Electroencephalogram response type nerve stimulation training system for dogs and control method

Abstract

The invention discloses an electroencephalogram response type nerve stimulation training system for dogs and a control method, and relates to the technical field of brain-computer interface application. A time-space alignment module; a feature extraction and identification module; a judgment and parameter self-adaption module; a nerve stimulation module; and a man-machine interaction module. Repeatable and low-time-delay alignment potential data are generated through real-time tracking and global clock alignment of the position of the dog head electrode in combination with high-quality electroencephalogram collection and pane-level feature extraction. Based on the judgment of the data, the real neural activity of the dog can be accurately corresponding in time and space, it is guaranteed that when the expected physiological event of the dog occurs, the award stimulation generated by the nerve stimulation module is accurate and in place, and the consistency and reliability of the stimulation-dog reaction time sequence in training are improved. The operating conditioned reflex principle is applied, and stimulation of the nerve stimulation module can be triggered by the remote controller when the dog has expected behaviors.

Description

Technical Field

[0001] This invention relates to the field of brain-computer interface technology, specifically to a canine EEG-responsive neural stimulation training system and control method. Background Technology

[0002] In recent years, with the development of brain-computer interfaces (BCIs), bioelectric signal processing, and neuromodulation technologies, closed-loop stimulation and behavioral intervention based on electroencephalography (EEG) have received widespread attention in medical rehabilitation, neuroscience research, and animal training. Non-invasive EEG acquisition hardware, real-time feature extraction algorithms, and programmable stimulation devices have matured, and combined with embedded control and wireless communication, they can now detect short-term neural events and apply external feedback to a certain extent, thus enabling applications such as enhanced learning, attention regulation, and behavioral correction. Simultaneously, multimodal reinforcement (such as sound and light, food delivery) and artificial interaction interfaces are also being used to improve training efficiency and safety.

[0003] Existing EEG-responsive training systems for dogs or small animals generally face the problem of insufficient spatiotemporal alignment accuracy in practical applications: electrode position drift, contact impedance changes, and subsystem clock asynchrony lead to inconsistencies in the spatial coordinates or timestamps of the sampled data, which in turn cause errors in feature extraction and trigger determination based on these data, resulting in false triggers or trigger times deviating from the animal's actual neural events, reducing training effectiveness and potentially increasing unnecessary stimulation burden.

[0004] In response to this problem, this application proposes a canine EEG-responsive neurostimulation training system and control method to solve the above-mentioned issues. Summary of the Invention

[0005] The purpose of this invention is to provide a canine EEG-responsive neurostimulation training system and control method to solve the problem of insufficient spatiotemporal alignment accuracy that is commonly faced in the practical application of existing EEG-responsive training systems for dogs or small animals.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A canine EEG-responsive neurostimulation training system includes:

[0008] The EEG acquisition module is used to acquire raw EEG data from the dog's head and output the raw EEG data.

[0009] The time-space alignment module is used to perform electrode position correction and global time synchronization processing on the raw EEG data to obtain aligned potential data.

[0010] The feature extraction and recognition module is used to perform noise reduction, spectral and time-frequency feature extraction on the alignment potential data, and to identify target behavior-related feature vectors based on the features.

[0011] The determination and parameter adaptation module is used to calculate the trigger determination based on the feature vector and output the stimulus parameter setting.

[0012] The neural stimulation module is used to output an appropriate neural stimulation signal to the dog according to the stimulation parameters and to feed back the stimulation response to the feature extraction and recognition module.

[0013] The human-computer interaction module is used to display the alignment potential data, feature vectors, trigger judgment and stimulus records, and supports manual intervention. The manual intervention includes manually triggering the neural stimulation module through the remote control subunit to achieve behavioral reinforcement based on the principle of operant conditioning.

[0014] Furthermore, the EEG acquisition module includes:

[0015] Flexible dry electrode array, differential signal acquisition unit and low-power wireless transmission submodule;

[0016] The signal-to-noise ratio of the differential signal acquisition unit is >40dB;

[0017] The power consumption of the wireless transmission submodule is <10mW;

[0018] The flexible dry electrode array is positioned to match the physiological structure of the canine skull and is placed behind the ear or in the parietal region.

[0019] The raw EEG data output by the EEG acquisition module corresponds one-to-one with the time-space alignment module.

[0020] Furthermore, the time-space alignment module corrects the original EEG data by fusing electrode position tracking information with a global time reference, thereby eliminating electrode positioning errors and clock drift, and outputting high-precision aligned potential data.

[0021] The aligned potential data maintains a correspondence between the number of samples and the timestamps of the original EEG data.

[0022] Furthermore, the feature extraction and recognition module incorporates a lightweight sequence classifier, which is used to calculate the features of the θ wave power spectral density, α-θ ratio, instantaneous phase synchronization and attention index based on the aligned potential data, and combines the features into a feature vector;

[0023] The recognition module outputs a confidence score for the target behavior based on the feature vector.

