CN113317798A - Electrocardiogram compressed sensing reconstruction system based on deep learning - Google Patents

Electrocardiogram compressed sensing reconstruction system based on deep learning Download PDF

Info

Publication number
CN113317798A
CN113317798A CN202110554209.2A CN202110554209A CN113317798A CN 113317798 A CN113317798 A CN 113317798A CN 202110554209 A CN202110554209 A CN 202110554209A CN 113317798 A CN113317798 A CN 113317798A
Authority
CN
China
Prior art keywords
signal
electrocardio
reconstruction
compressed sensing
compressed
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202110554209.2A
Other languages
Chinese (zh)
Inventor
张宏坡
董忠仁
孙梦雅
谷红壮
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhengzhou University
Original Assignee
Zhengzhou University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Zhengzhou University filed Critical Zhengzhou University
Priority to CN202110554209.2A priority Critical patent/CN113317798A/en
Publication of CN113317798A publication Critical patent/CN113317798A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7232—Signal processing specially adapted for physiological signals or for diagnostic purposes involving compression of the physiological signal, e.g. to extend the signal recording period
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235—Details of waveform analysis
    • A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems

Landscapes

  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Physics & Mathematics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physiology (AREA)
  • Psychiatry (AREA)
  • Signal Processing (AREA)
  • Veterinary Medicine (AREA)
  • Public Health (AREA)
  • Pathology (AREA)
  • General Health & Medical Sciences (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Medical Informatics (AREA)
  • Molecular Biology (AREA)
  • Surgery (AREA)
  • Animal Behavior & Ethology (AREA)
  • Mathematical Physics (AREA)
  • Evolutionary Computation (AREA)
  • Fuzzy Systems (AREA)
  • Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)

Abstract

The invention provides an electrocardio compressed sensing reconstruction system based on deep learning. According to the technical scheme, firstly, an observation matrix is utilized to simultaneously complete sampling and compression on the electrocardiosignals, then the compressed signals are transposed and projected, the transposed projection signals are ensured to be the same as the original electrocardiosignals in size, and meanwhile, the transposed projection signals are subjected to Z-Score standardization. And then, directly learning the mapping relation between the transposed projection signal and the original signal by using the CNN, and initially reconstructing the electrocardiosignal. And finally, performing secondary reconstruction on the CNN reconstructed signal by using the LSTM, and further improving the reconstruction quality of the signal. The invention provides a non-iterative electrocardio compressed sensing reconstruction algorithm (CSNet) by combining compressed sensing and deep learning, and electrocardio signals can be rapidly and accurately reconstructed.

