Derivative respiration signal acquisition method and device, computer equipment and storage medium

Abstract

The invention relates to the technical field of signal processing, in particular to a derivative respiration signal obtaining method and device, computer equipment and a storage medium, and the derivative respiration signal of a ballistocardiogram signal is obtained based on the ballistocardiogram signal of a sample user and a preset generative adversarial neural network. In combination with pectoral girdle sensing information synchronously collected with the ballistocardiogram signal, a deep learning method is adopted to train the generative adversarial neural network, so that the trained generative adversarial neural network can adaptively extract spatial-temporal characteristics in the ballistocardiogram signal, and the precision of obtaining the derived respiration signal is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of signal processing, and in particular to a method, device, computer equipment and storage medium for acquiring a derived respiratory signal. Background Art

[0002] Respiration is a key physiological parameter widely used in clinical practice and has practical diagnostic and prognostic value. The rhythm and effort of breathing are of great clinical reference significance for the diagnosis of respiratory diseases such as sleep apnea syndrome (SAS) and chronic obstructive pulmonary disease (COPD). In actual clinical scenarios, respiratory detection equipment usually evaluates the pressure or temperature and humidity changes of oral and nasal airflow, and the fluctuations of respiratory effort in the chest and abdomen. These methods generally have problems such as high equipment requirements, high usage thresholds, and discomfort in equipment use, and are not suitable for non-hospital scenarios.

[0003] Among them, the signal recorded by using the piezoelectric effect to sense the tiny body vibrations caused by the mechanical activity of the heart and the impact and rebound of blood vessels during blood flow is called Ballistocardiography (BCG). However, BCG is often modulated by respiratory and body motion artifacts. It is a type of aliased vital sign signal with high spectral complexity and large individual differences. The accuracy and reliability of the signal are not as good as those of contact clinical detection equipment based on oral and nasal airflow sensing and chest and abdominal sensing, resulting in low accuracy of the acquired respiratory signal. Therefore, the respiratory extraction algorithm for BCG signals needs to achieve more fine-grained feature extraction and analysis.

[0004] Current technical solutions usually adopt two types of methods, one based on time domain features or one based on frequency domain features, or a fusion of the two types, to obtain respiratory signals based on the ballistocardiogram signal obtained by piezoelectric sensing. Research on time domain features usually focuses on the time interval, amplitude, envelope, extreme value, signal energy, respiratory variability and other time domain features of the reference points in the signal, as well as statistical features based on these features. Algorithms based on frequency domain features often focus on the amplitude of the spectrum and related statistical features, usually by filtering or decomposing the original signal to extract the respiratory spectrum components or attenuate the non-respiratory spectrum components. The above methods usually require the selection of specific features or thresholds for design. When faced with clinical non-stationary and aliased physiological signals with large individual differences, the selected features may not reflect the specific attributes within the data, and there may be problems with insufficient adaptability and robustness, resulting in poor performance and efficiency of the algorithm, and it is difficult to accurately obtain respiratory signals from the ballistocardiogram signal. Summary of the invention

[0005] Based on this, the purpose of the present invention is to provide a method, device, equipment and storage medium for obtaining a derived breathing signal, based on the ballistocardiogram signal of a sample user and a preset generative adversarial neural network, to obtain a derived breathing signal of the ballistocardiogram signal, and to train the generative adversarial neural network using a deep learning method in combination with chest strap sensor information collected synchronously with the ballistocardiogram signal, so that the trained generative adversarial neural network can adaptively extract the spatiotemporal features in the ballistocardiogram signal, reconstruct the breathing signal, and improve the accuracy of obtaining the derived breathing signal.

[0006] In a first aspect, an embodiment of the present application provides a method for acquiring a derived respiratory signal, comprising the following steps:

[0007] Obtaining ballistocardiogram signals, chest strap sensor signals and a preset generative adversarial neural network synchronously collected by a number of sample users, wherein the generative adversarial neural network includes a generator network and a discriminator network;

[0008] Inputting the ballistocardiogram signals of several sample users into the generator network respectively to obtain the derived breathing signals of several sample users;

[0009] Combine chest strap sensor signals of the same sample user with corresponding ballistocardiogram signals and derived respiratory signals to obtain first training signal groups and second training signal groups of several sample users, input the first training signal groups and second training signal groups of several sample users into the discriminator network to obtain predicted label data of the first training signal groups and second training signal groups of several sample users;

[0010] Obtaining real label data of a first training signal group and a second training signal group of several sample users, and training the generative adversarial neural network according to the predicted label data and the real label data of the first training signal group and the second training signal group of several sample users to obtain a target generative adversarial neural network;

[0011] A ballistocardiogram signal of a user to be generated is obtained, and the ballistocardiogram signal of the user to be generated is input into the target generator network to obtain a derived breathing signal of the user to be generated.

[0012] In a second aspect, an embodiment of the present application provides a derived respiratory signal acquisition device, comprising:

[0013] A signal acquisition module, used to obtain the ballistocardiogram signals, chest strap sensor signals and a preset generative adversarial neural network synchronously collected by several sample users, wherein the generative adversarial neural network includes a generator network and a discriminator network;

[0014] A first signal generating module, used for inputting the ballistocardiogram signals of several sample users into the generator network respectively, to obtain the derived breathing signals of several sample users;

[0015] a label prediction module, used to combine the chest belt sensor signal of the same sample user with the corresponding ballistocardiogram signal and the derived respiratory signal, respectively, to obtain a first training signal group and a second training signal group of several sample users, and to input the first training signal group and the second training signal group of several sample users into the discriminator network, respectively, to obtain predicted label data of the first training signal group and the second training signal group of several sample users;

[0016] A generative adversarial neural network training module is used to obtain the real label data of the first training signal group and the second training signal group of several sample users, and train the generative adversarial neural network according to the predicted label data and the real label data of the first training signal group and the second training signal group of several sample users to obtain a target generative adversarial neural network;

[0017] The second signal generating module is used to obtain the ballistocardiogram signal of the user to be generated, input the ballistocardiogram signal of the user to be generated into the target generator network, and obtain the derived breathing signal of the user to be generated.

