Early warning system, early warning device and early warning method based on neural network and early warning model training method

Abstract

The application relates to a cognitive impairment early warning system, an early warning device and method, and an early warning model training method. The early warning method comprises: acquiring information related to a user behavior pattern; using a convolutional neural network and a Gaussian mixture model to extract features from the acquired information; using a recurrent neural network to sequence model the extracted features; classifying the output of the sequence modeling; and early warning of cognitive impairment based on the classification result.

Description

TECHNICAL FIELD

[0001] The present application relates to a neural network-based early warning system, early warning device and early warning method, and early warning model training method. More specifically, it relates to a light neural network-based early warning system, early warning device and early warning method for early cognitive impairment, and early warning model training method. BACKGROUND

[0002] Alzheimer's disease, also known as cognitive impairment, is a progressive neurodegenerative disease with insidious onset, and its clinical manifestations include memory loss and language deterioration, which seriously affects the normal work and life of patients. The older the person is, the greater the likelihood of developing Alzheimer's disease. Generally, the elderly population is divided into three categories by medicine, namely, normal (Normal Control, NC), early cognitive impairment (Mild Cognitive Impairment, MCI), and Alzheimer's disease. Therefore, it is of great significance to timely distinguish between NC and MCI populations and to warn early MCI. Although many related studies have been proposed to predict Alzheimer's disease, there is still no effective method and device that can provide detection and early warning in daily life.

[0003] A neural network is a mathematical or computational model that simulates the structure and function of a biological neural network. In the medical field, neural networks have been widely used in clinical diagnosis, medical image analysis and interpretation, signal analysis and judgment, etc. Among them, the neural network for wearable medical devices is facing two challenges to be solved: first, the traditional neural network has many parameters, and the energy consumption is too high when deployed in wearable devices; second, the intermediate variables of the traditional neural network are mostly high-dimensional space variables, which are difficult for general users to understand. Therefore, a lightweight and highly interpretable neural network model is needed.

[0004] The cognitive impairment early warning system, early warning device and early warning method, and early warning model training method proposed in the present application can perform early cognitive impairment warning based on a lightweight neural network. At the same time, by limiting the number of model parameters in the proposed early warning model, the device power consumption is effectively reduced, and by using a hybrid model for feature extraction, the device's interpretability is improved, so that the device (e.g., a wearable device) that implements the early warning method proposed in the present application is more suitable for use in daily life. SUMMARY

[0005] In order to solve or partially solve at least one of the above problems existing in the prior art, the present application provides a neural network-based early warning method, model and device to achieve low power consumption and high interpretability.

[0006] According to an aspect of the present application, there is provided a cognitive impairment early warning system, comprising: a feature extraction module, a sequence modeling module and a classification module, wherein the feature extraction module comprises a convolutional neural network and a Gaussian mixture model, for feature extraction on acquired data; the sequence modeling module is used for sequence modeling on the extracted features; and the classification module is used for classifying the output of the sequence modeling, for cognitive impairment early warning.

[0007] In an embodiment of the present application, the convolutional neural network is a single-layer convolutional neural network.

[0008] In an embodiment of the present application, the Gaussian mixture model comprises five Gaussian models respectively representing five behavior states.

[0009] In an embodiment of the present application, the five behavior states are sleep, sedentary, light activity, walking and exercise.

[0010] In an embodiment of the present application, the recurrent neural network comprises a long short-term memory model or a gated recurrent unit.

[0011] In an embodiment of the present application, the classification module comprises a logistic regression classifier.

[0012] According to another aspect of the present application, there is provided a method for training a cognitive impairment early warning model, comprising: acquiring information related to user behavior patterns; using a convolutional neural network and a Gaussian mixture model to extract features from the acquired information; using a recurrent neural network to sequence model the extracted features to generate an early warning initial model; classifying the output of the early warning initial model; inputting the classified results and a predetermined threshold into a loss function to train the early warning initial model until the loss function meets a convergence condition, to obtain an early warning model.

[0013] In an embodiment of the present application, the convolutional neural network used is a single-layer convolutional neural network.

[0014] In an embodiment of the present application, the Gaussian mixture model used uses five Gaussian models to represent five behavior states respectively.

[0015] In an embodiment of the present application, the five behavior states are sleep, sedentary, light activity, walking and exercise.

[0016] In an embodiment of the present application, after the step of using a convolutional neural network and a Gaussian mixture model to extract features from the acquired information, the method further comprises: inputting the features extracted by the convolutional neural network used and the Gaussian mixture model used into the recurrent neural network in parallel.

