Weak laryngeal electromyographic signal interpretation method and device based on cascade wavelet decomposition
The weak laryngeal electromyography signals are decomposed and characterized at multiple levels by the cascade wavelet decomposition method, which solves the problem of insufficient accuracy and robustness of laryngeal electromyography signal decoding in the existing technology and achieves a more efficient interpretation effect of weak laryngeal electromyography signals.
Patent Information
- Application Number
- CN202510523854.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-24
- Publication Date
- 2025-09-09
AI Technical Summary
Existing methods for decoding weak laryngeal electromyographic signals lack in-depth modeling of the nonlinear spatiotemporal dependencies of complex laryngeal electromyographic information, resulting in insufficient signal decoding accuracy and robustness, which limits their application in speech rehabilitation and human-computer interaction.
The cascade wavelet decomposition method is adopted to perform signal enhancement and feature extraction through the cascade wavelet transform weak laryngeal EMG signal enhancement network and the view-aware differential weak laryngeal EMG signal interpretation network, and a weak laryngeal EMG signal interpretation model based on cascade wavelet decomposition is constructed.
It improves the signal-to-noise ratio and interpretation accuracy of the signal, enhances the feature extraction ability and generalization performance of the model, can better adapt to the complex characteristics of laryngeal electromyography signals, and enhances the accuracy and robustness of the interpretation of weak laryngeal electromyography signals.
Smart Images
Figure CN120605024A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the fields of signal processing and human-computer interaction, and in particular to a method and device for interpreting weak laryngeal electromyographic signals based on cascaded wavelet decomposition. Background Art
[0002] In recent years, existing human-computer interaction methods have problems such as being single and inefficient, having poor universality, lacking confidentiality, and easily detected transmitted information, especially in application scenarios that require high confidentiality and low energy consumption. Therefore, it is imperative to study how to achieve safer, covert, and efficient human-computer interaction through biological signals. In this context, the interpretation and application of weak laryngeal electromyographic signals has gradually become a research hotspot.
[0003] By decoding laryngeal EMG signals, speech generation and control can be achieved, significantly improving the communication abilities of special populations (such as those with aphasia or vocal cord damage). At the same time, laryngeal EMG signals can be used for covert communication, enabling low-power, imperceptible human-computer interaction. However, interpreting laryngeal EMG signals still faces many challenges, such as how to extract high-signal-to-noise ratio signal features from noise, how to effectively capture weak and complex laryngeal muscle movement characteristics, and how to accurately map these signals to actual intentions.
[0004] Current methods for decoding weak laryngeal electromyographic signals mainly focus on feature extraction in a single time dimension and lack in-depth modeling of the nonlinear spatiotemporal dependencies in complex laryngeal electromyographic information. In addition, in the decoding process of weak laryngeal electromyographic signals, existing methods mostly stay at the shallow feature extraction level and cannot fully explore the deep correlations between the multi-dimensional data of laryngeal electromyographic signals, which limits the accuracy and robustness of decoding and restricts the further application of this technology in practical scenarios such as speech rehabilitation and human-computer interaction. Summary of the Invention
[0005] Based on this, it is necessary to propose a method and device for interpreting weak laryngeal electromyographic signals based on cascaded wavelet decomposition to address the above problems.
[0006] The present application provides a method for interpreting weak laryngeal electromyographic signals based on cascaded wavelet decomposition, the method comprising:
[0007] Acquire a weak laryngeal electromyographic signal set and a real command set, wherein the weak laryngeal electromyographic signal set includes a plurality of weak laryngeal electromyographic signals collected from the human larynx by an electromyographic sensor, and the real command set includes a plurality of commands representing thoughts and intentions, and each weak laryngeal electromyographic signal corresponds to a command in the real command set;
[0008] Constructing a cascade wavelet transform weak laryngeal electromyography signal enhancement network;
[0009] Construct a view-aware differential weak laryngeal electromyographic signal interpretation network;
[0010] Using the weak laryngeal electromyography signal set as a training set and the real instruction set as a label set, training a cascaded wavelet transform weak laryngeal electromyography signal enhancement network and a view-aware differential weak laryngeal electromyography signal interpretation network, and using the trained cascaded wavelet transform weak laryngeal electromyography signal enhancement network and the view-aware differential weak laryngeal electromyography signal interpretation network as a weak laryngeal electromyography signal interpretation model based on cascaded wavelet decomposition;
[0011] Obtain a weak laryngeal electromyographic signal T to be interpreted;
[0012] A weak laryngeal electromyographic signal to be interpreted is input into a weak laryngeal electromyographic signal interpretation model based on cascade wavelet decomposition to obtain an interpretation result.
[0013] In some embodiments, inputting a weak laryngeal electromyographic signal to be interpreted into a weak laryngeal electromyographic signal interpretation model based on cascade wavelet decomposition to obtain an interpretation result specifically includes:
[0014] performing signal enhancement on the weak laryngeal electromyography signal to be interpreted by using the cascaded wavelet transform weak laryngeal electromyography signal enhancement network to obtain an enhanced laryngeal electromyography signal;
[0015] The enhanced laryngeal electromyography signal is subjected to feature extraction and classification by the view-aware differential weak laryngeal electromyography signal interpretation network to obtain a final interpretation result.
