Traction stretching motion state monitoring method based on signal processing
By constructing a multi-channel signal feature extraction framework and utilizing self-evolving feature maps and convolutional mapping strategies, the problem of insufficient recognition accuracy of signal coordinated changes in traction and stretching motion state monitoring was solved, achieving high-precision and stable state monitoring.
Patent Information
- Application Number
- CN202511374403.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-25
- Publication Date
- 2025-10-31
- Estimated Expiration
- 2045-09-25
AI Technical Summary
Existing technologies are insufficient for effectively characterizing the coordinated changes in phase amplitude of multi-channel signals and the structural evolution relationship between segments in traction and tensile motion monitoring. This results in insufficient accuracy in state recognition, poor continuity of feature expression, and low stability of monitoring results.
A multi-channel signal feature extraction framework is constructed. Phase and amplitude components are extracted through normalization and filtering. A signal spectrum dataset is built, and a self-evolving feature map is established. A convolutional mapping strategy and an energy binding factor are introduced to perform inter-channel difference propagation and response combination to generate a multi-channel deep feature map. Iterative training is then performed to achieve accurate modeling and optimization.
It improves the identification accuracy and stability of traction and stretching motion state monitoring, effectively captures the cooperative evolution law between channels, and improves the dynamic modeling capability of signals and the accuracy of monitoring results.
Smart Images

Figure CN120873496A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of signal processing technology, and in particular to a method for monitoring the state of traction and stretching motion based on signal processing. Background Technology
[0002] Currently, traction stretching training is widely used in neuromuscular function reconstruction, skeletal rehabilitation, and posture correction. Its state monitoring is related to the rehabilitation effect. During traction, four types of multi-channel signals are generated: traction force, electromyography, displacement, and conductivity. These signals have non-stationary and strongly coupled characteristics, which increases the complexity of state recognition. Existing monitoring methods mostly rely on template matching and simple feature extraction, which cannot characterize the dynamic evolution of signals and the synergistic relationship between channels. This results in low recognition accuracy and easy neglect of abnormal responses, making it difficult to meet the needs of precise rehabilitation. Therefore, there is an urgent need for a traction stretching motion state monitoring method based on signal processing to achieve high-precision modeling and optimized monitoring of complex dynamic signals.
[0003] Publication No. CN112843563A discloses a traction training assessment system based on the fusion of force sensing and angle sensing. It constructs a training trajectory by collecting displacement and tension information and evaluates it by combining sensor data, thereby realizing trajectory reconstruction and state quantification analysis of the training process. Publication No. CN113989529A proposes an electromyographic signal state recognition method for rehabilitation training. It constructs a feature set by collecting electromyographic signals and uses a classifier model to identify training movements, thereby realizing the classification and evaluation of movement states during rehabilitation training.
[0004] However, existing technologies do not fully consider the coordinated changes in phase amplitude of traction stretching signals in continuous time segments and the structural evolution relationship between segments. They neglect the dynamic coupling characteristics between multi-channel signals, the correlation mechanism between trajectory changes and energy distribution, and the need for cross-channel state adaptive adjustment. As a result, the proposed method suffers from insufficient state recognition accuracy, poor continuity of feature expression, and low stability of monitoring results in scenarios with complex traction training states and significant individual differences. Summary of the Invention
[0005] This invention provides a traction and stretching motion state monitoring method based on signal processing. It aims to construct a multi-channel signal feature extraction framework with temporal modeling and structural awareness capabilities. The method acquires multi-channel raw signal data collected by a traction and stretching device under different operating states. Phase and amplitude components are extracted through normalization and filtering to construct a signal spectrum dataset containing trajectory change structure and temporal discrete sampling features. Based on the phase components in the signal spectrum dataset, a structural response mapping relationship is established under a time-series index to construct a self-evolving feature map reflecting the traction and stretching state change process. The self-evolving feature map and amplitude components are temporally aligned to construct a channel-level feature matrix group, generating a multi-channel feature representation map to capture the collaborative evolution law exhibited by channels during the traction and stretching process. Based on the multi-channel feature representation map, a convolutional mapping strategy based on structural path awareness is established, introducing... A path dissipation factor is used to adjust the response intensity of high-density paths. Channel difference propagation and response combination operations are performed to obtain a multi-channel depth feature map characterizing deep structural changes. Based on this, the energy distribution of each channel in the depth feature map is calculated in the time dimension. An energy binding factor is introduced to compress and adjust the energy response intensity, an inter-channel scheduling correlation matrix is established, and the original features are reconstructed directionally to generate a directionally enhanced energy-adjusted feature map. The feature vectors of each channel in this directionally enhanced feature map are sequentially concatenated at continuous time steps to form a feature sequence reflecting the dynamic changes in traction and stretching states. By constructing a traction and stretching motion state monitoring signal processing model, inputting a signal spectrum dataset, and sequentially performing self-evolutionary modeling, channel-level alignment, depth feature extraction, and directional enhancement operations, and iteratively training until convergence, accurate modeling and optimized processing of the traction and stretching motion state monitoring signal are achieved.
[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for monitoring traction and tensile motion state based on signal processing, the specific steps of which are as follows: S1. Collect multi-channel raw signals of the traction and tensioning device under different operating conditions, perform normalization and filtering processing, extract phase components and amplitude components, and construct a signal spectrum dataset containing trajectory change structure and time-series discrete features. S2. Based on the phase components in the signal spectrum dataset, establish the structural response mapping relationship under the time series index and construct a self-evolutionary feature map that reflects the traction and stretching state change process. S3. Align the self-evolving feature map with the amplitude components in time, construct a channel-level feature matrix group, and generate a multi-channel feature representation map; S4. Based on the multi-channel feature representation map, a structure path-aware convolutional mapping strategy is adopted, a path dissipation factor is introduced, and inter-channel difference propagation and response combination operations are performed to obtain a multi-channel deep feature map. S5. Calculate the temporal energy distribution of each channel of the depth feature map, introduce an energy binding factor, establish a scheduling correlation matrix between channels, reconstruct the original features in a directional manner, and generate a directionally enhanced energy-adjusted feature map. S6. Concatenate the feature vectors of each channel in the direction enhancement feature map according to the time step to obtain a feature sequence that reflects the changes in traction and tension state. S7. Construct a traction and stretching motion state monitoring signal processing model, input the signal spectrum dataset, and iteratively train through steps S2 to S6 until convergence, thereby achieving optimized processing of the traction and stretching motion state monitoring signal.
