A single-channel steady-state visual evoked intent decoding and false touch suppression method and system based on display frame acknowledgement and quality evidence accumulation

Abstract

The application relates to the field of brain-computer interface and display timing synchronization, and discloses a single-channel steady-state visual evoked intention decoding and false touch inhibition method and system. A desired stimulation sequence of a plurality of visual targets and frame-by-frame feedback of a display device are acquired, and an effective stimulation track reflecting frame dropping, refresh change and display delay is constructed; a mapping of a display clock and an electroencephalogram acquisition clock is established, fundamental frequency, harmonic and phase evidence are extracted from a single-channel electroencephalogram window of an occipital region, and a phase correction reference is generated according to the effective stimulation track; signal quality, synchronization residual and candidate evidence are written into a sequential accumulator containing a no-control hypothesis, a window is dynamically adjusted, and only when evidence separation degree, phase consistency and continuous stability meet entering conditions, an intention is output; after output, a locking state is entered, and until the evidence is lower than exiting conditions or a stimulation set is updated. The scheme reduces false judgment and repeated triggering caused by frame dropping and timing deviation, and improves single-channel SSVEP interaction reliability.

Description

Technical Field

[0001] This invention relates to the fields of brain-computer interfaces, electroencephalogram (EEG) signal processing, visual stimulus presentation, and asynchronous human-computer interaction. In particular, it relates to a method, system, and electronic device for acquiring actual frame-by-frame presentation feedback from display devices, correcting stimulus phase deviations caused by frame drops, variable refresh rates, and display delays in consumer-grade displays, performing quality-weighted sequential evidence accumulation under single-channel EEG conditions in the occipital region, and using uncontrolled states, hysteresis, and output locking to suppress continuous accidental touches. Background Technology

[0002] Steady-state visual evoked potentials (SSVEPs) are the electroencephalographic responses generated in the occipital visual cortex when a user fixates on periodic visual stimuli, containing the fundamental frequency of the stimulus and its harmonic components. SSVEPs have advantages such as low training burden, high information transmission rate, and suitability for target selection, and can be used for high-level intent input in AR menus, AIoT devices, digital avatars, and embodied intelligent agents.

[0003] Existing SSVEP systems typically generate multiple blinking targets on the display at nominal frequencies, use a fixed trigger signal to capture EEG windows, and then identify the user's gaze target through spectrum analysis, canonical correlation analysis, filter banks, template matching, or neural networks. To improve real-time performance, some schemes employ confidence thresholds, continuous time windows, control and idle state recognition, or dynamic stopping.

[0004] There are significant differences between consumer-grade products and laboratory equipment. Display devices may use 60Hz, 90Hz, 120Hz, or variable refresh rates; operating system graphics rendering, GPU load, wireless projection, and mobile power saving can cause frame drops, latency, or refresh rate switching. The stimulus frame sequence requested by the application is not equivalent to the brightness changes actually seen by the user. If the reference signal is still generated according to the nominal frequency and theoretical phase, a phase drift will occur between the EEG window and the reference, and short-window decoding is particularly susceptible to this.

[0005] Furthermore, display devices and EEG acquisition devices typically have independent clocks. Sending a software trigger only once at the start of the trial cannot reflect continuous clock drift and frame-by-frame display errors. If a frame is dropped but the window is still used as a normal label for calibration, display malfunctions may be mistaken for differences in the user's neural responses.

[0006] Single-channel Oz or adjacent pillow region devices lack multi-channel spatial filtering capabilities and are more sensitive to reference electrode fluctuations, contact impedance, motion artifacts, and power frequency interference. A fixed window that is too short increases false positives, while a fixed window that is too long reduces the interaction speed. Outputting only when the maximum classification score exceeds a threshold also fails to distinguish between candidate targets' separation, phase consistency, and current signal quality.

[0007] Continuous interaction also suffers from repeated triggering issues. After a user selects a target, the line of sight and SSVEP response do not disappear instantly, and the sliding window may output the same command multiple times in a row. Simply setting a fixed cooldown time may cause premature recovery while the response is still present, or create unnecessary waiting after the user's attention has shifted.