[0016] In some embodiments, the step of performing signal enhancement on the weak laryngeal electromyography signal to be interpreted by the cascaded wavelet transform weak laryngeal electromyography signal enhancement network to obtain an enhanced laryngeal electromyography signal specifically includes:
[0017] according to Perform wavelet transform on a weak laryngeal electromyographic signal T to be interpreted and obtain the low-frequency component of the signal Sum signal high frequency component
[0018] Then the low frequency component of the signal Perform wavelet transform and repeat the above decomposition process until the preset wavelet decomposition layer number i is reached to obtain the low-frequency component of the signal Sum signal high frequency component
[0019] according to The high frequency component of the signal Use one-dimensional depth-separable convolution for enhancement processing to obtain the enhanced high-frequency components of the signal
[0020] according to The enhanced high-frequency component of the signal and the low-frequency component of the signal Perform reorganization to obtain the enhanced laryngeal electromyographic signal Z;
[0021] Among them, WT is wavelet transform, W (i) is the weight of the i-th one-dimensional depth-wise separable convolution, Z (i) is the enhanced high-frequency component of the signal Sum signal low frequency component The enhanced laryngeal electromyographic signals were reconstructed by inverse wavelet transform (IWT).
[0022] In some embodiments, the feature extraction and classification of the enhanced laryngeal electromyography signal by the view-aware differential weak laryngeal electromyography signal interpretation network to obtain a final interpretation result specifically includes:
[0023] The enhanced laryngeal electromyographic signal Z is converted from one dimension to a two-dimensional structure by stacking signal channels to obtain a two-dimensional enhanced laryngeal electromyographic signal Z′;
[0024] Performing feature extraction on the two-dimensionally enhanced laryngeal electromyographic signal Z′ to obtain feature information H;
[0025] The feature information H is globally pooled to extract the global feature vector, and a classification result is generated through a linear classifier to obtain the final interpretation result A.
[0026] In some embodiments, extracting features from the two-dimensionally enhanced laryngeal electromyographic signal Z′ to obtain feature information H specifically includes:
[0027] The two-dimensional enhanced laryngeal electromyographic signal Z′ is subjected to horizontal and vertical difference calculations by a bidirectional differential convolution block to obtain a gradient enhancement feature G;
[0028] Performing feature extraction on the gradient enhancement feature G through a view conversion feature enhancement layer to obtain a view enhancement feature V;
[0029] Batch normalization and nonlinear activation are performed on the view enhancement feature V to obtain feature information H.
[0030] In some embodiments, the bidirectional differential convolution block is composed of a horizontal differential convolution, a vertical differential convolution and a standard two-dimensional convolution in parallel, wherein the horizontal differential convolution is used to extract the horizontal gradient enhancement feature of the laryngeal electromyography signal through horizontal differential calculation, the vertical differential convolution is used to extract the vertical gradient enhancement feature of the laryngeal electromyography signal through vertical differential calculation, and the standard convolution is used to extract the original explicit features of the laryngeal electromyography signal, and the above features are fused by convolution weighting to obtain the gradient enhancement feature G.
[0031] In some embodiments, the view conversion feature enhancement layer performs a dimensional conversion on the gradient enhancement feature G, so that the gradient enhancement feature G is converted from a multi-sensor perspective under different views to a multi-view perspective under different sensors, and extracts the features under single sensor multi-view through standard two-dimensional convolution, and then performs a dimensional inversion to obtain the view enhancement feature V.
[0032] In some embodiments, the weak laryngeal electromyography signal set is used as a training set, the real instruction set is used as a label set, the cascade wavelet transform weak laryngeal electromyography signal enhancement network and the view-aware differential weak laryngeal electromyography signal interpretation network are trained, and the trained cascade wavelet transform weak laryngeal electromyography signal enhancement network and the view-aware differential weak laryngeal electromyography signal interpretation network are used as the weak laryngeal electromyography signal interpretation model based on cascade wavelet decomposition, specifically comprising: the weak laryngeal electromyography signal set T {Set} As a training set, the real instruction set R {Set} As the label set, the cascade wavelet transform weak laryngeal electromyography signal enhancement network and the view-aware differential weak laryngeal electromyography signal interpretation network are trained. During training, the loss function L is calculated based on the interpretation results and the real instructions after each iteration. cls , update the model parameters and proceed to the next iteration.
[0033] In some embodiments, the loss function L cls for y ij Indicates the real instruction, x ij represents the interpretation result, C represents the number of instruction categories of weak laryngeal electromyographic signals, and n is the number of samples.
