Electromyographic signal classification detection method and system combining spiking neural network and super-dimensional calculation

Abstract

The invention belongs to the technical field of electromyographic signal classification detection, discloses an electromyographic signal classification detection method combining SNN and HDC, provides an electromyographic signal classification detection framework combining SNN and HDC, and aims to realize ultra-low power consumption operation. In the framework, the SNN executes event-driven feature extraction by using a random untrained weight, so that the calculation overhead is reduced to the greatest extent; and the HDC realizes anti-noise classification through high-dimensional representation. The integration of the two not only realizes real-time detection of energy conservation, but also is particularly suitable for being applied to resource-limited scenes such as wearable equipment and the like. The method provided by the invention realizes 95% of average accuracy (the peak accuracy is 96.44%) in three types of fatigue recognition tasks, and the training speed is 5.7 times faster than that of a one-dimensional convolutional neural network (1D-CNN) and 45 times faster than that of a five-dimensional long-short-term memory network (5D-LSTM). Even under the condition that only 20% of training data is used, the method can still keep the accuracy of 90% or above, and the high efficiency and robustness of the method in actual deployment are fully proved.

Description

Technical Field

[0001] The present invention belongs to the technical field of myoelectric signal classification detection, and in particular relates to a myoelectric signal classification detection method and system combining a pulse neural network and hyperdimensional computing. Background Art

[0002] Muscle fatigue is a common physiological phenomenon that usually occurs during continuous or repetitive muscle contractions. Its main characteristics include a gradual decrease in muscle strength, slower movement speed, and reduced endurance. The occurrence of this phenomenon is related to various factors, such as the depletion of energy reserves, the accumulation of metabolic byproducts, and the reduction in the efficiency of nerve impulse transmission. Muscle fatigue not only affects daily activities and work efficiency, but also may increase the risk of sports injuries and chronic musculoskeletal diseases. Therefore, accurately identifying and assessing muscle fatigue is of great significance for optimizing physical function, preventing health problems, and improving the practical effects of sports medicine, rehabilitation science, and ergonomics.

[0003] Surface electromyography (sEMG) signals are a non-invasive technique that effectively measures the electrical activity of muscle motor units. Due to their high sensitivity to changes in muscle state and their real-time, multi-channel nature, sEMG signals are valuable in assessing muscle fatigue. However, with the increasing demand for low-power edge devices in complex applications, traditional processing systems face significant challenges in handling the data-intensive tasks associated with sEMG signal analysis. These challenges include limited computing resources, high energy consumption, and the need for real-time and robustness.

[0004] In recent years, deep learning-based methods have been widely used in EMG signal classification. Most studies have demonstrated the effectiveness of these methods in EMG signal classification by extracting time-domain and frequency-domain features for model training. However, despite the excellent classification performance of deep learning models, their high computational cost and resource requirements have become major bottlenecks in practical applications. An ideal EMG signal classifier for edge devices needs to meet the following requirements: lightweight design to reduce computational and memory overhead, high energy efficiency to optimize battery life, strong robustness to noise and hardware failures, small-sample learning capabilities to reduce dependence on labeled data, and local processing capabilities to ensure data privacy. Traditional deep learning models struggle to meet these requirements simultaneously.

[0005] Muscle fatigue, defined here as localized neuromuscular fatigue, can be effectively assessed using surface electromyography (sEMG), a non-invasive biosignal that captures neuromuscular activity. sEMG reflects fatigue by shifting the spectrum toward higher frequencies during muscle movement, offering the advantage of real-time monitoring compared to invasive biochemical or imaging methods. While alternative methods such as mechanical electromyography exist, sEMG remains the most established method for fatigue characterization due to its reliability and minimal signal attenuation.

[0006] EMG signal classification method based on sEMG

[0007] When local muscles fatigue, significant changes in the temporal and frequency domain characteristics of surface electromyography (sEMG) signals can be observed. These characteristic changes in sEMG signals have become the focus of much research on EMG signal classification. However, most existing classification methods rely on single neural networks, which often suffer from high computational complexity and insufficient robustness.

[0008] Wang et al. collected sEMG signals from participants during cycling and used an LSTM network for EMG signal classification. While LSTM excels at capturing temporal dependencies, its gating mechanism and long-term memory require more parameters and computational resources, resulting in increased computational overhead. X. Wang et al. addressed this issue by dividing the processed EMG signals into seven segments, extracting features, and then applying various algorithms, including nearest neighbor, logistic regression, and support vector machines (SVM), to achieve three-level fatigue classification. While SVM achieved the best classification results among these methods, it requires a large amount of space to store training samples and kernel matrices. Furthermore, the computational complexity of SVM increases when processing high-dimensional features, resulting in limited robustness. Similarly, Q. Liu et al. proposed an improved whale optimization algorithm combined with SVM to improve the accuracy of dynamic fatigue classification. Despite this optimization, SVM still faces fundamental challenges in computational efficiency and robustness when processing high-dimensional data.

[0009] Spiking Neural Networks

[0010] Spiking neural networks (SNNs) are computational models that simulate the communication processes between neurons in biological nervous systems. They differ significantly from traditional artificial neural networks. While traditional neural networks represent and transmit information through continuous values, SNNs use discrete pulses or impulses at specific time points as information carriers, more closely resembling how the brain processes information in the real world. This unique operating mechanism enables SNNs to excel in applications requiring low energy consumption and fast response times.