[0008] Therefore, a technical solution is needed that combines the actual frame trajectory of the display end, cross-clock synchronization, single-channel quality assessment, phase correction reference, sequential evidence, uncontrolled state, and output latch-up into a complete closed loop. Summary of the Invention I. Purpose of the Invention

[0009] This invention aims to solve the problems of inconsistency between the actual stimulus trajectory and the nominal frequency in consumer-grade display devices, display and acquisition clock drift, single-channel signal quality variation, difficulty in balancing speed and reliability in fixed time windows, and repeated output caused by continuous gaze, thereby improving the accuracy, low false triggering and reproducibility of single-channel SSVEP in natural environments and continuous human-computer interaction. II. Technical Solution

[0010] The stimulus presentation module generates the desired stimulus sequence for each visual target, while simultaneously acquiring frame-by-frame feedback such as VSync, presentation completion, image composition, frame number, refresh rate, dropped frames, or photoelectric detection. The system reconstructs the effective stimulus trajectory based on the actual presentation time and brightness status, and does not consider frames that are not presented or have excessive time deviations as normal stimuli.

[0011] The cross-clock synchronization module establishes an affine mapping between the display clock and the EEG acquisition clock through common triggers, GPIO, network messages, reception time, or photoelectric events, continuously estimating offsets and drifts, and generating synchronization residuals and synchronization quality levels.

[0012] The decoding module extracts a single-channel EEG window from the occipital region based on the mapped actual stimulus trajectory and calculates quality indicators such as contact, reference stability, signal-to-noise ratio, artifacts, power frequency, effective data, and synchronization residuals. The system reconstructs the reference waveform and phase trajectory of each target based on the effective stimulus trajectory rather than the nominal frequency, and then extracts fundamental frequency, harmonics, phase lock, correlation, or template evidence.

[0013] The sequential evidence accumulator stores accumulated evidence for each visual target and uncontrolled state. Each new window provides only an increment, modulated by signal quality and synchronization quality; time slices with low quality or containing uncertain frames are downweighted or masked. The system continues to accumulate or extend the window when evidence is insufficient, pauses output when synchronization or quality is unsatisfactory, and only outputs intent when evidence for the highest target, separation from the second highest target, phase consistency, and continuity stability are all simultaneously satisfied.

[0014] After output, the system enters a locked state. During the locked state, it continuously monitors target evidence and uncontrolled evidence. The next output is only allowed after the original target evidence falls below the exit threshold, the uncontrolled state is established, the stimulus set version changes, or a new interaction cycle indicator is received. If the entry threshold is higher than the exit threshold, hysteresis is formed to avoid jitter near the threshold. III. Beneficial Effects

[0015] This invention can detect and compensate for reference phase errors caused by frame drops, display delays, and refresh rate changes; it can align the EEG window with the stimulus trajectory actually seen by the user; it can dynamically extend or stop decoding based on single-channel quality and synchronization reliability; it can accumulate uncontrolled states with multiple stimulus targets; it can prevent repeated commands caused by continuous fixation; and it can isolate display fault or synchronization error windows to avoid contaminating calibration data across users and individuals. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the overall system architecture of the present invention.

[0017] Figure 2 This is a schematic diagram illustrating the process of generating the desired stimulus sequence, displaying receipts frame by frame, and generating effective stimulus trajectories.

[0018] Figure 3 This diagram illustrates the mapping between the clock and the EEG acquisition clock, as well as the EEG windowing process.

[0019] Figure 4 This is a schematic diagram of the process for single-channel EEG quality assessment and phase correction evidence generation.

[0020] Figure 5 This is a schematic diagram of the sequential evidence accumulation and dynamic stopping process that includes an uncontrolled state.