[0007] Preferably, in step S1, for constructing the traction stretching state monitoring signal atlas dataset, firstly, multi-channel raw signal data collected by the traction stretching device under different operating states are acquired. The raw signal data includes traction force response signals, physiological parameter monitoring signals, electromyographic signals, acceleration signals, and attitude angle sensing signals. Considering the differences in physical quantity units and sampling methods of different channel signals, a channel normalization strategy is used to standardize the amplitude of the raw data. Anti-interference preprocessing is then performed based on a sliding window and filter combination design to remove electromagnetic noise, body motion artifacts, and baseline drift interference. Subsequently, the processed signal sequence is time-aligned to ensure that different channels form synchronous sampling results on the same time axis. In the feature extraction stage, temporal amplitude features and phase changes are extracted from each channel signal to construct a basic feature vector containing instantaneous response and periodic perturbation information. Furthermore, the feature vectors are grouped into time periods according to time windows to construct structurally stable time period response units. These units are then stacked along the channel dimensions to form a three-dimensional signal atlas dataset, used to reflect the trajectory change structure and discrete sampling pattern of the signal under traction conditions.
[0008] Preferably, the multi-channel signals acquired during the traction training process exhibit nonlinear coupling and dynamic mutation characteristics at the structural level, with significant phase coordination and amplitude linkage between different channel signals. Based on this characteristic, this invention proposes a modeling strategy that integrates phase components and amplitude features. Through a unified sequence alignment and collaborative enhancement mechanism, multi-source signals are constructed into an analytical multidimensional structure. This method enhances the structural expressive ability while maintaining the physical properties of the signals, providing a stable and reliable input foundation for subsequent self-evolutionary feature extraction, multi-channel modeling, and orientation enhancement processing.
[0009] Furthermore, in step S2, the specific process of constructing the self-evolving feature map includes: S21. Based on the phase component information of each channel in the signal spectrum dataset, extract the channel-level phase component sequences under the time series index to construct the phase component matrix. ; S22, Based on the phase component matrix Establish a metric relationship for structural response changes between adjacent time points and construct a structural response mapping vector. The mathematical model is: ; in, This represents the structural response intensity at time index t; S23. Map the structural response vectors under all time indices. Combining them in chronological order constitutes a self-evolutionary feature map. The mathematical model is: ; in, This is the final self-evolutionary feature map.
[0010] Furthermore, in step S2, regarding the changing patterns of phase information in multi-channel signals under traction and stretching conditions, this invention constructs a self-evolving feature map to characterize the signal structure response change process based on the evolution characteristics of the phase components of each channel in the signal spectrum dataset in the time dimension. Specifically, firstly, the phase distribution features of each channel within a continuous time period are extracted, and a time series index structure is constructed to record the phase evolution trajectory at different time points. Subsequently, by comparing the phase change trends of each channel between adjacent time points, the local response mutations and cross-channel disturbance features that occur between channels during the traction and stretching process are captured, thereby forming a response mapping that reflects the dynamic evolution relationship of the structure. On this basis, considering the synchronicity between channels, the phase drift amplitude, and the interference correlation strength, a time-varying structural response expression framework is established to generate a self-evolving feature map to describe the traction and stretching state change process. This feature map can dynamically reflect the state transition, disturbance propagation, and structural cooperative change behavior of the signal structure during traction, using time as an index, providing a highly dynamic and sensitive expression basis for subsequent signal modeling and state recognition.
[0011] Preferably, during traction stretching training, the multi-channel physiological and structural response signals exhibit frequent short-term disturbances, asymmetric cross-channel response structures, and nonlinear state transition processes in the temporal dimension. The coupling relationship between signals is jointly driven by traction intensity, movement amplitude, and tissue stress feedback factors, showing significant dynamic evolution characteristics. Based on this type of unstable and highly coordinated signal behavior, this invention proposes a disturbance-aware structural response mapping mechanism. Combining phase difference trends and inter-channel synchronization relationships, it models fine-grained structural changes during the traction stretching motion state change process and constructs a self-evolutionary feature representation map that reflects the signal co-evolution law, providing structural consistency support and response expression basis for subsequent feature encoding, structural enhancement, and state classification processing.
[0012] Furthermore, in step S3, the specific method for constructing a channel-aligned multi-channel feature representation map includes: S31. Based on the self-evolutionary feature map The amplitude components extracted from the signal spectrum dataset Establish a time index mapping function Align the amplitude components with the structural response time index in the time dimension to construct the aligned amplitude matrix. The mathematical model is as follows: , ; S32, According to the amplitude matrix Self-evolving eigenvalues in each channel Aligned amplitude By combining them sequentially along the time dimension, a channel-level feature matrix group is constructed. The feature matrix of each channel The joint feature column vector is represented as: ; S33, Channel-level feature matrices corresponding to all channels Stacking along the dimension of the channel set C yields a multi-channel feature representation map. The mathematical model is: .
[0013] Furthermore, in step S3, to construct a multi-channel feature representation with structural consistency and temporal correlation during the traction and stretching motion, firstly, the self-evolving feature map obtained in the previous stage is precisely aligned with the amplitude components of the original signal in the time dimension to ensure the response synchronization of each channel at continuous time points. Secondly, based on the aligned data, the amplitude response trend and perturbation change pattern of each channel at different time points are extracted to construct channel-level feature response units. Subsequently, using the time index as the main axis, combined with the structural dependence and perturbation synchronization behavior between channels, a multi-channel feature matrix group is assembled to capture the co-evolution law between different channels during training. Next, the feature matrix group is structurally rearranged and the response relationship is enhanced through matrix reconstruction operations to improve the expressive power of channel-level features at the temporal and structural levels. Finally, the processed feature matrix group is mapped into a multi-channel feature representation map in a unified format, providing basic data support for subsequent structural modeling and state recognition.
[0014] Preferably, the multi-channel signals in the traction training process exhibit uneven time response rates, inconsistent perturbation location distributions, and differences in channel excitation intensity, resulting in non-uniform cooperative characteristics in the response structure of different channels. Traditional methods based on simple superposition or local convolution are difficult to effectively express the structural dependencies between channels. This invention constructs a channel feature matrix group with temporal alignment and response relationship as the core, which can enhance the system's ability to model the dynamic response cooperation between channels while maintaining the temporal integrity of the original signal, and provide a stable and reliable structural input for subsequent traction state perception.
