IFNet achieves the highest average accuracy (57.3%) for six classifications by modeling cross-frequency interactions through interactive frequency convolution.

[0018] Preferably, the feature decoding module further includes: executing an optimized subclassification strategy to solidify the most stable discrete control instruction channel for the physiological characteristics of different subjects.

[0019] Furthermore, the optimized sub-classification strategies include binary and tri-classification combinations. Binary combinations include thumb-little finger and thumb-wrist; tri-classification combinations include thumb-middle finger-little finger and thumb-little finger-wrist. These optimized sub-combinations further improve decoding accuracy, achieving 86.6±7.3% in binary classification tasks. Interpretability analysis was conducted through… Pattern and model attribution map analysis confirmed that the classification features were mainly concentrated on the CP region electrodes corresponding to the somatosensory cortex, ensuring the physiological significance of the decoding results.

[0020] Compared with the prior art, the present invention has the following beneficial effects: This invention, without occupying the body's movement and visual channels, extracts highly discriminative and stable EEG spatial features at the single-trial level by applying local tactile stimulation to specific spatial locations of a unilateral limb and combining it with the user's active intention mapping. This successfully fills the gap in the field of non-motor multidimensional input control for unilateral limbs. The system is highly adaptable to beginners; users do not need to endure the visual fatigue easily caused by traditional paradigms, nor do they require complex long-term training. They can quickly establish a stable mapping between tactile location sequences and discrete control commands, achieving classification performance far exceeding random probability. By cleverly utilizing the finger-specific EEG patterns without increasing the natural degrees of freedom of movement, it significantly expands the input dimensions and control command types of unilateral limbs. This not only lays the core technological foundation for building high-dimensional active interaction systems that operate in parallel multitasking, but also demonstrates enormous application potential in areas such as multi-channel input in confined spaces, complex working environments (such as multi-threaded operations), neural prosthesis control, motor rehabilitation, and dexterous human-computer interaction. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a schematic diagram of the experimental scenario and the structure of the tactile glove device in an embodiment of the present invention.

[0023] Figure 2 This is a schematic diagram of the experimental paradigm design and single trial process of an embodiment of the present invention.

[0024] Figure 3 This is a diagram showing the spatiotemporal dynamic characteristics analysis results of tactile-evoked EEG in an embodiment of the present invention.

[0025] Figure 4 This is a graph showing the performance evaluation results of six-class decoding of vibration tactile stimulation according to an embodiment of the present invention.

[0026] Figure 5 This is a confusion matrix diagram for six-class decoding of vibration tactile stimulation in an embodiment of the present invention.

[0027] Figure 6 This is a diagram showing the results of the optimal combination identification and neural interpretability analysis in an embodiment of the present invention. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] It should be noted that, unless otherwise specified, the features in the following embodiments and implementation methods can be combined with each other.

[0030] To verify the effectiveness of the designed non-invasive tactile brain-computer interface paradigm and multi-classification decoding system, this invention designs experiments and collects data based on the following assumptions: If the system can perform high-resolution, fine-grained tactile evoked EEG signal decoding, then when the human body receives independent mechanical vibration tactile stimulation in different fingers and wrists of one side of the hand, the neurophysiological responses evoked in the sensorimotor cortex of the brain should have sufficient separability and stable spatial representation patterns. Traditional brain-computer interface systems are mostly limited to large joint control or coarse somatosensory feedback, or rely on motor imagery paradigms with high latency and steady-state visual evoked paradigms that easily induce visual fatigue. Therefore, this invention systematically proposes an active somatosensory brain-computer interface system based on tactile mapping to extend the unilateral limb input dimension.

[0031] This invention conducted a rigorous tactile sensory evoked experiment on 13 healthy participants with normal vision and no history of neurological or psychiatric illness. Inclusion criteria included individuals whose tactile perception was confirmed by professional assessment to be normal and who were right-handed. Participants wore specially designed tactile stimulation gloves and multi-lead EEG caps, and completed multiple rounds of tactile sensory tasks in a soundproof room with strict external interference shielding. Before the experiment officially began, the experimental procedure was explained in detail to the participants, and each participant underwent personalized glove calibration and tactile intensity testing to ensure that each stimulation site produced clear and comfortable tactile feedback. This resulted in targeted and high signal-to-noise ratio EEG characteristics, better ensuring the participants' concentration during the extended experiment and guaranteeing the quality of data collection.

