Physiological signal compression system based on depth auto-encoder
Through a physiological signal compression system based on a deep autoencoder, combined with a fast Fourier transform and discriminator, the problems of low compression rate and insufficient reconstruction accuracy in the prior art are solved, and an efficient 360-fold compression rate and good Signal reconstruction quality, suitable for ECG data compression.
Patent Information
- Application Number
- CN202410159173.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-02-04
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2044-02-04
AI Technical Summary
The existing physiological signal compression methods have problems such as low compression rate, slow speed or low reconstruction accuracy, which is difficult to meet actual needs.
Adopt a physiological signal compression system based on deep autoencoder, converting the input signal into latent variables through the encoder, the decoder reconstructs the original signal, combining the fast Fourier transform module and the discriminator, the optimizer updates the neural network weight and improves compression rate and reconstruction quality.
With the guarantee that mean square error and normalized RMS difference percentage are acceptable, the compression rate is improved to 360 times and the signal reconstruction quality is significantly improved. It can run on a single-core CPU and meet the ECG data compression requirements in actual scenarios. .
Smart Images

Figure CN119030548A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of physiological signal compression, and in particular to a physiological signal compression system based on a deep autoencoder. Background Art
[0002] Existing physiological signal compression methods include compressed sensing, deep learning autoencoders, and transformation-based compression methods. Compressed sensing method: Randomly sample the input signal at a sampling rate lower than the Nyquist rate to obtain a compressed data representation. After the receiver receives the randomly sampled signal, it reconstructs the compressed data through the reconstruction matrix. This process is lossy compression. Deep autoencoder method: Construct a codec based on the encoder-decoder architecture in deep learning. The input signal is converted into a latent variable form by an encoder (i.e., encoder), and then the latent variable is restored to the original signal by a decoder (i.e., decoder). This process is lossy compression. Transformation-based compression methods: such as DCT (discrete cosine transform), DWT (discrete wavelet transform), and fast Fourier transform. This process is also lossy compression. The above three methods all have the problems of low compression rate, slow compression speed, or high compression rate but low reconstruction accuracy. Summary of the invention
[0003] In view of the shortcomings of the prior art, the present invention proposes a physiological signal compression system based on a deep autoencoder, which can increase the compression rate to 360 times while ensuring that the RMS (mean square error) and PRDN (normalized RMS percentage difference) are acceptable, and improve the signal reconstruction quality by adding a deep discriminator for training during the training phase.
[0004] To implement the above technical solution, the present invention provides a physiological signal compression system based on a deep autoencoder, including: an encoder, used to convert the input physiological signal into a latent variable Z; a decoder, used to reconstruct the original signal based on the latent variable Z; a fast Fourier transform module, used to calculate the frequency components of the input and output signals and obtain the spectrum; a discriminator, used to judge the degree of difference between the input signal characteristics and the output signal characteristics; an optimizer, used to update the neural network weights to obtain a better compression and reconstruction effect.
[0005] Preferably, the encoder consists of two convolutional layers at the head and tail, four convolutional blocks and an LSTM layer. After the input signal enters the encoder, the number of channels of the first convolutional layer input is consistent with the number of channels of the input signal, and the convolution kernel size is 7. It is followed by four convolutional blocks with the same structure. The number of channels doubles after each convolution block. The step lengths of the four convolutional blocks are 2, 4, 5, and 8, respectively, to achieve downsampling of the input signal. The downsampled signal will pass through the LSTM layer to extract timing information. The number of channels of the last convolutional layer is consistent with the number of channels of the last convolution block, the convolution kernel size is 3, and the output is the required latent variable Z.
[0006] Preferably, the convolution block consists of three residual units and one one-dimensional convolution layer, the input channel of the residual unit is 1 / 2 of the input channel of the convolution block, the dilation parameter increases successively to 1, 3, and 9, respectively, to obtain temporal features of different scales, the convolution kernel size of the convolution layer is twice the step length, and the number of channels and the step length are consistent with the input channel of the convolution block.