[0011] To more accurately simulate biological neural systems, the leaky integrate-and-fire (LIF) model is adopted as one of the basic architectures. The LIF model provides a simplified but effective mathematical description of the behavior of a single neuron. In this model, each neuron acts as a capacitor with leaky characteristics: when the accumulated stimulation signals from other neurons or the external environment reach a certain threshold, the neuron triggers a "discharge" event, generating a spike output. After the discharge, the neuron enters a brief refractory period, during which it does not respond to new stimuli until it recovers. This mechanism not only reflects the basic working principle of biological neurons, but also enables networks built based on this model to process information with high temporal precision, making it well-suited for tasks requiring strict real-time performance.

[0012] Hyperdimensional computing

[0013] Hyperdimensional computing (HDC) is a computing paradigm inspired by neurons in the brain. Unlike traditional numerical computing, HDC uses high-dimensional vectors to represent data, with dimensions reaching hundreds of thousands or even millions. This representation gives HDC a unique advantage in handling complex pattern recognition and memory tasks.

[0014] Traditional numerical computing typically represents data as scalars or low-dimensional vectors, which leads to the "curse of dimensionality" when processing high-dimensional data. As the number of dimensions increases, data sparsity rises rapidly, leading to a sharp increase in computational complexity. HDC alleviates this problem by representing data as signal patterns in a high-dimensional space. In HDC, each data item is encoded as a high-dimensional binary vector, called a supervector. The dimensions of these supervectors are typically much larger than those used in traditional numerical computing. Supervectors from the same class are combined to form a supervector for a specific class. To improve accuracy, HDC performs self-checking during the iterative process: vectors that are misclassified are added to the correct supervector class, while they are removed from the originally incorrect class. This classification check uses the Hamming distance to query the most similar supervector.

[0015] Through the above analysis, the problems and defects of the existing technology are as follows:

[0016] While deep learning models offer impressive classification performance, their high computational cost and resource requirements pose a significant bottleneck in practical applications. An ideal EMG classifier for edge devices needs to meet the following requirements: lightweight design to reduce computational and memory overhead, high energy efficiency to optimize battery life, robustness to noise and hardware failures, small-sample learning capabilities to reduce reliance on labeled data, and local processing capabilities to ensure data privacy. Traditional deep learning models struggle to simultaneously meet these requirements. Summary of the Invention

[0017] In view of the problems existing in the prior art, the present invention provides a myoelectric signal classification and detection method combining SNN and HDC.

[0018] The present invention is implemented as follows: a myoelectric signal classification detection method combining SNN and HDC includes:

[0019] Step 1, preprocessing;

[0020] Step 2: Extract features using SNN;

[0021] Step 3, encoding;

[0022] Step 4: Train with HDC.

[0023] Furthermore, the pre-processing:

[0024] A 20-350 Hz bandpass filter was applied to improve signal quality while preserving fatigue-related signal components; a 150-250 ms time window was used to balance classification latency and accuracy.

[0025] In terms of feature extraction, time domain and frequency domain analysis are combined to capture the non-stationary characteristics of EMG signals; eight clinically significant features are extracted; time domain features include integrated EMG and mean absolute value and zero-crossing rate; frequency domain features include mean frequency and median frequency; peak frequency and frequency ratio; and wavelet energy.

[0026] Furthermore, the feature extraction using SNN is:

[0027] A spiking neural network is used as the core feature extractor to extract temporal dynamic features from raw electromyographic signals. The network design integrates the biomimetic properties of biological neurons with engineering optimization strategies.

[0028] After the data is transferred to the memory, it will be sent to the SNN for processing. The LIF neuron model is used as the core computing unit of the SNN. The membrane potential V (t) It evolves according to the following differential equation:

[0029]

[0030] where τ is the membrane time constant, I (t) represents the synaptic input current; when V (t) When a predefined threshold is exceeded, the neuron fires a spike and the membrane potential is reset to its baseline value, V rest ; After discretization, the membrane potential update rule becomes

[0031] Where τ controls the potential decay rate, W∈R N×Mis the weight matrix of the fully connected layer, input X [t] is the myoelectric signal after standardization and row-level normalization; when the membrane potential exceeds the threshold V th When , the neuron triggers a pulse to reset the potential to the resting state; this mechanism simulates the action potential discharge process of biological neurons and converts continuous signals into sparse pulse trains; captures the instantaneous muscle activation pattern of the electromyographic signal, and the input data is repeatedly input into the SNN in T time steps; the pulse output s for each time step is s [t] ∈{0,1} M Accumulate and calculate the average issuance rate:

[0032]

[0033] Eigenvector F∈R M Each element of represents the activation frequency of the corresponding neuron in the time window, compressing the dynamic spike activity into a static statistic;

[0034] SNN uses inverse tangent instead of gradient to achieve end-to-end training:

[0035]

[0036] This function provides a smooth gradient approximation near the impulse trigger point (x=0) so that the error can be used to update the weight matrix W through the standard backpropagation algorithm; the weights are initialized using a Gaussian distribution with mean 0 and standard deviation 1.

[0037] Furthermore, the encoding:

[0038] The encoding module maps the spatiotemporal features extracted by the SNN hidden layer into a high-dimensional binary space suitable for HDC; let the feature vector output by the SNN hidden layer be f∈R H , where H is the hidden layer dimension, using the random projection matrix W∈{-1,+1} H×D (D is the dimension of the hyperdimensional space) to perform linear transformation:

[0039] hV=f·W#(5)

[0040] Furthermore, the HDC training:

[0041] The training process is divided into two main stages: encoding and hyperdimensional classification;

[0042] The first stage (hyperdimensional coding): During the training process, the feature vector f∈R extracted by SNN H By random projection matrix W∈{-1,+1} H×D Mapping to D-dimensional hyperdimensional space:

[0043] hv=f·W∈R D #(6)

[0044] The projection result is binarized using the symbolic function to generate a hyperdimensional vector H∈{-1,+1} D :

[0045] H=sign(h)#(7)

[0046] The second stage (hyperdimensional classification): During the training process, the system accumulates hyperdimensional vectors of the same class into corresponding class prototypes based on sample labels:

[0047]

[0048] Among them C k ∈Z D is the prototype vector of class k.