[0024] Furthermore, the determination and parameter adaptation module is based on the following triggering scoring function. Determine if a stimulus has been triggered:

[0025]

[0026] in, Let θ be the change in θ-wave power relative to the baseline over time t. Baseline theta wave power; The attention index is calculated based on time-domain and frequency-domain features (normalized to 0–1). denoted as the recent target behavior success rate (0–1); α, β, γ are weighting coefficients and α+β+γ=1; δ is a positive adjustment constant; the stimulus is triggered when St≥Θt, where the threshold Θt is updated online smoothly based on training progress and safety constraints;

[0027] After initializing each parameter in the formula, the weights can be adjusted in small steps during training through the judgment and parameter adaptation module to achieve an individualized triggering strategy.

[0028] Furthermore, the neural stimulation module is a non-invasive neural stimulation unit or a minimally invasive patch structure;

[0029] The minimally invasive patch structure includes an adjustable current source and a bipolar / multipolar electrode patch. The current source has an output range of 0.1–1.0 mA, an adjustable frequency of 1–20 Hz, a sine / pulse waveform, and a pulse width range of 100–500 μs.

[0030] The neurostimulation module implements a gradual increase safety strategy before output and monitors the stimulus response in real time after output to achieve closed-loop intensity / duration adjustment.

[0031] Furthermore, the human-computer interaction module includes a mobile terminal application and a remote control subunit;

[0032] The mobile terminal application is used to display the alignment potential data, feature vectors, trigger confidence, stimulus records and training curves, and can set the maximum stimulus value, safety strategy, manual triggering and intervention.

[0033] The remote control subunit supports two manual modes: single pulse and continuous stimulation, as well as delay time adjustment. It is used by the trainer to manually trigger the neural stimulation module when the dog exhibits the desired behavior, so as to achieve positive reinforcement intervention based on the principle of operant conditioning.

[0034] The system constructs a closed-loop training mechanism based on the principle of operant conditioning: when the dog spontaneously exhibits the desired behavior, the remote control subunit of the human-computer interaction module can manually trigger the neural stimulation module to output reward stimuli, or the judgment and parameter adaptation module can automatically trigger stimuli based on EEG characteristics, establishing a positive reinforcement signal with the dog's behavior, prompting the dog to gradually learn and consolidate the target behavior; the two triggering methods complement each other, supporting fully automatic closed-loop training, and also allowing trainers to manually intervene and reinforce at key behavioral nodes, improving the flexibility and adaptability of training.

[0035] Furthermore, the system also includes a multimodal adapter interface for communicating with external training devices (such as clickers or food dispensers), enabling the triggering of the neural stimulation module to be superimposed with or mutually triggered by signals from traditional feedback devices. The adapter interface ensures that the timestamp is aligned with the alignment potential data during triggering.

[0036] Secondly, this application provides a canine EEG-responsive neural stimulation training control method, applied to the system described in the first aspect, the control method comprising the following steps:

[0037] Collect raw EEG data from the dog's head and output the raw EEG data;

[0038] Electrode position correction and global time synchronization processing are performed on the raw EEG data to obtain aligned potential data;

[0039] The alignment potential data is subjected to noise reduction, spectral and time-frequency feature extraction, and target behavior-related feature vectors are identified based on the features.

[0040] Based on the feature vector, a trigger determination is calculated and the stimulus parameter setting is output.

[0041] The appropriate neural stimulation signal is output to the dog according to the stimulation parameters and the stimulation response is fed back to the feature vector.

[0042] Display the alignment potential data, feature vectors, trigger determination, and stimulus records, and perform manual intervention.

[0043] Furthermore, the stimulation parameters are adjusted in a stepwise, individualized manner:

[0044] The initial current is 0.5 mA, and the current increases by 0.1 mA after each successful alarm until it does not exceed 1.0 mA.

[0045] The delay time of the stimulus signal is gradually extended from instantaneous triggering to no more than 3 seconds to simulate real task scenarios;

[0046] Multimodal reinforcement strategies are implemented through manual intervention via remote control using human-computer interaction or by receiving superimposed trigger signals from external training devices. The manual intervention includes manually triggering the neural stimulation module through the remote control subunit to achieve behavioral reinforcement based on the principle of operant conditioning.

[0047] Compared with existing technologies, this invention provides a canine EEG-responsive neurostimulation training system and control method. By real-time tracking and global clock alignment of the dog's head electrode positions, combined with high-quality EEG acquisition and pane-level feature extraction, repeatable and low-latency aligned potential data is generated. Based on this data, the system can accurately correspond to the dog's actual neural activity in time and space, thereby reducing false triggers caused by electrode drift or clock asynchrony. This ensures that stimulation is accurately delivered when the dog exhibits the desired physiological event, improving the consistency and reliability of the "stimulus-dog response" timing during training. Furthermore, this invention applies the principle of operant conditioning, allowing the neurostimulation module to be triggered by a remote control when the dog exhibits the desired behavior. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0049] Figure 1 A block diagram of a canine EEG-responsive neurostimulation training system provided in an embodiment of the present invention;

[0050] Figure 2 This is a block diagram of the EEG acquisition module provided in an embodiment of the present invention;

[0051] Figure 3 This is a system application scenario diagram provided in the embodiments of the present invention;

[0052] Figure 4 This is a block diagram showing the connection components of the human-computer interaction module provided in an embodiment of the present invention;

[0053] Figure 5 A flowchart of a canine EEG-responsive neural stimulation training and control method provided in an embodiment of the present invention. Detailed Implementation

[0054] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings.