[0015] Furthermore, in step S4, the specific methods for constructing a structure-path-aware convolutional mapping strategy and generating multi-channel deep feature maps include: S41. Based on the multi-channel feature representation diagram We construct a structure-path-aware convolutional mapping strategy, define the path mapping kernel tensor, and the mathematical model is as follows: : in, Indicates the first The channel at the r-th feature dimension and time position s is directed towards the r-th feature dimension. Channel transmission structure perceives convolution weights; S42. Construct a path dissipation factor based on the exponential decay relationship between path information density and channel propagation intensity. The mathematical model is: ; in, This represents the path accessibility strength between channels c1 and c2, with α>0 being a regulatory factor. S43. Perform inter-channel difference propagation operation, define the channel response difference at each time t, and construct the inter-channel difference propagation modulation factor matrix. The mathematical model is: ; in, This represents the adjustment factor for the difference in response between channels c1 and c2 at time t; S44, Path Mapping Weight Tensor Path dissipation factor With the propagation regulation factor matrix Combine to generate multi-channel depth feature maps Mathematical model: ; in, This represents the depth feature response value of the c2th channel at time point t.
[0016] Furthermore, in step S4, for the deep modeling method of multi-channel features, firstly, based on the aforementioned multi-channel feature representation map, the path connection relationship between channels is extracted, and a structure-aware convolutional mapping strategy is established to capture the channel interaction characteristics under traction and stretching conditions. Secondly, a path dissipation factor is introduced, and an adjustment function is constructed by combining channel structure density and propagation path strength to weaken the dominant effect of redundant paths in feature propagation and improve the information coupling degree between effective channels. Subsequently, a differential propagation operation between channels is performed, the response deviation of each channel in the path mapping is calculated, and local path features are integrated through a response combination function. Next, the convolutional mapping strategy is embedded in the feature propagation structure, and channel-level deep convolution processing is performed to extract response features with path-aware characteristics. Finally, a multi-channel deep feature map is output, providing a high-order feature foundation for structure-aware enhancement for subsequent energy adjustment and orientation modeling.
[0017] Preferably, during traction stretching training, each sensing channel exhibits uneven path accessibility, dynamic fluctuations in response distribution, and significant differences in coupling characteristics between channels in terms of force path, structural impedance, and feedback response. Based on these structural inhomogeneities and heterogeneous response characteristics, this invention employs a path-aware convolutional mapping mechanism and introduces a path dissipation factor to weaken the dominant role of regions with excessively high path response density during propagation and increase the proportion of structural awareness in low-response regions. This achieves a balanced expression of differences between channels, providing structural enhancement support for stable modeling of multi-channel deep feature maps and downstream state discrimination.
[0018] Furthermore, in step S5, the specific methods for constructing the directional enhancement energy regulation feature map include: S51. Calculate the multi-channel depth feature map. The energy distribution of each channel in the time dimension is used to construct an energy mapping matrix. The mathematical model is: ; in, This represents the value of channel c in the depth feature map at time point t. This represents the energy response value at the corresponding time point; S52, Based on Energy Mapping Matrix Calculate the intensity of energy coordination and variation between channels over time, and construct a scheduling correlation matrix. The mathematical model is: , ; in, This represents the energy co-factor between channel i and channel j; S53. Based on the scheduling association matrix Construct a set of directional adjustment node coefficients The original deep features are reconstructed directionally to generate a directionally enhanced feature map. The mathematical model is: ; S54. Based on the inverse decay relationship between the energy term at time points and the adjustment sensitivity, an energy binding factor is constructed. The mathematical model is: ; Where β>0 represents the energy regulation sensitivity coefficient. ∈(0,1] represents the degree of compression of the directional enhancement feature update at time point t; S55, Enhance the orientation feature map With energy mapping matrix Perform channel-dimensional weighted fusion to generate a direction-enhanced energy-modulated feature map. Its mathematical model is: ; in, This represents the directional enhancement feature value at time point t in channel c after fusion.
[0019] Furthermore, in step S5, regarding the construction method of the directional enhancement energy regulation feature map, firstly, based on the aforementioned multi-channel depth feature map, the energy change trajectory of each channel in continuous time series is extracted, and its local energy distribution in the time dimension is calculated to characterize the dynamic evolution intensity of the response of different channels during traction stretching training; secondly, an energy binding factor is introduced, embedding an energy amplitude adjustment structure in the energy time series of each channel to adjust the upper and lower limits of the energy distribution range, reducing the interference of local abnormal energy peaks on the overall scheduling weight allocation and improving the stability of energy feature expression; subsequently, a scheduling correlation matrix between channels is constructed, and by integrating the channel energy synchronization relationship and time scheduling synergy, a directional linkage structure reflecting the guidance and response relationship between each channel is established; next, based on the scheduling correlation matrix, a directional reconstruction operation is performed on the original depth feature map to adjust the distribution structure of features in the time domain and channel domain, strengthening significant response paths; finally, a directional enhancement energy regulation feature map is output, providing a feature foundation with directional and energy integrated expression capabilities for subsequent structural convolution mapping and dynamic trend modeling, effectively improving the cross-channel structural modeling accuracy and response focusing performance during the traction stretching state change process.
[0020] Preferably, during traction stretching training, the physiological response sites corresponding to different signal channels have an asymmetric coupling relationship on the structural path, resulting in the energy distribution of the signal in the time dimension exhibiting multi-stage activation, unstable peak perturbation, and fluctuations in the degree of coordination between channels. Based on this dynamic response characteristic, this invention introduces an energy binding factor to construct a time-energy constraint mechanism and combines it with the channel scheduling correlation matrix to model the directional relationship between channels. While maintaining the stability of the signal energy structure, it achieves directional reconstruction of the original features, effectively improving the expressive ability of cross-channel signal coupling relationship and the discrimination accuracy under directional scheduling under traction stretching state, providing more directionally focused and energy-constrained structural support for subsequent state change detection and trend prediction.