Description

Electrocardiogram compressed sensing reconstruction system based on deep learning
Technical Field
The invention belongs to the technical field of electrocardiogram monitoring, and particularly relates to an electrocardiogram compressed sensing reconstruction system based on deep learning.
Background
The sudden onset of cardiovascular disease leads to an increasing mortality rate of cardiovascular disease. Based on 2016's mortality statistics, it is estimated that 1700 thousands of people die from cardiovascular disease, accounting for 31% of the total number of deaths worldwide. For patients with cardiovascular diseases, wearable device-based remote electrocardiographic monitoring plays a very important role in preventing and treating cardiovascular diseases. However, long-term monitoring of cardiac electricity produces large amounts of data. For example, when the sampling rate is 400Hz and the resolution is 12 bits, 26MB of data needs to be stored or transmitted in a single lead electrocardiosignal; when the resolution is 16 bits, two leads are required to transmit 138MB of data. In addition, some scholars have demonstrated that the wireless transmission portion consumes most of the energy in a real-time monitoring environment.
Lossy compression techniques based on transform domains (e.g., discrete fourier transform, discrete cosine transform, discrete wavelet transform, etc.) are popular in the compression of cardiac electrical signals, but these methods are all performed by fully sampling the cardiac electrical signals, then performing sparse transform, discarding a large number of small coefficients obtained by transform, and retaining a small number of large coefficients, which is an inefficient practice. The compressed sensing samples signals by a sub-Nyquist sampling method, less data are collected as simply as possible, and under the condition that the signals are sparse, the collected signals are perfectly reconstructed by utilizing optimization algorithms such as L1 and the like. Based on the characteristic, the CS can be applied to the remote electrocardiographic monitoring problem, and because the compression is simple linear operation, the computational complexity of the acquisition end can be reduced.
The electrocardio compression sensing process generally comprises two steps: firstly, a fixed observation matrix is used at an acquisition end to simultaneously acquire and compress the electrocardiosignals, and the compressed signals are transmitted to a remote server end. This step requires only one matrix multiplication, and the computational cost is low. Secondly, at the server side, the received compressed signal is reconstructed using a reconstruction algorithm. The reconstruction algorithms are mainly classified into three categories: based on a greedy algorithm, based on an optimization algorithm and based on a Bayesian learning algorithm. The above algorithms are based on sparse prior knowledge and iteratively reconstruct the original signal by solving an optimization problem. However, the system requiring real-time performance cannot perform the corresponding task.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides an electrocardio compressed sensing reconstruction system based on deep learning. The specific scheme is as follows:
step 1: the wearable equipment end carries out random projection on the original electrocardiosignal to obtain a compressed signal Y; the process is as follows:
Figure RE-GDA0003111002450000021
wherein,
y(i)is the cardiac electrical signal after compression and is,
x(i)is the i-th raw ECG signal,
Figure RE-GDA0003111002450000022
is an observation matrix with dimensions n × m;
step 2: the wearable device side transmits the compressed data to a remote server side;
and step 3: the server side performs transposition projection operation on the compressed signal; the process is as follows:
Figure RE-GDA0003111002450000023
wherein,
Figure RE-GDA0003111002450000024
is an observation matrix
Figure RE-GDA0003111002450000025
The transpose of (a) is performed,
Figure RE-GDA0003111002450000026
is a transposed projection signal;
and 4, step 4: the server side performs normalization operation on the transposed projection signal; the process is as follows:
Figure RE-GDA0003111002450000027
Figure RE-GDA0003111002450000028
is the average value of the average of the values,
Figure RE-GDA0003111002450000029
is the standard deviation of the measured data to be measured,
r(i)is the data after normalization;
and 5: the server inputs the normalized data into CNN in the electrocardio compressed sensing reconstruction algorithm, and outputs an initially reconstructed electrocardiosignal; the process is as follows:
Figure RE-GDA00031110024500000210
wherein,
g ═ 1,2) is the index number of the convolutional layer,
cgis the output of the g-th convolutional layer,
f is a non-linear activation function,
b(i)is a bias term for the feature map,
m is the size of the convolution kernel,
wmis the weight of the mth feature map;
step 6: the server inputs the initial reconstruction signal into an LSTM in an electrocardio compressed sensing reconstruction algorithm and outputs a second reconstruction electrocardio signal; the process is as follows:
Figure RE-GDA0003111002450000031
Figure RE-GDA0003111002450000032
Figure RE-GDA0003111002450000033
Figure RE-GDA0003111002450000034
Figure RE-GDA0003111002450000035
wherein,
Figure RE-GDA0003111002450000036
is the input sequence (also the output of CNN) at time t,
it，ft，ot，
Figure RE-GDA0003111002450000037
and htRespectively an input gate, a forgetting gate, an output gate, a unit state and a hidden state,