[0049] Another object of the present invention is to provide a myoelectric signal classification and detection system combining SNN and HDC, comprising:

[0050] The preprocessing module is used to extract k types of time domain and frequency domain features from the original signal.

[0051] SNN feature extraction module for processing these features through temporal dynamics using a spiking neural layer with leaky integrate-and-fire neurons;

[0052] The encoding module is used to map the spatiotemporal features extracted by the SNN hidden layer into a high-dimensional binary space suitable for HDC;

[0053] The training module encodes the pulse patterns into a D-dimensional supervector through a bundling operation, followed by Hamming distance-based classification and testing based on pre-trained prototypes, where the system validation evaluates the classification robustness under signal variations.

[0054] Another object of the present invention is to provide a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the electromyographic signal classification and detection method combining SNN and HDC.

[0055] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, enables the processor to perform the steps of the electromyographic signal classification and detection method combining SNN and HDC.

[0056] Another object of the present invention is to provide an information data processing terminal, which is used to implement the electromyographic signal classification and detection system combining SNN and HDC.

[0057] In combination with the above technical solutions and the technical problems solved, please analyze the advantages and positive effects of the technical solutions to be protected by the present invention from the following aspects:

[0058] First, emerging neuromorphic computing models offer a highly promising solution. These models can perform complex computing tasks with ultra-low power consumption, making them particularly well-suited for edge computing scenarios. Furthermore, this invention addresses the core issues of traditional deep learning models in myoelectric signal classification, such as high computational cost, large resource requirements, insufficient robustness, and reliance on large amounts of labeled data. By combining this with edge computing, neuromorphic models can enable local data processing, reduce reliance on cloud computing, enhance data privacy protection, and significantly reduce overall energy consumption.

[0059] During the preprocessing phase, a 20-350Hz bandpass filter and a 150-250ms time window are combined to effectively suppress noise while preserving fatigue-related signal components, resolving the trade-off between signal fidelity and computational latency inherent in traditional methods. A multi-domain feature extraction strategy, integrating eight clinical features—time-domain integrated EMG, zero-crossing rate, frequency-domain median frequency, and wavelet energy—addresses the limitations of single-domain analysis in characterizing non-stationary EMG signals, providing more discriminative feature primitives for subsequent processing.

[0060] Among them, spiking neural networks (SNNs) and hyperdimensional computing (HDC) demonstrate unique advantages: SNNs achieve efficient and energy-efficient computing through event-driven processing and temporal information encoding, while HDC excels at robust pattern recognition using high-dimensional distributed representations. Unlike the continuous computing paradigm of traditional artificial neural networks, SNNs' biologically inspired architecture enables efficient processing of temporal and sparse signals, making them particularly suitable for processing dynamic data such as electromyographic signals.

[0061] This paper proposes a framework for myoelectric signal classification and detection that combines SNNs and HDCs, aiming to achieve ultra-low power operation. In this framework, the SNN utilizes random, untrained weights to perform event-driven feature extraction, minimizing computational overhead; while the HDC achieves noise-resilient classification through high-dimensional representation. This integration not only enables energy-efficient real-time detection but is also particularly well-suited for resource-constrained scenarios such as wearable devices.

[0062] To validate the effectiveness of the framework, experimental evaluations were conducted on a self-collected biceps fatigue dataset and a public dataset. Experimental results show that the proposed method achieves an average accuracy of 95% (with a peak accuracy of 96.44%) in three fatigue recognition tasks, and is trained 5.7 times faster than a one-dimensional convolutional neural network (1D-CNN) and 45 times faster than a five-dimensional long short-term memory network (5D-LSTM). Notably, even when using only 20% of the training data, the method still maintains an accuracy of over 90%, fully demonstrating its efficiency and robustness in practical deployments.

[0063] The main contributions of the present invention are as follows:

[0064] 1. Compared with traditional models, the proposed deep spiking neural network (DSN) combining SNN and HDC achieves 95.31% classification accuracy in 1.25 seconds, which is 5.7 times faster than 1D-CNN (7.16 seconds) and 45.2 times faster than LSTM (56.51 seconds) in three-category tasks. At the same time, it achieves 98.53% accuracy in two-class tasks, which is comparable to the peak performance of LSTM.

[0065] 2. Unlike most studies that focus solely on a single fatigue stage, our method utilizes a data extraction strategy that tracks the entire fatigue process, from relaxation to exertion. This comprehensive analysis significantly improves the accuracy and efficiency of assessments while ensuring research reliability.

[0066] 3. Compared with traditional deep learning models, the DSN method of the present invention uses a single-layer SNN with random weights as the feature extractor of the HDC layer, avoiding the complex weight training process and significantly reducing computational complexity and energy consumption. This feature makes it particularly suitable for real-time EMG signal classification and detection in resource-constrained embedded systems and edge devices.

[0067] 4. By segmenting multi-channel EMG data into segments of equal length and performing feature extraction and classification on each segment, the model's robustness and classification accuracy are further improved. This multi-segment feature processing method can more comprehensively reflect the dynamic changes in muscle fatigue, especially in complex and diverse exercise environments.