[0021] Furthermore, in step S6, the state sequence construction method based on the direction-enhanced energy regulation mechanism includes: S61, Energy modulation feature map with enhanced directionality The time feature vector is constructed by combining the channel response values at each time step. Its mathematical model is: ; in, This represents the vector-like response composed of the features of each channel at the current time step; S62. Convert the time feature vectors corresponding to each time step. Construct a feature sequence matrix of the traction-stretching state evolution process by assembling the sequences sequentially. The mathematical model is: ; in, It represents the ordered expression structure of the multi-channel dynamic characteristic response during traction and stretching.
[0022] Furthermore, in step S6, to improve the temporal continuity and evolutionary expression of state change modeling during traction and stretching motion, this invention proposes a temporal feature sequence construction mechanism based on directional enhancement feature maps. First, based on the directional enhancement energy regulation feature map, feature vectors of each channel within a continuous time window are extracted and structurally organized according to the time index order to form channel-level temporal feature units. Second, by performing a sequence splicing operation on the feature vectors within the time step, the structural consistency between channels is maintained while highlighting the state evolution trend and directional response change characteristics, constructing a multi-channel fused temporal feature trajectory. Finally, the feature change context in the temporal direction is retained in the splicing result, forming a feature sequence that can reflect the continuous evolution of the traction and stretching state, providing a high temporal resolution structural expression basis for subsequent state discrimination, trend modeling, and dynamic recognition modules.
[0023] Preferably, in step S7, for the signal optimization processing of traction and tensile motion state monitoring, a traction and tensile motion state monitoring signal processing model is constructed, and the structural response modeling, differential convolutional mapping, energy direction enhancement, and temporal feature reconstruction stages are executed sequentially. First, in the structural response modeling stage, the preprocessed signal spectrum dataset is input, its phase components are extracted, and a structural response mapping relationship is established based on the time series index to construct a self-evolving feature map reflecting the dynamic evolution process of the traction and tensile state, capturing the continuous change characteristics of the structural response during traction. Subsequently, in the differential convolutional mapping stage, the self-evolving feature map is aligned with the amplitude components to generate a multi-channel feature representation map, a convolutional mapping strategy based on structural path awareness is constructed, and a path dissipation factor is introduced to perform inter-channel differential propagation and response combination operations to generate a multi-channel depth feature map with deep discriminative capabilities. Then, in the energy direction enhancement stage, the phase components are extracted, and ... In the direction enhancement stage, the temporal energy distribution characteristics of each channel in the depth feature map are calculated. A channel scheduling correlation matrix is constructed by combining the energy binding factor. The original features are then reconstructed directionally to generate a direction-enhanced energy regulation feature map, which is used to improve the response sensitivity under the changing direction of the traction path. Then, in the temporal feature reconstruction stage, the feature vectors of each channel in the enhanced feature map are concatenated according to time steps to obtain a feature sequence reflecting the evolution law of traction stretching state. By integrating the above stages end-to-end, and combining iterative training mechanism and structural regulation optimization strategy, the model can achieve joint optimization of structural response modeling, energy direction enhancement and temporal feature construction based on the dynamic characteristics of state evolution during traction stretching. Finally, it outputs a high-quality signal expression result reflecting the state characteristics of the traction stretching process, effectively improving the state recognition accuracy and response stability under complex traction environment.
[0024] Compared with existing technologies, this invention focuses on the spatiotemporal evolution characteristics of multi-channel signals during traction and stretching training. It constructs a feature modeling mechanism oriented towards trajectory changes and state responses, proposes a signal map construction strategy, and combines phase and amplitude component extraction methods to achieve structural representation of traction and stretching signals during state changes, effectively improving the feature map's ability to preserve continuous dynamic information of the traction process. In the structural response modeling stage, a self-evolving feature map is constructed to reflect the response pattern under time-series indexing, and a channel-level alignment mechanism is introduced to generate feature representation maps, enhancing the model's ability to characterize the synergy and phase synchronization of changes between channels. In the depth mapping modeling stage, a structure-path-aware convolutional mapping strategy is designed, and a path dissipation factor is introduced to adjust high density. By combining path propagation intensity with inter-channel differential propagation and response operations, a structure-aware multi-channel deep feature map is obtained, significantly improving the directional consistency and structural discriminability of feature mapping. In the direction enhancement stage, an energy binding factor is introduced, and a scheduling correlation matrix is established to model the energy response relationship between channels, completing the construction of the energy adjustment map for direction enhancement, thus solving the problems of weak directionality and energy leakage in the original features. Finally, by constructing a traction and stretching motion state monitoring signal processing model and integrating self-evolutionary modeling, convolutional mapping, energy adjustment, and sequence splicing strategies, combined with an iterative training mechanism for optimization, stable modeling and precise prediction of multi-channel signals under traction and stretching conditions are achieved, effectively improving the accuracy and robustness of traction state recognition. Attached Figure Description
[0025] Figure 1 This is a flowchart of a traction and stretching motion state monitoring method based on signal processing provided by the present invention.
[0026] Figure 2 This is a structural diagram of the self-evolutionary feature map generation process based on signal spectrum coupling and clustering phase components provided by the present invention.
[0027] Figure 3 This invention provides a flowchart of a process that aligns amplitude components and self-evolving feature maps in a time series to generate a channel-level feature matrix group and obtain a multi-channel feature representation map.
[0028] Figure 4 This is a structural diagram of a multi-channel evolution feature map generated by combining path dissipation factor, convolutional mapping strategy and differential transmission response, provided by the present invention.
[0029] Figure 5 This invention provides a structural diagram for generating energy regulation feature maps by combining energy binding factors with amplitude modulation relationship matrices and directional construction.
[0030] Figure 6 This is a flowchart illustrating the generation process of a feature sequence from an energy regulation feature map provided by the present invention, which is then spliced together over time steps.
[0031] Figure 7 This is a comparison chart of channel feature lines provided by the present invention, showing the temporal feature changes of the original signal and the enhanced signal on multiple channels.
[0032] Figure 8 This is a feature grayscale image comparison image provided by the present invention, wherein... Figure 8 (a) is the original feature grayscale image. Figure 8 (b) is a grayscale image with enhanced features. Detailed Implementation
[0033] This invention provides a traction and stretching motion state monitoring method based on signal processing, aiming to achieve structural evolution modeling and state recognition construction of multi-channel physiological signals. The method acquires multi-channel raw signal data of a traction and stretching device under different operating states; performs normalization and filtering preprocessing on the raw signals, and extracts phase and amplitude components reflecting the internal modulation rules of the signals to construct a signal spectrum dataset with trajectory change structural features and temporal sampling attributes; based on the spectrum data, it performs structural response modeling, feature map construction, structural convolution mapping, orientation enhancement adjustment, and temporal sequence splicing processing, ultimately outputting a feature sequence reflecting the traction state change process, supporting subsequent state recognition and dynamic analysis tasks.
[0034] Please see Figure 1 As shown in the figure, a traction and stretching motion state monitoring method based on signal processing is described in the following steps in an embodiment of this application.
[0035] S1. Acquire multi-channel raw signal data collected by the traction and tensioning device under different operating conditions, perform normalization and filtering processing on the signal data, extract phase components and amplitude components, and construct a signal spectrum dataset containing trajectory change structure and time-series discrete sampling features.
[0036] Further, in step S1, a multi-channel physiological signal acquisition device (BIOPACMP160) was used to collect traction and stable traction training data from the subjects on a traction stretching training platform. The number of signal acquisition channels was set to 8, with each channel acquiring four types of physical quantity data: electrocardiogram (ECG) signals, traction force sensing signals, displacement sensing signals, and skin conductance signals. The sampling rate per channel was set to 512 Hz, the sampling duration for a single signal segment was 5 seconds, the acquisition window step size was 1 second, and the total number of sampling points was 3000. The acquired raw signals were uniformly processed and encoded using floating-point coded formats. The data is stored in a specific format. Before sampling, amplitude normalization is performed to compress the amplitude of each channel to the [-1,1] interval. A third-order bandpass filter is used for filtering, with the cutoff frequency set to 0.5Hz to 45Hz to eliminate baseline drift and high-frequency interference. Subsequently, the phase component is extracted through uniform sampling in the time domain, while the amplitude component is retained to form a signal representation object with physical meaning. Finally, the phase and amplitude components are concatenated according to the time index to construct a standardized signal spectrum dataset of trajectory change structure and time-series discrete sampling features, providing basic input for subsequent traction and stretching motion state recognition.
[0037] S2. Based on the phase components in the signal spectrum dataset, establish the structural response mapping relationship under the time series index, and construct a self-evolutionary feature map reflecting the traction and stretching state change process.
[0038] Furthermore, in step S2, to characterize the structural perturbation changes of the signal in the time dimension during the traction and stretching process, a structural response mapping mechanism is constructed based on the phase components of each channel in the signal spectrum dataset. The process is as follows: Figure 2 As shown, the specific steps for constructing a self-evolving feature map include:
[0039] S21. Extract the time index based on the structural feature extraction results of each channel in the signal spectrum dataset. The following channel numbers The corresponding phase component values form the phase component matrix of the channel-time two-dimensional structure. The mathematical model is: , ; in, Let be the phase value of the c-th channel at time t, in radians, with a range of [-π, π]. In this embodiment, T = 200. In this embodiment, and Let be the real and imaginary parts of the complex signal of channel c at time t, respectively, and their calculation formula is: , ; in, The sampling frequency of channel c is [value to be filled in]. N is the signal frame length, with a value of N=3000; t is the time index, determined by the sampling point n according to the conversion... get; In this embodiment, The amplitude value of the original discrete signal of channel c at the nth sampling point is calculated using the following formula: ; in, Let be the initial amplitude modulation coefficient of channel c. In this embodiment, λ ∈ [0.8, 1.2]. The time decay rate of channel c is given in this embodiment. = 0.05, The modulation frequency parameters of channel c, This represents the phase offset of channel c, which in this embodiment ranges from -π / 2 to π / 2.
[0040] S22, According to the phase component matrix The sum of the squares of the phase differences of each channel between adjacent times t and t+1 is defined as a measure of the structural disturbance amplitude, and a structural response time quantity is constructed. Its mathematical model is: ; in, This represents the structural response intensity at time index t, with values mainly concentrated in the range of [0.01, 0.25]. When a sudden change in state occurs, the local value exceeds 0.4, indicating a phase difference. After standardization, the values are set to [-1, 1] to maintain comparability of changes in the time neighborhood.
[0041] S23. Map the structural response vectors under all time indices. Combining them in chronological order constitutes a self-evolutionary feature map. The mathematical model is: ; in, The final self-evolving feature map is set to a dimension of 1×T, therefore The specific dimension is 1×200, which represents the structural response intensity sequence at 200 consecutive moments in the time series.
[0042] S3. Perform time-series alignment processing on the self-evolving feature map and amplitude components to construct a channel-level feature matrix group and generate a multi-channel feature representation map.
[0043] Furthermore, in step S3, to achieve joint modeling of the structural evolution characteristics and amplitude variation trends among different signal channels during traction and tension motion, the self-evolution feature map and the amplitude components of each channel need to be time-aligned, and a multi-channel feature representation map needs to be constructed. The process is as follows: Figure 3 As shown, the construction process includes the following steps.
[0044] S31. Based on the self-evolutionary feature map The amplitude components extracted from the signal spectrum dataset Establish a time index mapping function Align the amplitude components with the structural response time index in the time dimension to construct the aligned amplitude matrix. The mathematical model is as follows: , ; in, This represents the amplitude value of channel c at structural response index t and is normalized to the interval [0,1]. In this embodiment The calculation formula is: ; in, This represents the original time-domain signal of channel c at sampling point n. To calculate the number of sampling points in the window, this embodiment takes... The calculation results have been normalized. In this embodiment, ; In this embodiment, the time index mapping function The calculation formula is as follows: ; Where t∈{1,2,…,T} represents the time series position under the structural response index, and in this embodiment L=3000.
[0045] S32, According to the amplitude matrix The self-evolving eigenvalues of each channel in the matrix With the corresponding amplitude By combining them sequentially along the time dimension, a channel-level feature matrix group is constructed. The feature matrix of each channel The joint feature column vector is represented as: ; in, This represents the feature matrix of the c-th channel in the time series dimension, composed of structural response and amplitude components, preserving dual feature information of structural driving and signal energy; self-evolutionary feature map. The elements can be represented as Used to describe the structural response intensity of channel c at time point t, self-evolutionary eigenvalues Amplitude Obtained through amplitude envelope.
