Health data detection method based on convolution auto-encoder Gaussian mixture model

A Gaussian mixture model, convolutional self-encoding technology, applied in the field of healthcare, can solve problems such as high computational overhead, achieve the effect of improving accuracy and avoiding computational overhead

CN112509696APending Publication Date: 2021-03-16JIANGNAN UNIV
0 Cites 1 Cited by

Patent Information

Authority / Receiving Office
CN · China
Current Assignee / Owner
Publication Date
2021-03-16

Smart Images

  • Figure 1
    Figure 1
  • Figure 2
    Figure 2
  • Figure 3
    Figure 3
Patent Text Reader

Abstract

The invention discloses a health data detection method based on a convolution auto-encoder Gaussian mixture model, and belongs to the technical field of medical care. According to the method, multi-dimensional data is converted into low-dimensional feature representation in a self-adaptive, non-linear and multi-layer coding mode, and the problem that the detection accuracy is reduced due to largecalculation expenditure caused by dimensionality disasters is effectively solved; a convolution and de-convolution neural network layer is added for multi-stage characteristics of human body activitydata, data features are effectively recognized and extracted, and the detection precision is further improved; a dimension reduction process and a density estimation process are organically combined together, and the situation that two models are independent and fall into a locally optimal awkward situation is avoided; meanwhile, in consideration of the singular point problem of the matrix and theproblem that the inverse of the covariance matrix may not be solved, the cholesky decomposition of the covariance matrix is constructed by utilizing the hybrid probability, the mean value and the covariance, so that the sample density is calculated, and the problem that the inverse of the covariance matrix cannot be solved is avoided.
Need to check novelty before this filing date? Find Prior Art

Description

Technical field

[0001] The present invention relates to a health data detection method based on convolutionary self-encoder Gaussian mixed model, belonging to a health care technique. Background technique

[0002] With the continuous development of the modern economy, people have more attention to their own health, so some health data is more and more attention to their own health data, and more and more people start using some to monitor sleep data, sports steps And the wearable device of the resting heart rate, such as the bracelet. At the same time, if you do medical treatment, the patient can provide its personal daily health information, which will greatly speed up the diagnosis speed and improve the quality of diagnosis. Especially for chronic diseases such as sleep disorders, obstructive sleep apnea syndrome, if it is possible to screen and monitor such chronic diseases in advance according to daily health data will be a major advancement in the medical field.

[0003] Due...

Examples

Embodiment 1

[0059] This embodiment provides a health data detection method based on a convolutionary self-encoder Gaussian mixed model, the method comprising:

[0060] Step 1: Training the collected original high-dimensional health data is trained from the encoder to minimize the reconstruction error, combined with the BP optimization algorithm to obtain the low-dimensional representation of the original high-dimensional health data and corresponding Reconstruction error;

[0061] Step 2: The sample density corresponding to the original high-dimensional health data is calculated as the original high-dimensional health data of the original high-dimensional health data, and the maximum value of the sample density is not only the density threshold, and the EM algorithm is combined to the Gaussian mixture model. Training, to obtain the optimal Gaussian mixed model parameters;

[0062] During the calculation, the CHOLESKY decomposition of the covariance matrix is ​​constructed by the mixing probab...

Embodiment 2

[0066] This embodiment provides a health data detection method based on convolutionautoencododer gaussian mixture model (CAE-GMM), which first uses MIN-Max normalization to normalize data samples. Due to the "dimension disaster", the density estimate is very difficult by conventional methods, so data samples are trained on the convolution from the encoder until the reconstruction error reaches the minimum, and this nonlinear manner is reduced in reducing "dimension disaster" While accurately acquiring the potential spatial representation of data samples as much as possible;