[0017] In an embodiment of the present application, the sequence modeling of the extracted features using the recurrent neural network comprises: sequence modeling of the extracted features using a gated recurrent unit; or sequence modeling of the extracted features using a long short-term memory model.

[0018] In an embodiment of the present application, the classification of the output of the early warning initial model comprises: binary classification of the output of the early warning initial model using a logistic regression classifier.

[0019] In an embodiment of the present application, the information is a multivariate time series.

[0020] In an embodiment of the present application, the multivariate time series comprises three-direction displacement information and illumination information.

[0021] According to another aspect of the present application, a cognitive impairment early warning method is provided, which comprises: determining target information; and determining an early warning result corresponding to the target information based on the target information and an early warning model obtained through the training method.

[0022] In an embodiment of the present application, the step of determining a label corresponding to the target information comprises: feature extraction of the target information; inputting the extracted features into the trained early warning model, and inputting an output result of the early warning model into a classifier; and classifying the output result of the early warning model in the classifier to obtain a label corresponding to the output result.

[0023] According to another aspect of the present application, a cognitive impairment early warning device is provided, which comprises: a memory storing computer executable instructions; and a processor configured to execute the computer executable instructions stored in the memory to implement the method according to any one of the training methods.

[0024] According to another aspect of the present application, a computer storage medium comprising computer executable instructions is provided, which, when executed by a processor, causes the implementation of the method according to any one of the training methods.

[0025] Compared with a DTW (dynamic time warping) model, a BOSS model, an SMTS model, etc., the early warning model in the present application has more excellent performance in AUC. The early warning model, method and device provided in the present application can guarantee the light weight of the neural network to reduce power consumption, and have high interpretability. BRIEF DESCRIPTION OF DRAWINGS

[0026] In order to make the technical solutions in the embodiments of the present application or the prior art clearer, the accompanying drawings needed in the description of the embodiments or the prior art will be briefly introduced. Obviously, the accompanying drawings in the following description only need to be some embodiments of the present application, and for those skilled in the art, other related drawings can be obtained without creative work on the basis of these drawings.

[0027] Figure 1 A structural block diagram of a neural network-based early warning system according to an embodiment of the present application is shown.

[0028] Figure 2 A flowchart of a training method of a neural network-based early warning model according to an embodiment of the present application is shown.

[0029] Figure 3 A distribution diagram of K parameters of a GMM model based on Bayesian information criterion according to an embodiment of the present application is shown.

[0030] Figure 4A A flowchart of an early warning method using a neural network-based early warning model according to an embodiment of the present application is shown.

[0031] Figure 4B A flowchart of a sub-step in an early warning method using a neural network-based early warning model according to an embodiment of the present application is shown.

[0032] Figure 5 An exemplary system architecture diagram to which an embodiment of the present application can be applied is shown.

[0033] Figure 6 A block diagram of an electronic device for implementing an early warning method according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0034] Exemplary embodiments of the present application will be described below with reference to the accompanying drawings, which include various details of the embodiments of the present application to assist in understanding, and should be considered as merely exemplary. Thus, those skilled in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the present application. Also, in order to make the description clear and concise, the description of well-known functions and structures will be omitted in the following description.

[0035] It should be noted that the embodiments and features in the embodiments of the present application can be combined with each other without conflict. The present application will be described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.

[0036] Figure 1A structural block diagram of a neural network based early warning system according to an embodiment of the present application is shown. The early warning system 100 is for a CGTC system C NN and G MM T imeseries C lassification). And, the early warning system 100 comprises a feature extraction module 110, a sequence modeling module 120 and a classification module 130.

[0037] The feature extraction module 110 comprises a convolutional neural network (CNN) 111 and a Gaussian mixture model (GMM) 112, which are used for feature extraction of the acquired data. The convolutional neural network is a kind of feedforward neural network containing convolutional computation, and the Gaussian mixture model can be regarded as a model mixed by multiple single Gaussian models.

[0038] In the feature extraction module 110, a single layer CNN is used as a supervised feature extraction operator to extract features from the acquired information related to user behavior patterns. And a GMM is used as an unsupervised feature extraction operator to extract features from the acquired information related to user behavior patterns. The GMM can establish a mapping between visible time series and hidden behavior states (such as sleep, walking, exercise, etc.), so that the model has higher interpretability.

[0039] The sequence modeling module 120 is used for sequence modeling of the extracted features. The sequence modeling module 120 comprises a recurrent neural network, which can be a long short-term memory model (LSTM) or a gated recurrent unit (GRU), but is not limited thereto. According to an embodiment of the present application, since the early warning system 100 is applied to a wearable device, the parameters of the model cannot be too many, therefore, the sequence modeling module 120 used can comprise a GRU sequence model, because the GRU model has fewer parameters than the LSTM, and the modeling effect of the GRU is similar to that of the LSTM and easier to calculate under the same conditions.