[0046] S33, Channel-level feature matrices corresponding to all channels Stacking along the dimension of the channel set C yields a multi-channel feature representation map. The mathematical model is: ; in, The three-dimensional multi-channel feature representation structure of the traction and stretching process is given by G, which has dimensions of C×2×T.
[0047] S4. Based on the multi-channel feature representation map, establish a structure path-aware convolutional mapping strategy, introduce a path dissipation factor, and perform inter-channel difference propagation and response combination operations to obtain a multi-channel depth feature map.
[0048] Furthermore, in step S4, to enhance the ability to model the structural coupling between multi-dimensional channels under traction and tension motion, a structural path mapping perception strategy needs to be constructed based on the multi-channel feature representation map to extract a depth map reflecting the propagation characteristics of response differences between channels. The process is as follows: Figure 4 As shown, the construction process includes the following steps.
[0049] S41. Based on the multi-channel feature representation diagram We construct a structure-path-aware convolutional mapping strategy, define the path mapping kernel tensor, and the mathematical model is as follows: ; in, Indicates the first The channel at the r-th feature dimension and time position s is directed towards the r-th feature dimension. Channel transmission structure perceives convolution weights; In this embodiment, The calculation formula is: ; Where S is the size of the perception window, taken as... convolution kernel tensor The initialization uses a uniform distribution ; In this embodiment, g(·) is the underlying similarity function, and its calculation formula is: ; in, This represents the sampled value of channel c at time point s+n; In this embodiment, The local weight kernel on feature dimension r is calculated using the following formula: ; Where n is the index within the local window. In this embodiment, the Gaussian smoothing amplitude adjustment coefficient is used for the feature dimension r. =1.0, In this embodiment, a new Gaussian smoothing kernel decay factor is used. =0.1, This is the amplitude adjustment coefficient for the cosine oscillation term, in this embodiment... =0.3, The frequency parameter of the cosine modulation term is used in this embodiment. , where r is the feature dimension index.
[0050] S42. Construct a path dissipation factor based on the exponential decay relationship between path information density and channel propagation intensity. The mathematical model is: ; Where α>0 is the regulatory factor, and in this embodiment α=2.0, This indicates the path accessibility between channels c1 and c2; In this embodiment, The calculation formula is: ; in, For Euclidean distance, Set as the distance attenuation factor. =0.5, S is the total number of feature dimensions, with a value of S=32; In this embodiment, the embedding vector and Let C1 and C2 represent the features in feature dimension s, respectively. The calculation formula is as follows: ; in, In this embodiment, the scaling factor is used. =0.8, where K is the length of the embedding vector, and in this embodiment K=5; In this embodiment, This represents the k-th local feature component of channel c in feature dimension s, calculated using the following formula: ; Where K is the local neighborhood index. Let be the sampled value of channel c at time point s + k. In this embodiment, the smoothing adjustment coefficient for feature dimension s is... =1.0, The exponential decay factor controls the decay rate of the neighborhood weights as k increases. In this embodiment... =0.1, The amplitude of the oscillation modulation is, in this embodiment... =0.2, The oscillation frequency is used in this embodiment. .
[0051] S43. Perform inter-channel difference propagation operation, define the channel response difference at each time t, and construct the inter-channel difference propagation modulation factor matrix. The mathematical model is: ; in, In this embodiment, the convolution combination effect adjustment factor is used. =1.0.
[0052] S44, Path Mapping Kernel With the propagation regulation factor matrix Combine to generate multi-channel depth feature maps Its mathematical model is: , ; in, This represents the depth feature response value of the c2th channel at time point t; In this embodiment, The calculation formula is: ; in, This represents the input signal value of channel C1 at time tn. Here are the convolution kernel weights corresponding to the feature dimension r, with initial values following a uniform distribution U(0.01, 0.01). In this embodiment, the length of the convolutional sensing window is... .
[0053] S5. Calculate the temporal energy distribution of each channel in the depth feature map, establish a scheduling correlation matrix between channels, perform directional reconstruction processing on the original features, and generate a directional enhanced energy-adjusted feature map.
[0054] Furthermore, in step S5, to enhance the directional modeling capability of the multi-channel signal response structure during traction and stretching motion, it is necessary to construct a directionally enhanced energy modulation feature map to improve the channel representation and discrimination power of the signal under complex motion states. The construction process is as follows: Figure 5 As shown, the specific steps include:
[0055] S51. Calculate the multi-channel depth feature map. The energy distribution of each channel in the time dimension is used to construct an energy mapping matrix. The mathematical model is: , ; in, This represents the energy response value at the corresponding time point.
[0056] S52, Based on Energy Mapping Matrix Calculate the intensity of energy coordination and variation between channels over time, and construct a scheduling correlation matrix. The mathematical model is: , , ; in, The energy coordination factor of channel i and channel j reflects the consistency of their response modes during traction. To enhance the robustness of the calculation, a sliding window mechanism is adopted in this embodiment. The window length is 200 and the step size is 50. The energy coordination factor is calculated for each local window and averaged to reduce the interference of local anomalies on the overall structure judgment.
[0057] S53. Based on the scheduling association matrix Construct a set of directional adjustment node coefficients The original deep features are reconstructed directionally to generate a directionally enhanced feature map. The mathematical model is: , ; in, This represents the reconstructed eigenvalue of channel j at time point t after directional enhancement. This represents the feature value of the original depth feature map at the channel time point.
[0058] S54. Based on the inverse decay relationship between the energy term at time points and the adjustment sensitivity, an energy binding factor is constructed. The mathematical model is: ; Wherein, β>0 represents the energy regulation sensitivity coefficient, and in this embodiment, β=0.6; In this embodiment, The calculation formula is: ; Where H is the historical difference accumulation window length, and in this embodiment H=12. To enhance the eigenvalues of channel c in the direction at time point t, For historical characteristic values of time th, In this embodiment, the differential intensity adjustment factor is used. =1.8, In this embodiment, the differential energy index is used. =1.5, In this embodiment, the time decay factor is used. =0.25.
[0059] S55, Enhance the orientation feature map With energy mapping matrix Perform channel-dimensional weighted fusion to generate a direction-enhanced energy-modulated feature map. Its mathematical model is: , ; in, This represents the directional enhancement feature value at time point t in channel c after fusion;
[0060] S6. The feature vectors of each channel in the feature map of the enhanced direction are concatenated sequentially over time steps to obtain a feature sequence that reflects the changes in traction and stretching state.
[0061] Furthermore, in step S6, to achieve sequential modeling of the traction and stretching motion state, the energy regulation feature map after direction enhancement needs to be... The channel-level features at each time step are structurally combined to construct a dynamic feature representation under a unified temporal structure. This process includes the following two steps: Figure 6 As shown.
[0062] S61, Energy modulation feature map with enhanced directionality The time feature vector is constructed by combining the channel response values at each time step. Its mathematical model is: ; in, This represents the vector response composed of the features of each channel at the current time step. This represents the directional enhancement eigenvalue of channel c∈{1,…,C} at time point t.
[0063] S62. Convert the time feature vectors corresponding to each time step. Construct a feature sequence matrix of the traction-stretching state evolution process by assembling the sequences sequentially. The mathematical model is: ; in, It represents the ordered expression structure of the multi-channel dynamic characteristic response during traction and stretching.
[0064] S7. Construct a traction and stretching motion state monitoring signal processing model, input the signal spectrum dataset, and sequentially go through steps S2 to S6. Then, iterate and train until convergence to complete the optimized processing of the traction and stretching motion state monitoring signal.
[0065] In step S7, for the traction and stretching motion state monitoring signal processing model, a signal spectrum dataset containing trajectory change structure and temporal discrete sampling features is first input, and the processing stages defined in steps S2 to S6 are executed sequentially. Specifically, in the self-evolutionary feature modeling stage, a structural response mapping relationship under the time series index is constructed based on the signal phase components to obtain a self-evolutionary feature map reflecting the dynamic changes of the traction and stretching state, providing a temporal structure prior for subsequent multi-channel feature representation; in the multi-channel deep feature extraction stage, through the introduction of convolutional mapping strategy and path dissipation factor, the inter-channel difference propagation and response combination operation are completed, enhancing the expressive power of features under cross-channel coupling conditions; in the energy... In the adjustment modeling stage, an energy binding factor is introduced by calculating the temporal energy distribution, establishing a channel scheduling correlation matrix, and reconstructing the original features directionally to generate a directionally enhanced energy adjustment feature map. Subsequently, in the feature sequence construction stage, the directionally enhanced feature maps are sequentially spliced according to time steps to obtain a feature sequence that can completely reflect the evolution process of the traction and stretching state. By integrating the above stages end-to-end and combining iterative training mechanisms and structured optimization strategies, the model can adaptively adjust the feature modeling and energy adjustment strategies for different traction and stretching states, and finally output an optimized traction and stretching motion state monitoring signal, which significantly improves the recognition accuracy and stability under complex motion states.
[0066] Furthermore, in step S7, the constructed traction and stretching motion state monitoring signal processing model was written in Python and implemented using the PyTorch deep learning framework. The training process used the Adam optimizer, with an initial learning rate of 0.0003, a batch size of 64, and a total number of training epochs of 200. The cross-entropy loss function was used as the target optimization index. During training, parameter sharing and updates were performed on the structured convolution module and the orientation modulation module, and a linear annealing learning rate strategy was adopted. The overall loss function of the model tended to stabilize around the 160th epoch, indicating that the feature mapping and sequence fusion process had reached a convergence state and could accurately recover the state evolution characteristics of the signal during the traction process.
[0067] Furthermore, in step S7, the traction tensile state signal data is input into the constructed traction tensile state monitoring model for processing, and the experimental results are as follows: Figure 7 and Figure 8 As shown; Figure 7 This is a line graph comparing the channel characteristics, showing the temporal characteristics of the original and enhanced signals across multiple channels; the horizontal axis represents the sampling point number (sampling interval = 20), and the vertical axis represents the characteristic value. The dashed line represents the trend of the original signal, and the solid line represents the trend of the enhanced signal. Figure 7 It can be seen that the original signal fluctuates greatly between sampling points and has obvious noise interference, while the overall trend of the enhanced signal is clearer and more continuous, and the cross-channel features show higher consistency, indicating that the model effectively suppresses noise and enhances the key state evolution features. Figure 8 It consists of two grayscale heatmaps: the original feature grayscale image (a) and the enhanced feature grayscale image (b); the horizontal axis represents the time step, and the vertical axis represents the channel index; the grayscale patches represent the feature intensity distribution at different time points; from Figure 8 As can be seen in (a), the grayscale distribution in the original feature grayscale image is rather chaotic, the temporal evolution pattern is blurred, and the boundaries of state changes are not obvious; while Figure 8 In (b), the enhanced feature grayscale image exhibits clear grayscale transitions, regular feature distribution, and a prominent black-and-white boundary in the state change region, with high consistency in grayscale patterns between channels; in summary, from Figure 7 and Figure 8 It can be verified that the traction and tension state monitoring model proposed in this invention can effectively reduce noise interference while retaining and strengthening the core characteristics of the traction state, and achieve a clear presentation of the state evolution process, thus proving the effectiveness and robustness of this invention in the task of traction and tension signal monitoring.
[0068] The above description is only a preferred embodiment of the present invention. Those skilled in the art can make several substitutions, adjustments and combinations without departing from the concept of the present invention, and these modifications and improvements should be considered within the scope of protection of the present invention.
Claims
1. A method for monitoring the state of traction and tensile motion based on signal processing, characterized in that, Includes the following steps: S1. Collect multi-channel raw signals of the traction and tensioning device under different operating conditions, perform normalization and filtering processing, extract phase components and amplitude components, and construct a signal spectrum dataset containing trajectory change structure and time-series discrete features. S2. Based on the phase components in the signal spectrum dataset, establish the structural response mapping relationship under the time series index and construct a self-evolutionary feature map that reflects the traction and stretching state change process. S3. Align the self-evolving feature map with the amplitude components in time, construct a channel-level feature matrix group, and generate a multi-channel feature representation map; S4. Based on the multi-channel feature representation map, a structure path-aware convolutional mapping strategy is adopted, a path dissipation factor is introduced, and inter-channel difference propagation and response combination operations are performed to obtain a multi-channel deep feature map. S5. Calculate the temporal energy distribution of each channel of the depth feature map, introduce an energy binding factor, establish a scheduling correlation matrix between channels, reconstruct the original features in a directional manner, and generate a directionally enhanced energy-adjusted feature map. S6. Concatenate the feature vectors of each channel in the direction enhancement feature map according to the time step to obtain the feature sequence reflecting the changes in traction and tension state. S7. Construct a traction and stretching motion state monitoring signal processing model, input the signal spectrum dataset, and iteratively train through steps S2 to S6 until convergence, thereby achieving optimized processing of the traction and stretching motion state monitoring signal.
