CN112233785B - Intelligent identification method for Parkinson's disease - Google Patents

Intelligent identification method for Parkinson's disease Download PDF

Info

Publication number
CN112233785B
CN112233785B CN202010649619.0A CN202010649619A CN112233785B CN 112233785 B CN112233785 B CN 112233785B CN 202010649619 A CN202010649619 A CN 202010649619A CN 112233785 B CN112233785 B CN 112233785B
Authority
CN
China
Prior art keywords
parkinson
neural network
layer
network model
output
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.)
Active
Application number
CN202010649619.0A
Other languages
Chinese (zh)
Other versions
CN112233785A (en
Inventor
张智军
孙健声
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.)
South China University of Technology SCUT
Original Assignee
South China University of Technology SCUT
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 South China University of Technology SCUT filed Critical South China University of Technology SCUT
Priority to CN202010649619.0A priority Critical patent/CN112233785B/en
Publication of CN112233785A publication Critical patent/CN112233785A/en
Application granted granted Critical
Publication of CN112233785B publication Critical patent/CN112233785B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

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
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/40—Detecting, measuring or recording for evaluating the nervous system
    • A61B5/4076—Diagnosing or monitoring particular conditions of the nervous system
    • A61B5/4082—Diagnosing or monitoring movement diseases, e.g. Parkinson, Huntington or Tourette
    • 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
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
    • G10L25/27—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the analysis technique
    • G10L25/30—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the analysis technique using neural networks
    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
    • G10L25/48—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use
    • G10L25/51—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use for comparison or discrimination
    • G10L25/66—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use for comparison or discrimination for extracting parameters related to health condition

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biomedical Technology (AREA)
  • Public Health (AREA)
  • Computational Linguistics (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Molecular Biology (AREA)
  • Neurology (AREA)
  • Biophysics (AREA)
  • Artificial Intelligence (AREA)
  • Medical Informatics (AREA)
  • Acoustics & Sound (AREA)
  • General Engineering & Computer Science (AREA)
  • Neurosurgery (AREA)
  • Mathematical Physics (AREA)
  • Pathology (AREA)
  • General Physics & Mathematics (AREA)
  • Epidemiology (AREA)
  • Computing Systems (AREA)
  • Signal Processing (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Human Computer Interaction (AREA)
  • Software Systems (AREA)
  • Multimedia (AREA)
  • Developmental Disabilities (AREA)
  • Primary Health Care (AREA)
  • Databases & Information Systems (AREA)
  • Physiology (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Surgery (AREA)
  • Animal Behavior & Ethology (AREA)
  • Veterinary Medicine (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses an intelligent identification method of Parkinson's disease, which comprises the following steps: s1, acquiring voice data of the person to be identified; s2, preprocessing the sound data; s3, inputting the processed sound data into the trained dynamic convergence differential neural network model for recognition; and S4, outputting a corresponding result by the dynamic convergence differential neural network model. The dynamic convergence differential neural network model updates the parameter matrix by using a neurodynamic formula, and the whole neural network only comprises a hidden layer.