[0040] The classification module 130 is used for classifying the output of the sequence modeling, so as to perform early warning based on the classification result. According to an embodiment of the present application, the classification module 130 comprises a logistic regression classifier and is composed of full connection and a loss function, which is used for binary classification of the information input into the classification module 130 and obtaining the corresponding label. The loss function can be a cross-entropy (CE) loss function, but is not limited thereto. According to an embodiment of the present application, the labels of the binary classification are NC and MCI respectively, but are not limited thereto. According to another embodiment of the present application, the classification module 130 can be other classifiers, such as a Bayesian classifier, a nearest neighbor classifier, etc., for performing multi-classification, but is not limited thereto.

[0041] Figure 2 A flow chart of a method for training a neural network-based early warning model is shown according to an embodiment of the present application. As shown, the training method 200 comprises: Figure 2

[0042] In step S210, information related to user behavior patterns is obtained.

[0043] In step S220, the obtained information is feature extracted using a convolutional neural network and a Gaussian mixture model.

[0044] In step S230, the extracted features are sequence modeled using a recurrent neural network to obtain an early warning initial model.

[0045] In step S240, the output of the early warning initial model is classified.

[0046] In step S250, the classified result is input into a loss function with a predetermined threshold to train the early warning initial model until the loss function meets a convergence condition, and then an early warning model is obtained.

[0047] The steps S210 to S250 will be described in detail respectively.

[0048] In step S210, first, information related to user behavior patterns can be obtained by a sensor at a certain frequency. Specifically, the sensor can be a displacement sensor, a light-sensitive sensor, an acceleration sensor, etc., but is not limited thereto. The user behavior patterns can be one or a combination of multiple of sleep, sedentary, light activity, walking, exercise, etc., but are not limited thereto. According to an embodiment of the present application, the obtained information related to user behavior patterns is a multivariate time series, for example, the obtained information x i may be three-direction displacement information and light information, i.e., x i = [acc x ,acc y ,acc z , l]. Wherein, acc x , acc y , acc z represent the displacement information in three directions respectively, and l represents light.

[0049] In step S220, the obtained information is supervised feature extracted using a convolutional neural network, and unsupervised feature extracted using a Gaussian mixture model.

[0050] ​Compared with the commonly used CNN feature extraction method, only a single layer of CNN is used in step S220 to reduce the model parameters, which helps to realize the lightweight of the model. And because GMM is an unsupervised clustering algorithm, it can establish a mapping between the visible time series and the implicit behavior states (such as sleep, walking, exercise, etc.), so that the model has better interpretability, therefore, GMM is also used for feature extraction in step S220. The mixed use of GMM and CNN can make the mixed extracted features more rich, and does not require too many parameters, which also helps to realize the lightweight of the model.

[0051] The GMM model has two types of parameters, namely, the mixing parameter and the mean μ of the Gaussian model i and the variance Σ i . Among them, for a GMM model composed of K states, the sum of the mixing parameters is 1, that is, wherein, may be the probability of a certain behavior state at the current time.

[0052]

[0053] G i = Cat(n, p(X i )) (2)

[0054] wherein, represents a single Gaussian model, Cat() represents a multinomial distribution, n = 24 (indicating that there are 24 hours in a day), represents the state distribution of 24 hours. It is worth noting that the entire GMM model is optimized using the expectation maximization algorithm.

[0055] According to one embodiment of the present application, when the Bayesian Information Criterion (Bayesian Information Criterion) is used as the standard, the distribution of the K parameter of the obtained GMM is as shown in Figure 3 When the K parameter is 5, the Bayesian Information Criterion value is the minimum value. Since the Bayesian Information Criterion value increases with the increase of the model complexity, and decreases with the increase of the likelihood, in order to more accurately reflect the behavior state, the K parameter when the Bayesian Information Criterion value is the minimum value is selected, that is, the K parameter of the GMM is set to K = 5.

[0056] According to one embodiment of the present application, when the K parameter of the GMM is set to K = 5, the five behavior states are respectively mapped to easily understandable behavior models, that is, sleep, sedentary, light activity, walking, and exercise, but not limited thereto.

[0057] According to an embodiment of the present application, in order to make the model lightweight, the parameters of the CNN can be set to 64, and the parameters of the GMM model can be set to 15, but not limited thereto.