[0034] The embodiment of the present application further provides a weak laryngeal electromyographic signal interpretation device based on cascade wavelet decomposition, the weak laryngeal electromyographic signal interpretation device based on cascade wavelet decomposition comprising:
[0035] an acquisition module, configured to acquire a weak laryngeal electromyographic signal set and a real instruction set, wherein the weak laryngeal electromyographic signal set includes a plurality of weak laryngeal electromyographic signals collected from the human larynx by an electromyographic sensor, and the real instruction set includes a plurality of instructions representing thoughts and intentions, and each weak laryngeal electromyographic signal corresponds to an instruction in the real instruction set;
[0036] Enhanced network building module, used to construct a cascaded wavelet transform weak laryngeal electromyography signal enhancement network;
[0037] An interpretation network construction module, used to construct a view-aware differential weak laryngeal electromyography signal interpretation network;
[0038] Using the weak laryngeal electromyography signal set as a training set and the real instruction set as a label set, training a cascaded wavelet transform weak laryngeal electromyography signal enhancement network and a view-aware differential weak laryngeal electromyography signal interpretation network, and using the trained cascaded wavelet transform weak laryngeal electromyography signal enhancement network and the view-aware differential weak laryngeal electromyography signal interpretation network as a weak laryngeal electromyography signal interpretation model based on cascaded wavelet decomposition;
[0039] An electrical signal acquisition module, used to acquire a weak laryngeal electromyographic signal T to be interpreted;
[0040] The interpretation module is used to input a weak laryngeal electromyographic signal to be interpreted into a weak laryngeal electromyographic signal interpretation model based on cascade wavelet decomposition to obtain an interpretation result.
[0041] An embodiment of the present application further provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the following steps:
[0042] Acquire a weak laryngeal electromyographic signal set and a real command set, wherein the weak laryngeal electromyographic signal set includes a plurality of weak laryngeal electromyographic signals collected from the human larynx by an electromyographic sensor, and the real command set includes a plurality of commands representing thoughts and intentions, and each weak laryngeal electromyographic signal corresponds to a command in the real command set;
[0043] Constructing a cascade wavelet transform weak laryngeal electromyography signal enhancement network;
[0044] Construct a view-aware differential weak laryngeal electromyographic signal interpretation network;
[0045] Using the weak laryngeal electromyography signal set as a training set and the real instruction set as a label set, training a cascaded wavelet transform weak laryngeal electromyography signal enhancement network and a view-aware differential weak laryngeal electromyography signal interpretation network, and using the trained cascaded wavelet transform weak laryngeal electromyography signal enhancement network and the view-aware differential weak laryngeal electromyography signal interpretation network as a weak laryngeal electromyography signal interpretation model based on cascaded wavelet decomposition;
[0046] Obtain a weak laryngeal electromyographic signal T to be interpreted;
[0047] A weak laryngeal electromyographic signal to be interpreted is input into a weak laryngeal electromyographic signal interpretation model based on cascade wavelet decomposition to obtain an interpretation result.
[0048] The present application also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor performs the following steps:
[0049] Acquire a weak laryngeal electromyographic signal set and a real command set, wherein the weak laryngeal electromyographic signal set includes a plurality of weak laryngeal electromyographic signals collected from the human larynx by an electromyographic sensor, and the real command set includes a plurality of commands representing thoughts and intentions, and each weak laryngeal electromyographic signal corresponds to a command in the real command set;
[0050] Constructing a cascade wavelet transform weak laryngeal electromyography signal enhancement network;
[0051] Construct a view-aware differential weak laryngeal electromyographic signal interpretation network;
[0052] Using the weak laryngeal electromyography signal set as a training set and the real instruction set as a label set, training a cascaded wavelet transform weak laryngeal electromyography signal enhancement network and a view-aware differential weak laryngeal electromyography signal interpretation network, and using the trained cascaded wavelet transform weak laryngeal electromyography signal enhancement network and the view-aware differential weak laryngeal electromyography signal interpretation network as a weak laryngeal electromyography signal interpretation model based on cascaded wavelet decomposition;
[0053] Obtain a weak laryngeal electromyographic signal T to be interpreted;
[0054] A weak laryngeal electromyographic signal to be interpreted is input into a weak laryngeal electromyographic signal interpretation model based on cascade wavelet decomposition to obtain an interpretation result.
[0055] The embodiments of the present application have the following beneficial effects:
[0056] In the modeling method for interpreting weak laryngeal electromyography signals based on cascaded wavelet decomposition provided in the embodiment of the present application, a weak laryngeal electromyography signal set and a real instruction set are obtained. The weak laryngeal electromyography signal usually has a low signal-to-noise ratio and contains rich physiological information. The real instruction set is an instruction corresponding to the electromyography signal that represents the thought and intention. A cascaded wavelet transform weak laryngeal electromyography signal enhancement network is constructed. The network decomposes the signal through cascaded wavelet transform, extracts high-frequency and low-frequency components, and enhances the high-frequency components to improve the signal-to-noise ratio of the signal. A view-aware differential weak laryngeal electromyography signal interpretation network is constructed: the network extracts the feature information of the signal through differential convolution and view-aware mechanism, and generates interpretation results through a classification module. By combining the structural design of the cascaded wavelet transform signal enhancement network and the view-aware differential convolution network, it can better adapt to the complex characteristics of laryngeal electromyography signals and improve the feature extraction ability and generalization performance of the model. This method not only improves the signal-to-noise ratio and interpretation accuracy of the signal, but also provides new ideas and methods for the development of weak laryngeal electromyography signal interpretation technology. BRIEF DESCRIPTION OF THE DRAWINGS
[0057] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.
[0058] in:
[0059] Figure 1 1 is a flow chart of a method for interpreting weak laryngeal electromyographic signals based on cascaded wavelet decomposition in one embodiment;
[0060] Figure 2 1 is a schematic diagram of the overall process of a cascaded wavelet transform weak laryngeal electromyographic signal enhancement network when the number of layers is 3 in one embodiment;
[0061] Figure 3 is a schematic diagram of a bidirectional differential convolution module in an embodiment;
[0062] Figure 4 is a schematic diagram of a view conversion feature enhancement layer in an embodiment;
[0063] Figure 5 1 is a structural diagram of a weak laryngeal electromyographic signal interpretation device based on cascaded wavelet decomposition in one embodiment;
[0064] Figure 6 is a schematic structural diagram of a computer device in one embodiment;
[0065] Figure 7FIG. 1 is a schematic diagram of the structure of a computer-readable storage medium in one embodiment. DETAILED DESCRIPTION
[0066] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.
[0067] In the embodiment of the present application, a method for interpreting weak laryngeal electromyographic signals based on cascade wavelet decomposition is provided. Figure 1 , Figure 1 1 is a flow chart of a method for interpreting weak laryngeal electromyographic signals based on cascaded wavelet decomposition in one embodiment; the method for interpreting weak laryngeal electromyographic signals based on cascaded wavelet decomposition includes steps S1 to S6.
[0068] Step S1, obtaining a weak laryngeal electromyographic signal set and a real instruction set, wherein the weak laryngeal electromyographic signal set includes multiple weak laryngeal electromyographic signals collected from the human larynx by an electromyographic sensor, and the real instruction set includes multiple instructions representing thoughts and intentions, and each weak laryngeal electromyographic signal corresponds to an instruction in the real instruction set;
[0069] Specifically, the instruction R representing the thought intention is such as "forward", "backward" and the like.
[0070] Step S2, constructing a cascade wavelet transform weak laryngeal electromyography signal enhancement network;
[0071] Specifically, the cascaded wavelet transform weak laryngeal electromyography signal enhancement network includes a signal decomposition module, a signal enhancement module and a signal recombination module. The weak laryngeal electromyography signal T is enhanced by the cascaded wavelet transform weak laryngeal electromyography signal enhancement network to obtain an enhanced laryngeal electromyography signal Z.
[0072] Step S3, constructing a view-aware differential weak laryngeal electromyography signal interpretation network;
[0073] Specifically, the view-aware differential weak laryngeal electromyography signal interpretation network includes a dimensionality conversion layer, a view-aware differential convolution module and a classification module. The view-aware differential weak laryngeal electromyography signal interpretation network is used to perform feature extraction and classification on the enhanced laryngeal electromyography signal to obtain the final interpretation result A.
[0074] Step S4, using the weak laryngeal electromyography signal set as a training set and the real instruction set as a label set, training a cascaded wavelet transform weak laryngeal electromyography signal enhancement network and a view-aware differential weak laryngeal electromyography signal interpretation network, and using the trained cascaded wavelet transform weak laryngeal electromyography signal enhancement network and the view-aware differential weak laryngeal electromyography signal interpretation network as a weak laryngeal electromyography signal interpretation model based on cascaded wavelet decomposition;
[0075] Specifically, the weak laryngeal electromyographic signal set T {Set} As a training set, the real instruction set R {Set} As the label set, the cascade wavelet transform weak laryngeal electromyography signal enhancement network and the view-aware differential weak laryngeal electromyography signal interpretation network are trained. During training, the loss function L is calculated based on the interpretation results and the real instructions after each iteration. cls , update the model parameters and proceed to the next iteration.
[0076] Step S5: obtaining a weak laryngeal electromyographic signal T to be interpreted;
[0077] Step S6: inputting a weak laryngeal electromyographic signal to be interpreted into a weak laryngeal electromyographic signal interpretation model based on cascade wavelet decomposition to obtain an interpretation result.
[0078] Specifically, if Figure 2 As shown, the weak laryngeal electromyography signal to be interpreted is enhanced by the cascade wavelet transform weak laryngeal electromyography signal enhancement network to obtain an enhanced laryngeal electromyography signal;
[0079] The enhanced laryngeal electromyography signal is subjected to feature extraction and classification by the view-aware differential weak laryngeal electromyography signal interpretation network to obtain a final interpretation result.