Description

Intelligent identification method for Parkinson's disease
Technical Field
The invention relates to the technical field of neural network model identification, in particular to an intelligent identification method for Parkinson's disease.
Background
The early identification of parkinson is still a difficult problem at present, the traditional method for diagnosing according to clinical symptoms has long diagnosis time and low accuracy, the method using tracer is too high in cost and cannot be popularized, and the number of patients with parkinson is large, so that a low-cost, high-accuracy and early diagnosis method is urgently needed to be provided. The slight change in sound is an early expression of the Parkinson's disease, and research has proved that the change can be used for the identification of the Parkinson's disease, but the direct analysis has the defects of large workload, long identification time and possibility of error caused by human reasons, so that the sound characteristics for the identification of the Parkinson's disease are not practically applied at present.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides an intelligent identification method for the Parkinson's disease, which has high identification speed and high accuracy.
The purpose of the invention is realized by the following technical scheme:
an intelligent identification method for Parkinson's disease comprises the following steps:
s1, acquiring voice data of the person to be identified;
s2, preprocessing the sound data;
s3, inputting the processed sound data into the trained dynamic convergence differential neural network model for recognition;
and S4, outputting a corresponding result by the dynamic convergence differential neural network model.
Preferably, the training of the dynamically converged differential neural network model in step S3 includes:
s31, acquiring a Parkinson data set, and dividing the Parkinson data set into a training set and a testing set; the specific operation is that 70% of the Parkinson data set is used as a training set, 30% of the Parkinson data set is used as a test set, and the positive class and the negative class are uniformly distributed.
S32, preprocessing the divided Parkinson data set;
and S33, inputting the processed training set into a dynamic convergence differential neural network to train the network, and continuously adjusting parameters to obtain a trained dynamic convergence differential neural network model.
Preferably, step S31 includes: dividing the Parkinson data set into a training set and a testing set according to the positive and negative uniform distribution; the 44 acoustic attributes of the Parkinson data set are divided into eight groups, namely fundamental frequency disturbance, amplitude disturbance, harmonic noise ratio, 0-12 order spectral measurement and derivative thereof based on Mel cepstrum coefficients, recurrence period density entropy, detrending fluctuation analysis, fundamental frequency period entropy and glottal noise excitation ratio.
Preferably, the preprocessing the partitioned parkinson data set comprises: carrying out zero-mean standardization treatment (Z-score method) on the Parkinson data set, wherein the mean value is 0, the variance is 1, and carrying out feature enhancement treatment on the standardized Parkinson data set by utilizing statistical features in the statistical pool technology; the characteristic-enhanced Parkinson data set extracts key characteristics with the contribution rate of 80% by using a principal component analysis technology, wherein the statistical characteristics comprise average value, standard deviation, energy, entropy, autocorrelation, absolute average value, kurtosis, skewness, median, minimum value, maximum value, variation coefficient, root mean square, shape factor, peak value factor, margin factor, pulse factor, difference between the maximum value and the minimum value and difference between the maximum value and the average value.
Preferably, the processed training set is input into a dynamic convergence differential neural network to train the network, and the continuously adjusting parameters includes:
s331, obtaining a hidden dynamics formula of a parameter to be updated of the neural network by using derivative information of the error according to the neural dynamics formula;
Figure BDA0002574420620000031
wherein λ and
Figure BDA0002574420620000032
the network learning rate and the error activation function are respectively used, f' is the derivative of the activation function of the output layer, o (k) and y are labels of the actual output and the training set of the output layer, w (k) and b (k) respectively represent the weight and the bias from the hidden layer to the output layer, h (k) ═ g (v (k) x + a (k)) is output from the hidden layer, v (k) and a (k) are the weight and the bias from the input layer to the hidden layer, and x is training data (the training set);
s332, in the hidden dynamics formula, fixing the hidden layer output h (k) to ensure that
Figure BDA0002574420620000033
Then the updated formula of weight w (k) from hidden layer to output layer and bias b (k) is obtained
Figure BDA0002574420620000034
Wherein
Figure BDA0002574420620000035
H(k)＝[h(k)；1]，H+(k) Is the generalized inverse of H (k);
s333, in the implicit dynamics formula, fixing the parameters w (k) and b (k) of the implicit layer to the output layer so as to ensure that
Figure BDA0002574420620000036
Then obtain the updated formula of hidden layer output h (k) in feedback propagation
Figure BDA0002574420620000037
S334, according to the updated formula of the hidden layer output h (k), combining the hidden layer output h (k) g (v (k) x + a (k)) in the feed forward propagation, the derivative thereof is
Figure BDA0002574420620000038
Wherein X ═ X; 1]、W1(k)＝[v(k),a(k)]And g' is the derivative of the hidden layer activation function, an update formula of the weight v (k) and the bias a (k) of the input layer to the hidden layer is obtained.
Preferably, step S4 includes: and according to the classification threshold value of the two classification neural networks in advance, the dynamic convergence differential neural network model outputs a corresponding result. Examples are:
the classification threshold of the pre-binary neural network is 0. If the output of the dynamic convergent differential neural network model is 1 in step S4, it is determined that the person to be identified has parkinson' S disease. If the output of the dynamic convergence differential neural network model in the step S4 is-1, it is determined that the person to be identified is normal and does not suffer from parkinson' S disease.
Preferably, step S33 is followed by: inputting the processed test set into a dynamic convergence differential neural network model to train the network, continuously adjusting the parameter test set, outputting an identification result by the dynamic convergence differential neural network model, comparing the identification result with a known Parkinson's disease conclusion of the test set, and judging the accuracy of the dynamic convergence differential neural network model.
Preferably, the parkinson data set is from Naranjo in the UCI database. The parkinson data set contains the acoustic signals of parkinson patients and normal persons.
Compared with the prior art, the invention has the following advantages:
the method inputs processed sound data into a trained dynamic convergence differential neural network model for identification; the dynamic convergence differential neural network model updates the parameter matrix by using a neurodynamic formula, and the whole neural network only comprises a hidden layer. Aiming at a Parkinson data set given by Naranjo in a UCI database, the accuracy rate of 97.22% is realized, and the highest accuracy rate of the existing Parkinson identification based on machine learning is 91.25%, so that the method is greatly improved compared with the existing method.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this application, illustrate embodiments of the invention and, together with the description, serve to explain the invention and not to limit the invention. In the drawings:
fig. 1 is a schematic flow chart of the intelligent parkinson's disease recognition method of the present invention.
Fig. 2 is a network structure diagram of the intelligent parkinson's disease recognition method of the present invention;
FIG. 3 is a detailed flow chart of the present invention for training a dynamic convergent differentiation neural network model;
fig. 4 is a flow chart of the statistical pool part of the intelligent parkinson's disease recognition method of the present invention.
Detailed Description
The invention is further illustrated by the following figures and examples.
Referring to fig. 1, an intelligent parkinson's disease identification method includes:
s1, acquiring voice data of the person to be identified;
s2, preprocessing the sound data;
s3, inputting the processed sound data into the trained dynamic convergence differential neural network model for recognition;
and S4, outputting a corresponding result by the dynamic convergence differential neural network model.
The derivation process of the dynamic convergence differential neural network model is as follows:
in the first step, the input to the neural network is assumed to be a sample vector x (training set). For the input layer and the hidden layer, the weight matrix is defined as v (k) and the offset is a (k). When the output of the hidden layer is
h(k)＝g(v(k)x+a(k)) (1)
Assuming that the dimension of the input sample vector is m and the number of hidden layer nodes is n, v (k) is an n × m matrix, and a (k) is an n × 1 vector. g (-) is the softsign activation function. The specific formula is as follows:
Figure BDA0002574420620000051
here, softsign is selected as an activation function, and the output results of the dynamic convergence differential neural network model are preset to be-1 and 1.
For the hidden layer and the output layer, defining the weight matrix as w (k), and the bias as b (k), then the output of the output layer at this time is
o(k)＝f(w(k)h(k)+b(k)) (3)
Assuming that the number of output nodes is p, w (k) is a matrix of p × n, b (k) is a vector of p × 1, and f (·) is an activation function, where the softsign function is still used.
Then, the network updating step is given, and an error function is defined in the first step
ε(k)＝f(w(k)h(k)+b(k))-y (4)
Where k represents the number of times the weight is iteratively updated and y represents the class label of the input data. Unlike the scalar error of the conventional gradient descent method, here the error function is of the vector type. When a plurality of data are input simultaneously, the error is iteratively updated to each sample data, so that the error is converged to the global minimum. The network adjusts the weight and the bias through a neurodynamics formula, and the specific formula is as follows:
Figure BDA0002574420620000061
the lambda is larger than 0 and is used as a learning rate, and the value of the lambda is continuously adjusted through experiments during network training so that convergence is more stable.
Figure BDA0002574420620000062
Should be a monotonically increasing odd function. In this patent, a power-sigmoid function is used as the activation function
Figure BDA0002574420620000063
The formula is as follows:
Figure BDA0002574420620000064
here, n.gtoreq.2 and r.gtoreq.2 are required to be integers. Here, n ═ 4 and r ═ 2 are given.
Second, referring to fig. 2, the formula of the error function (derivative information of the deviation function) is substituted into the formula of the neurodynamics (5), and the hidden dynamics equation is obtained as follows:
Figure BDA0002574420620000065
where f' is the derivative of the output layer activation function,
Figure BDA0002574420620000066
respectively, represent the corresponding derivatives.
And thirdly, updating the weights v (k), w (k) and the offsets a (k) and b (k) in a step-by-step iterative manner. To facilitate the formulation derivation, the following four matrices are defined: x ═ X; 1]、W1(k)＝[v(k),a(k)]、W2(k)＝[w(k),b(k)]、H(k)＝[h(k)；1]. Subsequently, another equivalent of equation (7) can be obtained:
Figure BDA0002574420620000071
a specific update procedure will be given below.
First, H (k) is fixed, and at this time, it can be considered that
Figure BDA0002574420620000072
Then, the formula (8) can obtain
Figure BDA0002574420620000073
Wherein H+(k) Is the generalized inverse of H (k). Thereby obtaining a matrix W2The update formula of (2):
W2(k+1)＝W2(k)+ΔW2 (10)
wherein
Figure BDA0002574420620000074
The matrix W is then aligned in the same manner1And (6) updating. Fixed W2(k) At this time have
Figure BDA0002574420620000075
I.e. is equivalent to
Figure BDA0002574420620000076
And
Figure BDA0002574420620000077
from equation (7), the following equation
Figure BDA0002574420620000078
And can be obtained from the formula
Figure BDA0002574420620000079
Where g' is the derivative of the hidden layer activation function. Can be obtained by combining the above two types
Figure BDA00025744206200000710
Wherein X+Is the generalized inverse of the input matrix. From this, a matrix W can be obtained1Is updated to
W1(k+1)＝W1(k)+ΔW1 (14)
In the above formula
Figure BDA00025744206200000711
Continuously repeating the above operations for iteration until the error is stably converged, and obtaining the optimal weight and bias matrix W1And W2And finally finishing the training process.
The above is the mathematical derivation process of model building. From fig. 3, it is known that the dynamically converged differential neural network model needs to preprocess data to work normally, and the preprocessing process will be given below.
The pretreatment is divided into three steps, wherein the first step is standardization, namely, original data are converted into data under the same scale, the average value of the data after the model is standardized is 0, and the standard deviation is 1.
The second step is the statistical pool technique. Since the data set that was tested at the time of this patent contained duplicate samples from one person, statistical pooling techniques were used to increase the differences between the data for better classification, as shown in fig. 4. The original data is divided into four, two and one blocks with non-overlapping attributes, and each block is calculated by using the formula in table 1 to obtain blocks containing 19 attributes. Together with the original dataset, the new dataset contains a total of 177 attributes.
Watch 1
Figure BDA0002574420620000081
Figure BDA0002574420620000091
The third step is Principal Component Analysis (PCA), which requires a reduction in the number of attributes to allow our model to converge to the best, since there are too many attributes added. The new data set contains only 9 attributes after PCA is used with 80% contribution.
So far, the establishment of the neural network model and the data preprocessing are all introduced.
The process of preprocessing the sound data in step S2 is the same as the preprocessing sub-process described above. The process of inputting the processed sound data into the trained dynamic convergent-differential neural network model for recognition in step S3 is the same as the second step and the third step in the derivation process of the model.
The above-mentioned embodiments are preferred embodiments of the present invention, and the present invention is not limited thereto, and any other modifications or equivalent substitutions that do not depart from the technical spirit of the present invention are included in the scope of the present invention.

Claims (6)

1. An intelligent identification method for Parkinson's disease is characterized by comprising the following steps:
s1, acquiring voice data of the person to be identified;
s2, preprocessing the sound data;
s3, inputting the processed sound data into the trained dynamic convergence differential neural network model for recognition; the training of the dynamic convergent-differential neural network model comprises the following steps:
s31, acquiring a Parkinson data set, and dividing the Parkinson data set into a training set and a testing set;
s32, preprocessing the divided Parkinson data set;
s33, inputting the processed training set into a dynamic convergence differential neural network to train the network, and continuously adjusting parameters to obtain a trained dynamic convergence differential neural network model; the method comprises the following steps:
s331, obtaining a hidden dynamics formula of a parameter to be updated of the neural network by using derivative information of the error according to the neural dynamics formula;
Figure FDA0003468241520000011
wherein λ and
Figure FDA0003468241520000012
respectively, a network learning rate and an error activation function, f' is a derivative of an output layer activation function, o (k) and y are labels of an output layer actual output and a training set, w (k) and b (k) respectively represent weights and biases of an implicit layer to an output layer, and an implicit layer output h (k) ═ g (v (k) x + a (k)) is shown, wherein v (k) and a (k) are the weights and biases of an input layer to the implicit layer, and x is the training set;
s332, in the hidden dynamics formula, fixing the hidden layer output h (k) to ensure that
Figure FDA0003468241520000013
Then the updated formula of weight w (k) from hidden layer to output layer and bias b (k) is obtained
Figure FDA0003468241520000014
Wherein
Figure FDA0003468241520000015
H(k)＝[h(k)；1]，H+(k) Is the generalized inverse of H (k);
s333, in the implicit dynamics formula, fixing the parameters w (k) and b (k) of the implicit layer to the output layer so as to ensure that
Figure FDA0003468241520000021
Then obtain the updated formula of hidden layer output h (k) in feedback propagation
Figure FDA0003468241520000022
S334, according to the updated formula of the hidden layer output h (k), combining the hidden layer output h (k) g (v (k) x + a (k)) in the feed forward propagation, the derivative thereof is
Figure FDA0003468241520000023
Wherein X ═ X; 1]、W1(k)＝[v(k),a(k)]G' is the hidden layer laserObtaining the updated formula of the weight v (k) from the input layer to the hidden layer and the bias a (k) by using the derivative of the function;
and S4, outputting a corresponding result by the dynamic convergence differential neural network model.
2. The intelligent parkinson' S disease recognition method of claim 1, wherein step S31 includes:
dividing the Parkinson data set into a training set and a testing set according to the positive and negative uniform distribution; the 44 acoustic attributes of the Parkinson data set are divided into eight groups, namely fundamental frequency disturbance, amplitude disturbance, harmonic noise ratio, 0-12 order spectral measurement and derivative thereof based on Mel cepstrum coefficients, recurrence period density entropy, detrending fluctuation analysis, fundamental frequency period entropy and glottal noise excitation ratio.
3. The intelligent parkinson's disease recognition method of claim 1, wherein the preprocessing of the partitioned parkinson's data set comprises:
the method comprises the steps of carrying out zero-mean standardization on a Parkinson data set, carrying out feature enhancement on the standardized Parkinson data set by utilizing statistical features in a statistical pool technology, and extracting key features from the feature-enhanced Parkinson data set at a preset contribution rate by using a principal component analysis technology, wherein the statistical features comprise a mean value, a standard deviation, energy, entropy, autocorrelation, an absolute mean value, kurtosis, skewness, a median value, a minimum value, a maximum value, a variation coefficient, a root-mean-square, a shape factor, a peak value factor, a margin factor, a pulse factor, a difference between the maximum value and the minimum value, and a difference between the maximum value and the mean value.
4. The intelligent parkinson' S disease recognition method of claim 1, wherein step S4 includes:
and according to the preset classification threshold value of the two classification neural networks, the dynamic convergence differential neural network model outputs a corresponding result.
5. The intelligent parkinsonism recognition method according to claim 1, further comprising, after step S33: inputting the processed test set into a dynamic convergence differential neural network model to test the network, comparing the recognition result output by the dynamic convergence differential neural network model with the known Parkinson's disease result of the test set, and judging the accuracy of the dynamic convergence differential neural network model.
6. The intelligent parkinson's disease recognition method of claim 1, wherein the parkinson's data set is from Naranjo in the UCI database.
CN202010649619.0A 2020-07-08 2020-07-08 Intelligent identification method for Parkinson's disease Active CN112233785B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010649619.0A CN112233785B (en) 2020-07-08 2020-07-08 Intelligent identification method for Parkinson's disease

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010649619.0A CN112233785B (en) 2020-07-08 2020-07-08 Intelligent identification method for Parkinson's disease

Publications (2)

Publication Number Publication Date
CN112233785A CN112233785A (en) 2021-01-15
CN112233785B true CN112233785B (en) 2022-04-22

Family

ID=74111731

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010649619.0A Active CN112233785B (en) 2020-07-08 2020-07-08 Intelligent identification method for Parkinson's disease

Country Status (1)

Country Link
CN (1) CN112233785B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113887374B (en) * 2021-09-27 2024-04-16 华南理工大学 Brain control water drinking system based on dynamic convergence differential neural network
CN114788687B (en) * 2022-06-23 2022-09-27 中国科学院自动化研究所 Quantitative assessment method and device for Parkinson myotonia symptoms

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109285551A (en) * 2018-09-18 2019-01-29 上海海事大学 Disturbances in patients with Parkinson disease method for recognizing sound-groove based on WMFCC and DNN
CN110070888A (en) * 2019-05-07 2019-07-30 颐保医疗科技(上海)有限公司 A kind of Parkinson's audio recognition method based on convolutional neural networks
CN110703603A (en) * 2019-10-28 2020-01-17 华南理工大学 Control method of multi-layer recursive convergence neural network controller of unmanned aerial vehicle
CN111037550A (en) * 2019-12-03 2020-04-21 华南理工大学 Solution method for motion control of redundant manipulator
CN111210846A (en) * 2020-01-07 2020-05-29 重庆大学 Parkinson voice recognition system based on integrated manifold dimensionality reduction
CN111210912A (en) * 2020-01-14 2020-05-29 上海恩睦信息科技有限公司 Parkinson prediction method and device
CN111354338A (en) * 2020-02-26 2020-06-30 重庆大学 Parkinson speech recognition system based on PSO convolution kernel optimization sparse transfer learning

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1417742A (en) * 2002-12-16 2003-05-14 武汉大学 Dynamic cognitive nerve network and its implementation
CN107368091B (en) * 2017-08-02 2019-08-20 华南理工大学 A kind of stabilized flight control method of more rotor unmanned aircrafts based on finite time neurodynamics
JP2020537579A (en) * 2017-10-17 2020-12-24 ラオ、サティシュ Machine learning-based system for identifying and monitoring neuropathy
CA3124755A1 (en) * 2018-12-26 2020-07-02 Analytics For Life Inc. Methods and systems to configure and use neural networks in characterizing physiological systems

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109285551A (en) * 2018-09-18 2019-01-29 上海海事大学 Disturbances in patients with Parkinson disease method for recognizing sound-groove based on WMFCC and DNN
CN110070888A (en) * 2019-05-07 2019-07-30 颐保医疗科技(上海)有限公司 A kind of Parkinson's audio recognition method based on convolutional neural networks
CN110703603A (en) * 2019-10-28 2020-01-17 华南理工大学 Control method of multi-layer recursive convergence neural network controller of unmanned aerial vehicle
CN111037550A (en) * 2019-12-03 2020-04-21 华南理工大学 Solution method for motion control of redundant manipulator
CN111210846A (en) * 2020-01-07 2020-05-29 重庆大学 Parkinson voice recognition system based on integrated manifold dimensionality reduction
CN111210912A (en) * 2020-01-14 2020-05-29 上海恩睦信息科技有限公司 Parkinson prediction method and device
CN111354338A (en) * 2020-02-26 2020-06-30 重庆大学 Parkinson speech recognition system based on PSO convolution kernel optimization sparse transfer learning

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
基于语音声谱图的帕金森症检测研究与***开发;焦嘉烽;《中国优秀博硕士学位论文全文数据库(硕士)工程科技Ⅱ辑》;20190215(第2期);第C030-86页 *

Also Published As

Publication number Publication date
CN112233785A (en) 2021-01-15

Similar Documents

Publication Publication Date Title
Ding et al. Group Latent Embedding for Vector Quantized Variational Autoencoder in Non-Parallel Voice Conversion.
CN110390955B (en) Cross-database speech emotion recognition method based on depth domain adaptive convolutional neural network
CN112233785B (en) Intelligent identification method for Parkinson's disease
CN112465040B (en) Software defect prediction method based on class unbalance learning algorithm
CN103366733A (en) Text to speech system
Ke et al. Speech emotion recognition based on SVM and ANN
CN109671423B (en) Non-parallel text-to-speech conversion method under limited training data
CN110047501B (en) Many-to-many voice conversion method based on beta-VAE
Ilyas Pseudo-colored rate map representation for speech emotion recognition
JPH08227408A (en) Neural network
CN103077408A (en) Method for converting seabed sonar image into acoustic substrate classification based on wavelet neutral network
Lai et al. Phone-aware LSTM-RNN for voice conversion
Zhiyan et al. Speech emotion recognition based on deep learning and kernel nonlinear PSVM
Yulita et al. Fuzzy Hidden Markov Models for Indonesian Speech Classification.
CN117409819A (en) Human voice detection and analysis method based on artificial intelligence
Yang et al. Speech emotion analysis of netizens based on bidirectional lstm and pgcdbn
Aishwarya et al. Kannada speech recognition system for Aphasic people
Gowda et al. Affective computing using speech processing for call centre applications
Li et al. Spectro-Temporal Modelling with Time-Frequency LSTM and Structured Output Layer for Voice Conversion.
Anindya et al. Development of Indonesian speech recognition with deep neural network for robotic command
Hashimoto et al. Parallel-Data-Free Many-to-Many Voice Conversion Based on DNN Integrated with Eigenspace Using a Non-Parallel Speech Corpus.
Wen et al. Learning auxiliary categorical information for speech synthesis based on deep and recurrent neural networks
Poddar et al. Data capturing and modeling by speech recognition: roles demonstrated by artificial intelligence, A survey
CN114357414B (en) Emotion speaker authentication method based on cross-gradient training
Chelliah et al. Robust Hearing-Impaired Speaker Recognition from Speech using Deep Learning Networks in Native

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
GR01 Patent grant
GR01 Patent grant