[0055] Example 1:

[0056] like Figures 1-4As shown, a canine EEG-responsive neurostimulation training system includes:

[0057] EEG acquisition module 100 is used to acquire raw EEG data from the head of a dog and output the raw EEG data.

[0058] Specifically, the EEG acquisition module 100 includes:

[0059] Flexible dry electrode array, differential signal acquisition unit and low-power wireless transmission submodule;

[0060] The signal-to-noise ratio of the differential signal acquisition unit is >40dB;

[0061] The power consumption of the wireless transmission submodule is <10mW;

[0062] The flexible dry electrode array is positioned to match the physiological structure of the canine skull and is placed behind the ear or in the parietal region.

[0063] The raw EEG data output by the EEG acquisition module 100 corresponds one-to-one with the time-space alignment module 200.

[0064] Furthermore, the EEG acquisition module 100 is responsible for acquiring raw potential signals from the dog's head, and includes an electrode array 101 and a signal receiving box 102. The signal receiving box 102 integrates a differential amplifier, an analog-to-digital converter, and a low-power wireless / wired transmission component. The electrode array 101 includes a flexible dry electrode or hydrogel wet electrode array (arranged according to the canine skull shape), a differential amplifier (with power frequency notch, configurable high-pass 0.5 Hz, low-pass 250–500 Hz), a 16–24 bit ADC, and a sampling rate of 500–2000 Hz.

[0065] The output is a raw sample stream with timestamps and includes electrode metadata (coordinates / impedance), which is used as input to the time-space alignment module; the device achieves ≥40 dB SNR, low motion artifact suppression, and real-time packet loss detection to ensure downstream processing quality.

[0066] The time-space alignment module 200 is used to perform electrode position correction and global time synchronization processing on the raw EEG data to obtain aligned potential data.

[0067] Specifically, the time-space alignment module 200 corrects the original EEG data by fusing electrode position tracking information with a global time reference, thereby eliminating electrode positioning errors and clock drift, and outputting high-precision aligned potential data.

[0068] The aligned potential data maintains a correspondence between the number of samples and the timestamps with the original EEG data;

[0069] Furthermore, the time-space alignment module 200 performs electrode position correction and global clock synchronization on the original sampling stream to eliminate positioning errors and clock drift. Implementation measures may include: optical or inertial navigation electrode position tracking (real-time return of electrode 3D coordinates), online contact impedance estimation for electrode failure detection, and master clock (PTP-like) synchronization of timestamps for each subsystem.

[0070] Processing steps: Remap the coordinates of each channel using positioning data, recalibrate the timestamps according to the global time base, and label and interpolate / remove motion artifact segments. Output aligned potential data, with each sample containing both corrected coordinate information and a precise timestamp, ensuring that subsequent modules can reference each sample individually.

[0071] The feature extraction and recognition module 300 is used to perform noise reduction, spectral and time-frequency feature extraction on the alignment potential data, and to identify target behavior-related feature vectors based on the features.

[0072] Specifically, the feature extraction and recognition module 300 has a built-in lightweight sequence classifier, which is used to calculate the features of the theta wave power spectral density, α-θ ratio, instantaneous phase synchronization and attention index based on the aligned potential data, and combine the features into a feature vector;

[0073] The identification module outputs a confidence score for the target behavior based on the feature vector;

[0074] Furthermore, this module preprocesses and constructs features for the alignment potential data: first, it performs filtering (bandpass / notch filtering), adaptive power frequency and electromyography artifact suppression, and then calculates frequency domain (θ, α, β wave power), time frequency (short-time Fourier or wavelet) and phase / synchronization index, time domain energy / amplitude statistics, etc., based on a sliding window (e.g., 500 ms).

[0075] These quantification metrics are then synthesized into feature vectors (which can be normalized, dimensionality reduced such as PCA or lightweight autoencoders), and the target behavior and its confidence are output by the built-in classifier / sequence model (lightweight LSTM, 1D-CNN or tree model).

[0076] The feature vectors and confidence scores are the direct inputs to the decision module, maintaining a one-to-one correspondence with time alignment.

[0077] The determination and parameter adaptation module 400 is used to calculate the trigger determination based on the feature vector and output the stimulus parameter setting.

[0078] Specifically, the determination and parameter adaptation module is based on the following triggering scoring function. Determine if a stimulus has been triggered:

[0079]

[0080] in, Let θ be the change in θ-wave power relative to the baseline over time t. Baseline theta wave power; The attention index is calculated based on time-domain and frequency-domain features (normalized to 0–1). denoted as the recent target behavior success rate (0–1); α, β, γ are weighting coefficients and α+β+γ=1; δ is a positive adjustment constant; the stimulus is triggered when St≥Θt, where the threshold Θt is updated online smoothly based on training progress and safety constraints;

[0081] After initializing each parameter in the formula, the weights can be adjusted in small steps during training through the judgment and parameter adaptation module to achieve an individualized triggering strategy.

[0082] Furthermore, this module calculates trigger scores and determines stimulus parameters based on feature vectors, including scoring function implementation, threshold management, and online adaptive logic.

[0083] Functionality: Calculate the real-time score St according to a predetermined formula (e.g., weighted θ change + attention index + recent success rate); compare it with the dynamic threshold Θt to determine the trigger, and adjust Θt and the weight coefficient (small step gradient or regular step) according to short-term / long-term window statistics.