[0007] Preferably, the residual unit is composed of a one-dimensional hole convolution layer, a one-dimensional convolution layer and a residual connection. The input signal outputs features after passing through two convolution layers, and the output features are added to the input signal itself as the final output of the residual unit.
[0008] Preferably, the decoder structure is a mirror image of the encoder, and the operation of the decoder is the inverse process of the encoder operation. The input is the latent variable Z output by the encoder. First, a one-dimensional convolution layer is input, the convolution kernel size is 7, and the number of channels is 16 times the number of channels of the original signal; then the same LSTM layer as the encoder is input to restore the timing information, and then four decoding convolution blocks with the same structure are passed, with step lengths of 8, 5, 4, and 2 respectively. The number of channels is halved after each convolution block, and finally the original signal is restored through a one-dimensional convolution layer with a convolution kernel size of 7 and a channel number of 1.
[0009] Preferably, a single discriminator consists of an STFT block and 6 two-dimensional convolutional layers, Real represents the real part of the stft output result, Im represents the imaginary part of the stft output result, and the two parts are spliced to form a time-frequency feature map, which is then input into a two-dimensional convolutional layer with a convolution kernel size of 24 and converted into a 32-channel feature map, and then passes through three 32-channel, convolution kernel size of 3*8, and step length of (1, 2) two-dimensional convolutional layers, the dilation parameters are (1, 1), (2, 1), (4, 1), and finally passes through two convolutional layers with a convolution kernel size of 3*3 to obtain the final output result.
[0010] Preferably, the operation process of the system is divided into a training phase and an inference phase, wherein the training phase learns the features of the ECG signals of different patients based on the MIT-BIH ECG data set, maps the learned features to a lower dimension, removes the correlation to obtain the latent variable z, and the decoder reconstructs the ECG signal based on the acquired latent variable z. The encoder and decoder are trained together in the training phase to obtain the optimal encoder and decoder. The inference phase includes encoding and decoding processes, and the encoder and decoder are used separately. When compressing, the input signal is input into the encoder to obtain the compressed representation. When decompressing, the decoder is used to decompress the received compressed signal and reconstruct the original signal.
[0011] Preferably, the training process is as follows: the input raw data is first preprocessed and then input into the codec network to calculate the output reconstructed signal. According to each loss function, the corresponding loss is calculated, and the parameters of the codec and the discriminator are updated respectively according to the loss. The parameters of the encoder and decoder are updated together, and the discriminator parameters are updated separately according to the adversarial loss.
[0012] Preferably, the data should be preprocessed in the training stage. The data set used is the open source ECG database of MIT-BIH. The original data contains 48 ECG data, including the results collected under different leads, 650,000 data points for each lead, and the sampling rate is 360 Hz. The data is first divided into windows of 1 s, and a segment of data is cut into multiple 360-point data segments. The segmented data is inverted in time domain and amplitude to obtain the inverted data to obtain ECG features at different spatial scales and time scales. This operation is used to enhance the data and expand the data set to avoid the influence of different lead characteristics and reverse connection of positive and negative electrodes during ECG collection. In the training stage, the encoder and decoder are connected together, and the input data is directly input into the decoder after passing through the encoder to obtain the output, and the input is compared with the output. After calculating the three losses, the optimizer is used to update the weights of the neural network to obtain a better compression and reconstruction effect.
[0013] Preferably, the training of the discriminator is carried out simultaneously with the encoder and decoder. The task of the discriminator is to determine whether the reconstructed data is real data. Adversarial loss is used for training. Whenever the encoder and decoder calculate and output reconstructed data, the reconstructed data is input into the discriminator, the output of the discriminator is obtained, and the adversarial loss is calculated based on the discriminator output. The gradient of all parameters of the discriminator is calculated based on the loss, and the parameters of each layer inside the discriminator are updated.
[0014] Compared with the prior art, the present invention has the following advantages:
[0015] (1) The present invention can increase the compression rate to 360 times while ensuring that the rms (mean square error) and PRDN (normalized RMS percentage difference) are acceptable, and has a lower reconstruction loss and better fits the original ECG curve.
[0016] (2) The present invention improves the signal reconstruction quality by adding a deep discriminator for training during the training phase.
[0017] (3) This system can run on a single-core CPU with high operating efficiency, meeting the ECG data compression requirements in actual scenarios. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] Figure 1 FIG. 4 is a system architecture diagram of the present invention in the training phase.
[0019] Figure 2 This is a system architecture diagram of the present invention at the reasoning stage.
[0020] Figure 3 Figure 2 is a diagram of the one-dimensional hole convolution calculation process.
[0021] Figure 4 Graph of the two-dimensional convolution input and output process.
[0022] Figure 5 Diagram of the two-dimensional hole convolution calculation process.
[0023] Figure 6 The overall and internal structure diagram of LSTM.
[0024] Figure 7 This is the encoder structure diagram.
[0025] Figure 8 This is the structure diagram of the encoder convolution block.
[0026] Fig. 9 This is the residual unit structure diagram.
[0027] Fig.10 This is the decoder structure diagram.
[0028] Fig.11 This is the structure diagram of the decoder convolution block.
[0029] Fig.12 This is the structure diagram of the discriminator.
[0030] Fig.13 This is the flow chart of the training phase. DETAILED DESCRIPTION
[0031] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary persons in the art without creative work are within the scope of protection of the present invention.
[0032] Embodiment: A physiological signal compression system based on deep autoencoder.
[0033] Background knowledge:
[0034] One-dimensional convolution:
[0035] k is the convolution kernel size, N is the number of input channels, S (stride) is the step length, Coutj is the number of output channels, and the number of convolution kernels required for the following operations depends on the number of output channels:
[0036]
[0037] The one-dimensional dilated convolution adds a dilation parameter to the ordinary one-dimensional convolution, such as Figure 3 As shown, (b) in the figure is the case of dilation=1.
[0038] The output length is
[0039]
[0040] Padding is the amount of input completion, and the completion method can be zero padding or mean padding.
[0041] 2D Convolution:
[0042] The input matrix is I, the kernel matrix is K, and the shape is m*n, then the two-dimensional convolution formula can be defined as:
[0043]
[0044] Process Figure 4 shown.
[0045] The two-dimensional dilated convolution adds a dilation parameter to the ordinary one-dimensional convolution, such as Figure 5 shown.
[0046] Compared with ordinary convolution, dilated convolution skips some pixels during operation to obtain a larger receptive field. Figure 4 The convolution kernel sizes from left to right are 3*3, 5*5, 7*7, and the dilation parameters are all (1, 1).
[0047] Residual Module:
[0048] Each convolution block contains a residual module to avoid the gradient vanishing problem during training, which can be expressed as:
[0049] y=x+f(x)
[0050] Here f(x) can be considered as the input x, the output generated by the one-dimensional convolution layer, that is, the final output y is the sum of the one-dimensional convolution output and the input x.
[0051] STFT is short-time Fourier transform, which is used to calculate the time-frequency information of the input signal. Different window lengths can be used to calculate the change of the amplitude corresponding to the frequency over time. The formula is as follows:
[0052]
[0053] LSTM
[0054] i t =σ(W ii x t +b ii +W hi h t-1 +b hi )
[0055] f t =σ(W if x t +b if +W hf h t-1 +b hf )
[0056] g t =tanh(W ig x t +b ig +W hg h t-1 +b hg )
[0057] o t =σ(W io x t +b io +W ho h t-1 +b ho )
[0058] c t =f t ⊙c t-1 +i t ⊙g t
[0059] h t =o t ⊙tanh(c t)
[0060] In the above formula, ft is the output of the forget gate, ct is the output of the memory gate, ot is the output of the output gate, and gt and it are both intermediate variables used to calculate the output ct of the memory gate.
[0061] LSTM is a special recurrent neural network with an overall structure like Figure 6 shown.
[0062] Reference Figure 1 and Figure 2 As shown, a physiological signal compression system based on a deep autoencoder includes: an encoder, used to convert an input physiological signal into a latent variable Z; a decoder, used to reconstruct the original signal based on the latent variable Z; a fast Fourier transform module, used to calculate the frequency components of the input and output signals and obtain the spectrum; a discriminator, used to judge the degree of difference between the input signal characteristics and the output signal characteristics; an optimizer, used to update the weights of the neural network to obtain a better compression and reconstruction effect.
[0063] The operation process of this system is divided into two stages, the training stage and the inference stage; the inference stage includes the encoding and decoding process. In the training stage, taking ECG compression as an example, the features of ECG signals of different patients are learned based on the MIT-BIH ECG dataset, and the learned features are mapped to a lower dimension, and the correlation is removed to obtain the latent variable z. The decoder reconstructs the ECG signal based on the acquired latent variable z. The encoder and decoder are trained together in the training stage to obtain the optimal encoder and decoder. In the inference stage, the encoder and decoder are used separately. When compressing, the input signal is input into the encoder to obtain the compressed representation; when decompressing, the decoder is used to decompress the received compressed signal and reconstruct the original signal.
[0064] Overall structure Figure 1 and Figure 2 As shown, Figure 1 is the overall structure of the training phase, Figure 2 The system structure used in the reasoning phase. Figure 1 In the example, Encoder is an encoder, which is used to convert the input ECG into a latent variable Z; Decoder is a decoder, which is used to reconstruct the ECG signal based on the latent variable Z; FFT is a fast Fourier transform module, which is used to calculate the frequency components of the input and output signals and obtain the spectrum; module D is a discriminator, which is used to judge the degree of difference between the input signal characteristics and the output signal characteristics. Loss_t is the loss function of the input and output in the time domain, Loss_f is the loss function of the input and output in the frequency domain, and Loss_adv is the adversarial loss function of the input and output signals.
[0065] The structure of the encoder is as follows Figure 7 As shown in Figure 1, the encoder consists of two convolutional layers at the head and tail, four convolutional blocks, and one LSTM layer. After the input signal enters the encoder, it goes through Figure 7 The structure shown above obtains the latent variable z output. The number of channels of the first convolution layer is consistent with the number of channels of the input signal (usually 1), and the convolution kernel size is 7; followed by four convolution blocks with the same structure, the number of channels doubles after each convolution block, and the step lengths of the four convolution blocks are 2, 4, 5, and 8 respectively, which realizes the downsampling of the input signal; the downsampled signal will pass through the LSTM layer to extract the timing information; the number of channels of the last convolution layer is consistent with the number of channels of the last convolution block, the convolution kernel size is 3, and the output is the required latent variable.
[0066] The structure of the convolutional block is as follows Figure 8 As shown in the figure, the convolution block consists of three residual units and a one-dimensional convolution layer. The input channel of the residual unit is 1 / 2 of the input channel of the convolution block, and the dilation parameter increases successively to 1, 3, and 9 to obtain temporal features of different scales. The convolution kernel size of the convolution layer is twice the step length, and the number of channels and step length are consistent with the input channel of the convolution block.
[0067] The structure of the residual unit is as follows Fig. 9 As shown in the figure, the residual unit is composed of a one-dimensional hole convolution layer, a one-dimensional convolution layer and a residual connection. The input signal passes through the above two convolution layers and outputs features. The output features are added to the input signal itself as the final output of the residual unit.
[0068] The decoder structure is a mirror image of the encoder, such as Fig.10 As shown in the figure. The operation of the decoder is the inverse process of the encoder operation. The input is the latent variable Z output by the encoder. First, it is input into a one-dimensional convolution layer with a convolution kernel size of 7 and a number of channels that is 16 times the number of channels of the original signal. Then, it is input into the same LSTM layer as the encoder to restore the timing information. After that, it passes through four decoding convolution blocks with the same structure, with step lengths of 8, 5, 4, and 2 respectively. The number of channels is halved after each convolution block. Finally, the original signal is restored through a one-dimensional convolution layer with a convolution kernel size of 7 and a channel number of 1.
[0069] The decoder convolutional block structure is as follows Fig.11 As shown, Conv1D T It is a transposed one-dimensional convolution, and the operation method is similar to the one-dimensional convolution Conv1D, except that the output length is greater than the input length. ResidualUnit is consistent with the residual unit described above.
[0070] The discriminator structure is as follows Fig.12As shown in , the complete network structure will contain K discriminators, corresponding to STFTs of different scales / window lengths. Here K = 5, and the window lengths are 128, 256, 512, 1024, and 2048 respectively. A single discriminator consists of an STFT block and 6 two-dimensional convolutional layers. Fig.12 As shown in the figure, Real represents the real part of the stft output result, and Im represents the imaginary part of the stft output result. The two parts are concatenated to form a time-frequency feature map, which is then input into a two-dimensional convolution layer with a convolution kernel size of 24 to be converted into a 32-channel feature map. It then passes through three two-dimensional convolution layers with 32 channels, a convolution kernel size of 3*8, and a step length of (1, 2) (representing a step length of 1 in the x direction and a step length of 2 in the y direction). The dilation parameters are (1, 1), (2, 1), and (4, 1), respectively. The definition of two-dimensional atrous convolution is shown in the previous text. Finally, it passes through two convolution layers with a convolution kernel size of 3*3 to obtain the final output result.
[0071] Loss function: Loss_t is the loss function in the time domain, using L1 Norm as the output loss, that is, the sum of the absolute values of the difference between each data point of the input and output data. The formula is as follows:
[0072] Loss t =‖x recons -x in ‖1
[0073] Loss_f is the loss function in the frequency domain. First, perform a 360-point FFT (Fast Fourier Transform) on the input and output data to obtain the frequency amplitude spectra of the two signals. The linear combination of L1 Norm and mean square error (MSE) is used as the loss function, that is, the linear weighted sum of the absolute value of the difference between the two frequency amplitude spectra and the sum of the square values of the difference between the amplitude spectra. The formula is as follows:
[0074]
[0075] Where F(·) is the fast Fourier transform function, and α and β are custom weight parameters.
[0076] Loss_feat is the loss function of the discriminator output. The input and output data are simultaneously input into the discriminator's neural network to obtain the corresponding feature output. The KL divergence between the two feature outputs is calculated to obtain the feature loss. The formula is as follows:
[0077]
[0078] where Dk(·) is the kth discriminator and K is the number of discriminators.
[0079] Loss_g is the discriminator adversarial loss. The input data is the reconstructed data, and the output is the difference between the reconstructed data and the original data. The formula is as follows. Note: When the reconstructed data is exactly the same as the original data, the discriminator D(x recons ) output is 1.
[0080]
[0081] Where K is the number of discriminators, G(·) represents the generator, i.e. the entire encoder-decoder structure dec(enc(x)), and Dk(·) is the kth discriminator.
[0082] The training process is as follows Fig.13 As shown in Figure 1, the input raw data is first preprocessed and then input into the encoder-decoder network described above to calculate the output reconstructed signal. According to the various loss functions described above, the corresponding losses are calculated, and the parameters of the encoder-decoder and the discriminator are updated according to the losses. Note that the parameters of the encoder and decoder are updated together, and the discriminator parameters are updated separately according to the adversarial loss.
[0083] Data preprocessing: The training phase in the present invention needs to preprocess the data first, and the data set used is the open source ECG database of MIT-BIH. The original data contains 48 ECG data, including the results collected under different leads, 650,000 data points for each lead, and the sampling rate is 360Hz. In the present invention, the data is first divided into windows of 1s, that is, a segment of data is cut into multiple 360-point data segments. The segmented data must be reversed in time domain and amplitude to obtain the inverted data, so as to obtain ECG features on different spatial scales and time scales, and this operation is used to perform data enhancement and expand the data set. In addition, the influence of different lead characteristics and reverse connection of positive and negative electrodes during ECG acquisition is avoided.
[0084] End-to-end codec training: During the training phase, the encoder and decoder are connected together. The input data passes through the encoder and is directly input into the decoder to obtain the output. The input and output are compared. After calculating the three losses mentioned above, the Adam optimizer is used to update the neural network weights to obtain better compression and reconstruction effects.
[0085] The overall training loss function is:
[0086] Loss = α * Loss t +β*Loss f +γ*Loss feat
[0087] Among them, α=1, β=5, γ=1, which are all specified parameters.
[0088] The training of the discriminator is carried out simultaneously with the encoder and decoder. The task of the discriminator is to determine whether the reconstructed data is real data. It is trained using adversarial loss, i.e., Loss_g. The definition can be found in the previous text. Whenever the encoder and decoder calculate and output reconstructed data, the reconstructed data is input into the discriminator, the output of the discriminator is obtained, and the adversarial loss Loss_g is calculated based on the output of the discriminator. The gradient of all parameters of the discriminator is calculated based on the loss, and the parameters of each layer inside the discriminator are updated. The optimizer is consistent with the Adam optimizer used in the previous encoder and decoder training.
[0089] The physiological signal compression system proposed in the present invention can increase the compression rate to 360 times while ensuring that the rms (mean square error) and PRDN (normalized RMS difference percentage) are acceptable, and has a lower reconstruction loss and better fits the original ECG curve. At the same time, the system can run on a single-core CPU with high operating efficiency, meeting the ECG data compression requirements of actual scenarios. In addition, the present invention improves the quality of ECG signal reconstruction by adding a deep discriminator for training during the training phase.
[0090] The above description is only a preferred embodiment of the present invention, but the present invention should not be limited to the contents disclosed in the embodiment and the drawings, so any equivalent or modification completed without departing from the spirit disclosed in the present invention shall fall within the scope of protection of the present invention.
Claims
1. A physiological signal compression system based on deep autoencoder, characterized in that include: Encoder, used to convert the input physiological signal into latent variable Z; Decoder, used to reconstruct the original physiological signal based on the latent variable Z; Fast Fourier transform module, used to calculate the frequency components of input and output signals and obtain the spectrum; The discriminator is used to determine the degree of difference between the input signal characteristics and the output signal characteristics; The optimizer is used to update the weights of the neural network to obtain better compression and reconstruction effects.
2. The physiological signal compression system based on deep autoencoder according to claim 1, characterized in that: The encoder consists of two convolutional layers at the head and tail, four convolutional blocks and an LSTM layer. After the input signal enters the encoder, the number of channels of the first convolutional layer is consistent with the number of channels of the input signal, and the convolution kernel size is 7. It is followed by four convolutional blocks with the same structure. The number of channels doubles after each convolution block. The step lengths of the four convolutional blocks are 2, 4, 5, and 8, respectively, to achieve downsampling of the input signal. The downsampled signal will pass through the LSTM layer to extract timing information. The number of channels of the last convolutional layer is consistent with the number of channels of the last convolution block, the convolution kernel size is 3, and the output is the required latent variable Z.
3. The physiological signal compression system based on deep autoencoder according to claim 2, characterized in that: The convolution block consists of three residual units and a one-dimensional convolution layer. The input channel of the residual unit is 1 / 2 of the input channel of the convolution block. The dilation parameter increases successively to 1, 3, and 9, respectively, to obtain temporal features of different scales. The convolution kernel size of the convolution layer is twice the step length, and the number of channels and step length are consistent with the input channel of the convolution block.
4. The physiological signal compression system based on deep autoencoder according to claim 3, characterized in that: The residual unit consists of a one-dimensional hole convolution layer, a one-dimensional convolution layer and a residual connection. The input signal passes through two convolution layers and outputs features. The output features are added to the input signal itself as the final output of the residual unit.
5. The physiological signal compression system based on deep autoencoder according to claim 4, characterized in that: The decoder structure is a mirror image of the encoder. The operation of the decoder is the inverse process of the encoder operation. The input is the latent variable Z output by the encoder. First, a one-dimensional convolution layer is input with a convolution kernel size of 7 and a number of channels that is 16 times the number of channels of the original signal. Then, the same LSTM layer as the encoder is input to restore the timing information. After that, four decoding convolution blocks with the same structure are passed, with step lengths of 8, 5, 4, and 2 respectively. The number of channels is halved after each convolution block. Finally, the original signal is restored through a one-dimensional convolution layer with a convolution kernel size of 7 and a channel number of 1.
6. The physiological signal compression system based on deep autoencoder according to claim 1, characterized in that: A single discriminator consists of an STFT block and six two-dimensional convolutional layers. Real represents the real part of the stft output result, and Im represents the imaginary part of the stft output result. The two parts are concatenated to form a time-frequency feature map, which is then input into a two-dimensional convolutional layer with a convolution kernel size of 24 and converted into a 32-channel feature map. It then passes through three two-dimensional convolutional layers with 32 channels, a convolution kernel size of 3*8, and a step length of (1, 2). The dilation parameters are (1, 1), (2, 1), and (4, 1), respectively. Finally, it passes through two convolutional layers with a convolution kernel size of 3*3 to obtain the final output result.
7. The physiological signal compression system based on deep autoencoder according to claim 1, characterized in that: The operation process of this system is divided into a training phase and an inference phase, wherein the training phase is based on the required compressed physiological signal data set for training. Taking ECG compression as an example, the features of ECG signals of different patients are learned based on the MIT-BIH ECG dataset, and the learned features are mapped to a lower dimension, and the correlation is removed to obtain the latent variable z. The decoder reconstructs the ECG signal based on the acquired latent variable z. The encoder and decoder are trained together in the training phase to obtain the optimal encoder and decoder. The inference phase includes the encoding and decoding processes, and the encoder and decoder are used separately. When compressing, the input signal is input into the encoder to obtain the compressed representation. When decompressing, the decoder is used to decompress the received compressed signal and reconstruct the original signal.
8. The physiological signal compression system based on deep autoencoder according to claim 7, characterized in that: The training process is as follows: the input raw data is first preprocessed and then input into the codec network to calculate the output reconstructed signal. According to each loss function, the corresponding loss is calculated, and the parameters of the codec and discriminator are updated separately according to the loss. The parameters of the encoder and decoder are updated together, and the discriminator parameters are updated separately according to the adversarial loss.
9. The physiological signal compression system based on deep autoencoder according to claim 8, characterized in that: In the training phase, the data must be preprocessed first. The sample data set used is the open source ECG database of mit-bih. The original data in mit-bih contains 48 ECG data, including the results collected under different leads, 650,000 data points for each lead, and a sampling rate of 360Hz. The data is first segmented into windows of 1s, and a segment of data is cut into multiple 360-point data segments. The segmented data is inverted in time domain and amplitude to obtain the inverted data to obtain ECG features at different spatial scales and time scales. This operation is used to enhance the data and expand the data set to avoid the influence of different lead characteristics and reverse connection of positive and negative electrodes during ECG acquisition. In the training phase, the encoder and decoder are connected together, and the input data is directly input into the decoder after passing through the encoder to obtain the output, and the input is compared with the output. After calculating the three losses, the optimizer is used to update the weights of the neural network to obtain a better compression and reconstruction effect.
10. The physiological signal compression system based on deep autoencoder according to claim 9, characterized in that: The training of the discriminator is carried out simultaneously with the encoder and decoder. The task of the discriminator is to determine whether the reconstructed data is real data. Adversarial loss is used for training. Whenever the encoder and decoder calculate and output reconstructed data, the reconstructed data is input into the discriminator, the output of the discriminator is obtained, and the adversarial loss is calculated based on the output of the discriminator. The gradient of all parameters of the discriminator is calculated based on the loss, and the parameters of each layer inside the discriminator are updated.
Citation Information
Patent Citations
Time sequence signal efficient denoising and high-precision reconstruction modeling method and system
CN111046824A
Electrocardiosignal quality evaluation method based on auto-encoder
CN116350233A
Cognitive load judgment method based on brain intelligence technology
CN117171658A
Device and method for monitoring driver's biometric information abnormality through multi-sampling sensor fusion
KR102440752B1
KR20220142679A
Cited By
Electrocardio compression method and system based on learnable semantic distillation and layered residual quantization
CN122004881A