w is a matrix of weights that is,
b is a bias vector;
compared with the prior art, the invention has outstanding substantive characteristics and remarkable progress, and particularly provides a non-iterative electrocardio compressed sensing reconstruction algorithm (CSNet), which combines a compressed sensing and deep learning method and can rapidly and accurately reconstruct electrocardiosignals. The reconstruction error of the invention is lower than that of the traditional method under the condition of high compression ratio (CR is more than or equal to 50 percent). Meanwhile, for an electrocardiosignal with the duration of 30 minutes, reconstruction can be completed only by about 0.12s, and the reconstruction speed is improved by at least 45 times compared with that of the traditional method, which is enough to support real-time application.
Drawings
FIG. 1 is an overall architecture of the electrocardiographic compressive sensing reconstruction system of the present invention.
Fig. 2 shows the reconstruction performance of different reconstruction algorithms under different compression ratios.
Fig. 3 shows QRS detection performance of different reconstruction algorithms corresponding to the reconstructed signal.
Detailed Description
The technical solution of the present invention is further described in detail by the following embodiments.
As shown in fig. 1, an electrocardiographic compressed sensing reconstruction system based on deep learning includes:
compressing the original electrocardiosignals according to different compression ratios to obtain compressed data;
performing transposition projection on the compressed signal and normalizing transposition projection data;
and inputting the processed data into the CNN and LSTM network models for electrocardiosignal reconstruction.
Given a dataset X { (X)(1)，z(1))，...，(x(i)，z(i))，...，(x(n)，z(n)) -said raw signal compression is done by:
Figure RE-GDA0003111002450000041
x(i)is the ith electrocardiosignal, and is the first electrocardiosignal,
Figure RE-GDA0003111002450000042
is an observation matrix fixed in the experiment, the dimension is n x m,
y(i)is a compressed signal obtained by random projection of an observation matrix.
The transpose projection operation of the compressed data is completed by the following steps:
Figure RE-GDA0003111002450000043
Figure RE-GDA0003111002450000044
is an observation matrix
Figure RE-GDA0003111002450000045
The transpose of (a) is performed,
Figure RE-GDA0003111002450000046
is to turn toAnd (4) setting a projection signal.
The normalization operation of the inverted projection signal is completed by the following steps:
Figure RE-GDA0003111002450000047
Figure RE-GDA0003111002450000051
is the average value of the average of the values,
Figure RE-GDA0003111002450000052
is the standard deviation of the measured data to be measured,
r(i)is the data after normalization.
The step of inputting the processed data into the CNN and LSTM network models for atrial fibrillation screening is completed by the following steps:
Figure RE-GDA0003111002450000053
Figure RE-GDA0003111002450000054
Figure RE-GDA0003111002450000055
Figure RE-GDA0003111002450000056
Figure RE-GDA0003111002450000057
Figure RE-GDA0003111002450000058
g ═ 1,2) is the index number of the convolutional layer,
cgis the output of the g-th convolutional layer,
f is a non-linear activation function,
b(i)is a bias term for the feature map,
m is the size of the convolution kernel,
wmis the weight of the mth feature map,
Figure RE-GDA0003111002450000059
is the input sequence (also the output of CNN) at time t,
it，ft，ot，
Figure RE-GDA00031110024500000510
and htRespectively an input gate, a forgetting gate, an output gate, a unit state and a hidden state,
w is a matrix of weights that is,
b is a bias vector.
The initial reconstruction network structure parameters in the CSNet provided by the invention are shown in Table 1, the input of the network is a transposed projection signal with the shape of 256 multiplied by 1, the size of a first layer convolution kernel is 1 multiplied by 11, 64 characteristic graphs are output, the step length is 1, and the activation function is ReLU; the size of a second layer convolution kernel is 1 multiplied by 11, 32 characteristic graphs are output, the step length is 1, and the activation function is ReLU; the third layer of convolution kernel size is 1 × 11, and 1 feature map (initial reconstructed signal) is output.
TABLE 1 initial reconstructed network parameters
Figure RE-GDA0003111002450000061
The secondary reconstruction network in the CSNet provided by the invention is provided with a hidden layer, the hidden layer has 250 units, and the activation function is tanh. The density layer output is 256 and the activation function is a linear function. The final output shape is controlled to 256 × 1.
It is not necessary to find the true global minimum during deep network model learning. Thus, when accuracy is lost during learning or enters a relatively flat region, Early stopping (Early stopping) techniques may be selected for use. Early stopping may be considered an unobtrusive form of normalization without affecting learning motivation, and this strategy may be used in conjunction with other normalization techniques. The technology can prevent the model from being over-fitted and accelerate the learning speed. By default, the maximum number of early stop periods may be set to 50 in view of efficiency and efficiency. The default settings for the model parameters are given in table 2.
TABLE 2 model parameter settings
Figure RE-GDA0003111002450000062
Verification experiment