[0084] In addition, it includes a safety constraint submodule (maximum current, minimum stimulation interval) and a manual intervention interface. The output is a safety-verified set of stimulation parameters for use by the neural stimulation module.

[0085] In this embodiment, the principle of operant conditioning is implemented through two triggering paths:

[0086] Automatic triggering pathway: After the judgment and parameter adaptation module identifies the target neural activity of the dog based on EEG characteristics, it automatically outputs stimulation parameters to the neural stimulation module, and provides immediate feedback as a positive reinforcement signal, so that the dog can gradually establish the association between "neural activity and reward stimulus".

[0087] Manual triggering pathway: When the trainer observes the dog exhibiting the desired behavior (such as sitting down or looking at the target) through the remote control subunit of the human-computer interaction module, the trainer manually triggers the neural stimulation module to output a reward stimulus, directly binding the behavior in the operant conditioning reflex to the stimulus.

[0088] The two triggering pathways operate in parallel, with stimulus responses being fed back to the feature extraction and recognition module for updating feature vectors and assessing the confidence level of trigger decisions. Manual trigger records also serve as supervisory signals, which can be used for subsequent calibration and optimization of the automatic triggering threshold, enabling personalized training strategies for human-machine collaboration.

[0089] The neural stimulation module 500 is used to output an appropriate neural stimulation signal to the dog according to the stimulation parameters and to feed back the stimulation response to the feature extraction and recognition module.

[0090] Specifically, the nerve stimulation module 500 is a non-invasive nerve stimulation unit or a minimally invasive patch structure;

[0091] The minimally invasive patch structure includes an adjustable current source and a bipolar / multipolar electrode patch. The current source has an output range of 0.1–1.0 mA, an adjustable frequency of 1–20 Hz, a sine / pulse waveform, and a pulse width range of 100–500 μs.

[0092] The neurostimulation module implements a slow-increase safety strategy before output and monitors the stimulation response in real time after output to achieve closed-loop intensity / duration adjustment.

[0093] Furthermore, the neurostimulation module 500 is responsible for outputting physical stimulation signals according to the parameters of the judgment module, and can use non-invasive patch or minimally invasive patch electrodes and adjustable current sources.

[0094] Implementation details: Current amplitude (0.1–1.0 mA), frequency (1–20 Hz), pulse width (100–500 μs), waveform selection, and ramp-up strategy (e.g., 200 ms linear ramp-up) can all be programmed; the module must monitor the output current / voltage and contact impedance in real time and shut down immediately in case of abnormality.

[0095] After stimulation, the module transmits the stimulation time, parameters, and immediate EEG / physiological response (short-term power changes, behavioral confirmation) back to the feature extraction module, forming a closed-loop feedback data stream.

[0096] The human-computer interaction module 600 is used to display the alignment potential data, feature vector, trigger judgment and stimulus record and supports manual intervention. The manual intervention includes manually triggering the neural stimulation module through the remote control subunit to achieve behavioral reinforcement based on the principle of operant conditioning.

[0097] Specifically, the human-computer interaction module 600 includes a mobile terminal application 601 and a remote control subunit 602;

[0098] The mobile terminal application 601 is used to display the alignment potential data, feature vector, trigger confidence, stimulus record and training curve, and can set the maximum stimulus value, safety strategy, manual trigger and intervention;

[0099] The remote control subunit 602 supports two manual modes: single pulse and continuous stimulation, as well as delay time adjustment. It is used to allow trainers to manually trigger the neural stimulation module when the dog exhibits the desired behavior, so as to achieve positive reinforcement intervention based on the principle of operant conditioning.

[0100] Furthermore, the human-computer interaction module 600 provides visualization, parameter configuration, and manual intervention interfaces, including mobile / desktop applications and remote control subunits.

[0101] Key features include: real-time display of alignment potentials, feature vectors, trigger confidence, stimulus logs, and training curves; support for threshold / weight / safety limit configuration, single / continuous manual triggering, and remote termination.

[0102] This module simultaneously records operation logs and allows experts to replay raw and aligned data for offline annotation and model retraining; its display / command output communicates bidirectionally with the judgment module, ensuring that manual operations become part of the system data flow and are recorded.

[0103] Specifically, the system also includes a multimodal adapter interface for communicating with external training devices (such as clickers or food dispensers), enabling the triggering of the neural stimulation module to be superimposed with or mutually triggered by signals from traditional feedback devices. The adapter interface ensures that the timestamp is aligned with the alignment potential data during triggering.

[0104] All modules use a unified event sequence number and master control timestamp. Data packets carry source channels, coordinates, correction flags, and quality indicators. Each processing step adds an immutable log entry (time, input summary, output summary) for auditing and online learning.

[0105] As shown above, by real-time tracking of the electrode positions on the dog's head and alignment with a global clock, combined with high-quality EEG acquisition and pane-level feature extraction, repeatable and low-latency aligned potential data can be generated. Based on this data, the determination can accurately correspond to the dog's actual neural activity in time and space, thereby reducing false triggers caused by electrode drift or clock asynchrony, ensuring that stimuli are accurately delivered when the dog experiences the desired physiological event, and improving the consistency and reliability of the "stimulus-dog response" timing during training.

[0106] Example 2:

[0107] Non-invasive, modular closed-loop adaptation;

[0108] like Figure 3 As shown, applicable scenarios: short-term training programs in dog behavior training and attention enhancement laboratories / dog training centers (mainly for medium-sized dogs).