[0032] In addition, this invention also incorporates a variety of classic machine learning and cutting-edge deep learning decoding algorithms to extract and verify the spatiotemporal features of tactile EEG in a single trial. Through data analysis of multi-channel EEG signals, time-frequency dynamic response, and classification accuracy during the intervention period, the effectiveness and reliability of this finger-level fine tactile brain-computer interface system can be fully demonstrated.

[0033] The present invention discloses an active somatosensory brain-computer interface system based on tactile mapping to extend the input dimension of a single limb. The specific implementation and data acquisition process is as follows: First, the experimental environment and hardware equipment are set up. The experiment is conducted in a quiet laboratory with good electromagnetic shielding. The subject sits upright in front of the screen with his right arm horizontally supported on a custom-made armrest to ensure muscle relaxation. Figure 1 This is a schematic diagram of the experimental scenario and tactile glove device structure of an embodiment of the present invention. As shown in the figure, the present invention uses an integrated tactile actuation glove. The glove has six miniature linear resonant actuator motors embedded in the fingertips of the five independent fingers on the right side (i.e., thumb-D1, index finger-D2, middle finger-D3, ring finger-D4, and little finger-D5) and the dorsal side of the wrist W1. The physical deformation generated by these motors precisely activates the deep mechanoreceptors in the skin, converting them into neural electrical signals transmitted to the brain, providing physiological support for subsequent precise EEG decoding. In terms of stimulation parameter settings, the vibration frequency is strictly locked at 170 Hz, and the amplitude at each point remains constant.

[0034] Synchronous acquisition of EEG signals relies on a 64-channel high-density EEG system, with electrode arrangement conforming to the internationally extended 10-20 standard. The signal is referenced by the average value of the bilateral earlobe electrodes, grounded at AFz, with a sampling frequency of 512 Hz, and electrode impedance strictly controlled below the threshold to ensure the acquisition quality of weak EEG signals.

[0035] Once the hardware platform is ready, this invention designs a rigorous experimental paradigm for tactile sensation to induce high-quality single-trial EEG responses. Figure 2This is a schematic diagram of the experimental paradigm design and single trial process of an embodiment of the present invention. In this experimental paradigm, the complete experiment contains 24 identical test blocks, each block containing 30 pseudo-randomized single trials. This means that the six independent stimulation sites will be randomly triggered 5 times in each block, so that each subject can accumulate 120 high-quality repetitive stimulation data for each specific stimulation site throughout the entire experimental period. The specific timeline design of a single trial is as follows: Two seconds before the start of each trial, the screen prompts relaxation, and the subject keeps his whole body relaxed and looks at the center of the screen; when the time reaches -1 second, a white fixation point appears in the center of the screen, accompanied by a short prompt sound lasting 200 milliseconds, as a warning signal to prompt the subject to concentrate highly; when the time advances to 0 seconds, the screen visual cue is updated, indicating that the tactile stimulation has officially begun. At this time, the tactile glove triggers continuous vibration tactile stimulation for 4 seconds at a designated site on the subject's right hand according to a preset sequence. During this 4-second task phase, participants need to focus all their attention on the vibration sensation and try to clearly identify its specific location in their minds. After the stimulus ends, the system provides a fixed 1-second rest period, allowing participants to blink naturally, followed by a trial interval of random duration ranging from 0 to 1 second. This design of introducing random duration effectively prevents participants from developing neural adaptive fatigue and ensures the independence of the evoked signal for each trial.

[0036] To address the massive amount of raw EEG data acquired in the experiment, this invention implemented a sophisticated offline preprocessing and spatiotemporal feature extraction computational process. Data preprocessing first involved removing power frequency noise using a 48-52 Hz notch filter, retaining core frequency information from 0.1 to 60 Hz using a bandpass filter, and re-referencing to the bilateral earlobe average potentials. Subsequently, independent component analysis was introduced, combined with automated tools to automatically screen and remove artifacts related to electrooculography (EOG) and electromyography (EMG) with a probability threshold of 0.9. Continuous data was segmented into segments from 2 seconds before stimulation to 4 seconds after stimulation, and any segment with an absolute amplitude exceeding 150 microvolts was discarded.

[0037] Figure 3This figure shows the spatiotemporal dynamic characteristics analysis results of tactile-evoked EEG in an embodiment of the present invention. To characterize non-phase-locked oscillatory changes, event-related spectral perturbations were calculated. Energy changes within the core window of 0.5 to 3.5 seconds after stimulation were extracted and normalized to obtain event-related desynchronization indices. The results show that the mean data from all participants reveal the dynamic evolution of the C3 channel. Significant and persistent alpha and beta frequency power desynchronization occurred in the contralateral sensorimotor cortex after tactile stimulation was initiated. Strong alpha frequency desynchronization spread to the prefrontal cortex, suggesting the involvement of higher somatosensory networks. Furthermore, whole-brain spatial topology confirmed that stimulation of different fingers induced spatially specific desynchronization patches, while the activation range at the wrist was relatively weak.