[0080] The step of performing signal enhancement on the weak laryngeal electromyography signal to be interpreted by the cascaded wavelet transform weak laryngeal electromyography signal enhancement network to obtain an enhanced laryngeal electromyography signal specifically includes:
[0081] according to Perform wavelet transform on a weak laryngeal electromyographic signal T to be interpreted and obtain the low-frequency component of the signal Sum signal high frequency component
[0082] Then the low frequency component of the signal Perform wavelet transform and repeat the above decomposition process until the preset wavelet decomposition layer number i is reached to obtain the low-frequency component of the signal Sum signal high frequency component
[0083] Specifically, after the first wavelet transform of the laryngeal electrical signal, the low-frequency component of the signal is obtained. and high frequency components Subsequently, the low-frequency component of the obtained signal is subjected to a second wavelet transform and decomposed into the low-frequency component of the signal again and high frequency components Repeat the above steps for multiple decompositions until the preset decomposition times i is reached. The high-frequency components obtained from each decomposition do not participate in the wavelet decomposition process, but are only used for the subsequent depthwise separable convolution enhancement processing and subsequent recombination operations.
[0084] according to The high frequency component of the signal Use one-dimensional depth-separable convolution for enhancement processing to obtain the enhanced high-frequency components of the signal
[0085] according to The enhanced high-frequency component of the signal and the low-frequency component of the signal Perform reorganization to obtain the enhanced laryngeal electromyographic signal Z;
[0086] Among them, WT is wavelet transform, W (i) is the weight of the i-th one-dimensional depth-wise separable convolution, Z (i) is the enhanced high-frequency component of the signal Sum signal low frequency component The enhanced laryngeal electromyographic signals were reconstructed by inverse wavelet transform (IWT).
[0087] The process of extracting and classifying the enhanced laryngeal electromyography signal through the view-aware differential weak laryngeal electromyography signal interpretation network to obtain the final interpretation result specifically includes:
[0088] The enhanced laryngeal electromyographic signal Z is converted from one dimension to a two-dimensional structure by stacking signal channels to obtain a two-dimensional enhanced laryngeal electromyographic signal Z′;
[0089] Performing feature extraction on the two-dimensionally enhanced laryngeal electromyographic signal Z′ to obtain feature information H;
[0090] The feature information H is globally pooled to extract the global feature vector, and a classification result is generated through a linear classifier to obtain the final interpretation result A.
[0091] The step of extracting features from the two-dimensionally enhanced laryngeal electromyographic signal Z′ to obtain feature information H specifically includes:
[0092] The two-dimensional enhanced laryngeal electromyographic signal Z′ is subjected to horizontal and vertical difference calculations by a bidirectional differential convolution block to obtain a gradient enhancement feature G;
[0093] Performing feature extraction on the gradient enhancement feature G through a view conversion feature enhancement layer to obtain a view enhancement feature V;
[0094] Batch normalization and nonlinear activation are performed on the view enhancement feature V to obtain feature information H.
[0095] Among them, such as Figure 3 As shown in FIG, the bidirectional differential convolution block is composed of a horizontal differential convolution, a vertical differential convolution and a standard two-dimensional convolution in parallel, wherein the horizontal differential convolution is used to extract the horizontal gradient enhancement feature of the laryngeal electromyography signal by horizontal differential calculation, the vertical differential convolution is used to extract the vertical gradient enhancement feature of the laryngeal electromyography signal by vertical differential calculation, and the standard convolution is used to extract the original explicit features of the laryngeal electromyography signal, and the above features are fused by convolution weighting to obtain the gradient enhancement feature G.
[0096] Among them, such as Figure 4 As shown, the view conversion feature enhancement layer performs a dimensional conversion on the gradient enhancement feature G, so that the gradient enhancement feature G is converted from a multi-sensor perspective under different views to a multi-view perspective under different sensors, and extracts the features under single sensor multi-view through standard two-dimensional convolution, and then performs a dimensional inversion to obtain the view enhancement feature V.
[0097] By adopting the technical solution of this embodiment, the weak laryngeal electromyography signal is subjected to multi-level decomposition processing through cascaded wavelet transform, the high-frequency information representation in the signal is further extracted, and the signal is effectively enhanced in fine granularity, overcoming the problem that the laryngeal electromyography signal is weak and it is difficult to capture key features in the prior art, so that the present application has the advantage of being able to better extract the feature expression of the weak laryngeal electromyography signal. The horizontal gradient and vertical gradient of the laryngeal electromyography signal are enhanced by horizontal and vertical difference calculations, the implicit gradient information in the laryngeal electromyography signal is mined, and the feature expression under different views of the signal is further fused through the view conversion feature enhancement layer, effectively improving the mining and fusion of laryngeal electromyography signal information, overcoming the problem of insufficient signal information utilization and poor feature extraction capability in the prior art, so that the present invention has the advantage of being able to more comprehensively extract and fuse feature information in the weak laryngeal electromyography signal. By constructing an interpretation model that combines a cascaded wavelet transform signal enhancement network and a view-aware differential convolutional network, the global and local information of the laryngeal electromyography signal is captured, and the interpretation results are further generated through global pooling and linear classifiers. This overcomes the problem in the existing technology that it is impossible to simultaneously take into account the global pattern and local details of the signal, making this application have the advantage of being able to more accurately interpret weak laryngeal electromyography signals.