The data set used for the experiment was the MIT-BIH arrhythmia data set. This data set contains 48 30-minute length recordings with two lead (MLII and V5) signals each, and MLII data was used for this experiment at a data sampling rate of 360 Hz. Where 100,101,102,107,109,111,115,117,118 and 119 are selected as the test set and the remaining records are the training set. In addition, 12.5% was randomly taken from the training set as the validation set, with the ratio of training set, validation set, and test set being approximately 7:1: 2.
The method aims to complete reconstruction of the electrocardio-compressed signal. The error of the reconstructed signal from the original signal is crucial.
The present invention uses PRD and SNR to estimate the error between the reconstructed signal and the original signal. Different methods the PRD and SNR at different CR per record averaged over the test set is shown in figure 2. The results in FIG. 2(a) show that the average PRD of our proposed method at high CR (CR ≧ 50%) is lower than that of the conventional algorithm. In addition, the overall increase trend of the reconstruction error of the CSNet is relatively gentle. Similarly, the experimental results in FIG. 2(b) show that the average SNR of CSNet is better than that of the conventional algorithm at high CR (CR ≧ 50%).
The purpose of reconstruction of the cardiac signal is for further analysis and diagnosis. Therefore, the requirements of clinical medicine on the quality of the reconstructed signal are worth further analysis. The invention takes PRD 9% as the qualified signal reconstruction quality standard of clinical examination. The different tests were recorded under CSNet and the four conventional methods when PRD was 9% corresponding CR is listed in table 3. The results of the experiment in Table 3 show that the mean CR of CSNet in the test record is 85%, which means that even if 85% of the electrocardiosignal is lost, the signal reconstructed by CSNet still meets the clinical requirements. Meanwhile, the reconstructed signals of other methods meet the clinical requirements of CR of 52%, 69%, 72% and 78% respectively, and are at least lower than CSNet 7%. The CSNet can compress the electrocardiosignals to the maximum extent under the condition that the reconstructed signals of different methods meet the clinical requirements.
Table 3 compression ratio of different reconstruction algorithms in case of PRD 9%
Figure RE-GDA0003111002450000071
The signal reconstruction time is a key factor of electrocardio compressed sensing, and the invention compares the average execution time of the reconstruction process of each record (about 30 minutes) in the test records of different algorithms under different CRs. The execution times of the different algorithms are listed in table 4. Under different CR, the CSNet provided by the invention takes about 20 to 88 times faster than BP algorithm, about 54 to 1810 times faster than OMP algorithm, about 671 to 1515 times faster than BSBL-BO algorithm and about 98 to 196 times faster than R-SVD + BP to reconstruct long-time electrocardiogram data in 30 minutes. For example, when CR is 50%, the average time for reconstructing one record by BP, OMP, BSBL-BO, R-SVD + BP, and CSNet is 5.64s, 50.71s, 143.43s, 16.60s, and 0.1244s, respectively. Notably, the non-iterative nature of CSNet is well suited for parallelization, making implementation very fast.
TABLE 4 reconstruction time per record (approximately 30 minutes per record)
Figure RE-GDA0003111002450000081
The QRS wave is an important component of the ECG signal. Generally, the reconstructed cardiac signal requires further analysis (e.g. arrhythmia classification), the first step of which is QRS wave detection. In addition, QRS wave detection analysis is another method for evaluating the quality of the reconstructed signal, i.e., how the compression method stores clinically relevant information in an electrocardiographic signal, rather than simply using energy-based PRD index evaluation. Therefore, the performance of the QRS detection algorithm on reconstructed signals at different CRs is worth analyzing. Based on this goal, the well-known Pan-Tompkins method is used to perform QRS detection on the reconstructed signal. The test set (MIT-BIH dataset already provided clinical QRS markers) was used to assess whether QRS waves at different CRs could be correctly detected. QRS detection performance of the reconstructed signal is evaluated with recall (recall) and precision (precision). Fig. 3 shows QRS detection performance under different CR for different methods. Experimental results show that Recall and Precision of reconstructed signals of CSNet are higher than those of other methods. Specifically, the experimental results of FIG. 3(a) show that when CR is from 10% to 90%, the Recall of CSNet is greater than 98.86%; when CR is 70%, Reacll of BP, BSBL-BO and OMP begins to be greatly reduced, and R-SVD + BP is better in overall performance, but Recall is still lower than CSNet. The experimental results in FIG. 3(b) show that Precision of different methods also shows similar trend, and when CR is from 10% to 90%, Precision of CSNet is highest (more than 99.24% overall).
Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention and not to limit it; although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art will understand that: modifications to the specific embodiments of the invention or equivalent substitutions for parts of the technical features may be made; without departing from the spirit of the present invention, it is intended to cover all aspects of the invention as defined by the appended claims.