2. The method for monitoring traction and tensile motion state based on signal processing according to claim 1, characterized in that, The signal spectrum dataset includes multi-channel raw signal data acquired by the traction and stretching device under different operating conditions, including traction force response signals, physiological monitoring signals, electromyographic signals, motion acceleration signals, and posture angle sensing signals. A time-series response sequence under a continuous time window is established through time-division sampling. Normalization and filtering are performed on the signal data to extract amplitude features and phase changes. A channel feature spectrum characterizing the energy distribution structure and dynamic change trend is constructed and stacked according to the channel dimension to form a signal spectrum dataset containing trajectory change structure and time-series discrete sampling features.
3. The method for monitoring traction and tensile motion state based on signal processing according to claim 2, characterized in that, Constructing the self-evolutionary feature map includes: S21. Based on the phase component information of each channel in the signal spectrum dataset, extract the channel-level phase component sequences under the time series index to construct the phase component matrix. ; S22. Based on the phase component matrix, establish a metric relationship for the structural response change between adjacent time points, and construct a structural response mapping vector. The mathematical model is: ; in, This represents the structural response intensity at time index t; S23. Map the structural response vectors under all time indices. Combining them in chronological order constitutes a self-evolutionary feature map. The mathematical model is: ; Among them, represents the final self-evolutionary feature map.
4. The method for monitoring traction and tensile motion state based on signal processing according to claim 3, characterized in that, The process of constructing a multi-channel 3D feature map includes: S31. Based on the self-evolutionary feature map The amplitude components extracted from the signal spectrum dataset Establish a time index mapping function Align the amplitude components with the structural response time index in the time dimension to construct the aligned amplitude matrix. The mathematical model is as follows: , ; S32, According to the amplitude matrix Self-evolving eigenvalues in each channel Aligned amplitude By combining them sequentially along the time dimension, a channel-level feature matrix group is constructed. The channel-level feature matrix of each channel The joint feature column vector is represented as: ; S33, Channel-level feature matrices corresponding to all channels Stacking along the dimension of the channel set C yields a multi-channel feature representation map. The mathematical model is: 。 5. The method for monitoring traction and tensile motion state based on signal processing according to claim 4, characterized in that, The process of constructing a multi-channel deep feature map includes: S41. Based on the multi-channel feature representation diagram We construct a path-aware convolutional mapping strategy, define the path mapping kernel tensor, and the mathematical model is as follows: ; in, Indicates the first The channel is directed to the r-th feature dimension and time position s. Channel transmission structure perceives convolution weights; S42. Construct a path dissipation factor based on the exponential decay relationship between path information density and channel propagation intensity. The mathematical model is: ; in, This represents the path accessibility strength between channels c1 and c2, with α>0 being a regulatory factor. S43. Perform inter-channel difference propagation operation, define the channel response difference at each time t, and construct the inter-channel difference propagation modulation factor matrix. The mathematical model is: ; in, This represents the adjustment factor for the difference in response between channels c1 and c2 at time t; S44, Path Mapping Weight Tensor Path dissipation factor With the propagation regulation factor matrix Combine to generate multi-channel depth feature maps Mathematical model: ; in, This represents the depth feature response value of the c2th channel at time point t.
6. The method for monitoring traction and tensile motion state based on signal processing according to claim 5, characterized in that, The process of constructing a directionally enhanced energy-modulated feature map includes: S51. Calculate the multi-channel depth feature map. The energy distribution of each channel in the time dimension is used to construct an energy mapping matrix. The mathematical model is: ; in, This represents the value of channel c in the depth feature map at time point t. This represents the energy response value at the corresponding time point; S52, Based on Energy Mapping Matrix Calculate the intensity of energy coordination and variation between channels over time, and construct a scheduling correlation matrix. The mathematical model is: , ; in, This represents the energy co-factor between channel i and channel j; S53. Based on the scheduling association matrix Construct a set of directional adjustment node coefficients The original deep features are reconstructed directionally to generate a directionally enhanced feature map. The mathematical model is: 。 7. The method for monitoring traction and tensile motion state based on signal processing according to claim 6, characterized in that, The process of constructing a directionally enhanced energy-modulated feature map includes: S501 constructs an energy binding factor based on the inverse decay relationship between the energy term at time points and the adjustment sensitivity. The mathematical model is: ; Where β>0 represents the energy regulation sensitivity coefficient. ∈(0,1] represents the degree of compression of the directional enhancement feature update at time point t; S502, Enhance the directional feature map With energy mapping matrix Perform channel-dimensional weighted fusion to generate a direction-enhanced energy-modulated feature map. Its mathematical model is: ; in, This represents the directional enhancement feature value at time point t in channel c after fusion.
8. The method for monitoring traction and tensile motion state based on signal processing according to claim 7, characterized in that, The process of constructing the traction-stretch state evolution feature sequence includes: S61, Energy modulation feature map with enhanced directionality The time feature vector is constructed by combining the channel response values at each time step. Its mathematical model is: ; in, This represents the vector-like response composed of the features of each channel at the current time step; S62. Convert the time feature vectors corresponding to each time step. Construct a feature sequence matrix of the traction-stretching state evolution process by assembling the sequences sequentially. The mathematical model is: , in, It represents the ordered expression structure of the multi-channel dynamic characteristic response during traction and stretching.
Citation Information
Patent Citations
Intelligent fire-fighting device
CN112843563A
Intelligent detection method and system for grafting of sunshine rose grapes
CN113989529A
Forklift driver behavior monitoring method based on point cloud data processing and edge extraction
CN119251806A
Electroencephalogram signal real-time decoding and intention recognition method and equipment for attention glasses
CN119740102A
Attention state recognition neural network modeling and reasoning method based on electroencephalogram sequence
CN120123752A
Cited By
Feature extraction and identification method for radio frequency signal of unmanned aerial vehicle
CN122065131A