[0098] In the embodiment of the present application, a weak laryngeal electromyographic signal interpretation device based on cascade wavelet decomposition is provided. Figure 5 , Figure 5 1 is a structural diagram of a weak laryngeal electromyographic signal interpretation device based on cascade wavelet decomposition in one embodiment. The weak laryngeal electromyographic signal interpretation device based on cascade wavelet decomposition includes: an acquisition module 201, an enhancement network construction module 202, an interpretation network construction module 203, an interpretation model construction module 204, an electrical signal acquisition module 205, and an interpretation module 206.
[0099] The acquisition module is configured to acquire a weak laryngeal electromyographic signal set and a real instruction set, wherein the weak laryngeal electromyographic signal set includes a plurality of weak laryngeal electromyographic signals collected from the human larynx by an electromyographic sensor, and the real instruction set includes a plurality of instructions representing thoughts and intentions, and each weak laryngeal electromyographic signal corresponds to an instruction in the real instruction set;
[0100] Enhanced network building module, used to construct a cascaded wavelet transform weak laryngeal electromyography signal enhancement network;
[0101] An interpretation network construction module, used to construct a view-aware differential weak laryngeal electromyography signal interpretation network;
[0102] an interpretation model construction module, configured to use the weak laryngeal electromyography signal set as a training set and the real instruction set as a label set, train a cascaded wavelet transform weak laryngeal electromyography signal enhancement network and a view-aware differential weak laryngeal electromyography signal interpretation network, and use the trained cascaded wavelet transform weak laryngeal electromyography signal enhancement network and the view-aware differential weak laryngeal electromyography signal interpretation network as a weak laryngeal electromyography signal interpretation model based on cascaded wavelet decomposition;
[0103] An electrical signal acquisition module, used to acquire a weak laryngeal electromyographic signal T to be interpreted;
[0104] The interpretation module is used to input a weak laryngeal electromyographic signal to be interpreted into a weak laryngeal electromyographic signal interpretation model based on cascade wavelet decomposition to obtain an interpretation result.
[0105] For other details of the implementation of the above technical solution by each module in the weak laryngeal electromyography signal interpretation device based on cascade wavelet decomposition, please refer to the description of the weak laryngeal electromyography signal interpretation method based on cascade wavelet decomposition provided above, which will not be repeated here.
[0106] In the embodiment of the present application, a computer device is provided. Figure 6 , Figure 63 is a schematic diagram of the structure of a computer device in one embodiment. The device includes a memory 301 and a processor 302. The memory 301 stores a computer program. When the computer program is executed by the processor 302, the processor 302 performs the following steps:
[0107] Acquire a weak laryngeal electromyographic signal set and a real command set, wherein the weak laryngeal electromyographic signal set includes a plurality of weak laryngeal electromyographic signals collected from the human larynx by an electromyographic sensor, and the real command set includes a plurality of commands representing thoughts and intentions, and each weak laryngeal electromyographic signal corresponds to a command in the real command set;
[0108] Constructing a cascade wavelet transform weak laryngeal electromyography signal enhancement network;
[0109] Construct a view-aware differential weak laryngeal electromyographic signal interpretation network;
[0110] Using the weak laryngeal electromyography signal set as a training set and the real instruction set as a label set, training a cascaded wavelet transform weak laryngeal electromyography signal enhancement network and a view-aware differential weak laryngeal electromyography signal interpretation network, and using the trained cascaded wavelet transform weak laryngeal electromyography signal enhancement network and the view-aware differential weak laryngeal electromyography signal interpretation network as a weak laryngeal electromyography signal interpretation model based on cascaded wavelet decomposition;
[0111] Obtain a weak laryngeal electromyographic signal T to be interpreted;
[0112] A weak laryngeal electromyographic signal to be interpreted is input into a weak laryngeal electromyographic signal interpretation model based on cascade wavelet decomposition to obtain an interpretation result.
[0113] Among them, the processor 302 can also be called a CPU (Central Processing Unit), and the processor 302 may be an integrated circuit chip with signal processing capabilities; the processor 302 can also be a general-purpose processor, DSP (Digital Signal Processor), ASIC (Application Specific Integrated Circuit), FPGA (Field Programmable Gate Array) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, among which the general-purpose processor can be a microprocessor or the processor 302 can also be any conventional processor, etc.