[0021] Figure 6 This is a schematic diagram of the gating process for intention output, locking, exiting, and training samples. Detailed Implementation I. System Overall Architecture

[0022] like Figure 1 As shown, the system includes a stimulus presentation and frame feedback module 110, an effective stimulus trajectory module 120, a cross-clock synchronization module 130, a single-channel EEG acquisition and quality assessment module 140, a phase correction decoding module 150, a sequential evidence accumulation module 160, and a reliable output, locking, and sample gating module 170.

[0023] Visual stimuli can be presented via computer monitors, mobile phones, tablets, AR glasses, VR headsets, or LED stimulators. The EEG acquisition device may include an occipital region signal channel, a reference electrode, and a ground electrode or bias electrode. The embodiment uses a signal electrode and reference electrode near Oz, but the invention is also applicable to O1, O2, POz, or their adjacent regions. II. Expected Stimulus Sequence

[0024] For a display device with a refresh rate of F_r, the desired brightness sequence for target i can be generated by sinusoidal sampling: s_i[n] = (1 + sin(2πf_i n / F_r + phi_i)) / 2, or it can be generated using square wave, duty cycle coding, joint frequency-phase coding, or a predetermined frame sequence. The system assigns a frame_id to each frame and a stimulus_set_version to each stimulus set.

[0025] The expected sequence recording application plans to submit frames and target brightness, which are not considered as actual stimulus truth values. The presentation module simultaneously requests the graphics interface to return the actual presentation time or complete the fence; reads the VSync or TE signal when the display hardware allows; and optionally sets an invisible or low-interference synchronization block in the display area, with photodiodes recording actual brightness changes. III. Displaying receipts and valid stimulus trajectories frame by frame.

[0026] like Figure 2 As shown, each display receipt is associated with at least the frame_id, target brightness state, expected time, and actual time. Receipts can originate from the operating system presentation timestamp, GPU fence, VSync, display controller, frame rate counter, or photodetector. The system constructs r_i(t) based on the receipt, where r_i(t) represents the brightness state of target i at actual time t or its confidence interval.

[0027] If a frame is not displayed, the actual presentation interval exceeds a multiple of the expected interval, or the system reports a dropped frame, the frame is marked as missing. If it is only known that the frame was presented within a time interval, it is marked as uncertain. For missing frames, the previous brightness can be maintained, the transition can be determined based on the next valid frame, or the corresponding time slice can be directly blocked. The selected strategy is written to trajectory_version.

[0028] During variable refresh rate switching, the system reconstructs the trajectory using the actual frame time and does not continue to use the theoretical phase of the old refresh rate. If the acknowledgment completeness is below the threshold, the stimulus set enters an unverifiable state, and the system suspends the output of valid intentions. IV. Cross-clock synchronization

[0029] The display device clock t_d and the EEG acquisition clock t_e can be represented as t_e = a * t_d + b. The initial offset b can be estimated from the initial triggering event, and the drift a can be estimated by linear fitting, robust regression, or Kalman filtering of multiple synchronization events during operation. Synchronization events can be the photoelectric edge of the display synchronization block, GPIO triggering, audio / video synchronization packets, wireless reception time, or correlation peaks between the effective stimulus trajectory and the phase of the EEG frequency.

[0030] For the k-th synchronization event, the residual e_k = t_e,k - (a t_d,k + b). The system uses the root mean square of the residuals, the maximum residual, the interval between the most recent events, and the variance estimated by the drift to form the synchronization quality q_sync. If the residual exceeds a preset proportion of the acceptable window, the EEG window crossing the abnormal point is discarded or the phase uncertainty interval is expanded. V. Single-channel EEG mass vector

[0031] The quality assessment module generates q=[q_contact,q_ref,q_sat,q_drift,q_line,q_motion,q_snr,q_valid,q_sync] for each window. Contact and reference stability can be obtained from impedance, common-mode, or baseline statistics; saturation ratio is the proportion of samples reaching the boundary of the acquisition range; power frequency index is the energy ratio near 50Hz or 60Hz to the energy in the neighborhood; motion artifacts can be determined from amplitude transitions, low-frequency energy, or synchronous inertial data; target band signal-to-noise ratio is calculated based on the fundamental frequency and harmonic energy of the candidate stimulus.