[0109] Subjects and Ethics:

[0110] Subjects: 12 adult medium-sized dogs (weighing 15–25 kg), mixed sex, aged 2–6 years.

[0111] Experiment approval: Approved by the animal experiment ethics committee of the institution.

[0112] Follow-up period: Baseline data was collected for each dog for 72 hours, followed by 4 weeks of training (2 training sessions per day, 20 minutes each time).

[0113] Hardware and Specifications:

[0114] EEG acquisition module: 8-channel array of flexible dry electrodes (layout covering the parietal bone and postauricular region), differential acquisition unit, sampling rate 1000 Hz, 16-bit ADC.

[0115] The target SNR is ≥ 40 dB (nominal value of the device), and the actual average measured value is 42 dB (see table).

[0116] Wireless transmission power consumption < 10 mW (8.2 mW measured under no-load conditions).

[0117] Electrode positioning and tracking (used for time-space alignment): optical markers + real-time positioning with a monocular structured light camera (positioning accuracy 1.5–3.0 mm, timestamp error <1 ms).

[0118] Alignment reference: The master base station provides a PTP-like local clock with a synchronization accuracy of ±1 ms.

[0119] Neurostimulation module (non-invasive electrical stimulation patch):

[0120] Current range 0.1–1.0 mA, adjustable; starting current 0.5 mA;

[0121] Frequency 1–20 Hz; pulse width adjustable from 100–500 μs (default 300 μs); waveform supports bipolar square wave.

[0122] Gradual escalation strategy: escalation time 200 ms; maximum single stimulus duration ≤ 2 s; stimulus interval ≥ 5 s (safety first).

[0123] Single-phase charge (1.0 mA, 500 μs) = 0.5 μC (far below the safety threshold for minimally invasive procedures).

[0124] Algorithm and Initialization:

[0125] Feature set (feature vector): Extracted every 500 ms window: theta wave power (4–8 Hz), α-θ ratio, instantaneous phase synchronization (across electrodes), short-time energy fluctuation (RMS), and attention index AI (time-frequency normalized synthesis, range 0–1).

[0126] Trigger the scoring function:

[0127]

[0128] Initial experimental parameters (initial settings in Example 1): α=0.40, β=0.40, γ=0.20, δ=5; threshold Θ0 =0.55 (initial).

[0129] Calculated by taking the median over a baseline 72 hours; The difference between the baseline and the baseline at time t; The success rate of the most recent 10 tasks is determined by a sliding window. Weighted normalization by time / frequency domain.

[0130] Online adaptive: When the success rate is > 0.7 in 20 consecutive triggers, the judgment module increases Θt by a step size of 0.02 (to make the triggering more conservative);

[0131] If the success rate is less than 0.4, then reduce Θt by the synchronization length;

[0132] The weights α / β / γ are finely adjusted every 2 days with an amplitude ≤ 0.05 to ensure smooth convergence.

[0133] Data Acquisition Instructions:

[0134] Raw EEG data: Flexible array → Differential acquisition → Wireless synchronization to main control (sample rate 1000 Hz).

[0135] Electrode position: Structured light camera + electrode optical markers, recording (x,y,z) coordinates and timestamps; used for alignment.

[0136] Behavioral labels and training results (used to calculate Ct and evaluate performance): high-resolution video (30 fps) + two independent annotators (behavioral event labels); supplemented by a food dispenser trigger sensor (recording dispensing timestamps) when necessary.

[0137] Stimulus response measurement: Changes in EEG characteristics within 2 seconds after stimulation, heart rate (collar heart rate sensor), and immediate behavior (video) were jointly assessed.

[0138] Training process:

[0139] S1 Baseline Acquisition: EEG was acquired during 72 hours of free movement to generate Pθ,base, AI baseline and electrode position baseline curves.

[0140] S2 Real-time Acquisition and Alignment: Real-time sampling → Upload acquired data along with positioning data → Time-space alignment module correction (position error correction, timestamp synchronization), output alignment potential data.

[0141] S3 Feature Extraction and Recognition: Calculate feature vectors every 500 ms and output behavioral confidence scores by the sequence classifier.

[0142] S4 Decision and Trigger: Calculate St. If St ≥ Θt, then trigger the stimulus and write it to the stimulus log (time, parameters).

[0143] S5 Feedback and Adaptation: Stimulus response data and behavioral results are fed back to the feature module and decision module, Θt and weights are updated online according to rules, and the training curve is recorded on the HMI.

[0144] Results (Experimental observation in Example 1, 12 dogs, 4 weeks of training):

[0145] Results: Compared with similar baselines (routine laboratory EEG recognition + fixed threshold triggering), Example 1 showed robust but moderate improvements in detection accuracy, false triggering rate and training success rate; the triggering delay was significantly shortened (improving user experience and training timeliness), and the stimulation safety was good (no adverse events were recorded).

[0146] In the experimental procedure of this embodiment, in addition to automatic triggering, trainers monitor the dog's behavior and EEG characteristics in real time through a mobile terminal application or remote control subunit. When the judgment module does not trigger but the trainer confirms that the dog has exhibited the desired behavior, a single pulse stimulus can be manually sent through the remote control subunit, with the stimulus parameters executed according to the current step-wise settings. Manually triggered events and automatically triggered events are recorded together in the stimulus log for subsequent analysis of trigger consistency and model fine-tuning.