[0038] To automate and efficiently identify various fine tactile evoked EEG patterns, this invention integrates three representative brain-computer interface decoding algorithms within its intervention and decoding framework. These include a traditional classification method combining filter bank co-space patterns with support vector machines, and two lightweight convolutional neural network models, EEGNet and IFNet, specifically designed for EEG signals. These deep models can automatically learn complex spatiotemporal representations related to the sensorimotor cortex through cascaded spatiotemporal convolutional kernels. For each subject, 80% of the data in a single category is used as the training set, and the remaining 20% ​​is used as an independent test set to evaluate generalization performance.

[0039] Figure 4 This is a graph showing the performance evaluation results of six-class decoding under vibration-tactile stimulation according to an embodiment of the present invention. The bar chart clearly shows that the average classification accuracy of all three methods significantly exceeded the level of random guessing on the subjects, proving that the tactile neural representations of different fingers and wrists are sufficiently separable in a single trial. Among them, the deep learning model performed better, with IFNet achieving the highest average accuracy. Figure 5 This is a confusion matrix diagram for six-class decoding of vibrational tactile stimulation according to an embodiment of the present invention. The confusion matrix given for the best subjects and the average level of the group shows a clear diagonal clustering trend, with the vast majority of responses being correctly classified. The few misclassifications are mainly concentrated between anatomically adjacent fingers, while the recognition rate of the thumb and little finger is extremely high. This is highly consistent with the spatial arrangement topology of hand representations in the dwarfism map of the brain's sensory cortex.

[0040] Given the complexity of the six-class classification task, this study systematically analyzed the spatial discrimination and decoding performance of finger and wrist tactile brain-electronic combinations by using optimal identification at the individual level and consistent statistical dimensions at the group level. In the individual-level assessment, all possible binary and tri-class stimulus site arrangements were explored for each subject. Table 1 shows the accuracy of the optimal sub-combination: the accuracy of each subject in the binary and tri-class tasks within the optimal sub-combination. As shown in the table, in the individually customized binary classification task, the average accuracy of the optimal combination reached 86.6%, while in the tri-class task, the average accuracy of the optimal combination reached 71.7%. This significant high performance demonstrates that personalized stimulus site selection and combination for different users can overcome inefficiency and achieve robust brain-controlled output. Building on the breakthrough at the individual level, this invention extends to the group level to find highly discriminative universal combinations.

[0041] Table 1. Binary and Tri-class Decoding Accuracy Based on Individual Optimal Sub-combinations for Each Subject

[0042] Figure 6 This paper presents the analysis results of group-level optimal combination identification and neural interpretability in embodiments of the present invention. By conducting in-depth statistical analysis and mining of the frequency of high-performance combinations generated by all subjects during the experiment, this study successfully identified and extracted highly stable feature subsets. In the binary classification task, the experiment found that the pure finger group (thumb-little finger) and the finger-wrist group (thumb-wrist) had the highest recognition and most stable performance. In the more challenging tri-class classification task, universal core combinations represented by the pure finger group (thumb-middle finger-little finger) and the finger-wrist group (thumb-little finger-wrist) were further extracted. The identification of these combinations not only reveals the common patterns of physiological electrical signals in hand movement representation, but also provides a solid neurophysiological basis for constructing a standardized model across subjects.

[0043] To verify the universal value of the aforementioned optimal sub-combinations for the entire population, Table 2 details the distribution of classification accuracy among all subjects under the condition of fixing these optimal combinations. Experimental data shows that when a fixed binary classification combination is uniformly used, the average population accuracy remains consistently high at 80.2%; even with a fixed three-class classification combination where task complexity increases, the average population accuracy still reaches an ideal level of 67.2%. This evaluation result strongly confirms the feasibility and robustness of the universal combination across subjects in practical applications. By adopting such pre-defined optimal combinations, the system can effectively overcome the differences in physiological signals between individuals, thereby significantly reducing the calibration time cost for new users before use, providing crucial technical support for the rapid deployment and plug-and-play functionality of the electromyography control system.