[0068] The EMG signal classification framework combining SNN and HDC proposed in this paper shows significant advantages in accuracy, efficiency and energy consumption, providing an efficient and reliable solution for real-time EMG signal analysis in wearable devices and edge computing scenarios.

[0069] Second, the architecture of this invention can provide low-cost intelligent monitoring technology for wearable healthcare or muscle monitoring devices. Medical-grade electromyography diagnostic patches can be developed, leveraging real-time muscle fatigue analysis to expand into the fitness market and deploy on domestic hardware.

[0070] Existing international EMG classification technologies have long suffered from low accuracy, high power consumption, and weak robustness. While traditional SNN solutions offer the advantage of low power consumption (such as the Intel Loihi chip), they are limited by their inadequate spatiotemporal feature extraction capabilities and susceptibility to interference. While HDC, when used alone, offers noise immunity, it cannot effectively process the dynamic changes in EMG signals. This invention achieves a deep fusion of SNN feature extraction and HDC's hyperdimensional computing and classification, achieving an average accuracy of 95% on real-world datasets. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 This is a flow chart of the myoelectric signal classification and detection method combining SNN and HDC provided by an embodiment of the present invention.

[0072] Figure 2 This is a structural block diagram of the electromyographic signal classification and detection system combining SNN and HDC provided in an embodiment of the present invention.

[0073] Figure 3 It is a design diagram of the algorithm structure provided by an embodiment of the present invention.

[0074] Figure 4 This is a graph showing changes in accuracy for different data amounts under different noise probabilities provided by an embodiment of the present invention.

[0075] Figure 5 This is a diagram showing the changes in accuracy of different models provided by an embodiment of the present invention under different bit error rates. DETAILED DESCRIPTION

[0076] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0077] like Figure 1 As shown, an embodiment of the present invention provides a myoelectric signal classification and detection method combining SNN and HDC, comprising the following steps:

[0078] S101, pretreatment;

[0079] S102, extract features using SNN;

[0080] S103, coding;

[0081] S104, trained with HDC.

[0082] 1. During the preprocessing phase, the method first filters the raw electromyographic (EMG) signal using a bandpass filter (passband range 20-350Hz) to eliminate low-frequency baseline drift and high-frequency noise, while retaining mid- and high-frequency components closely associated with muscle fatigue and contraction. The filtered signal is then segmented into 150-250 millisecond time windows to minimize classification latency while maintaining signal integrity, meeting the needs of real-time detection.

[0083] 2. In terms of feature extraction, this method comprehensively considers the non-stationary nature of electromyographic signals, combining time-domain and frequency-domain analysis to extract eight clinically significant features. Time-domain features include integrated EMG, mean absolute value, and zero-crossing rate, which measure the intensity and periodicity of muscle activity. Frequency-domain features include parameters such as mean frequency, median frequency, peak frequency, and frequency ratio to characterize the signal's spectral distribution and energy concentration. Furthermore, wavelet transforms can characterize the transient characteristics of signals at multiple scales, providing richer feature representations for subsequent classification.

[0084] 3. The extracted multidimensional feature vectors undergo deep encoding and feature enhancement using a spiking neural network (SNN). The SNN captures the dynamic changes in the EMG signal through the timing characteristics of pulse emission, mapping continuous feature vectors into pulse trains and preserving the key information of the EMG signal through sparse emission in the time domain. The core of this stage is to utilize membrane potential updates and synaptic plasticity mechanisms to achieve adaptive learning of the input signal, thereby improving robustness to noise and variability and compressing redundant features.

[0085] 4. The final step uses high-dimensional computing (HDC) to train and classify the feature sequences processed by the SNN. HDC maps features into an ultra-high-dimensional space, encodes the characteristic patterns corresponding to the pulse sequence into supervectors, and distinguishes different categories based on vector operations. Compared with traditional low-dimensional classification methods, HDC is more tolerant to slight feature perturbations in high-dimensional space and can achieve stable classification performance with fewer training samples, effectively improving the robustness of EMG signal classification and detection.

[0086] The preprocessing provided by the embodiment of the present invention is as follows:

[0087] A 20-350 Hz bandpass filter was applied to improve signal quality while preserving fatigue-related signal components; a 150-250 ms time window was used to balance classification latency and accuracy.

[0088] In terms of feature extraction, time domain and frequency domain analysis are combined to capture the non-stationary characteristics of EMG signals; eight clinically significant features are extracted; time domain features include integrated EMG and mean absolute value and zero-crossing rate; frequency domain features include mean frequency and median frequency; peak frequency and frequency ratio; and wavelet energy.

[0089] The embodiment of the present invention provides the use of SNN to extract features:

[0090] A spiking neural network is used as the core feature extractor to extract temporal dynamic features from raw electromyographic signals. The network design integrates the biomimetic properties of biological neurons with engineering optimization strategies.

[0091] After the data is transferred to the memory, it will be sent to the SNN for processing. The LIF neuron model is used as the core computing unit of the SNN. The membrane potential V (t) It evolves according to the following differential equation:

[0092]

[0093] where τ is the membrane time constant, I (t) represents the synaptic input current; when V (t) When a predefined threshold is exceeded, the neuron fires a spike and the membrane potential is reset to its baseline value, V rest ; After discretization, the membrane potential update rule becomes

[0094] Where τ controls the potential decay rate, W∈R N×M is the weight matrix of the fully connected layer, input X [t] is the myoelectric signal after standardization and row-level normalization; when the membrane potential exceeds the threshold V th When , the neuron triggers a pulse to reset the potential to the resting state; this mechanism simulates the action potential discharge process of biological neurons and converts continuous signals into sparse pulse trains; captures the instantaneous muscle activation pattern of the electromyographic signal, and the input data is repeatedly input into the SNN in T time steps; the pulse output s for each time step is s [t] ∈{0,1} M Accumulate and calculate the average issuance rate:

[0095]

[0096] Eigenvector F∈R M Each element of represents the activation frequency of the corresponding neuron in the time window, compressing the dynamic spike activity into a static statistic;

[0097] SNN uses inverse tangent instead of gradient to achieve end-to-end training:

[0098]

[0099] This function provides a smooth gradient approximation near the impulse trigger point (x=0) so that the error can be used to update the weight matrix W through the standard backpropagation algorithm; the weights are initialized using a Gaussian distribution with mean 0 and standard deviation 1.