[0147] Example 3:

[0148] Multimodal enhancement, differentiated adaptive strategies, and peripheral device interaction;

[0149] Applicable scenarios: Dog training centers and behavioral rehabilitation centers; training programs requiring multimodal reinforcement (stimuli + sound and light + food). Compared to Example 2, Example 3 further innovates and enhances the judgment strategy, peripheral device linkage, and parameter adaptation.

[0150] Subjects: 16 adult dogs (different from those in Example 2), with other conditions similar; baseline data collection was also conducted at 72 hours.

[0151] Key differences (relative to Implementation Example 2):

[0152] Multimodal adaptation interface: The system is linked with the food dispenser and clicker (acoustic stimulator) through a low-latency wired / wireless TTL interface to ensure that the timestamps of all peripherals are aligned with the alignment potential data (timing error <2 ms).

[0153] Enhanced electrode position correction: Combining optical positioning with online estimation of electrode-scalp contact resistance further reduces effective spatial error (actual 1.2–2.0 mm).

[0154] Parameter adjustment of the judgment function: Example 2 adopts a weight combination that emphasizes baseline changes and a higher δ to improve the sensitivity to short-term theta wave changes. Initial parameters: α=0.45, β=0.35, γ=0.20, δ=8; threshold Θ0 = 0.6.

[0155] A more refined stepwise stimulation strategy: the initial current is 0.4 mA, and after a successful alarm, it is increased by 0.08 mA (more refined than Example 1), with a maximum of no more than 0.9 mA; the stimulation delay time is initially 0.5 s, and gradually extended to no more than 3 s to improve task similarity.

[0156] Enhanced online learning mechanism: Adaptive dual-indicator approach: a short window (last 10 times) for rapid threshold fine-tuning; a long window (last 200 times) for weight fine-tuning, with a weight adjustment step size ≤ 0.03.

[0157] Data acquisition and measurement methods:

[0158] EEG, positioning, video, and peripheral (food / acoustic) trigger timestamps are all synchronized using the same main controller local clock (error <1 ms).

[0159] Training effectiveness evaluation includes safety indicators such as behavioral task completion rate, stimulus false trigger rate, detection accuracy, trigger delay, SNR, electrode alignment error, single-phase charge quantity and heart rate fluctuation.

[0160] Training process

[0161] The process is the same as in Example 2, but the following is added to S4 determination and triggering: the determination is made through multimodal logic (e.g., food delivery + micro-stimulus superposition is only executed when St exceeds the threshold and peripheral events (such as sound prompts) / behavioral triggers are consistent), and all peripheral timings are recorded for subsequent analysis.

[0162] In step S5 feedback and adaptation, in addition to EEG feedback, behavioral sensors (food dispenser trigger recording) are used first to determine whether the current stimulus produces positive reinforcement, thereby deciding whether to upregulate the stepwise stimulus.

[0163] In this embodiment, the remote control subunit is linked with the multimodal adaptation interface. When the trainee manually triggers the system via the remote control, in addition to outputting neural stimulation, the system can simultaneously trigger external training devices such as a food dispenser or a clicker, forming a multimodal positive reinforcement combination. The timestamp of the manual trigger is strictly synchronized with the alignment potential data. The system automatically compares the changes in EEG characteristics before and after the manual trigger to optimize the weighting coefficients in the automatic trigger scoring function, so that the automatic trigger gradually approaches the timing and confidence level of human judgment.

[0164] Results (Experimental observation in Example 3, 16 dogs, 4 weeks of training);

[0165] Results: Example 3 shows a slight improvement in detection accuracy, false trigger rate, trigger latency, and training success rate compared to Example 2; multimodal linkage results in a higher success rate for positive reinforcement and faster training convergence; the online adaptive strategy demonstrates better long-term stability.

[0166] Comprehensive comparison data (Table: Baseline vs. Example 2 vs. Example 3);

[0167] Note: Baseline = Existing standard laboratory system (8-channel dry electrodes, fixed threshold triggering, electrodeless optical positioning correction, no multimodal linkage). All data in the table are average values ​​(mean ± SD) obtained from experiments conducted with 12–16 dogs according to the above experimental setup.

[0168] Table 1

[0169] Indicators (definition / measurement description below) Baseline system (existing) Example 2 (Invention) Example 3 (Enhanced Version of the Invention) EEG SNR (dB) 30 ± 3 42 ± 2 44 ± 2 Effective electrode positioning error (mm, after optical correction) 8.0 ± 2.1 2.2 ± 0.6 1.6 ± 0.5 Alignment timestamp error (ms) 5.2 ± 1.8 <1.0 ± 0.3 <1.0 ± 0.2 Target behavior detection accuracy (label matching, %) 75.0 ± 4.5 82.3 ± 3.8 85.1 ± 3.2 False alarm rate (false trigger % / total number of triggers) 18.0 ± 3.0 10.2 ± 2.6 8.1 ± 2.2 Average trigger delay (from feature window boundary to stimulus, ms) 250 ± 40 120 ± 25 95 ± 20 Training task success rate (training task completed / attempted, %) 58.0 ± 5.0 68.5 ± 4.2 72.0 ± 3.8 Stimulation safety indicator: Maximum single-phase charge (μC) 0.6 ±0.05 0.5 ±0.02 0.45 ±0.02 Heart rate abnormalities (beats / 1000 stimulations) 2.8 0 (recorded) 0 (recorded) System end-to-end data packet loss rate (%) 1.5 ±0.6 0.2 ±0.1 0.2 ±0.1

[0170] Meaning and acquisition methods of indicators:

[0171] EEG SNR (dB): Logarithmic ratio of signal power to noise power. Acquisition method: Calculate the average power of the target frequency band as the signal during the resting phase (dog stationary, no significant movement); estimate the noise during the moving / no-task phase. The measuring equipment is calibrated and averaged through repeated sampling.