[0018] In a third aspect, an embodiment of the present application provides a computer device, comprising: a processor, a memory, and a computer program stored in the memory and executable on the processor; when the computer program is executed by the processor, the steps of the derived respiratory signal acquisition method described in the first aspect are implemented.

[0019] In a fourth aspect, an embodiment of the present application provides a storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the derived respiratory signal acquisition method as described in the first aspect are implemented.

[0020] In an embodiment of the present application, a method, apparatus, device and storage medium for acquiring a derived breathing signal are provided. Based on a ballistocardiogram signal of a sample user and a preset generative adversarial neural network, a derived breathing signal of the ballistocardiogram signal is obtained. In combination with chest strap sensor information collected synchronously with the ballistocardiogram signal, a deep learning method is used to train the generative adversarial neural network, so that the trained generative adversarial neural network can adaptively extract the spatiotemporal features in the ballistocardiogram signal and reconstruct the breathing signal, thereby improving the accuracy of acquiring the derived breathing signal.

[0021] For better understanding and implementation, the present invention is described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 A schematic diagram of a flow chart of a method for acquiring a derived respiratory signal provided in one embodiment of the present application;

[0023] Figure 2 A schematic diagram of the flow of S2 in the method for acquiring a derived respiratory signal provided in one embodiment of the present application;

[0024] Figure 3 A schematic diagram of the flow of S3 in the method for acquiring a derived respiratory signal provided in one embodiment of the present application;

[0025] Figure 4 A schematic diagram of the flow of S4 in the method for acquiring a derived respiratory signal provided in one embodiment of the present application;

[0026] Figure 5 A schematic diagram of the flow of S41 in the method for acquiring a derived respiratory signal provided in one embodiment of the present application;

[0027] Figure 6 A schematic diagram of the flow of S42 in the method for acquiring a derived respiratory signal provided in one embodiment of the present application;

[0028] Figure 7 A schematic diagram of the structure of a derived respiratory signal acquisition device provided in one embodiment of the present application;

[0029] Figure 8 A schematic diagram of the structure of a computer device provided for one embodiment of the present application. DETAILED DESCRIPTION

[0030] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0031] The terms used in this application are for the purpose of describing specific embodiments only and are not intended to limit this application. The singular forms of "a", "said" and "the" used in this application and the appended claims are also intended to include plural forms unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0032] It should be understood that, although the terms first, second, third, etc. may be used in the present application to describe various information, these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of the present application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the words "if" / "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determination".

[0033] See also Figure 1 , Figure 1 A flowchart of a method for acquiring a derived respiratory signal provided in one embodiment of the present application is provided, wherein the method comprises the following steps:

[0034] S1: Obtaining the ballistocardiogram signals, chest strap sensor signals and a preset generative adversarial neural network synchronously collected by several sample users.

[0035] The execution subject of the derived respiratory signal acquisition method is an acquisition device of the derived respiratory signal acquisition method (hereinafter referred to as acquisition device). In an optional embodiment, the acquisition device can be a computer device, a server, or a server cluster composed of multiple computer devices.

[0036] In this embodiment, the acquisition device can use a sub-occipital sensor device with an integrated piezoelectric sensor module to acquire weak body vibration signals caused by the tester's heartbeat, breathing, etc. in a non-contact measurement method, wherein the signal contains mixed vital signs of heartbeat, breathing, body movement and other noise signals. The acquisition device passes the above-collected signal through a 0.1-10Hz second-order Butterworth bandpass filter to filter out interference noise and acquire the ballistocardiogram signals of several testers as the ballistocardiogram signals of the sample users.

[0037] At the same time, the acquisition device adopts the chest strap sensor in the polysomnography PSG to obtain the mechanical sensing signal caused by the change of the chest strap when the tester breathes by the contact measurement method. The acquisition device passes the above-mentioned acquisition signal through a second-order Butterworth bandpass filter of 0.1Hz-0.5Hz to filter out the interference noise, and obtains the chest strap sensor signals of several testers as the chest strap sensor signals of the sample users.

[0038] Since the tester will inevitably move or get out of bed during the measurement process, this will introduce interference to the signal acquisition, resulting in some signals being accompanied by extreme amplitudes and morphological deformations, destroying the respiratory component of the signal. In an optional embodiment, the acquisition device uses an empirical threshold cutting method to remove the body motion signals in the ballistocardiogram signal and the chest strap sensor signal, respectively. Specifically, the acquisition device downsamples the ballistocardiogram signal and the chest strap sensor signal to 100Hz and normalizes them to the [-1,1] interval. The spectrum of the ballistocardiogram signal is mainly concentrated between 0.1Hz-10Hz, while the spectrum of the chest strap sensor signal is mainly concentrated between 0.1Hz-0.5Hz. Therefore, downsampling the original signal to 100Hz can reduce the number of parameters calculated by the model while retaining most of the signal information, thereby improving the accuracy of the derived respiratory signal generation. In addition, for different patients, the amplitude range obtained by the piezoelectric sensor will be different, so the signal needs to be normalized to reduce the signal amplitude to improve the calculation efficiency of the model.

[0039] The acquisition device obtains a preset generative adversarial neural network (GAN), wherein the generative adversarial neural network includes a generator network and a discriminator network.