[0032] The quality vector is used not only for results reporting but also for harmonic weighting, window increments, evidence caps, and output eligibility. When the reference electrode is unstable, it can simultaneously reduce the evidence increment for all targets and increase uncontrolled evidence, rather than forcibly selecting a particular target. VI. Phase Correction Reference

[0033] The system maps the effective stimulus trajectory r_i(t) to the acquisition time and resamples it at the EEG sampling points to obtain the reference waveform x_i[m]. The Fourier basis, Hilbert phase, fundamental frequency and sine and cosine components of multiple harmonics can be calculated on x_i[m], or x_i[m] can be directly input into the template encoder.

[0034] Candidate evidence can be generated from any one or a combination of the following: correlation coefficients between the EEG window and the phase-corrected sine and cosine references; amplitude and phase differences of each harmonic complex spectrum; phase-locked values; matching scores with the user template; and lightweight neural network outputs. If CCA or FBCCA is used, the reference matrix is ​​reconstructed from the actual trajectory, rather than fixed sin(2πf_it) and cos(2πf_it).

[0035] For the h-th harmonic, the weight can be set as w_h = q_snr,h q_sync / (h^p + epsilon). The sample corresponding to the dropped frame is given a mask[m] = 0 or a lower weight. The candidate evidence E_i is a weighted sum of multiple harmonic evidences, and the phase deviation delta_phi_i and cross-window phase consistency are recorded. VII. Uncontrolled states and sequential accumulation

[0036] The sequential accumulator maintains multiple target accumulators L_i and uncontrolled accumulators L_0. Each new window j generates increments d_i,j and d_0,j, which are updated to L_i,j = lambda L_i,j-1 + w(q_j)d_i,j, where lambda is the forgetting or overlap correction coefficient, and w(q_j) is determined by quality and synchronization.

[0037] The uncontrolled increment can be obtained from model scores based on low evidence for all targets, insufficient separation between the highest and second-highest values, phase instability, low signal quality that does not reach hard rejection, or the user being in an idle state. The uncontrolled state is not a name for a post-failure threshold, but rather an assumption of participating in sequential competition.

[0038] The system begins its evaluation after reaching the minimum window length. If the highest L_i has not reached the entry threshold but the quality is valid, the window is increased; if the quality can be recovered through a short wait, the update is paused once; if no target meets the conditions even after reaching the maximum window length, the system outputs "no control" or "rejected" and resets the accumulator. VIII. Reliable Output Conditions

[0039] The output target i must simultaneously satisfy the following: L_i is not lower than the entry threshold T_in; L_i-L_second is not lower than the separation threshold T_sep; phase consistency is not lower than T_phase; signal quality and synchronization quality are higher than the corresponding thresholds; target i ranks first in k consecutive decision times; and the stimulus set version and the effective trajectory version have not undergone incompatible changes during the accumulation period.

[0040] Thresholds can be adjusted based on user quality zoning, target frequency, and task risk, but the adjustments must take effect at the start of the session or upon explicit version switching; thresholds cannot be arbitrarily lowered using currently unconfirmed results. High-risk targets may also require a second selection, blink confirmation, EMG confirmation, or application-side confirmation. IX. Output Lockout and Resumption

[0041] After the target output is obtained, the state machine transitions from READY to LOCKED. In the LOCKED state, evidence continues to be computed, but the output interface ignores candidate outputs generated by the same sustained gaze. It only enters RECOVERY when L_i falls below T_out, L_0 reaches the recovery threshold, a new interaction cycle indicator is received, or the stimulus set version is updated. After a minimum idle window, it returns to READY.

[0042] A hysteresis occurs when T_in is higher than T_out. Compared to a fixed cooldown time, this mechanism requires that the neural evidence or interaction state actually leaves the original choice, thus avoiding repeated triggering at the tail of a long response and allowing the user to quickly move to the next target and recover early. 10. Training Sample Gating

[0043] Each candidate output saves `stimulus_set_version`, `trajectory_version`, `sync_version`, `window_id`, quality vector, target evidence, uncontrolled evidence, phase consistency, state machine state, and task acknowledgment. The window is only written to the calibration cache if the acknowledgment is complete, the synchronization residual is acceptable, the EEG quality is acceptable, and the task acknowledgment is attributable.