[0100] The encoding provided by the embodiment of the present invention is:

[0101] The encoding module maps the spatiotemporal features extracted by the SNN hidden layer into a high-dimensional binary space suitable for HDC; let the feature vector output by the SNN hidden layer be f∈R H , where H is the hidden layer dimension, using the random projection matrix W∈{-1,+1} H×D (D is the dimension of the hyperdimensional space) to perform linear transformation:

[0102] hV=f·W#(5).

[0103] The HDC training provided by the embodiment of the present invention:

[0104] The training process is divided into two main stages: encoding and hyperdimensional classification;

[0105] The first stage (hyperdimensional coding): During the training process, the feature vector f∈R extracted by SNN H By random projection matrix W∈{-1,+1} H×D Mapping to D-dimensional hyperdimensional space:

[0106] hv=f·W∈R D #(6)

[0107] The projection result is binarized using the symbolic function to generate a hyperdimensional vector H∈{-1,+1} D :

[0108] H=sign(h)#(7)

[0109] The second stage (hyperdimensional classification): During the training process, the system accumulates hyperdimensional vectors of the same class into corresponding class prototypes based on sample labels:

[0110]

[0111] Among them C k ∈Z D is the prototype vector of class k.

[0112] like Figure 2 As shown, an embodiment of the present invention provides an electromyographic signal classification and detection system combining SNN and HDC, including:

[0113] The preprocessing module is used to extract multi-domain features from the original signal and generate an N-dimensional feature vector for each sample;

[0114] SNN feature extraction module for processing these features through temporal dynamics using a spiking neural layer with leaky integrate-and-fire neurons;

[0115] The encoding module is used to map the spatiotemporal features extracted by the SNN hidden layer into a high-dimensional binary space suitable for HDC;

[0116] The training module encodes the pulse patterns into a D-dimensional supervector through a bundling operation, followed by Hamming distance-based classification and testing based on pre-trained prototypes, where the system validation evaluates the classification robustness under signal variations.

[0117] The raw electromyographic (sEMG) signal is first preprocessed through a bandpass filter (typical bandwidth 20–450 Hz) to remove baseline drift and power frequency interference. The signal is then windowed (e.g., 200 ms sliding window with 50% overlap). Within each time window, statistical features are extracted in the time domain (e.g., root mean square value (RMS) and zero-crossing rate (ZC), frequency domain (e.g., mean frequency (MNF) and power spectral density (PSD), and time-frequency domain (e.g., short-time Fourier transform or wavelet packet energy). A multi-domain feature vector set of dimension N is constructed, which serves as the encoding starting point for the SNN input space.

[0118] The N-dimensional feature vectors are mapped to spike trains and fed into the SNN model. The spiking neural network uses Leaky Integrate-and-Fire (LIF) neurons as the basic computational unit, combined with temporal coding mechanisms such as Latency Coding or Population Rate Coding to achieve temporal serialization of features. After the input layer is activated through static mapping or Poisson coding, multiple layers of spike convolution or fully connected layers are used to extract event-driven dynamic features across time scales. Ultimately, a spatiotemporally coupled neural activity pattern is formed in the hidden layer, reflecting the temporal dependencies and amplitude variations in the EMG signal.

[0119] The temporal pattern of the SNN hidden layer output is reduced within a time window and converted into a sparse or binary high-dimensional vector using a high-dimensional mapping function (such as one based on random projection or orthogonal transformation). Through the binding and accumulation operations of HDC, multiple time segments are encoded into D-dimensional supervectors based on the temporal structure and category labels of the signal. Hadamard multiplication or XOR operations in HD space are used to strengthen the distributed representation between features, preventing the impact of local perturbations on the overall expression.

[0120] During the inference phase, the system uses the trained class prototype vectors in HDC as a matching benchmark. For each test sample, the generated supervector performs approximate nearest neighbor classification by calculating the Hamming distance or cosine similarity with the prototype set. This process requires no backpropagation or gradient optimization, resulting in high interpretability and low computational resource consumption. The system further introduces signal perturbations (such as EMG intensity fluctuations and noise injection) for robustness verification, demonstrating stable classification performance under low signal-to-noise ratio conditions.

[0121] The present invention is specifically implemented:

[0122] 1. DSN architecture for EMG signal classification

[0123] The proposed DSN architecture for fatigue classification of EMG consists of four core modules, such as Figure 1 As shown in the figure, the preprocessing module extracts multi-domain features from the original signal and generates an N-dimensional feature vector for each sample; SNN feature extraction, in which the spiking neural layer with leaky integration and excitation neurons processes these features through time dynamics; in the training module, the spike pattern is encoded into a D-dimensional supervector through a binding operation, and then classified and tested based on the Hamming distance according to the pre-trained prototype, where the system verification evaluates the classification robustness under signal changes.

[0124] Figure 3 Algorithm architecture design. The proposed framework integrates spiking neural networks for spatiotemporal feature extraction and high-dimensional computation for efficient classification. SNNs capture dynamic EMG patterns through biologically inspired spike coding, while HDCs map features into a high-dimensional space for classification.

[0125] 1.1 Preprocessing

[0126] Electromyographic (EMG) signals are widely used in various fields, such as prosthetic control and human-robot interaction. Effective classification of these signals requires careful signal preprocessing and feature extraction. A 20-350 Hz bandpass filter was used to improve signal quality while preserving fatigue-related signal components. Furthermore, a 150-250 millisecond time window was used to balance classification latency and accuracy.

[0127] In terms of feature extraction, time domain and frequency domain analysis were combined to capture the non-stationary nature of EMG signals. Eight clinically significant features were extracted. Time domain features included integrated EMG and mean absolute value (used to quantify signal strength) and zero crossing rate (used to approximate frequency content). Frequency domain features included mean frequency and median frequency (used to monitor spectral changes and indicate fatigue); peak frequency and frequency ratio (used to capture neuromuscular adaptability); and wavelet energy (providing information for time-frequency analysis).

[0128] IEMG and MAV reflect increasing muscle activation intensity, while MNF and MDF capture progressive spectral compression. Their complementary trends validate the ability of multi-domain approaches to characterize fatigue dynamics and provide discriminative input for subsequent SNN-based feature extraction and HDC classification. This set of features addresses the inherent non-stationarity of EMG signals and offers significant advantages over single-domain approaches.

[0129] 1.2 Feature Extraction Using SNN

[0130] This paper uses a spiking neural network (SNN) as the core feature extractor, whose main task is to extract temporal dynamic features from raw electromyographic (EMG) signals. The network design combines the biomimetic properties of biological neurons with engineering optimization strategies to efficiently capture the time-varying characteristics of EMG signals.

[0131] After the data is transferred to the memory, it will be sent to the SNN for processing. The LIF neuron model is used as the core computing unit of the SNN. The membrane potential V (t) It evolves according to the following differential equation:

[0132]

[0133] where τ is the membrane time constant, I (t) Represents the synaptic input current. When V (t) When a predefined threshold is exceeded, the neuron fires a spike and the membrane potential is reset to its baseline value, V rest After discretization, the membrane potential update rule becomes

[0134] Where τ controls the potential decay rate, W∈R N×M is the weight matrix of the fully connected layer, input X [t] is the electromyographic signal after standardization and row-level normalization. When the membrane potential exceeds the threshold V th When , the neuron triggers a pulse to reset the potential to the resting state. This mechanism simulates the action potential discharge process of biological neurons and converts continuous signals into sparse pulse trains. In order to capture the instantaneous muscle activation pattern of the electromyographic signal, the input data is repeatedly input to the SNN in T time steps. The pulse output s for each time step is [t] ∈{0,1} M Accumulate and calculate the average issuance rate:

[0135]

[0136] Eigenvector F∈R MEach element of represents the activation frequency of the corresponding neuron in the time window, compressing the dynamic pulse activity into a static statistic. This design not only preserves the temporal characteristics of muscle contraction (such as burst activation and relaxation periods), but also provides dimensionally regular input for subsequent classifiers.

[0137] To overcome the non-differentiability of the impulse function, SNN uses inverse tangent instead of gradient to achieve end-to-end training:

[0138]

[0139] This function provides a smooth gradient approximation near the impulse trigger point (x=0) so that the error can be used to update the weight matrix W through the standard backpropagation algorithm. The weights are initialized using a Gaussian distribution with mean 0 and standard deviation 1 to ensure diversity of the initial response and stability of training.

[0140] 1.3 Encoding

[0141] The encoding module maps the spatiotemporal features extracted by the SNN hidden layer into a high-dimensional binary space suitable for HDC. Let the feature vector output by the SNN hidden layer be f∈R H , (where H is the hidden layer dimension), using the random projection matrix W∈{-1,+1} H×D (D is the dimension of the hyperdimensional space) to perform linear transformation:

[0142] hV=f·W#(5)

[0143] To enhance robustness to noise and enable energy-efficient bitwise operations, the projection vectors are binarized using a sign function. The random projection matrix W is fixed during training, eliminating backpropagation overhead while ensuring approximate orthogonality between different feature maps. This is a key property for achieving HDC error resilience. Through this encoding scheme, the temporal dynamics captured by SNN spikes are embedded in a stable geometric relationship within a hyperdimensional space, allowing the subsequent HDC classifier to perform efficient similarity-based inference via a hardware-friendly XOR operation.

[0144] 1.4 Training with HDC

[0145] This paper's training process achieves the classification task of EMG signals through HDC. The core idea is to combine symbolic operations in high-dimensional space with biomimetic feature extraction of SNN to build an efficient and robust classifier. The training process is divided into two main stages: encoding and hyperdimensional classification.

[0146] The first stage (hyperdimensional coding): During the training process, the feature vector f∈R extracted by SNN H By random projection matrix W∈{-1,+1} H×D Mapping to D-dimensional hyperdimensional space:

[0147] hv=f·W∈R D #(6)

[0148] The projection result is binarized using the symbolic function to generate a hyperdimensional vector H∈{-1,+1} D :

[0149] H=sign(h)#(7)

[0150] Binarization not only compresses storage space but also enhances the model's noise immunity, making it suitable for deployment on resource-constrained edge devices. Electromyographic signals exhibit significant individual variability and noise interference. Hyperdimensional coding exploits the randomness and symbolic manipulation of high-dimensional space to construct a feature representation that is insensitive to distribution shifts.