[0040] S2: Inputting the ballistocardiogram signals of several sample users into the generator network respectively to obtain the derived breathing signals of several sample users.

[0041] In this embodiment, the acquisition device inputs the ballistocardiogram signals of several sample users into the generator network respectively to obtain the derived breathing signals of several sample users.

[0042] Specifically, the generator network includes a first encoder module, a temporal feature extraction module and a decoder module connected in sequence; wherein the first encoder module includes a plurality of stacked first encoder units, the temporal feature extraction module includes a plurality of stacked temporal feature extraction units; the decoder module includes a plurality of stacked decoder units, see Figure 2 , Figure 2 The flowchart of S2 in the derived respiratory signal acquisition method provided in one embodiment of the present application includes steps S21 to S23, which are specifically as follows:

[0043] S21: Using the ballistocardiogram signals of several sample users as input signals of the first first encoder unit of the first encoder module respectively, obtaining the first output signal of the first first encoder unit according to a preset first encoding algorithm, using the first output signal as the input signal of the next first encoder unit, repeating the encoding operation, and obtaining the first output signal of the last first encoder unit as the first encoded signals of several sample users.

[0044] In this embodiment, the acquisition device uses the ballistocardiogram signals of several sample users as input signals of the first first encoder unit of the first encoder module, and obtains the first output signal of the first first encoder unit according to a preset first encoding algorithm, wherein the first encoding algorithm is:

[0045] Y out1 =LeakyRelu(BN(K1·LeakyRelu(BN(K1·X n1 ))))

[0046] Where Y out1 is the first output signal, LeakyRelu(·) is the first activation function, BN(·) is the normalization function, K1 is the first convolution kernel parameter, X n1 Ballistocardiogram signal.

[0047] The acquisition device uses the first output signal as the input signal of the next first encoder unit, repeats the encoding operation, obtains the first output signal of the last first encoder unit as the first encoded signal of several sample users, and performs derived breathing space feature extraction from shallow to deep on the ballistocardiogram signal to improve the accuracy of constructing the derived breathing signal in the ballistocardiogram signal.

[0048] S22: respectively use the first coded signals of several sample users as the input signals of the first timing feature extraction unit of the timing feature extraction module, obtain the second output signal of the first timing feature extraction unit according to the preset long short-term memory algorithm, use the second output signal as the input signal of the next timing feature extraction unit, repeat the timing feature extraction operation, and obtain the second output signals of several sample users of the last timing feature extraction unit as the timing feature signals of several sample users.

[0049] The time series feature extraction module adopts a long short-term memory neural network (LSTM). LSTM is a neural network structure that can learn long-term dependencies in time series. It can selectively memorize important relationships in time series signals through memory gating, thereby establishing time series dependencies between signals, and extracting deep time features by stacking multiple LSTM layers.

[0050] In this embodiment, the acquisition device uses the first coded signals of several sample users as input signals of the first time series feature extraction unit of the time series feature extraction module, and extracts the time series features of the signal from the time dimension according to the preset long short-term memory algorithm to obtain the second output signal of the first time series feature extraction unit, wherein the long short-term memory algorithm is:

[0051] Y out2 =LSTM(Y out1 )

[0052] Where Y out2 is the second output signal, and LSTM(·) is the long short-term memory function.

[0053] The acquisition device uses the second output signal as the input signal of the next timing feature extraction unit, repeats the timing feature extraction operation, and obtains the second output signals of several sample users of the last timing feature extraction unit as the timing feature signals of several sample users.

[0054] In an optional embodiment, in order to fully consider the morphological characteristics of the input signal, the acquisition device performs residual connection processing on the first encoded signal before inputting it into the timing feature extraction unit, and uses the first encoded signal after the residual connection processing as the input signal of the first timing feature extraction unit of the timing feature extraction module.

[0055] S23: splicing the coded signal and the timing characteristic signal of the same sample user to obtain second spliced ​​signals of several sample users, respectively using the second spliced ​​signals of several sample users as input signals of the first decoder unit of the decoder module, obtaining the third output signal of the first decoder unit according to a preset decoding algorithm, using the third output signal as the input signal of the next decoder unit, repeating the decoding operation, and obtaining the third output signal of the last decoder unit as the derived breathing signals of the several sample users.

[0056] In this embodiment, the acquisition device splices the coded signal of the same sample user with the timing characteristic signal to obtain second spliced ​​signals of several sample users, and uses the second spliced ​​signals of several sample users as input signals of the first decoder unit of the decoder module respectively, and obtains the third output signal of the first decoder unit according to a preset decoding algorithm, wherein the decoding algorithm is:

[0057] Y out3 =LeakyRelu(BN(K2·LeakyRelu(BN(K2·Y concat ))))

[0058] Where Yout3 is the third output signal, K2 is the second convolution kernel parameter, Y concat is the second splicing signal.

[0059] The acquisition device uses the third output signal as the input signal of the next decoder unit, repeats the decoding operation, and obtains the third output signal of the last decoder unit as the derived breathing signal of several sample users, thereby realizing upsampling of the feature dimension of the second spliced ​​signal, reconstructing the breathing signal, and obtaining an accurate derived breathing signal.

[0060] S3: Combine the chest strap sensor signal of the same sample user with the corresponding ballistocardiogram signal and derived breathing signal respectively to obtain a first training signal group and a second training signal group of several sample users, input the first training signal group and the second training signal group of several sample users into the discriminator network respectively, and obtain predicted label data of the first training signal group and the second training signal group of several sample users.