[0044] Table 2. Distribution of cross-subject classification accuracy based on the population's universal optimal sub-combination

[0045] To explore the neurophysiological mechanisms behind the model's high accuracy, this invention introduces a rigorous population-level optimal combination of neural interpretability analysis. At the data-driven level, the spatial separability of binary classification features, such as those for the pure finger group and the finger-wrist group, was evaluated based on the R-squared coefficient. The relevant topology map shows that highly discriminative EEG features are densely clustered in the central parietal lobe region of the brain, perfectly corresponding to the anatomical location of the human sensorimotor cortex. At the model-driven level, an integrated gradient mapping algorithm was introduced for neural attribution analysis for the three-class classification task. This technique, by quantifying the contribution of each category to the model's decision, clearly reveals the spatiotemporal attention characteristics of deep networks to different tactile sites, proving that the model accurately captures the somatosensory evoked potentials induced by mechanical vibration and the rhythmic decay signals of related brain regions, rather than relying on environmental noise or muscle artifacts to make judgments.

[0046] The embodiments described above provide a detailed explanation of the technical solutions and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, additions, and equivalent substitutions made within the scope of the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An active somatosensory brain-computer interface system based on tactile mapping to extend the input dimension of a single limb, characterized in that, It includes a tactile stimulation module, an EEG signal acquisition module, and a feature decoding module; During the stimulation phase, the tactile stimulation module provides localized tactile vibration stimulation to designated areas of the fingers and wrist on one side of the subject's hand without occupying the visual and motor channels. During the stimulation process, the subject needs to focus their attention on the vibration stimulation and identify the location of the tactile vibration stimulation in their mind, thereby establishing a mapping relationship between the tactile location sequence and discrete control commands. During the acquisition phase, the EEG signal acquisition module is used to record and preprocess the EEG signals induced by tactile vibration stimulation at designated points on one side of the fingers and wrist in real time. In the decoding stage, the feature decoding module is used to identify the preprocessed EEG signal, extract and classify the spatiotemporal features in the EEG signal, so as to decode the tactile response of a specific part.

2. The active somatosensory brain-computer interface system based on tactile mapping to extend the unilateral limb input dimension according to claim 1, characterized in that, The tactile stimulation module delivers stimulation via a tactile glove equipped with a linear resonant actuator motor.

3. The active somatosensory brain-computer interface system based on tactile mapping to extend the unilateral limb input dimension according to claim 1, characterized in that, The specific finger and wrist points on one side of the hand are the tips of the thumb, index finger, middle finger, ring finger, little finger, and wrist.

4. The active somatosensory brain-computer interface system based on tactile mapping to extend the unilateral limb input dimension according to claim 1, characterized in that, During the stimulation phase, each trial consisted of the following steps: first, providing visual cues to indicate the start of tactile stimulation, followed by applying vibrational stimulation to a single designated site for 4 seconds; during these 4 seconds of vibrational stimulation, the subject was required to focus their attention on the vibrational stimulation and identify the location of the tactile vibrational stimulation in their mind; finally, there was a 1-second relaxation period and random 0- to 1-second intervals between trials.

5. The active somatosensory brain-computer interface system based on tactile mapping to extend the unilateral limb input dimension according to claim 1, characterized in that, Preprocessing of EEG signals includes: A 48-52Hz notch filter was applied to suppress power frequency noise, and a 0.1-60Hz bandpass filter was applied. Independent component analysis and ICLabel were used to automatically remove electrooculography and electromyography artifacts, and signal segments were extracted from 0 to 4 seconds relative to the start of tactile stimulation.

6. The active somatosensory brain-computer interface system based on tactile mapping to extend the unilateral limb input dimension according to claim 1, characterized in that, The feature decoding module includes a filter bank co-space pattern combined with a support vector machine (FBCSP-SVM), as well as deep learning models EEGNet and IFNet. These three models are used to learn stable spatiotemporal features of sensorimotor-related intentions and support six-class active tactile EEG signal decoding.

7. The active somatosensory brain-computer interface system based on tactile mapping to extend the unilateral limb input dimension according to claim 1, characterized in that, The feature decoding module also includes: executing an optimized subclassification strategy to solidify the most stable discrete control instruction channel for the physiological characteristics of different subjects.

8. The active somatosensory brain-computer interface system based on tactile mapping to extend the unilateral limb input dimension according to claim 7, characterized in that, The optimized subclassification strategies include binary and tri-classification combinations. The binary combinations include thumb-little finger and thumb-wrist; the tri-classification combinations include thumb-middle finger-little finger and thumb-little finger-wrist.

Images