[0151] The second stage (hyperdimensional classification): During the training process, the system accumulates hyperdimensional vectors of the same class into corresponding class prototypes based on sample labels:

[0152]

[0153] Among them C k ∈Z D is the prototype vector of class k. The essence of the accumulation mechanism is to construct the class centroid in high-dimensional space, enhance the common features of similar samples, and amplify the differences between different samples.

[0154] In the testing phase, the hyperdimensional vector f of the input sample n By using the Hamming distance with each class prototype C k Matching is performed for classification.

[0155] The dot product operation is equivalent to calculating the similarity between two vectors. The greater the similarity, the higher the degree of match between the sample and the category. Hamming distance matching is implemented through bitwise operations, which has high computational efficiency and is suitable for real-time classification tasks. The symbolic operation of HDC and Hamming distance matching not only ensures classification accuracy, but also greatly reduces computational complexity, meeting the real-time requirements of wearable devices.

[0156] 1. Specific application fields or related products of the present invention.

[0157] The technical solution of the present invention has wide applicability in the field of medical health, and is particularly suitable for intelligent interaction and physiological status monitoring based on surface electromyography signals. This technology can be integrated into an intelligent muscle monitoring system to achieve real-time monitoring and control of multiple degrees of freedom by analyzing muscle activity, helping users to restore their refined operational capabilities; it can also be applied to wearable rehabilitation monitoring equipment to dynamically track muscle fatigue status and provide real-time feedback for sports rehabilitation training, assisting clinical physicians in optimizing treatment plans. In addition, this technology can also be extended to the field of implantable medical devices, and through miniaturized design, it can achieve long-term acquisition and precise analysis of neuromuscular signals, providing low-power solutions for various complex monitoring scenarios. The lightweight architecture and anti-interference characteristics of this technology further support diversified applications in sports health monitoring, human-computer interaction interfaces and telemedicine, demonstrating the multi-dimensional technical value from clinical medicine to daily health management.

[0158] 2. Relevant evidence of the technical effects obtained by the embodiments of the present invention.

[0159] The technical effect of the embodiment of the present invention is fully verified by the following comparative experimental data: Compared with the current mainstream electromyographic signal classification method, this solution shows significant advantages in both accuracy and efficiency. In terms of classification performance, this method (DSN) achieved an accuracy of 95.31% in the three-category task, which is significantly better than the 82.3% of the LSTM model and the 91.96% of SVM, and is comparable to the top level of 98.53% of LSTM in the two-category scenario. In terms of computational efficiency, this method achieves a classification accuracy of more than 95% with a computing time of 1.25 seconds, which is 5.7 times faster than CNN's 7.16 seconds and 45 times faster than LSTM's 56.51 seconds for three-category classification. The hardware adaptability advantage is reflected in the following: the number of model parameters is compressed to one-third of that of the CNN solution (requiring only 3.072 million matrix operations), and the computational complexity is reduced to 20.3% of that of CNN (15,154,176 FLOPs → 3,072,000 operations), while maintaining real-time performance comparable to traditional machine learning methods such as random forest (RF) and XGBoost (1.25 seconds vs. RF / XGBoost, which is not specified but of the same order of magnitude). These data confirm that this solution achieves a substantial reduction in algorithmic complexity while maintaining high accuracy, providing key technical support for the embedded deployment of wearable medical devices.

[0160]

[0161] Robustness analysis ensures that a neural network maintains stable outputs despite small changes in its input data. Typically, if a bit in a floating-point number flips, the resulting numerical error can grow exponentially.

[0162] In this experiment, we systematically vary the supervector dimensionality over multiple values ​​of D and introduce different levels of noise into the supervector. We train a model for each dimension and evaluate its accuracy with increasing noise levels. Figure 4 As shown, the results show a clear trend: high-dimensional supervectors are significantly more robust to noise.

[0163] For example, when D = 6000, even when 40% of the elements are flipped, the model still maintains an accuracy of 90.12%, only 5.18% lower than the noise-free baseline. In contrast, the accuracy of low-dimensional supervectors (e.g., D = 1000) drops more significantly with increasing noise, dropping to 72.55% at 40% noise. This demonstrates that the neural network of the present invention is highly robust and highly tolerant to hardware errors.

[0164] Bit error rate (BER) is a key metric that quantifies the reliability of a data transmission or storage system by measuring the ratio of erroneous bits to the total number of bits processed. In emerging computing technologies such as resistive random access memory and other non-volatile memories, BER is a common challenge due to inherent hardware imperfections, low-voltage operation, or environmental noise. These bit-level errors can significantly degrade the performance of machine learning models, especially when applied to sensitive tasks such as classification.

[0165] To study the impact of bit error rate (BER) on various neural network architectures, the authors introduced simulated bit errors in the data and model parameters to accurately replicate challenging real-world scenarios where BER is a significant operational factor. This approach enables the authors to create controlled test conditions while maintaining experimental validity.

[0166] The present invention compares with different classifiers and evaluates their performance over a range of BER levels, starting from severe error conditions (0.8) and progressing to near-optimal scenarios (0.0001). Figure 5 The method of the present invention maintains high accuracy at all BER levels, especially under low to medium BER conditions, while the combination of SNN with MLP and SNN with VAE shows higher sensitivity to bit errors, especially at higher BER levels.

[0167] It should be noted that the embodiments of the present invention can be implemented by hardware, software, or a combination of software and hardware. The hardware portion can be implemented using dedicated logic; the software portion can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. Those skilled in the art will appreciate that the above-mentioned devices and methods can be implemented using computer-executable instructions and / or contained in processor control code, for example, such as a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., can also be implemented by software executed by various types of processors, or can be implemented by a combination of the above-mentioned hardware circuits and software, such as firmware.

[0168] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with this technical field within the technical scope disclosed by the present invention and within the spirit and principles of the present invention should be covered by the scope of protection of the present invention.

Claims

1. A myoelectric signal classification and detection method combining SNN and HDC, characterized in that: The following steps are involved: Step 1: Preprocessing: band-pass filter the EMG signal at 20-350 Hz and segment it into 150-250 ms time windows; Step 2: Use SNN to extract features. Extract feature vectors including integrated EMG, mean absolute value, zero-crossing rate, mean frequency, median frequency, peak frequency, frequency ratio, and wavelet energy from both the time and frequency domains of the filtered and segmented electromyographic signals. Input the feature vectors into a spiking neural network (SNN) for temporal encoding. Step 3: Encoding, feature representation of the pulse sequence processed by SNN; Step 4: Use HDC training to map the feature representation to a high-dimensional vector space, and use high-dimensional computing (HDC) methods to classify, train, and recognize different action categories or muscle activity patterns.

2. The electromyographic signal classification detection method according to claim 1, characterized in that: The bandpass filter used in step 1 is used to remove low-frequency baseline drift and high-frequency noise of the electromyographic signal, and the 150-250 millisecond time window is used to balance classification delay and feature integrity. In step 2, the key features of non-stationary EMG signals are obtained by combining time domain and frequency domain analysis, thereby improving the robustness of the subsequent SNN and HDC joint classification.

3. The myoelectric signal classification detection method combining SNN and HDC as claimed in claim 1, characterized in that: The use of SNN to extract features: Spiking neural network is used as the core feature extractor to extract temporal dynamic features from the original electromyographic signals; The network is designed to integrate the biomimetic properties of biological neurons with engineering optimization strategies; After the data is transferred into the memory, it is sent to the SNN for processing; The LIF neuron model is used as the core computing unit of SNN; the membrane potential V (t) According to the following differential equation evolution: where τ is the membrane time constant, I (t) represents the synaptic input current; when V (t) When a predefined threshold is exceeded, the neuron fires a spike and the membrane potential is reset to its baseline value, V rest ; After discretization, the membrane potential update rule becomes Where τ controls the potential decay rate, W∈R N×M is the weight matrix of the fully connected layer, input X [t] is the electromyographic signal after standardization and row-level normalization; When the membrane potential exceeds the threshold V th When , the neuron fires a pulse, resetting the potential to the resting state; This mechanism simulates the action potential discharge process of biological neurons and converts continuous signals into sparse pulse trains; To capture the instantaneous muscle activation pattern of the EMG signal, the input data is repeatedly fed into the SNN in T time steps; The pulse output s for each time step [t] ∈{0,1} M Accumulate and calculate the average issuance rate: Eigenvector F∈R M Each element of represents the activation frequency of the corresponding neuron in the time window, compressing the dynamic spike activity into a static statistic; SNN uses inverse tangent instead of gradient to achieve end-to-end training: This function provides a smooth gradient approximation near the impulse trigger point (x=0) so that the error can be used to update the weight matrix W through the standard backpropagation algorithm; the weights are initialized using a Gaussian distribution with mean 0 and standard deviation 1.

4. The myoelectric signal classification detection method combining SNN and HDC as claimed in claim 1, characterized in that: The encoding: The encoding module maps the spatiotemporal features extracted by the SNN hidden layer into a high-dimensional binary space suitable for HDC; let the feature vector output by the SNN hidden layer be f∈R H , where H is the hidden layer dimension, using the random projection matrix W∈{-1,+1} H×D (D is the dimension of the hyperdimensional space) to perform linear transformation: hV=f·W#(5).

5. The myoelectric signal classification detection method combining SNN and HDC as claimed in claim 1, characterized in that: The HDC training: The training process is divided into two main stages: encoding and hyperdimensional classification; The first stage (hyperdimensional coding): During the training process, the feature vector f∈R extracted by SNN H By random projection matrix W∈{-1,+1} H×D Mapping to D-dimensional hyperdimensional space: hv=f·W∈R D #(6) The projection result is binarized using the symbolic function to generate a hyperdimensional vector H∈{-1,+1} D : H=sign(h)#(7) The second stage (hyperdimensional classification): During the training process, the system accumulates hyperdimensional vectors of the same class into corresponding class prototypes based on sample labels: Among them C k ∈Z D is the prototype vector of class k.

6. A myoelectric signal classification detection system combining SNN and HDC that implements the myoelectric signal classification detection method combining SNN and HDC as described in any one of claims 1 to 5, characterized in that: The electromyographic signal classification and detection system combining SNN and HDC includes: The preprocessing module is used to extract multi-domain features from the original signal and generate an N-dimensional feature vector for each sample; SNN feature extraction module for processing these features through temporal dynamics using a spiking neural layer with leaky integrate-and-fire neurons; The encoding module is used to map the spatiotemporal features extracted by the SNN hidden layer into a high-dimensional binary space suitable for HDC; The training module encodes the pulse patterns into a D-dimensional supervector through a bundling operation, followed by Hamming distance-based classification and testing based on pre-trained prototypes, where the system validation evaluates the classification robustness under signal variations.

7. A computer device, characterized in that: The computer device includes a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the electromyographic signal classification and detection method combining SNN and HDC as described in any one of claims 1 to 6.

8. A computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the steps of the myoelectric signal classification and detection method combining SNN and HDC as described in any one of claims 1 to 6.

9. An information data processing terminal, characterized in that: The information data processing terminal is used to implement the electromyographic signal classification detection system combining SNN and HDC as described in claim 6.

Images