[0058] After the feature extraction is completed, since the extracted feature information has a complementary effect, the features extracted using the CNN and the GMM model are input into the sequence modeling tool in parallel to model using the features.

[0059] In step S230, the extracted features are sequence modeled using a recurrent neural network (RNN). According to an embodiment of the present application, in order to make the early warning method more suitable for wearable devices (such as smart watches, smart bracelets, etc.), it is necessary to ensure that the model parameters are not too many. Since the GRU model has fewer model parameters and is easier to calculate compared with the LSTM, the GRU model can be used for sequence modeling and generating an early warning initial model, but not limited thereto. According to an embodiment of the present application, in order to make the model lightweight, the parameters of the GRU model can be set to 66.

[0060] The output S of sequence modeling using the GRU model i The last state output of the GRU model, i.e.,

[0061] S i = GRU ([G i , C i ]) (3)

[0062] wherein,

[0063] According to another embodiment of the present application, the LSTM can be used for sequence modeling and generating an early warning initial model.

[0064] In step S240, the output of the early warning initial model can be classified using a classifier to determine the label corresponding to the output. The classifier includes a full connection and a loss function, and the loss function can be a cross-entropy (CE) loss function, but not limited thereto.

[0065] P i = Sigmoid (WS i ) (4)

[0066] wherein, W represents the parameters of the full connection.

[0067] According to an embodiment of the present application, the classifier is a logistic regression classifier for binary classification. The two labels of the binary classification are NC and MCI, but not limited thereto.

[0068] According to one embodiment of the present application, in order to make the model lightweight, the parameters of the full connection can be set to 26, but not limited thereto.

[0069] According to one embodiment of the present application, in order to make the model lightweight, the sum of the parameters of the neural network can be set to 171, but not limited thereto.

[0070] In step S250, the label determined in step S240 is input into a loss function in a classifier, and the pre-warning initial model generated in step S230 is trained based on a loss value of the loss function until the loss function meets a convergence condition. When the loss function meets the convergence condition, a pre-warning model that meets the condition is obtained.

[0071] Figure 4A is a flowchart of a pre-warning method using a pre-warning model based on a neural network according to one embodiment of the present application.

[0072] As shown in Figure 4A , a pre-warning method 400 includes the following steps:

[0073] Step S410, determining target information;

[0074] Step S420, determining a label corresponding to the target information based on the target information and the trained pre-warning model; and

[0075] Step S430, performing a corresponding pre-warning operation based on the determined label.

[0076] The steps S410 to S430 will be described in detail respectively.

[0077] In step S410, first, a sensor such as a displacement sensor, a light-sensitive sensor, etc. can be used to determine target information, which is information related to a user's behavior pattern. The user behavior pattern can be one or a combination of sleep, sedentary, light activity, walking, exercise, etc., but not limited thereto.

[0078] According to one embodiment of the present application, the acquired information related to the user behavior pattern is a multivariate time series, for example, the acquired information x i may be three-direction displacement information and light information, i.e., x i =[acc x ,acc y ,acc z ,l]. Wherein, acc x , acc y , acc z respectively represent the displacement information in three directions, and l represents light.

[0079] In step S420, as shown in FIG. 4B, determining the label corresponding to the target information based on the target information and the trained early warning model includes the following sub-steps: Figure 4B

[0080] In step S421, feature extraction is performed on the target information. According to an embodiment of the present application, CNN and GMM model can be used for feature extraction, but not limited thereto.

[0081] In step S422, the extracted features are input into the trained early warning model, and the output of the early warning model is input into the classifier.

[0082] In step S423, the output of the early warning model is classified in the classifier to obtain the label corresponding to the output. According to an embodiment of the present application, the classifier is a logistic regression classifier, i.e., binary classification is performed on the input. The two labels in the binary classification are NC and MCI, but not limited thereto.

[0083] In step S430, corresponding early warning operation is performed based on the determined label. According to an embodiment of the present application, when the label corresponding to the output of the early warning model is NC, no early warning is needed; when the label corresponding to the output of the early warning model is MCI, early warning is needed.

[0084] Figure 5 is an exemplary system architecture diagram applicable thereto according to an embodiment of the present application.

[0085] As shown in FIG. 5A, the system architecture 500 can include terminal devices 501, 502, 503, a network 504 and a server 505. The network 504 is used to provide a communication link medium between the terminal devices 501, 502, 503 and the server 505. The network 504 can include various connection types, such as wired or wireless communication links, etc. According to an embodiment of the present application, in order to maximize the detection time and limit the power consumption within a lower range, the terminal device can communicate with the server 505 through, for example, an offline socket or WIFI, but not limited thereto. Figure 5 A user can use the terminal devices 501, 502, 503 to interact with the server 505 through the network 504 to receive or send messages, etc. According to an embodiment of the present application, the terminal devices 501, 502, 503 can perform early warning classification based on sensor devices or local storage information, and transmit the early warning result to the server 505. According to an embodiment of the present application, a doctor can obtain the early warning result from the server 505, and take appropriate intervention measures according to the early warning result.