Claims (6)

1. The utility model provides an electrocardio compressed sensing reconfiguration system based on degree of depth study, includes wearable equipment end and remote server end, characterized by: the wearable equipment end carries out random projection on the original electrocardiosignals according to a compression perception theory to obtain compressed signals; the remote server side performs transposition projection and normalization processing on the compressed signal sent by the wearable equipment side, inputs the processed data into a pre-established electrocardio compression perception reconstruction model and outputs a well-reconstructed electrocardio signal; the reconstruction comprises the following specific steps:
step 1: the wearable equipment end carries out random projection on the original electrocardiosignal; the process is as follows:
Figure RE-FDA0003111002440000011
wherein,
y(i)is the cardiac electrical signal after compression and is,
x(i)is the i-th raw ECG signal,
Figure RE-FDA0003111002440000012
is an observation matrix with dimensions n × m;
step 2: the wearable device side transmits the compressed data to a remote server side;
and step 3: the server side performs transposition projection operation on the compressed signal; the process is as follows:
Figure RE-FDA0003111002440000013
wherein,
Figure RE-FDA0003111002440000014
is an observation matrix
Figure RE-FDA0003111002440000015
The transpose of (a) is performed,
Figure RE-FDA0003111002440000016
is a transposed projection signal;
and 4, step 4: the server side performs normalization operation on the transposed projection signal; the process is as follows:
Figure RE-FDA0003111002440000017
wherein,
Figure RE-FDA0003111002440000018
is the average value of the average of the values,
Figure RE-FDA0003111002440000019
is the standard deviation of the measured data to be measured,
r(i)is the data after normalization;
and 5: the server inputs the normalized data into CNN in the electrocardio compressed sensing reconstruction algorithm, and outputs an initially reconstructed electrocardiosignal; the process is as follows:
Figure RE-FDA0003111002440000021
wherein,
g ═ 1,2) is the index number of the convolutional layer,
cgis the output of the g-th convolutional layer,
f is a non-linear activation function,
b(i)is a bias term for the feature map,
m is the size of the convolution kernel,
wmis the weight of the mth feature map;
step 6: the server inputs the initial reconstruction signal into an LSTM in an electrocardio compressed sensing reconstruction algorithm and outputs a second reconstruction electrocardio signal; the process is as follows:
Figure RE-FDA0003111002440000022
Figure RE-FDA0003111002440000023
Figure RE-FDA0003111002440000024
Figure RE-FDA0003111002440000025
Figure RE-FDA0003111002440000026
wherein,
Figure RE-FDA0003111002440000027
is the input sequence (also the output of CNN) at time t,
it，ft，ot，
Figure RE-FDA0003111002440000028
and htRespectively an input gate, a forgetting gate, an output gate, a unit state and a hidden state,
w is a matrix of weights that is,
b is a bias vector;
2. the deep learning-based electrocardio compressed sensing reconstruction system according to claim 1, characterized in that: the entire CSNet is formed by combining a CNN and an LSTM. CNN consists of only three convolutional layers, the number of convolutional cores is 64, 32 and 1 in sequence; the convolution kernels are all 11 in size. In addition, no pooling layer is arranged for ensuring the length of the electrocardiosignal to be unchanged. Finally, only one convolution kernel is set so as to ensure that an initial reconstruction signal with the same size as the original signal is output.
3. The deep learning-based electrocardio compressed sensing reconstruction system according to claim 1, characterized in that: the entire CSNet is formed by combining a CNN and an LSTM. The LSTM is provided with 250 cells, and the last full-connection layer is provided with 256 neurons, so that the shape of a secondary reconstruction signal is the same as that of an original signal.
4. The deep learning-based electrocardio compressed sensing reconstruction system according to claim 1, characterized in that: in the training process of the electrocardio-compressed sensing reconstruction algorithm, the batch size is set to be 256, the loss function used in the training is MSE, Adam is adopted for optimization, the initial learning rate is set to be 0.0005, and the model parameters are updated in each loop iteration.
5. The deep learning-based electrocardio compressed sensing reconstruction system according to claim 1, characterized in that: and the wearable equipment end transmits the compressed data subjected to random projection to a remote server end through a wireless transmitting module.
6. The deep learning-based electrocardio compressed sensing reconstruction system according to claim 1, characterized in that: the wearable equipment terminal collects human body electrocardio data.
CN202110554209.2A 2021-05-20 2021-05-20 Electrocardiogram compressed sensing reconstruction system based on deep learning Pending CN113317798A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110554209.2A CN113317798A (en) 2021-05-20 2021-05-20 Electrocardiogram compressed sensing reconstruction system based on deep learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110554209.2A CN113317798A (en) 2021-05-20 2021-05-20 Electrocardiogram compressed sensing reconstruction system based on deep learning