[0172] Effective electrode positioning error (mm): The root mean square (RMS) of the difference between the actual electrode coordinates and the reference coordinates (structured light high-precision scanning). Acquisition method: Obtain a reference by scanning the scalp with structured light, then compare the positioning results with those obtained from the camera.

[0173] Alignment timestamp error (ms): The maximum absolute difference in timestamps between different acquisition sources (EEG, camera, peripherals) within the system. Acquisition method: Measuring the time difference recorded by the same event (TTL) in each subsystem.

[0174] Target behavior detection accuracy (%): The consistency rate between positive samples identified by the system and manually annotated samples (two independent annotators, with conflicts arbitrated by a third party). Acquisition method: Blind annotation comparison of 200–500 behavior event segments.

[0175] False Alarm Rate (%): The percentage of system-triggered events that, according to manual annotations, should not have been triggered. Obtained by comparing trigger logs with manual annotations.

[0176] Average trigger delay (ms): The average delay from the end of the characteristic time window to the output of the neural stimulation (including computational delay, communication delay, and electrical stimulation initiation time). Acquisition method: Recording and statistically analyzing timestamps within the system.

[0177] Training task success rate (%): The number of times the dog successfully completed the task / number of attempts in a given training task (e.g., "sit and hold for 5 seconds"). Acquisition method: Video annotation + external sensor (e.g., food dispenser) verification.

[0178] Stimulation safety indicator: Maximum single-phase charge (μC): Single pulse current × pulse width (A × s -> C), converted to μC, to ensure it is below the safety recommendation value. Acquisition method: Calculated from stimulation parameter logs.

[0179] Abnormal heart rate events (beats / 1000 stimuli): The number of events in which the measured heart rate exceeds baseline ±30% within 30 seconds after triggering (can be used as an indicator of stimulus-related adverse reactions). Acquisition method: Recorded by a collar-type heart rate sensor and aligned with the stimulus log.

[0180] System end-to-end data packet loss rate (%): The percentage of packet loss in wireless / wired transmission, affecting real-time performance. Acquisition method: The ratio of received packet loss counts to the total number of transmitted packets recorded by the main control unit.

[0181] As can be seen from the above, this invention uses the multidimensional EEG characteristics and behavioral feedback of dogs as input, employs a scoring function and short / long-term adaptive strategies to dynamically adjust the trigger threshold and stimulation parameters, and gradually converges to the optimal training strategy for the dog under the synergy of multimodal positive reinforcement (such as food or sound and light). This closed loop can accelerate the dog's learning convergence, and significantly reduce the risk of overstimulation of the dog by limiting the stimulation intensity through a gradual escalation strategy and real-time monitoring, thus balancing training effectiveness and animal welfare.

[0182] This invention applies the principle of operant conditioning, and the stimulation of the neural stimulation module can also be triggered by a remote control when the dog exhibits the desired behavior.

[0183] like Figure 5 As shown, in one embodiment, this application also provides a canine EEG-responsive neural stimulation training control method, the control method comprising the following steps:

[0184] Collect raw EEG data from the dog's head and output the raw EEG data;

[0185] Electrode position correction and global time synchronization processing are performed on the raw EEG data to obtain aligned potential data;

[0186] The alignment potential data is subjected to noise reduction, spectral and time-frequency feature extraction, and target behavior-related feature vectors are identified based on the features.

[0187] Based on the feature vector, a trigger determination is calculated and the stimulus parameter setting is output.

[0188] The appropriate neural stimulation signal is output to the dog according to the stimulation parameters and the stimulation response is fed back to the feature vector.

[0189] Display the alignment potential data, feature vectors, trigger determination, and stimulus records, and perform manual intervention.

[0190] Furthermore, the stimulation parameters are adjusted in a stepwise, individualized manner:

[0191] The initial current is 0.5 mA, and the current increases by 0.1 mA after each successful alarm until it does not exceed 1.0 mA.

[0192] The delay time of the stimulus signal is gradually extended from instantaneous triggering to no more than 3 seconds to simulate real task scenarios;

[0193] Multimodal reinforcement strategies are implemented through manual intervention via remote control using human-computer interaction or by receiving superimposed trigger signals from external training devices. The manual intervention includes manually triggering the neural stimulation module through the remote control subunit to achieve behavioral reinforcement based on the principle of operant conditioning.

[0194] Its beneficial effects are the same as those of the embodiment of the canine EEG-responsive neurostimulation training system, and will not be repeated here.