[0061] The significance of the discriminator is to enable it to distinguish the authenticity of the generator samples through training. In this embodiment, the acquisition device combines the chest strap sensor signal of the same sample user with the corresponding ballistocardiogram signal and derived breathing signal to obtain the first training signal group and the second training signal group of several sample users, and inputs the first training signal group and the second training signal group of several sample users into the discriminator network to obtain the predicted label data of the first training signal group and the second training signal group of several sample users.

[0062] Specifically, the discriminator network includes a second encoder module and a post-processing module connected in sequence; wherein the second encoder module includes a plurality of stacked second encoder units; see Figure 3 , Figure 3 The flowchart of S3 in the derived respiratory signal acquisition method provided in one embodiment of the present application includes steps S31 to S33, which are specifically as follows:

[0063] S31: respectively concatenate the signals in the first training signal group and the second training signal group to obtain concatenated signals of the first training signal group and the second training signal group of several sample users, use the concatenated signals as input signals of the discriminator network, and perform convolution processing on the input signals of the discriminator network according to a preset convolution algorithm to obtain convolution feature signals of the input signals.

[0064] In this embodiment, the acquisition device splices the signals in the first training signal group and the second training signal group respectively to obtain spliced ​​signals of the first training signal group and the second training signal group of several sample users.

[0065] The acquisition device uses the spliced ​​signal as the input signal of the discriminator network, and performs convolution processing on the input signal of the discriminator network according to a preset convolution algorithm to obtain a convolution feature signal of the input signal, wherein the convolution algorithm is:

[0066] Y conv =LeakyRelu(BN(K3·Y n1 ))

[0067] Where Y conv is the convolution feature signal, K3 is the third convolution kernel parameter, Y n1 is the input signal of the discriminator network.

[0068] S32: Use the convolution feature signal as the input signal of the first second encoder unit of the second encoder module, obtain the fourth output signal of the first second encoder unit according to a preset second encoding algorithm, use the fourth output signal as the input signal of the next second encoder unit, repeat the encoding operation, and obtain the fourth output signal of the last second encoder unit.

[0069] In this embodiment, the acquisition device uses the convolution feature signal as an input signal of the first second encoder unit of the second encoder module, and obtains a fourth output signal of the first second encoder unit according to a preset second encoding algorithm, wherein the second encoding algorithm is:

[0070] Y out4 =LeakyRlu(SN(K4·LeakyRelu(SN(K4·Y n2 )))+Y n2 )

[0071] Where Y out4 is the fourth output signal, SN(·) is the spectral normalization function, K4 is the fourth convolution kernel parameter, Y n2 is the input signal of the second encoder unit.

[0072] The acquisition device uses the fourth output signal as the input signal of the next second encoder unit, repeats the encoding operation, and obtains the fourth output signal of the last second encoder unit, ensuring that the discriminator network gradient can satisfy the Lipsitz continuity when used for loss calculation to stabilize the training of the model.

[0073] S33: Input the fourth output signal of the last second encoder unit into the post-processing module, and obtain predicted label data of the first training signal group and the second training signal group of several sample users according to a preset label prediction algorithm.

[0074] In this embodiment, the acquisition device inputs the fourth output signal of the last second encoder unit into the post-processing module, and uses the global maximum pooling and full connection method to perform label prediction according to the preset label prediction algorithm to obtain the predicted label data of the first training signal group and the second training signal group of several sample users for subsequent loss calculation and verify the reliability of the generator generation result, wherein the label prediction algorithm is:

[0075] Y out5 =Sigmoid(W·Maxpool(Y out4 ))

[0076] Where Y out5 To predict label data, Sigmoid(·) is the second activation function, W is the weight matrix, and Maxpool(·) is the maximum pooling function.

[0077] S4: Obtain the real label data of the first training signal group and the second training signal group of several sample users, and train the generative adversarial neural network according to the predicted label data and the real label data of the first training signal group and the second training signal group of several sample users to obtain a target generative adversarial neural network.

[0078] In this embodiment, the acquisition device obtains the real label data of the first training signal group and the second training signal group of several sample users, and trains the generative adversarial neural network according to the predicted label data and the real label data of the first training signal group and the second training signal group of several sample users to obtain the target generative adversarial neural network. Compared with the traditional neural network, the training process of the generative adversarial neural network is usually more complicated due to the game learning between two different network structures. In order to avoid premature fitting of one of the generator or the discriminator and thus affect the training of the other, the accuracy of the derived breathing signal generation is improved.

[0079] See also Figure 4 , Figure 4 The flowchart of S4 in the derived respiratory signal acquisition method provided in one embodiment of the present application includes steps S41 to S42, which are specifically as follows:

[0080] S41: Adopting a cross entropy algorithm, constructing an adversarial loss value according to the predicted label data and the real label data of the first training signal group and the second training signal group of several sample users.

[0081] In this embodiment, the acquisition device uses a cross entropy algorithm to construct an adversarial loss value based on the predicted label data and the real label data of the first training signal group and the second training signal group of several sample users, wherein the adversarial loss value is:

[0082] L GAN =Loss D (G(X),X,Y)+Loss G (G(X),X)

[0083] Where, L GAN To combat the loss value, Loss D (G(X), X, Y) is the cross entropy loss value of the discriminator network, D is the discriminator, G(X) is the derived respiratory signal, X is the ballistocardiogram signal, Y is the chest strap sensor signal, and Loss G (G(X),X) is the cross entropy loss value of the generator network, and G is the generator network.

[0084] The cross entropy loss value of the discriminator network is the result of averaging the cross entropy loss value obtained based on the predicted label data and true label data of the first training signal group of the sample user and the cross entropy loss value obtained based on the predicted label data and true label data of the second training signal group of the sample user; the cross entropy loss value of the generator network is the cross entropy loss value obtained based on the predicted label data and true label data regained from the second training signal group of the sample user.