[0114] In an embodiment of the present application, a computer readable storage medium is provided. Figure 7 , Figure 7FIG4 is a schematic diagram of the structure of a computer-readable storage medium in one embodiment, wherein the storage medium stores a readable computer program 401; wherein the computer program 401 may be stored in the storage medium in the form of a software product, and includes a number of instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) or a processor to perform the following steps:
[0115] Acquire a weak laryngeal electromyographic signal set and a real command set, wherein the weak laryngeal electromyographic signal set includes a plurality of weak laryngeal electromyographic signals collected from the human larynx by an electromyographic sensor, and the real command set includes a plurality of commands representing thoughts and intentions, and each weak laryngeal electromyographic signal corresponds to a command in the real command set;
[0116] Constructing a cascade wavelet transform weak laryngeal electromyography signal enhancement network;
[0117] Construct a view-aware differential weak laryngeal electromyographic signal interpretation network;
[0118] Using the weak laryngeal electromyography signal set as a training set and the real instruction set as a label set, training a cascaded wavelet transform weak laryngeal electromyography signal enhancement network and a view-aware differential weak laryngeal electromyography signal interpretation network, and using the trained cascaded wavelet transform weak laryngeal electromyography signal enhancement network and the view-aware differential weak laryngeal electromyography signal interpretation network as a weak laryngeal electromyography signal interpretation model based on cascaded wavelet decomposition;
[0119] Obtain a weak laryngeal electromyographic signal T to be interpreted;
[0120] A weak laryngeal electromyographic signal to be interpreted is input into a weak laryngeal electromyographic signal interpretation model based on cascade wavelet decomposition to obtain an interpretation result.
[0121] The aforementioned storage media include: USB flash drives, mobile hard drives, magnetic disks or optical disks, ROM (Read-Only Memory), RAM (Random Access Memory), and other media that can store program codes, or terminal devices such as computers, service machines, mobile phones, and tablets.
[0122] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program. The program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchl ink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
[0123] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0124] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.
Claims
1. A method for interpreting weak laryngeal electromyographic signals based on cascaded wavelet decomposition, characterized in that: The method includes: Acquire a weak laryngeal electromyographic signal set and a real command set, wherein the weak laryngeal electromyographic signal set includes a plurality of weak laryngeal electromyographic signals collected from the human larynx by an electromyographic sensor, and the real command set includes a plurality of commands representing thoughts and intentions, and each weak laryngeal electromyographic signal corresponds to a command in the real command set; Constructing a cascade wavelet transform weak laryngeal electromyography signal enhancement network; Construct a view-aware differential weak laryngeal electromyographic signal interpretation network; Using the weak laryngeal electromyography signal set as a training set and the real instruction set as a label set, training a cascaded wavelet transform weak laryngeal electromyography signal enhancement network and a view-aware differential weak laryngeal electromyography signal interpretation network, and using the trained cascaded wavelet transform weak laryngeal electromyography signal enhancement network and the view-aware differential weak laryngeal electromyography signal interpretation network as a weak laryngeal electromyography signal interpretation model based on cascaded wavelet decomposition; Obtain a weak laryngeal electromyographic signal T to be interpreted; A weak laryngeal electromyographic signal to be interpreted is input into a weak laryngeal electromyographic signal interpretation model based on cascade wavelet decomposition to obtain an interpretation result.
2. The method for interpreting weak laryngeal electromyographic signals based on cascaded wavelet decomposition according to claim 1, characterized in that: The step of inputting a weak laryngeal electromyographic signal to be interpreted into a weak laryngeal electromyographic signal interpretation model based on cascade wavelet decomposition to obtain an interpretation result specifically includes: performing signal enhancement on the weak laryngeal electromyography signal to be interpreted by using the cascaded wavelet transform weak laryngeal electromyography signal enhancement network to obtain an enhanced laryngeal electromyography signal; The enhanced laryngeal electromyography signal is subjected to feature extraction and classification by the view-aware differential weak laryngeal electromyography signal interpretation network to obtain a final interpretation result.
3. The method for interpreting weak laryngeal electromyographic signals based on cascaded wavelet decomposition according to claim 2, characterized in that: The step of performing signal enhancement on the weak laryngeal electromyographic signal to be interpreted by the cascaded wavelet transform weak laryngeal electromyographic signal enhancement network to obtain an enhanced laryngeal electromyographic signal specifically includes: according to Perform wavelet transform on a weak laryngeal electromyographic signal T to be interpreted and obtain the low-frequency component of the signal Sum signal high frequency component Then the low frequency component of the signal Perform wavelet transform and repeat the above decomposition process until the preset wavelet decomposition layer number i is reached to obtain the low-frequency component of the signal Sum signal high frequency component according to The high frequency component of the signal Use one-dimensional depth-separable convolution for enhancement processing to obtain the enhanced high-frequency components of the signal according to The enhanced high-frequency component of the signal and the low-frequency component of the signal Perform reorganization to obtain the enhanced laryngeal electromyographic signal Z; Among them, WT is wavelet transform, W (i) is the weight of the i-th one-dimensional depth-wise separable convolution, Z (i) is the enhanced high-frequency component of the signal Sum signal low frequency component The enhanced laryngeal electromyographic signals were reconstructed by inverse wavelet transform (IWT).