[0044] For example, a robot's failure to execute a command might stem from a security gate blockage rather than an SSVEP decoding error; when dropped frames and user rejection occur simultaneously, the cause cannot be determined. Such samples are marked as isolated and not used to update user templates or thresholds. This gating can be combined with an authorized data management system, but authorization credentials are not a necessary step in this invention. XI. Example 1: Four-Target Single-Channel SSVEP Menu

[0045] The display presents four visual targets at a refresh rate of 60Hz, with nominal frequencies set to 8Hz, 10Hz, 12Hz, and 15Hz, respectively. The application records a presentation timestamp for each frame and exchanges a synchronization timestamp with the acquisition device every second. Single-channel EEG is acquired by Oz relative to a reference electrode at a sampling rate of 250Hz.

[0046] The system updates evidence every 0.25 seconds, with a minimum window of 0.75 seconds and a maximum window of 3 seconds. Under normal quality conditions, it accumulates multi-harmonic phase correction correlation scores; when a frame is dropped, the reference is reconstructed based on the actual brightness trajectory; output is paused when there are consecutive frame drops or the synchronization residual exceeds 20 milliseconds. Output is initiated after the target meets three consecutive stabilization and separation conditions, then the system enters LOCKED mode until the original target evidence exits and the uncontrolled state continues for a preset duration.

[0047] The figures above are for illustrative purposes only and do not constitute limitations on frequency, sampling rate, or threshold. During research and development verification, parameters should be determined based on the actual device refresh rate, user safety, and ethical requirements. 12. Example 2: AR Glasses and Variable Refresh Rate

[0048] AR glasses may switch from 90Hz to 72Hz depending on the load. The frame receipt module detects the refresh rate switch and increments trajectory_version, discarding accumulated evidence that crosses the switch point, and regenerating the target trajectory based on the new frame time. Spatial target anchoring, embodied robot task planning, and shared autonomy can be provided by the application side; this case only outputs visual evoked intents with target identification, time, confidence level, quality, and version information. Thirteen, Example 3: Weak Subject Adaptive Window

[0049] For users with weak target frequency band energy but qualified contact and synchronization, the quality strategy lowers the upper limit of single-window evidence increment and extends the maximum window, without lowering the false trigger threshold; if the user's phase template stabilizes after a small amount of calibration, user template evidence can be added. For users who consistently fail to meet the separation conditions, no control or rejection is maintained, and multiple guesses are not counted as usable. XIV. Example 4: Continuous Interaction in Experience Stores

[0050] The experience store terminal continuously serves different users. Each session creates a new synchronized version and stimulus set version; the user's accumulated state is cleared at the end of the session. The system statistics include frame drop rate, synchronization residual, quality partition, first selection time, accidental touch, and lock recovery time. Only records that have obtained informed consent and meet the training sample gating requirements are included in the subsequent cross-user model dataset. XV. Evaluation Methods

[0051] The comparison baselines include: using only nominal frequency reference; using only start triggering; fixed window classification; maximum score threshold; fixed cooldown time; and the complete scheme of this invention. Key metrics include Accuracy, ITR, first reliable output time, False Activation Rate, repeated triggering rate, uncontrolled recognition rate, frame drop condition accuracy, synchronization error sensitivity, weak subject availability, and edge inference latency.