Publications (1)

Publication Number Publication Date
CN113317798A true CN113317798A (en) 2021-08-31

Family

ID=77416082

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110554209.2A Pending CN113317798A (en) 2021-05-20 2021-05-20 Electrocardiogram compressed sensing reconstruction system based on deep learning

Country Status (1)

Country Link
CN (1) CN113317798A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114098679A (en) * 2021-12-30 2022-03-01 中新国际联合研究院 Vital sign monitoring waveform recovery method based on deep learning and radio frequency perception
CN114841892A (en) * 2022-05-20 2022-08-02 天津大学 Sparse guided wave data recovery method based on full-connection network
CN115618201A (en) * 2022-10-09 2023-01-17 湖南万脉医疗科技有限公司 Breathing machine signal processing method based on compressed sensing and breathing machine

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106788885A (en) * 2016-12-07 2017-05-31 上海交通大学 A kind of compress ecg data rate Automatic adjusument wireless transmitting system and transmission method
CN107147397A (en) * 2017-04-24 2017-09-08 电子科技大学 Towards the Fast Compression sensing reconstructing method of wearable device
CN111643074A (en) * 2020-06-29 2020-09-11 郑州大学 Lossy compression electrocardiosignal atrial fibrillation screening system
CN112348914A (en) * 2020-11-12 2021-02-09 中国科学院合肥物质科学研究院 Deep learning image compression sensing algorithm and system based on Internet of vehicles