4. The method for interpreting weak laryngeal electromyographic signals based on cascaded wavelet decomposition according to claim 3, characterized in that: The method of extracting features and classifying the enhanced laryngeal electromyography signal through the view-aware differential weak laryngeal electromyography signal interpretation network to obtain a final interpretation result specifically includes: The enhanced laryngeal electromyographic signal Z is converted from one dimension to a two-dimensional structure by stacking signal channels to obtain a two-dimensional enhanced laryngeal electromyographic signal Z′; Performing feature extraction on the two-dimensionally enhanced laryngeal electromyographic signal Z′ to obtain feature information H; The feature information H is globally pooled to extract the global feature vector, and a classification result is generated through a linear classifier to obtain the final interpretation result A.
5. The method for interpreting weak laryngeal electromyographic signals based on cascaded wavelet decomposition according to claim 4, characterized in that: The feature extraction of the two-dimensionally enhanced laryngeal electromyographic signal Z′ to obtain feature information H specifically includes: The two-dimensional enhanced laryngeal electromyographic signal Z′ is subjected to horizontal and vertical difference calculations by a bidirectional differential convolution block to obtain a gradient enhancement feature G; Performing feature extraction on the gradient enhancement feature G through a view conversion feature enhancement layer to obtain a view enhancement feature V; Batch normalization and nonlinear activation are performed on the view enhancement feature V to obtain feature information H.
6. The method for interpreting weak laryngeal electromyographic signals based on cascaded wavelet decomposition according to claim 5, characterized in that: The bidirectional differential convolution block is composed of a horizontal differential convolution, a vertical differential convolution and a standard two-dimensional convolution in parallel, wherein the horizontal differential convolution is used to extract the horizontal gradient enhancement feature of the laryngeal electromyography signal through horizontal differential calculation, the vertical differential convolution is used to extract the vertical gradient enhancement feature of the laryngeal electromyography signal through vertical differential calculation, and the standard convolution is used to extract the original explicit features of the laryngeal electromyography signal, and the above features are fused by convolution weighting to obtain the gradient enhancement feature G.
7. The method for interpreting weak laryngeal electromyographic signals based on cascaded wavelet decomposition according to claim 6, characterized in that: The view conversion feature enhancement layer performs a dimensional conversion on the gradient enhancement feature G, so that the gradient enhancement feature G is converted from a multi-sensor perspective under different views to a multi-view perspective under different sensors, and extracts the features under single sensor multi-view through standard two-dimensional convolution, and then performs a dimensional inversion to obtain the view enhancement feature V.
8. The method for interpreting weak laryngeal electromyographic signals based on cascaded wavelet decomposition according to any one of claims 1 to 7, characterized in that: The method uses the weak laryngeal electromyography signal set as a training set and the real instruction set as a label set to train a cascade wavelet transform weak laryngeal electromyography signal enhancement network and a view-aware differential weak laryngeal electromyography signal interpretation network, and uses the trained cascade wavelet transform weak laryngeal electromyography signal enhancement network and the view-aware differential weak laryngeal electromyography signal interpretation network as a weak laryngeal electromyography signal interpretation model based on cascade wavelet decomposition, specifically comprising: using the weak laryngeal electromyography signal set T {Set} As a training set, the real instruction set R {Set} As the label set, the cascade wavelet transform weak laryngeal electromyography signal enhancement network and the view-aware differential weak laryngeal electromyography signal interpretation network are trained. During training, the loss function L is calculated based on the interpretation results and the real instructions after each iteration. cls , update the model parameters and proceed to the next iteration.
9. The method for interpreting weak laryngeal electromyographic signals based on cascaded wavelet decomposition according to any one of claims 1 to 7, characterized in that: The loss function L cls for y ij Indicates the real instruction, x ij represents the interpretation result, C represents the number of instruction categories of weak laryngeal electromyographic signals, and n is the number of samples.
10. A device for interpreting weak laryngeal electromyographic signals based on cascaded wavelet decomposition, characterized in that: The device includes: an acquisition module, configured to acquire a weak laryngeal electromyographic signal set and a real instruction set, wherein the weak laryngeal electromyographic signal set includes a plurality of weak laryngeal electromyographic signals collected from the human larynx by an electromyographic sensor, and the real instruction set includes a plurality of instructions representing thoughts and intentions, and each weak laryngeal electromyographic signal corresponds to an instruction in the real instruction set; Enhanced network building module, used to construct a cascaded wavelet transform weak laryngeal electromyography signal enhancement network; An interpretation network construction module, used to construct a view-aware differential weak laryngeal electromyography signal interpretation network; an interpretation model construction module, configured to use the weak laryngeal electromyography signal set as a training set and the real instruction set as a label set, train a cascaded wavelet transform weak laryngeal electromyography signal enhancement network and a view-aware differential weak laryngeal electromyography signal interpretation network, and use the trained cascaded wavelet transform weak laryngeal electromyography signal enhancement network and the view-aware differential weak laryngeal electromyography signal interpretation network as a weak laryngeal electromyography signal interpretation model based on cascaded wavelet decomposition; An electrical signal acquisition module, used to acquire a weak laryngeal electromyographic signal T to be interpreted; The interpretation module is used to input a weak laryngeal electromyographic signal to be interpreted into a weak laryngeal electromyographic signal interpretation model based on cascade wavelet decomposition to obtain an interpretation result.