[0085] See also Figure 5 , Figure 5 The flowchart of S41 in the derived respiratory signal acquisition method provided in one embodiment of the present application includes steps S411 to S412, which are specifically as follows:

[0086] S411: Based on the predicted label data and true label data of the first training signal group of the sample user and the predicted label data and true label data of the second training signal group, a cross entropy algorithm is used to obtain a cross entropy loss value of the first training signal group and a cross entropy loss value of the second training signal group. The cross entropy loss value of the first training signal group and the cross entropy loss value of the second training signal group are averaged to construct a cross entropy loss value of the discriminator network.

[0087] In this embodiment, the acquisition device sets the true label data of the first training signal group to 1, and the true label data of the second training signal group to 0, and uses the cross entropy algorithm according to the predicted label data and true label data of the first training signal group of the sample user, the predicted label data and true label data of the second training signal group, to obtain the cross entropy loss value Loss of the first training signal group. Dreal And the cross entropy loss value Loss of the second training signal group Dfake , wherein the cross entropy algorithm is:

[0088] BCELoss(p,q)=-[p·log(q)+(1-p)·log(1-q)]

[0089] In the formula, BCELoss(p,q) is the cross entropy loss value, p is the true label data, and q is the predicted label data.

[0090] The acquisition device performs mean processing on the cross entropy loss value of the first training signal group and the cross entropy loss value of the second training signal group to construct a cross entropy loss value of the discriminator network.

[0091] S412: According to the cross-entropy loss value of the discriminator network, back-propagate the discriminator network, input the second training signal group of the sample user into the discriminator network after back-propagation to perform label prediction, and regain the predicted label data of the second training signal group of the sample user; based on the regained predicted label data and true label data of the second training signal group of the sample user, adopt a cross-entropy algorithm to obtain the cross-entropy loss value of the second training signal group as the cross-entropy loss value of the generator network.

[0092] In this embodiment, the acquisition device performs back propagation on the discriminator network according to the cross entropy loss value of the discriminator network, freezes the parameters of the discriminator network after the back propagation is completed, and inputs the second training signal group of the sample user into the discriminator network after the back propagation is completed to perform label prediction, and re-obtains the predicted label data of the second training signal group of the sample user. According to the re-obtained predicted label data and true label data of the second training signal group of the sample user, a cross entropy algorithm is used to obtain the cross entropy loss value Loss of the second training signal group. Gfake , as the cross entropy loss value of the generator network.

[0093] S42: Obtain the distance loss value of the generator network, obtain a comprehensive loss value based on the adversarial loss value, the distance loss value and a preset comprehensive loss function, and train the generative adversarial neural network based on the comprehensive loss value to obtain a target generative adversarial neural network.

[0094] In this embodiment, the acquisition device obtains the distance loss value of the generator network, wherein the distance loss value is used to guide the optimization and update of the generator network, with the purpose of introducing the loss relationship between sample points.

[0095] The acquisition device obtains a comprehensive loss value according to the adversarial loss value, the distance loss value and a preset comprehensive loss function, and trains the generative adversarial neural network according to the comprehensive loss value to obtain a target generative adversarial neural network, wherein the comprehensive loss function is:

[0096] L=αL dis +βL GAN

[0097] In the formula, L is the comprehensive loss value, α is the first weight coefficient, β is the second weight coefficient, and L dis is the distance loss value.

[0098] See also Figure 6 , Figure 6 The flowchart of S42 in the derived respiratory signal acquisition method provided in one embodiment of the present application includes steps S421 to S423, which are as follows:

[0099] S421: Obtain a global distance loss value according to the second training signal group of the sample user and a preset global distance loss function.

[0100] The global distance loss function is:

[0101]

[0102] In the formula, l d is the global distance loss value, K is the time point, x k is the signal segment at the kth time point of a signal in the second training signal group, y k is a signal segment at the kth time point of another signal in the second training signal group.

[0103] In this embodiment, the acquisition device obtains the global distance loss value according to the second training signal group of the sample user and a preset global distance loss function.

[0104] S422: Obtain a local distance loss value according to the second training signal group of the sample user and a preset local distance loss function.

[0105] The local distance loss function is:

[0106] l m =max(|x1-y1|,…,|x k -y k |)-min(|x1-y1|,…,|x k -y k |)

[0107] In the formula, l m is the local distance loss value, max(·) is the maximum value function, and min(·) is the minimum value function.

[0108] In this embodiment, the acquisition device obtains the local distance loss value according to the second training signal group of the sample user and a preset local distance loss function.

[0109] S423: Obtain the distance loss value according to the global distance loss value, the local distance loss value and a preset distance loss function.

[0110] The distance loss function is:

[0111] L dis =γl d +ηl m

[0112] Wherein, γ is the third weight coefficient, and η is the fourth weight coefficient.

[0113] In this embodiment, the acquisition device obtains the distance loss value according to the global distance loss value, the local distance loss value and a preset distance loss function.

[0114] S5: Obtain the ballistocardiogram signal of the user to be generated, input the ballistocardiogram signal of the user to be generated into the target generator network, and obtain the derived breathing signal of the user to be generated.

[0115] In this embodiment, a ballistocardiogram signal of the user to be generated is obtained, and the ballistocardiogram signal of the user to be generated is input into the target generator network to obtain a derived breathing signal of the user to be generated.

[0116] Based on the ballistocardiogram signal of the sample user and the preset generative adversarial neural network, the derived respiratory signal of the ballistocardiogram signal is obtained. Combined with the chest strap sensor information collected synchronously with the ballistocardiogram signal, the generative adversarial neural network is trained using a deep learning method. The trained generative adversarial neural network can adaptively extract the spatiotemporal features in the ballistocardiogram signal, thereby improving the accuracy of obtaining the derived respiratory signal.

[0117] Please refer to Figure 7 , Figure 7 This is a schematic diagram of the structure of a derived breathing signal acquisition device provided by an embodiment of the present application. The device can implement all or part of the derived breathing signal acquisition device through software, hardware, or a combination of both. The device 7 includes:

[0118] A signal acquisition module 71, used to obtain the ballistocardiogram signals, chest strap sensor signals and a preset generative adversarial neural network synchronously collected by a plurality of sample users, wherein the generative adversarial neural network includes a generator network and a discriminator network;

[0119] A first signal generating module 72, for inputting the ballistocardiogram signals of several sample users into the generator network respectively, to obtain the derived breathing signals of several sample users;

[0120] The label prediction module 73 is used to combine the chest belt sensor signal of the same sample user with the corresponding ballistocardiogram signal and the derived respiratory signal to obtain the first training signal group and the second training signal group of several sample users, and input the first training signal group and the second training signal group of several sample users into the discriminator network to obtain the predicted label data of the first training signal group and the second training signal group of several sample users;

[0121] A generative adversarial neural network training module 74 is used to obtain the real label data of the first training signal group and the second training signal group of several sample users, and train the generative adversarial neural network according to the predicted label data and the real label data of the first training signal group and the second training signal group of several sample users to obtain a target generative adversarial neural network;

[0122] The second signal generating module 75 is used to obtain the ballistocardiogram signal of the user to be generated, input the ballistocardiogram signal of the user to be generated into the target generator network, and obtain the derived breathing signal of the user to be generated.

[0123] In an embodiment of the present application, a signal acquisition module is used to obtain a plurality of sample users' synchronously collected ballistocardiogram signals, chest strap sensor signals, and a preset generative adversarial neural network, wherein the generative adversarial neural network includes a generator network and a discriminator network; a first signal generation module is used to input the ballistocardiogram signals of the plurality of sample users into the generator network to obtain a plurality of sample users' derived breathing signals; a label prediction module is used to combine the chest strap sensor signals of the same sample user with the corresponding ballistocardiogram signals and the derived breathing signals to obtain a first training signal group and a second training signal group of the plurality of sample users; and a label prediction module is used to combine the chest strap sensor signals of the same sample user with the corresponding ballistocardiogram signals and the derived breathing signals to obtain a first training signal group and a second training signal group of the plurality of sample users. The signal groups are respectively input into the discriminator network to obtain the predicted label data of the first training signal group and the second training signal group of several sample users; the real label data of the first training signal group and the second training signal group of several sample users are obtained through the generative adversarial neural network training module, and the generative adversarial neural network is trained according to the predicted label data and the real label data of the first training signal group and the second training signal group of several sample users to obtain the target generative adversarial neural network; the ballistocardiogram signal of the user to be generated is obtained through the second signal generation module, and the ballistocardiogram signal of the user to be generated is input into the target generator network to obtain the derived breathing signal of the user to be generated. Based on the ballistocardiogram signal of the sample user and the preset generative adversarial neural network, the derived breathing signal of the ballistocardiogram signal is obtained, and combined with the chest strap sensor information collected synchronously with the ballistocardiogram signal, the generative adversarial neural network is trained by a deep learning method, so that the trained generative adversarial neural network can adaptively extract the spatiotemporal features in the ballistocardiogram signal, reconstruct the breathing signal, and improve the accuracy of the derived breathing signal acquisition.

[0124] Please refer to Figure 8 , Figure 8 The computer device 8 is a schematic diagram of a structure of a computer device provided in an embodiment of the present application. The computer device 8 includes: a processor 81, a memory 82, and a computer program 83 stored in the memory 82 and executable on the processor 81; the computer device may store multiple instructions, which are suitable for being loaded and executed by the processor 81. Figures 1 to 6 The method steps shown in the figure can be found in the specific execution process. Figures 1 to 6 The specific description shown will not be repeated here.

[0125] Among them, the processor 81 may include one or more processing cores. The processor 81 uses various interfaces and lines to connect various parts in the server, and executes various functions and processes data of the derived respiratory signal acquisition device 7 by running or executing instructions, programs, code sets or instruction sets stored in the memory 82, and calling the data in the memory 82. Optionally, the processor 81 can be implemented in at least one hardware form of digital signal processing (Digital Signal Processing, DSP), field programmable gate array (Field-Programmable Gate Array, FPGA), and programmable logic array (Programble Logic Array, PLA). The processor 81 can integrate one or more combinations of a central processing unit 81 (Central Processing Unit, CPU), an image processor 81 (Graphics Processing Unit, GPU) and a modem. Among them, the CPU mainly processes the operating system, user interface and application programs; the GPU is responsible for rendering and drawing the content to be displayed on the touch display; the modem is used to process wireless communications. It can be understood that the above-mentioned modem may not be integrated into the processor 81, and it can be implemented by a single chip.

[0126] Among them, the memory 82 may include a random access memory 82 (Random Access Memory, RAM), and may also include a read-only memory 82 (Read-Only Memory). Optionally, the memory 82 includes a non-transitory computer-readable storage medium. The memory 82 can be used to store instructions, programs, codes, code sets or instruction sets. The memory 82 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch instructions, etc.), instructions for implementing the above-mentioned various method embodiments, etc.; the data storage area may store data involved in the above-mentioned various method embodiments, etc. The memory 82 may also be optionally at least one storage device located away from the aforementioned processor 81.

[0127] The present application also provides a storage medium that can store multiple instructions, which are suitable for the processor to load and execute the above-mentioned Figures 1 to 6 The method steps shown in the figure can be found in the specific execution process. Figures 1 to 6 The specific description shown will not be repeated here.

[0128] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0129] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0130] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraint algorithm of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0131] In the embodiments provided by the present invention, it should be understood that the disclosed devices / terminal equipment and methods can be implemented in other ways. For example, the device / terminal equipment embodiments described above are only schematic. For example, the division of the modules or units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0132] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0133] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0134] If the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and when the computer program is executed by a processor, the steps of the above-mentioned method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form, etc.

[0135] The present invention is not limited to the above-mentioned embodiments. If various changes or modifications to the present invention do not depart from the spirit and scope of the present invention, and if these changes and modifications fall within the scope of the claims and equivalent technologies of the present invention, the present invention is also intended to include these changes and modifications.

Claims

1. A method for obtaining a derived respiratory signal, characterized in that: The following steps are involved: Obtaining ballistocardiogram signals, chest strap sensor signals and a preset generative adversarial neural network synchronously collected by a number of sample users, wherein the generative adversarial neural network includes a generator network and a discriminator network; Inputting the ballistocardiogram signals of several sample users into the generator network respectively to obtain the derived breathing signals of several sample users; Combine chest strap sensor signals of the same sample user with corresponding ballistocardiogram signals and derived respiratory signals to obtain first training signal groups and second training signal groups of several sample users, input the first training signal groups and second training signal groups of several sample users into the discriminator network to obtain predicted label data of the first training signal groups and second training signal groups of several sample users; Obtaining real label data of a first training signal group and a second training signal group of several sample users, and training the generative adversarial neural network according to the predicted label data and the real label data of the first training signal group and the second training signal group of several sample users to obtain a target generative adversarial neural network; A ballistocardiogram signal of a user to be generated is obtained, and the ballistocardiogram signal of the user to be generated is input into the target generator network to obtain a derived breathing signal of the user to be generated.

2. The derived respiratory signal acquisition method according to claim 1, characterized in that: The generator network includes a first encoder module, a temporal feature extraction module and a decoder module connected in sequence; wherein the first encoder module includes a plurality of stacked first encoder units, the temporal feature extraction module includes a plurality of stacked temporal feature extraction units; and the decoder module includes a plurality of stacked decoder units; The ballistocardiogram signals of several sample users are respectively input into the generator network to obtain the derived breathing signals of several sample users, including the steps of: The ballistocardiogram signals of several sample users are respectively used as input signals of the first first encoder unit of the first encoder module, and the first output signal of the first first encoder unit is obtained according to a preset first encoding algorithm, and the first output signal is used as the input signal of the next first encoder unit, and the encoding operation is repeated to obtain the first output signal of the last first encoder unit as the first encoded signals of several sample users, wherein the first encoding algorithm is: Y out1 =LeakyRelu(BN(K1 LeakyRelu(BN(K1 X n1 )))) Where Y out1 is the first output signal, LeakyRelu(·) is the first activation function, BN(·) is the normalization function, K1 is the first convolution kernel parameter, X n1 It is the ballistocardiogram signal; The first coded signals of several sample users are respectively used as the input signals of the first time series feature extraction unit of the time series feature extraction module, and the second output signal of the first time series feature extraction unit is obtained according to the preset long short-term memory algorithm, and the second output signal is used as the input signal of the next time series feature extraction unit, and the time series feature extraction operation is repeated to obtain the second output signals of several sample users of the last time series feature extraction unit as the time series feature signals of several sample users, wherein the long short-term memory algorithm is: Y out2 =LSTM(Y out1 ) Where Y out2 is the second output signal, LSTM(·) is the long short-term memory function; The coded signal of the same sample user is spliced ​​with the timing characteristic signal to obtain second spliced ​​signals of several sample users, and the second spliced ​​signals of several sample users are respectively used as input signals of the first decoder unit of the decoder module, and according to a preset decoding algorithm, a third output signal of the first decoder unit is obtained, and the third output signal is used as the input signal of the next decoder unit, and the decoding operation is repeated to obtain the third output signal of the last decoder unit as the derived breathing signals of several sample users, wherein the decoding algorithm is: Y out3 =LeakyRelu(BN(K2·LeakyRelu(BN(K2·Y concat )))) Where Y out3 is the third output signal, K2 is the second convolution kernel parameter, Y concat is the second splicing signal.

3. The derived respiratory signal acquisition method according to claim 2, characterized in that: The discriminator network comprises a second encoder module and a post-processing module connected in sequence; wherein the second encoder module comprises a plurality of stacked second encoder units; The first training signal group and the second training signal group of several sample users are respectively input into the discriminator network to obtain the predicted label data of the first training signal group and the second training signal group of several sample users, including the steps of: The signals in the first training signal group and the second training signal group are respectively concatenated to obtain concatenated signals of the first training signal group and the second training signal group of several sample users, and the concatenated signals are used as input signals of the discriminator network. According to a preset convolution algorithm, the input signals of the discriminator network are convolved to obtain convolution feature signals of the input signals, wherein the convolution algorithm is: Y conv =LeakyRelu(BN(K3·Y n1 )) Where Y conv is the convolution feature signal, K3 is the third convolution kernel parameter, Y n1 is the input signal of the discriminator network; The convolution feature signal is used as the input signal of the first second encoder unit of the second encoder module, and according to the preset second encoding algorithm, the fourth output signal of the first second encoder unit is obtained, and the fourth output signal is used as the input signal of the next second encoder unit, and the encoding operation is repeated to obtain the fourth output signal of the last second encoder unit, wherein the second encoding algorithm is: Y out4 =LeakyRelu(SN(K4·LeakyRelu(SN(K4·Y n2 )))+Y n2 ) Where Y out4 is the fourth output signal, SN(·) is the spectral normalization function, K4 is the fourth convolution kernel parameter, Y n2 is an input signal of the second encoder unit; The fourth output signal of the last second encoder unit is input into the post-processing module, and the predicted label data of the first training signal group and the second training signal group of several sample users are obtained according to a preset label prediction algorithm, wherein the label prediction algorithm is: Y out5 =Sigmoid(W·Maxpool(Y out4 )) Where Y out5 To predict label data, Sigmoid(·) is the second activation function, W is the weight matrix, and Maxpool(·) is the maximum pooling function.

4. The derived respiratory signal acquisition method according to claim 3, characterized in that: The method of training the generative adversarial neural network according to the predicted label data and the real label data of the first training signal group and the second training signal group of several sample users to obtain a target generative adversarial neural network includes the following steps: The cross entropy algorithm is used to construct an adversarial loss value based on the predicted label data and the real label data of the first training signal group and the second training signal group of several sample users, wherein the adversarial loss value is: L GAN =Loss D (G(X),X,Y)+Loss G (G(X),X) Where, L GAN To combat the loss value, Loss D (G(X), X, Y) is the cross entropy loss value of the discriminator network, D is the discriminator, G(X) is the derived respiratory signal, X is the ballistocardiogram signal, and Y is the chest strap sensor signal. G (G(X),X) is the cross entropy loss value of the generator network, G is the generator network; Obtain a distance loss value of the generator network, obtain a comprehensive loss value according to the adversarial loss value, the distance loss value and a preset comprehensive loss function, train the generative adversarial neural network according to the comprehensive loss value, and obtain a target generative adversarial neural network, wherein the comprehensive loss function is: L=αL dis +βL GAN In the formula, L is the comprehensive loss value, α is the first weight coefficient, β is the second weight coefficient, and L dis is the distance loss value.

5. The derived respiratory signal acquisition method according to claim 4, characterized in that: The cross entropy algorithm is used to construct an adversarial loss value according to the first training signal group, the predicted label data of the second training signal group and the real label data of several sample users, including the steps of: According to the predicted label data and the real label data of the first training signal group of the sample user, and the predicted label data and the real label data of the second training signal group, a cross entropy algorithm is used to obtain the cross entropy loss value of the first training signal group and the cross entropy loss value of the second training signal group, and the cross entropy loss value of the first training signal group and the cross entropy loss value of the second training signal group are averaged to construct the cross entropy loss value of the discriminator network, wherein the cross entropy algorithm is: BCELoss(p,q)=-[p·log(q)+(1-p)·log(1-q)] In the formula, BCELoss(p,q) is the cross entropy loss value, p is the real label data, and q is the predicted label data; According to the cross-entropy loss value of the discriminator network, the discriminator network is back-propagated, and the second training signal group of the sample user is input into the discriminator network after the back-propagation is completed to perform label prediction, and the predicted label data of the second training signal group of the sample user is recovered. According to the recovered predicted label data and true label data of the second training signal group of the sample user, a cross-entropy algorithm is used to obtain the cross-entropy loss value of the second training signal group as the cross-entropy loss value of the generator network.

6. The derived respiratory signal acquisition method according to claim 4, characterized in that: The obtaining of the distance loss value of the generator network comprises the steps of: According to the second training signal group of the sample user and a preset global distance loss function, a global distance loss value is obtained, wherein the global distance loss function is: In the formula, l d is the global distance loss value, K is the time point, x k is the signal segment at the kth time point of a signal in the second training signal group, y k is a signal segment at the kth time point of another signal in the second training signal group; According to the second training signal group of the sample user and a preset local distance loss function, a local distance loss value is obtained, wherein the local distance loss function is: l m =max(|x1-y1|,…,|x k -and k |)-min(|x1-y1|,…,|x k -and k |) In the formula, l m is the local distance loss value, max(·) is the maximum value function, and min(·) is the minimum value function; The distance loss value is obtained according to the global distance loss value, the local distance loss value and a preset distance loss function, wherein the distance loss function is: L dis =γl d +ηl m Wherein, γ is the third weight coefficient, and η is the fourth weight coefficient.

7. A derived respiratory signal acquisition device, characterized in that: include: A signal acquisition module, used to obtain the ballistocardiogram signals, chest strap sensor signals and a preset generative adversarial neural network synchronously collected by several sample users, wherein the generative adversarial neural network includes a generator network and a discriminator network; A first signal generating module, used for inputting the ballistocardiogram signals of several sample users into the generator network respectively, to obtain the derived breathing signals of several sample users; a label prediction module, used to combine the chest belt sensor signal of the same sample user with the corresponding ballistocardiogram signal and the derived respiratory signal, respectively, to obtain a first training signal group and a second training signal group of several sample users, and to input the first training signal group and the second training signal group of several sample users into the discriminator network, respectively, to obtain predicted label data of the first training signal group and the second training signal group of several sample users; A generative adversarial neural network training module is used to obtain the real label data of the first training signal group and the second training signal group of several sample users, and train the generative adversarial neural network according to the predicted label data and the real label data of the first training signal group and the second training signal group of several sample users to obtain a target generative adversarial neural network; The second signal generating module is used to obtain the ballistocardiogram signal of the user to be generated, input the ballistocardiogram signal of the user to be generated into the target generator network, and obtain the derived breathing signal of the user to be generated.

8. A computer device, characterized in that: include: A processor, a memory, and a computer program stored in the memory and executable on the processor; when the computer program is executed by the processor, the steps of the derived respiratory signal acquisition method according to any one of claims 1 to 6 are implemented.

9. A storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the derived respiratory signal acquisition method according to any one of claims 1 to 6 are implemented.

Images