[0052] Conditions such as no frame drops, random frame drops, continuous frame drops, refresh rate switching, clock drift, reference electrode disturbance, and motion artifacts should be reported separately and verified in reserved user and real natural sessions. Lockout samples must not be mixed with ordinary idle samples in the same evaluation category. XVI. Other Implementation Methods

[0053] This invention can be combined with CCA, FBCCA, TRCA, template transfer, lightweight neural networks, or other decoders; it can use any combination of software frame receipt, hardware synchronization, or photoelectric detection; and it can be extended to cVEP or other visual evoked paradigms where the reference trajectory is determined by the display frame sequence. Any equivalent scheme that corrects a single-channel visual evoked reference based on the actual frame-by-frame stimulus trajectory and combines quality sequential accumulation and evidence exit locking to suppress false triggers can adopt the technical concept of this invention.

Claims

1. A method for decoding and suppressing accidental touches in a single-channel steady-state visual evoked intent based on display frame receipts and quality evidence accumulation, characterized in that, include: The system acquires the desired stimulus sequence generated by the display device for multiple visual stimulus targets, and acquires the frame-by-frame display receipt generated by the display device when presenting the desired stimulus sequence. Based on the actual presentation time, presentation status and target brightness status of each frame, the system generates an effective stimulus trajectory for each visual stimulus target. Acquire the acquisition clock information of the single-channel EEG acquisition device in the occipital region, establish the time mapping relationship between the display clock of the display device and the acquisition clock, map the effective stimulus trajectory to the acquisition clock, and determine the stimulus synchronization residual; The single-channel EEG signal is windowed according to the mapped effective stimulus trajectory to generate an EEG signal quality vector. Fundamental frequency evidence, harmonic evidence, and phase evidence corresponding to each visual stimulus target are extracted from the EEG window, and a phase correction reference corresponding to each visual stimulus target is generated according to the effective stimulus trajectory. Based on the fundamental frequency evidence, harmonic evidence, phase evidence, phase correction reference, EEG signal quality vector, and stimulus synchronization residual, candidate evidence increments and uncontrolled state evidence increments are generated for each visual stimulus target, and the candidate evidence increments and uncontrolled state evidence increments are written into the sequential evidence accumulator. Based on the EEG signal quality vector, stimulus synchronization residual, the separation degree between the current highest candidate evidence and the second highest candidate evidence, and phase consistency, determine one of the following processing options: continue accumulation, extend the EEG window, pause output, or execute intention output. The visual evoked intent corresponding to the visual stimulus target is output only when the cumulative evidence for the same visual stimulus target meets the entry threshold, the separation degree meets the separation condition, the phase consistency meets the phase condition, the EEG signal quality vector and the stimulus synchronization residual meet the validity condition, and remain stable for multiple consecutive decision moments. After outputting the visual evoked intent, the system enters an output lock state, in which it prevents the repeated output of a valid intent for the same continuous gaze process until at least one of the following occurs: target evidence falls below the exit threshold, no control state evidence meets the recovery condition, the visual stimulus target set is updated, or a new interaction cycle identifier is received.

2. The method according to claim 1, characterized in that, The frame-by-frame display receipt includes at least three of the following: vertical synchronization signal timestamp, graphics interface presentation timestamp, display synthesis completion timestamp, frame number, refresh rate, frame drop flag, variable refresh rate status, target brightness status, or brightness change time obtained by a photodetector. When a frame lacks a display receipt, is marked as a dropped frame, or its actual presentation time deviates from the expected presentation time by more than a threshold, the corresponding frame is marked as an uncertain frame in the effective stimulus trajectory or the actual brightness state is reconstructed according to the adjacent effective frames.

3. The method according to claim 1, characterized in that, Establishing the time mapping relationship includes: estimating the offset coefficient and drift coefficient from the display clock to the acquisition clock based on at least one of the following: common trigger event, network clock message, display receipt and acquisition terminal reception time, photoelectric detection event or periodic related peak value; mapping the display time t_d to the acquisition time t_e = a t_d + b, where a is the clock drift coefficient and b is the clock offset; generating stimulus synchronization residuals and synchronization quality levels based on the mapping residuals of multiple synchronization events, and pausing the visual evoked intent output when the synchronization quality level is lower than the threshold.