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106788885A (en) * 2016-12-07 2017-05-31 上海交通大学 A kind of compress ecg data rate Automatic adjusument wireless transmitting system and transmission method
CN107147397A (en) * 2017-04-24 2017-09-08 电子科技大学 Towards the Fast Compression sensing reconstructing method of wearable device
CN111643074A (en) * 2020-06-29 2020-09-11 郑州大学 Lossy compression electrocardiosignal atrial fibrillation screening system
CN112348914A (en) * 2020-11-12 2021-02-09 中国科学院合肥物质科学研究院 Deep learning image compression sensing algorithm and system based on Internet of vehicles

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
K. YAMAMOTO: "ECG Signal Reconstruction via Doppler Sensor by Hybrid Deep Learning Model With CNN and LSTM", 《IEEE ACCESS》 *

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114098679A (en) * 2021-12-30 2022-03-01 中新国际联合研究院 Vital sign monitoring waveform recovery method based on deep learning and radio frequency perception
CN114098679B (en) * 2021-12-30 2024-03-29 中新国际联合研究院 Vital sign monitoring waveform recovery method based on deep learning and radio frequency sensing
CN114841892A (en) * 2022-05-20 2022-08-02 天津大学 Sparse guided wave data recovery method based on full-connection network
CN114841892B (en) * 2022-05-20 2023-10-17 天津大学 Sparse guided wave data recovery method based on fully connected network
CN115618201A (en) * 2022-10-09 2023-01-17 湖南万脉医疗科技有限公司 Breathing machine signal processing method based on compressed sensing and breathing machine

Similar Documents

Publication Publication Date Title
CN113317798A (en) Electrocardiogram compressed sensing reconstruction system based on deep learning
CN111814656B (en) Electrocardiosignal noise reduction method based on countermeasure generation network
CN109394205B (en) Electrocardiosignal analysis method based on deep neural network
Pandey et al. Quality controlled ECG data compression based on 2D discrete cosine coefficient filtering and iterative JPEG2000 encoding
CN112508110A (en) Deep learning-based electrocardiosignal graph classification method
CN112806977B (en) Physiological parameter measuring method based on multi-scale fusion network
CN112329609A (en) Feature fusion transfer learning arrhythmia classification system based on 2D heart beat
WO2020228420A1 (en) Denoising autoencoder training method, electrocardiography signal denoising method, and apparatuses
CN117814805B (en) Intelligent processing method for data of clinical care equipment
El B’charri et al. The ECG signal compression using an efficient algorithm based on the DWT
CN116361688A (en) Multi-mode feature fusion model construction method for automatic classification of electrocardiographic rhythms
CN114052744A (en) Electrocardiosignal classification method based on pulse neural network
CN111643074A (en) Lossy compression electrocardiosignal atrial fibrillation screening system
CN110537907B (en) Electrocardiosignal compression and identification method based on singular value decomposition
CN118044813B (en) Psychological health condition assessment method and system based on multitask learning
CN114711784A (en) Heart transplantation postoperation electrocardiogram abnormity analysis method and system
Yan et al. Wireless electrocardiograph monitoring based on wavelet convolutional neural network
CN111700609A (en) Atrial fibrillation detection method, device and equipment based on short-time electrocardiosignals
Huang et al. Sub-Nyquist sampling of ECG signals based on the extension of variable pulsewidth model
Allam et al. A deformable CNN architecture for predicting clinical acceptability of ECG signal
Huang et al. Sub-Nyquist sampling of ECG signals with differentiated VPW optimization model
Hakkak et al. Analysis of lossless compression techniques time-frequency-based in ECG signal compression
CN116671929A (en) Method, system, equipment and medium for determining optimal decomposition scale for electrocardiosignal denoising
CN112274159B (en) Compression domain electrocardiosignal quality assessment method based on improved band-pass filter
Pal et al. A new automated compression technique for 2D electrocardiogram signals using discrete wavelet transform

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination