CN112185543A - Construction method of medical induction data flow classification model - Google Patents
Construction method of medical induction data flow classification model Download PDFInfo
- Publication number
- CN112185543A CN112185543A CN202010925639.6A CN202010925639A CN112185543A CN 112185543 A CN112185543 A CN 112185543A CN 202010925639 A CN202010925639 A CN 202010925639A CN 112185543 A CN112185543 A CN 112185543A
- Authority
- CN
- China
- Prior art keywords
- classification model
- data flow
- data
- medical
- flow classification
- 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
Links
- 230000006698 induction Effects 0.000 title claims abstract description 57
- 238000013145 classification model Methods 0.000 title claims abstract description 56
- 238000010276 construction Methods 0.000 title claims abstract description 8
- 238000000034 method Methods 0.000 claims abstract description 41
- 230000002159 abnormal effect Effects 0.000 claims description 9
- 230000004913 activation Effects 0.000 claims description 6
- 230000011218 segmentation Effects 0.000 claims description 6
- 230000006870 function Effects 0.000 claims description 4
- 230000002547 anomalous effect Effects 0.000 claims description 3
- 238000005516 engineering process Methods 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 3
- 238000013501 data transformation Methods 0.000 claims 1
- 201000010099 disease Diseases 0.000 abstract description 5
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 abstract description 5
- 238000003745 diagnosis Methods 0.000 abstract description 4
- 230000005856 abnormality Effects 0.000 abstract description 3
- 239000010410 layer Substances 0.000 description 35
- 206010003658 Atrial Fibrillation Diseases 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 2
- 230000000295 complement effect Effects 0.000 description 2
- 208000024172 Cardiovascular disease Diseases 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 239000002346 layers by function Substances 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Biomedical Technology (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Biophysics (AREA)
- Public Health (AREA)
- Medical Informatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Databases & Information Systems (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Pathology (AREA)
- Probability & Statistics with Applications (AREA)
- Epidemiology (AREA)
- Primary Health Care (AREA)
- Measuring And Recording Apparatus For Diagnosis (AREA)
Abstract
The invention discloses a construction method of a medical induction data flow classification model, which comprises the following steps: the method for training and obtaining the medical induction data flow classification model based on the combination of CNN and LSTM by adopting an automatic supervision learning method comprises the following steps: inputting unmarked data into a sparse self-encoder of a medical induction data stream classification model for training so as to enable the input and the output of the model to be consistent; inputting the data with the labels into a trained sparse self-encoder, and outputting to obtain a corresponding classification result; training a medical induction data flow classification model based on a deep separable convolution method, obtaining a lightweight medical induction data flow classification model, and storing the obtained lightweight medical induction data flow classification model in a cloud service library or deploying the lightweight medical induction data flow classification model on edge equipment. According to the invention, by constructing an efficient and accurate lightweight model, higher abnormality diagnosis accuracy can be ensured, and high-precision disease classification can be realized. The method meets the requirement of light weight while meeting high reasoning accuracy, and is effectively deployed.
Description
Technical Field
The invention relates to a construction method of a medical induction data flow classification model, and belongs to the technical field of medical induction data processing.
Background
One major problem faced with anomaly detection based on medical sensing data is the lack of tagged data. The marking of data requires professional medical knowledge and the marking process is complex and time consuming. The large-scale high-precision deep learning model can ensure higher abnormality diagnosis accuracy, and occupies a large amount of memory and computing resources when actually deployed and used. For different application scenarios, medical sensing data flow classification models with different inference accuracy and different scales (i.e. different spatial and temporal complexities) need to be proposed.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides a construction method of a medical induction data flow classification model, which is based on an automatic supervision learning theory and designs an efficient and accurate medical induction data flow classification model; based on a deep separable convolution method, a lightweight class model of the medical induction data stream is designed, and the model can be effectively and quickly deployed on a cloud service library or various resource-limited devices.
The invention specifically adopts the following technical scheme to solve the technical problems:
a construction method of a medical induction data flow classification model comprises the following steps:
the method for training and obtaining the medical induction data flow classification model based on the combination of CNN and LSTM by adopting an automatic supervision learning method comprises the following steps: inputting unmarked data into a sparse self-encoder of a medical induction data stream classification model for training so as to enable the input and the output of the model to be consistent; inputting the data with the labels into a trained sparse self-encoder, and outputting to obtain a corresponding classification result;
training a medical induction data flow classification model based on a deep separable convolution method to obtain a lightweight medical induction data flow classification model, which specifically comprises the following steps:
segmenting a time collected by a sensor using a sliding window techniqueHuman body induction data flow in the time interval is divided into n sections of multidimensional time sequence data: x is the number of1,...,xnWherein n belongs to Z, and n is more than or equal to 0; z is an integer set;
inputting the segmented n segments of multi-dimensional time sequence data into a medical induction data stream classification model for classification: extracting features of the multi-dimensional time series data within each sliding window, and predicting time series data within a variable time length from the output of the unit time and the time series data of the next unit time; classifying whether the predicted time sequence data is abnormal or not, obtaining and outputting a classification result, and obtaining a lightweight medical induction data stream classification model;
and storing the obtained lightweight medical induction data flow classification model in a cloud service library or deploying the lightweight medical induction data flow classification model on edge equipment.
Further, as a preferred technical solution of the present invention, in the method, the medical sensing data flow classification model is composed of two LSTM layers, a full connection layer and a softmax output layer, and edges between the two connection layers have weights.
Further, as a preferred technical solution of the present invention, the data conversion between two cells in the LSTM layer is input to the bottom hidden layer
Wherein ilowerIs an input to the bottom hidden layer; f () is an activation function;
is the weight connecting the edges of the upper unit and the lower unit; n is the output number of the upper layer; onIs the output of a cell; blowerIs an offset value.
Further, as a preferred technical solution of the present invention, a depth separable convolutional network including a plurality of convolutional layers is disposed inside each cell of the LSTM layer; the convolutional layers include two types: 3 x 3Depth-wise convolutional layers and 1 x 1 convolutional layers, and the two types of convolutional layers are arranged at intervals, and each convolutional layer is followed by a batch standardized BN layer and a ReLU activation layer.
Further, as a preferred embodiment of the present invention, the method classifies whether the predicted time-series data is abnormal or not, specifically:
due to n predicted time series data { x1，...，xnThere are m digital signals for each data sample
And, the ith predicted time-series data xi∈xn；
Let the true value of the n time-series data be
The error of the jth element in the ith time-series data is
Defining time series data x for multivariate Gaussian ith predictioniError vector e ofiThe distribution p of (A) is:
wherein
Is an n x 1 vector and is,
is an n × n matrix;
if p (e)i(ii) a Mu; xi is less, then called xiIs a normal time series data, otherwise, xiIs anomalous time series data, in which is an error threshold.
Further, as a preferred technical solution of the present invention, the method further includes performing a segmentation removal on the collected human body sensing data stream by using a conventional segmentation removal method.
By adopting the technical scheme, the invention can produce the following technical effects:
the method of the invention obtains a high-precision medical induction data flow classification model based on CNN and LSTM based on self-supervision learning method training, and uses unlabelled and labeled data to train the medical induction data flow classification model, wherein the integrated model is the combination of CNN and LSTM, CNN can learn local characteristics and can share weight, LSTM can learn the front and back dependency relationship of data, and the two complement each other, thereby improving the accuracy of prediction. And moreover, a lightweight medical induction data stream classification model is obtained by training based on a depth separable convolution technology, and a lightweight full-precision model idea is adopted, so that a system deployed at the edge has two characteristics: light weight and high inference accuracy. The large-scale high-precision medical induction data flow classification model can ensure higher abnormal diagnosis accuracy and realize high-precision disease classification. The lightweight medical induction data flow classification model meets the requirement of lightweight while meeting high reasoning accuracy, and can be deployed on resource-limited equipment such as a cloud service library or edge mobile equipment.
Drawings
FIG. 1 is a schematic flow chart of a model training method for the self-supervised learning method adopted in the present invention.
FIG. 2 is a schematic diagram of a model trained based on the deep separable convolution method employed in the present invention.
Detailed Description
The following describes embodiments of the present invention with reference to the drawings.
The invention provides a construction method of a medical induction data flow classification model, which specifically comprises the following steps:
step one, training and obtaining a medical induction data flow classification model based on combination of CNN and LSTM by adopting a self-supervision learning method, which comprises the following steps:
for a disease domain, such as cardiovascular disease, a large-scale, full-precision depth model can be trained, deployed in the cloud. Considering that the initial period of practical collection of labeled medical data is very limited, the invention adopts a self-supervision learning method to train a medical induction data flow classification model based on combination of CNN and LSTM, as shown in FIG. 1, including:
firstly, inputting unlabelled data into a sparse self-encoder of a medical induction data stream classification model for training, so that the input and the output of the model are consistent. This allows learning of important features in the sample and obtaining a better characterization than the original input data. The output of the hidden layer obtained after the sparse self-encoder is trained is a better representation of the original input data.
The tagged data is then input into a trained sparse autoencoder to obtain a better representation of the features. The processed labeled data is used for training a medical induction data flow classification model, wherein the integrated model is the combination of CNN and LSTM. CNN can learn local characteristics and weight sharing, and LSTM can learn the front-back dependency relationship of data, and the front-back dependency relationship complement each other, so that the accuracy of prediction is improved. Thus, a prediction result output, i.e., a classification result of the corresponding sample, can be obtained.
And secondly, training a medical induction data flow classification model based on a deep separable convolution method to obtain a lightweight medical induction data flow classification model. On the basis of early work, a lightweight accurate convolution algorithm based on deep learning is developed and is released in a cloud service library in a micro-service mode or is deployed on edge equipment. The invention adopts the thought of a light-weight full-precision model, and ensures that a system deployed at the edge has two characteristics: light weight and high inference accuracy.
Training a medical induction data flow classification model based on a depth separable convolution method, extracting the characteristics of multi-dimensional time sequence data in each sliding window, and predicting the time sequence data in variable time length according to the output of unit time and the time sequence data of the next unit time; classifying whether the predicted time sequence data is abnormal or not, obtaining and outputting a classification result, and obtaining a lightweight medical induction data stream classification model; the specific process is as follows:
firstly, a sliding window technology is adopted to segment a human body induction data stream collected by a sensor in a time period, and the human body induction data stream is segmented into n segments of multidimensional time sequence data: x is the number of1，...，xnWherein n belongs to Z, and n is more than or equal to 0; z is an integer set; and inputting the segmented n segments of multi-dimensional time sequence data into a medical induction data flow classification model for classification.
The medical induction data flow classification model is an integrated model of CNN and LSTM for time series, the architecture of the model is shown in FIG. 2, and the model consists of four functional layers: two LSTM layers, one full link layer and one softmax output layer, where the edges between the two links have weights. One cell within the LSTM layer represents one time unit. Data conversion between two cells, i.e. input to the bottom hidden layer
Wherein ilowerIs an input to the bottom hidden layer; f is an activation function;
is the weight connecting the edges of the upper unit and the lower unit; n is the output number of the upper layer; onIs the output of a cell; blowerIs an offset value. Since each cell of the LSTM layer has a function of controlling the flow of sensing information and a unique multiplication structure of the model, the model can predict time series data in a variable time length according to the output of a unit time and time series data of the next unit time.
Deep separable convolutional networks are designed inside each LSTM cell, and the characteristics of the multi-dimensional time sequence data in each sliding window are extracted. The depth-in-cell separable convolutional network architecture comprises a plurality of convolutional layers, such as 25 layers. There are two types of convolutional layers: 3X 3Depth-wise convolutional layers and 1X 1 convolutional layers. The two types of convolution layers are arranged at intervals. Each convolutional layer is followed by a bulk standardized BN layer and a ReLU activation layer.
After the training of the model is completed, the method is usedIt is possible to classify whether the predicted time-series data is abnormal or not. For this reason, the present invention assumes that the error in predicting time series data is derived from the fact that the true time series fits a multivariate Gaussian distribution. Due to n predicted time series data { x1，...，xnThere are m digital signals for each sample
And ith predicted time-series data xi∈xn(ii) a . Let the true value of the n time-series data be
The error of the jth element in the ith time-series data is
By using
Time series data x defining multivariate Gaussian ith predictioniError vector e ofiDistribution p of (a). Wherein
Is an n x 1 vector and is,
is an n × n matrix. If p (e)i(ii) a Mu; xi is less, then called xiIs a normal time series data, otherwise, xiIs anomalous time series data, in which is an error threshold.
And step three, finally, storing the obtained lightweight medical induction data flow classification model in a cloud service library, and publishing the model in the cloud service library in a micro-service mode, or deploying the model on edge equipment, and deploying and storing the model on a cloud server or edge mobile equipment and other equipment with limited resources.
In addition, the method can also adopt a conventional segmentation removing method, and the conventional segmentation removing method is adopted to carry out segmentation removal on the acquired human body induction data stream, namely, before data is input into the model, useless segments are removed, and only segments with possible exceptions or exception prefixes are reserved. Here, the regular segment refers to a segment that maintains a daily state. For example, an abnormality of a patient with long-term atrial fibrillation refers to an abnormal ECG signal caused by other diseases, which are developed on the basis of atrial fibrillation in a conventional segment.
In conclusion, the method can ensure higher accuracy of abnormal diagnosis and realize high-precision disease classification by constructing an efficient and accurate lightweight medical induction data stream classification model. The lightweight medical induction data flow classification model meets the requirement of lightweight while meeting high reasoning accuracy, and can be deployed on a cloud service library or edge equipment.
The embodiments of the present invention have been described in detail with reference to the drawings, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge of those skilled in the art without departing from the gist of the present invention.
Claims (6)
1. A construction method of a medical induction data flow classification model is characterized by comprising the following steps:
the method for training and obtaining the medical induction data flow classification model based on the combination of CNN and LSTM by adopting an automatic supervision learning method comprises the following steps: inputting unmarked data into a sparse self-encoder of a medical induction data stream classification model for training so as to enable the input and the output of the model to be consistent; inputting the data with the labels into a trained sparse self-encoder, and outputting to obtain a corresponding classification result;
training a medical induction data flow classification model based on a deep separable convolution method to obtain a lightweight medical induction data flow classification model, which specifically comprises the following steps:
the method comprises the following steps of adopting a sliding window technology to segment a human body induction data stream in a time period collected by a sensor, wherein the human body induction data stream is segmented into n segments of multi-dimensional time sequence data: x is the number of1,...,xnWherein n belongs to Z, and n is more than or equal to 0; z is an integer set;
inputting the segmented n segments of multi-dimensional time sequence data into a medical induction data stream classification model for classification: extracting features of the multi-dimensional time series data within each sliding window, and predicting time series data within a variable time length from the output of the unit time and the time series data of the next unit time; classifying whether the predicted time sequence data is abnormal or not, obtaining and outputting a classification result, and obtaining a lightweight medical induction data stream classification model;
and storing the obtained lightweight medical induction data flow classification model in a cloud service library or deploying the lightweight medical induction data flow classification model on edge equipment.
2. The method for constructing the medical sensing data flow classification model according to claim 1, wherein the medical sensing data flow classification model in the method is composed of two LSTM layers, a full connection layer and a softmax output layer, and edges between the two connection layers have weights.
3. The method of claim 2, wherein the data transformation between two cells in the LSTM layer is input to the bottom hidden layer
Wherein ilowerIs an input to the bottom hidden layer; f () is an activation function;
is the weight connecting the edges of the upper unit and the lower unit; n is the output number of the upper layer; onIs the output of a cell; bl0werIs an offset value.
4. The method for constructing the medical sensing data flow classification model according to claim 2, wherein a deep separable convolution network including a plurality of convolution layers is arranged inside each cell of the LSTM layer; the convolutional layers include two types: 3 x 3Depth-wise convolutional layers and 1 x 1 convolutional layers, wherein the two types of convolutional layers are arranged at intervals, and a batch standardized BN layer and a ReLU activation layer are connected behind each convolutional layer.
5. The method for constructing the medical sensing data flow classification model according to claim 1, wherein the method classifies whether the predicted time-series data is abnormal or not, and specifically comprises the following steps:
due to n predicted time series data { x1,…,xnThere are m digital signals for each data sample
And, the ith predicted time-series data xi∈xn；
Let the true value of the n time-series data be
The error of the jth element in the ith time-series data is
Defining time series data x for multivariate Gaussian ith predictioniError vector e ofiThe distribution p of (A) is:
wherein
Is an n x 1 vector and is,
is an n × n matrix;
if p (e)i(ii) a Mu; xi is less, then called xiIs a normal time series data, otherwise, xiIs anomalous time series data, in which is an error threshold.
6. The method for constructing the medical sensing data flow classification model according to claim 5, characterized in that the method further comprises a conventional segmentation removal method for the acquired human sensing data flow.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010925639.6A CN112185543A (en) | 2020-09-04 | 2020-09-04 | Construction method of medical induction data flow classification model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010925639.6A CN112185543A (en) | 2020-09-04 | 2020-09-04 | Construction method of medical induction data flow classification model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112185543A true CN112185543A (en) | 2021-01-05 |
Family
ID=73925083
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010925639.6A Pending CN112185543A (en) | 2020-09-04 | 2020-09-04 | Construction method of medical induction data flow classification model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112185543A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113783716A (en) * | 2021-07-27 | 2021-12-10 | 国网冀北电力有限公司信息通信分公司 | Flow prediction method and device based on cloud edge collaborative framework |
Citations (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105530303A (en) * | 2015-12-15 | 2016-04-27 | 南京信息工程大学 | Linear network cache substitution method |
| CN106555788A (en) * | 2016-11-11 | 2017-04-05 | 河北工业大学 | Application of the deep learning based on Fuzzy Processing in hydraulic equipment fault diagnosis |
| CN107451552A (en) * | 2017-07-25 | 2017-12-08 | 北京联合大学 | A kind of gesture identification method based on 3D CNN and convolution LSTM |
| CN107729993A (en) * | 2017-10-30 | 2018-02-23 | 国家新闻出版广电总局广播科学研究院 | Utilize training sample and the 3D convolutional neural networks construction methods of compromise measurement |
| CN109977950A (en) * | 2019-03-22 | 2019-07-05 | 上海电力学院 | A kind of character recognition method based on mixing CNN-LSTM network |
| CN110119709A (en) * | 2019-05-11 | 2019-08-13 | 东南大学 | A kind of driving behavior recognition methods based on space-time characterisation |
| CN110322009A (en) * | 2019-07-19 | 2019-10-11 | 南京梅花软件系统股份有限公司 | Image prediction method based on the long Memory Neural Networks in short-term of multilayer convolution |
| CN110490863A (en) * | 2019-08-22 | 2019-11-22 | 北京红云智胜科技有限公司 | Whether there is or not the systems of total occlusion for detection coronary angiography based on deep learning |
| CN110490298A (en) * | 2019-07-25 | 2019-11-22 | 西安理工大学 | Lightweight depth convolutional neural networks model based on expansion convolution |
| CN110648527A (en) * | 2019-08-20 | 2020-01-03 | 浙江工业大学 | Traffic speed prediction method based on deep learning model |
| CN110688986A (en) * | 2019-10-16 | 2020-01-14 | 南京林业大学 | Attention branch guided 3D convolution behavior recognition network method |
| CN110801221A (en) * | 2019-12-09 | 2020-02-18 | 中山大学 | Sleep apnea fragment detection method and device based on unsupervised feature learning |
| CN110991219A (en) * | 2019-10-11 | 2020-04-10 | 东南大学 | Behavior identification method based on two-way 3D convolutional network |
| CN111382677A (en) * | 2020-02-25 | 2020-07-07 | 华南理工大学 | Human behavior identification method and system based on 3D attention residual error model |
-
2020
- 2020-09-04 CN CN202010925639.6A patent/CN112185543A/en active Pending
Patent Citations (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105530303A (en) * | 2015-12-15 | 2016-04-27 | 南京信息工程大学 | Linear network cache substitution method |
| CN106555788A (en) * | 2016-11-11 | 2017-04-05 | 河北工业大学 | Application of the deep learning based on Fuzzy Processing in hydraulic equipment fault diagnosis |
| CN107451552A (en) * | 2017-07-25 | 2017-12-08 | 北京联合大学 | A kind of gesture identification method based on 3D CNN and convolution LSTM |
| CN107729993A (en) * | 2017-10-30 | 2018-02-23 | 国家新闻出版广电总局广播科学研究院 | Utilize training sample and the 3D convolutional neural networks construction methods of compromise measurement |
| CN109977950A (en) * | 2019-03-22 | 2019-07-05 | 上海电力学院 | A kind of character recognition method based on mixing CNN-LSTM network |
| CN110119709A (en) * | 2019-05-11 | 2019-08-13 | 东南大学 | A kind of driving behavior recognition methods based on space-time characterisation |
| CN110322009A (en) * | 2019-07-19 | 2019-10-11 | 南京梅花软件系统股份有限公司 | Image prediction method based on the long Memory Neural Networks in short-term of multilayer convolution |
| CN110490298A (en) * | 2019-07-25 | 2019-11-22 | 西安理工大学 | Lightweight depth convolutional neural networks model based on expansion convolution |
| CN110648527A (en) * | 2019-08-20 | 2020-01-03 | 浙江工业大学 | Traffic speed prediction method based on deep learning model |
| CN110490863A (en) * | 2019-08-22 | 2019-11-22 | 北京红云智胜科技有限公司 | Whether there is or not the systems of total occlusion for detection coronary angiography based on deep learning |
| CN110991219A (en) * | 2019-10-11 | 2020-04-10 | 东南大学 | Behavior identification method based on two-way 3D convolutional network |
| CN110688986A (en) * | 2019-10-16 | 2020-01-14 | 南京林业大学 | Attention branch guided 3D convolution behavior recognition network method |
| CN110801221A (en) * | 2019-12-09 | 2020-02-18 | 中山大学 | Sleep apnea fragment detection method and device based on unsupervised feature learning |
| CN111382677A (en) * | 2020-02-25 | 2020-07-07 | 华南理工大学 | Human behavior identification method and system based on 3D attention residual error model |
Non-Patent Citations (3)
| Title |
|---|
| LE SUN 等: "A stacked LSTM for atrial fibrillation prediction based on multivariate ECGs", HEALTH INFORMATION SCIENCE AND SYSTEMS * |
| 张德正 等: "基于深度卷积长短时神经网络的视频帧预测", 计算机应用 * |
| 潘信亮 等: "基于卷积长短期记忆网络的NDVI预测方法研究", 地理信息世界 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113783716A (en) * | 2021-07-27 | 2021-12-10 | 国网冀北电力有限公司信息通信分公司 | Flow prediction method and device based on cloud edge collaborative framework |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| AU2020103905A4 (en) | Unsupervised cross-domain self-adaptive medical image segmentation method based on deep adversarial learning | |
| CN110598736B (en) | Power equipment infrared image fault positioning, identifying and predicting method | |
| Lamba et al. | GAN based image augmentation for increased CNN performance in Paddy leaf disease classification | |
| Aslan et al. | Multi-classification deep CNN model for diagnosing COVID-19 using iterative neighborhood component analysis and iterative ReliefF feature selection techniques with X-ray images | |
| CN112232407B (en) | Neural network model training method and device for pathological image samples | |
| Naik et al. | Weather classification and forecasting using back propagation feed-forward neural network | |
| CN107180155A (en) | A kind of disease forecasting method and system based on Manufacturing resource model | |
| CN111046961B (en) | Fault classification method based on bidirectional long-time and short-time memory unit and capsule network | |
| CN110837865A (en) | Domain adaptation method based on representation learning and transfer learning | |
| CN110458022A (en) | It is a kind of based on domain adapt to can autonomous learning object detection method | |
| CN114065307A (en) | Hardware Trojan horse detection method and system based on bipartite graph convolutional neural network | |
| CN114118138A (en) | Bearing composite fault diagnosis method based on multi-label field self-adaptive model | |
| Zou et al. | Exploring brain effective connectivity networks through spatiotemporal graph convolutional models | |
| CN116310647A (en) | Labor insurance object target detection method and system based on incremental learning | |
| CN112101418A (en) | Method, system, medium and equipment for identifying breast tumor type | |
| US11829442B2 (en) | Methods and systems for efficient batch active learning of a deep neural network | |
| CN112668633B (en) | Adaptive graph migration learning method based on fine granularity field | |
| CN112185543A (en) | Construction method of medical induction data flow classification model | |
| CN114295967A (en) | Analog circuit fault diagnosis method based on migration neural network | |
| Reda et al. | A transfer learning-enabled optimized extreme deep learning paradigm for diagnosis of covid-19 | |
| CN115834161A (en) | Power grid false data injection attack detection method of artificial intelligence four-layer architecture | |
| Sharma et al. | Robust and Uncertainty-Aware VAE (RU-VAE) for One-Class Classification | |
| CN115047313A (en) | ZPW-2000R track circuit fault diagnosis method and device based on OC-SVM and DNN | |
| Herok et al. | Cotton leaf disease identification using transfer learning | |
| CN115329906B (en) | Fault diagnosis method and device based on graph regularization network |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210105 |