4. The method according to claim 1, characterized in that, The single-channel EEG signal in the occipital region is acquired by a signal electrode located in or near Oz, O1, O2, POz, or a reference electrode. The EEG signal quality vector includes at least three of the following: contact impedance, reference electrode stability, saturation ratio, baseline drift, power frequency energy, motion artifacts, target band signal-to-noise ratio, effective data ratio, and stimulus synchronization residual. These parameters are used to determine the weight of each harmonic, the current EEG window length, the upper limit of evidence increment, or whether output is allowed.

5. The method according to claim 1, characterized in that, Generating the phase correction reference includes: reconstructing the temporal stimulus waveform for each visual stimulus target based on the actual brightness state and actual presentation time corresponding to each frame in the effective stimulus trajectory; calculating the orthogonal components, phase trajectory, or spectral template of the fundamental frequency and one or more harmonics of the temporal stimulus waveform; using the reconstructed orthogonal components, phase trajectory, or spectral template to perform correlation, coherence, phase locking, or template matching with the EEG window; when the display device switches refresh rates or the cumulative frame drop ratio exceeds a threshold, discarding the reference generated based on the nominal stimulus frequency and regenerating the phase correction reference.

6. The method according to claim 1, characterized in that, The sequential evidence accumulator stores the cumulative log-likelihood, cumulative correlation score, cumulative posterior probability, or cumulative evidence quality for each visual stimulus target and uncontrolled state. During each update, the incremental candidate evidence is scaled according to the quality of the EEG signal and the synchronization quality, and the time slices corresponding to uncertain frames are downweighted or masked. When the separation between the highest and second-highest candidate evidence is insufficient, the window is extended. When the total window length reaches the maximum length and still does not meet the output conditions, the uncontrolled state is output or the output is rejected.

7. The method according to claim 1, characterized in that, The entry threshold is higher than the exit threshold to create hysteresis; Continuous stability is determined based on at least two of the following: the ranking of the same target at consecutive decision times, phase deviation, evidence growth direction, and candidate set version; during the output lockout state, if the target continues to occupy the highest candidate position but its evidence does not first drop below the exit threshold, a second valid intent is not generated; for intents with a risk level higher than the threshold, at least one of repeated selection consistency, user confirmation, or second input modality confirmation is also required.

8. The method according to claim 1, characterized in that, Also includes: Record the visual stimulus set version, effective stimulus trajectory version, time-mapped version, EEG window label, signal quality vector, candidate evidence, uncontrolled evidence, output status, and task feedback; only when the frame-by-frame display feedback completeness, synchronization quality, EEG signal quality, and task feedback attributability meet the training sample conditions, add the corresponding EEG window and visual stimulus target label to the calibration cache; when displaying dropped frames, synchronization errors, execution failures, or user negative feedback cannot be distinguished, isolate the corresponding window and do not use it to update decoding parameters.

9. A single-channel steady-state visual evoked intent decoding and accidental touch suppression system based on display frame receipt and quality evidence accumulation, characterized in that, include: The stimulus presentation and frame feedback module is used to generate the desired stimulus sequence of multiple visual stimulus targets and obtain frame-by-frame display feedback to construct an effective stimulus trajectory. The cross-clock synchronization module is used to establish a time mapping between the display clock and the single-channel EEG acquisition clock and generate stimulus synchronization residuals; the single-channel quality assessment module is used to generate EEG signal quality vectors; the phase correction decoding module is used to generate phase correction references based on effective stimulus trajectories and extract fundamental frequency, harmonics and phase evidence; the sequential evidence accumulation module is used to accumulate evidence of each visual stimulus target and uncontrolled state and adjust the window according to quality and synchronization residuals. A credible output module is used to output a visually evoked intent when the evidence, separation, phase consistency, quality, and continuity stability meet the entry conditions. The lockout and resume module is used to prevent the same continuous watch process from being output repeatedly after the output has been completed, until the exit or resume conditions are met.

10. An electronic device, characterized in that, The device includes a processor, a memory, a display interface, and a neural signal acquisition interface. The memory stores a computer program, which, when executed by the processor, causes the electronic device to perform the method described in any one of claims 1 to 8.

Images