[0195] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A canine EEG-responsive neurostimulation training system, characterized in that, include: The EEG acquisition module is used to acquire raw EEG data from the dog's head and output the raw EEG data. The time-space alignment module is used to perform electrode position correction and global time synchronization processing on the raw EEG data to obtain aligned potential data. The feature extraction and recognition module is used to perform noise reduction, spectral and time-frequency feature extraction on the alignment potential data, and to identify target behavior-related feature vectors based on the features. The determination and parameter adaptation module is used to calculate the trigger determination based on the feature vector and output the stimulus parameter setting. The neural stimulation module is used to output an appropriate neural stimulation signal to the dog according to the stimulation parameters and to feed back the stimulation response to the feature extraction and recognition module. The human-computer interaction module is used to display the alignment potential data, feature vectors, trigger judgments and stimulus records, and to perform manual intervention. The manual intervention includes manually triggering the neural stimulation module through the remote control subunit to achieve behavioral reinforcement based on the principle of operant conditioning.

2. The canine EEG-responsive neurostimulation training system according to claim 1, characterized in that, The EEG acquisition module includes: Flexible dry electrode array, differential signal acquisition unit and wireless transmission submodule; The signal-to-noise ratio of the differential signal acquisition unit is >40dB; The power consumption of the wireless transmission submodule is <10mW; The flexible dry electrode array is positioned to fit the physiological structure of the canine skull and is placed behind the ear or in the parietal region.

3. The canine EEG-responsive neurostimulation training system according to claim 1, characterized in that, The time-space alignment module corrects the raw EEG data by fusing electrode position tracking information with a global time reference, thereby eliminating electrode positioning errors and clock drift, and outputting high-precision aligned potential data.

4. The canine EEG-responsive neurostimulation training system according to claim 1, characterized in that, The feature extraction and recognition module has a built-in lightweight sequence classifier, which is used to calculate the features of the theta wave power spectral density, α-θ ratio, instantaneous phase synchronization and attention index based on the aligned potential data, and combine the features into a feature vector.

5. The canine EEG-responsive neurostimulation training system according to claim 1, characterized in that, The judgment and parameter adaptation module is based on the following triggering scoring function. Determine if a stimulus has been triggered: in, Let θ be the change in θ-wave power relative to the baseline over time t. Baseline theta wave power; The attention index is calculated based on time-domain and frequency-domain features; denoted as the recent target behavior success rate (0–1); α, β, γ are weighting coefficients and α+β+γ=1; δ is a positive adjustment constant; the stimulus is triggered when St≥Θt, where the threshold Θt is updated online smoothly based on training progress and safety constraints.

6. The canine EEG-responsive neurostimulation training system according to claim 1, characterized in that, The neurostimulation module is a non-invasive neurostimulation unit or a minimally invasive patch structure. The minimally invasive patch structure includes an adjustable current source and a bipolar / multipolar electrode patch. The current source has an output range of 0.1–1.0 mA, an adjustable frequency of 1–20 Hz, a sine / pulse waveform, and a pulse width range of 100–500 μs.

7. The canine EEG-responsive neurostimulation training system according to claim 1, characterized in that, The human-computer interaction module includes a mobile terminal application and a remote control subunit; The mobile terminal application is used to display the alignment potential data, feature vectors, trigger confidence, stimulus records and training curves, and can set the maximum stimulus value, safety strategy, manual triggering and intervention. The remote control subunit supports two manual modes: single pulse and continuous stimulation, as well as delay time adjustment. It is used by the trainer to manually trigger the neural stimulation module when the dog exhibits the desired behavior, so as to achieve positive reinforcement intervention based on the principle of operant conditioning.

8. The canine EEG-responsive neurostimulation training system according to claim 1, characterized in that, The system also includes a multimodal adapter interface for communicating with external training devices, enabling the triggering of the neurostimulation module to be superimposed with or mutually triggered by signals from traditional feedback devices. The adapter interface ensures that the timestamp is aligned with the alignment potential data during triggering.

9. A method for training and controlling canine EEG-responsive neural stimulation, characterized in that, Applied to the system as described in any one of claims 1-8, the control method includes the following steps: Collect raw EEG data from the dog's head and output the raw EEG data; Electrode position correction and global time synchronization processing are performed on the raw EEG data to obtain aligned potential data; The alignment potential data is subjected to noise reduction, spectral and time-frequency feature extraction, and target behavior-related feature vectors are identified based on the features. Based on the feature vector, a trigger determination is calculated and the stimulus parameter setting is output. The appropriate neural stimulation signal is output to the dog according to the stimulation parameters and the stimulation response is fed back to the feature vector. Display the alignment potential data, feature vectors, trigger determination, and stimulus records, and perform manual intervention.

10. A canine EEG-responsive neural stimulation training and control method according to claim 9, characterized in that, The stimulation parameters are adjusted in a stepwise, individualized manner: The initial current is 0.5 mA, and the current increases by 0.1 mA after each successful alarm until it does not exceed 1.0 mA. The delay time of the stimulus signal is gradually extended from instantaneous triggering to no more than 3 seconds to simulate real task scenarios; Multimodal reinforcement strategies are implemented through manual intervention via remote control using human-computer interaction or by receiving superimposed trigger signals from external training devices. The manual intervention includes manually triggering the neural stimulation module through the remote control subunit to achieve behavioral reinforcement based on the principle of operant conditioning.

Images