[0086]

[0087] ​​The terminal devices 501, 502, 503 can be hardware, for example, can be various wearable devices including but not limited to smart watches, smart bands, and smart necklaces, etc. The server 505 can be a server providing various services, for example, after the terminal devices 501, 502, 503 send the early warning result to the server 505.

[0088] It should be noted that the server 505 can be hardware, can be implemented as a distributed server cluster composed of multiple servers, or can be implemented as a single server. When the server 505 is software, it can be implemented as multiple software or software modules (for example, to provide distributed services), or can be implemented as a single software or software module. Herein, no specific limitation is made.

[0089] It should be understood that Figure 5 The number of terminal devices, networks, and servers in the above-mentioned system is merely illustrative. According to the implementation needs, any number of terminal devices, networks, and servers can be provided.

[0090] Figure 5 A block diagram of an electronic terminal device 500 for implementing the early warning method of one embodiment of the present application is shown. The electronic device is intended to represent various forms of wearable devices, such as smart watches, smart bands, and smart necklaces, etc., but is not limited thereto. The components shown herein, their connections and relationships, and their functions are merely examples, and are not intended to limit the implementation of the present application described herein and / or claimed.

[0091] As shown in Figure 6 The terminal device 600 includes a computing unit 601 which can perform various appropriate operations and processes according to programs stored in a read-only memory (ROM) 602 or loaded from a storage unit 607 into a random access memory (RAM) 603. In the RAM 603, various programs and data required for the operation of the terminal device 600 can also be stored. The computing unit 601, the ROM 602, and the RAM 603 are connected to each other through a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0092] The plurality of components in the terminal device 600 include an input unit 605, for example, a sensor, etc.; an output unit 606, for example, various types of displays, speakers, etc.; a storage unit 607, for example, a memory card, etc.; and a communication unit 608, for example, a network card, a modem, a wireless communication transceiver, etc. The communication unit 608 allows the terminal device 600 to exchange information / data with a server through, for example, the Internet and / or various telecommunication networks.

[0093] The computing unit 601 can be various general and / or special purpose push information having processing and computing capabilities. Some examples of the computing unit 601 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The computing unit 601 performs various methods and processes described above, e.g., the early cognitive impairment warning method or the warning model training method. For example, in some embodiments, the early cognitive impairment warning method or the warning model training method can be implemented as an application which is tangibly embodied in a machine-readable medium, e.g., the storage unit 607. In some embodiments, part or all of the application can be loaded and / or installed onto the terminal device 600 via the ROM 602 and / or the communication unit 608. When the application is loaded onto the RAM 603 and executed by the computing unit 601, one or more steps of the early cognitive impairment warning method or the warning model training method described above can be performed. Alternatively, in other embodiments, the computing unit 601 can be configured to perform the early cognitive impairment warning method or the warning model training method by any other appropriate means, e.g., by means of firmware.

[0094] Program code for carrying out methods of the present application can be written in any combination of one or more programming languages. The program code can be provided to a processor or controller of a processing device, such that the program code, when executed by the processor or controller, causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code can be entirely on the machine or partially on the machine.

[0095] To provide for interaction with a user, the systems and techniques described here can be implemented on a device having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user and a keyboard and a pointing device (e.g., a mouse or a touchpad) by which the user can provide input to the device. Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form, including acoustic input, linguistic input, or tactile input.

[0096] The systems and techniques described here can be implemented in a computing system that includes a back end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front end component (e.g., a user computer having a graphical user interface or a Web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back end, middleware, or front end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), and the Internet.

[0097] The computer system can include clients and servers. This relationship can be

[0098] It should be understood that the procedures shown above can be re-ordered, added to, or removed from, while still falling within the scope of the technology disclosed in this application. For example, the steps recited in this application can be performed in parallel, in series, or in a different order, without limitation, as long as the desired results of the technology disclosed in this application are achieved.

[0099] The specific embodiments described above are not intended to be limiting, and any modifications, combinations, sub-combinations and alternatives are intended to fall within the scope of the claims. Changes can be made in the design, operation, composition and arrangement of the components depicted and described above without departing from the spirit and scope of the claims.

Claims

1. A cognitive impairment early warning system, the system comprising: The feature extraction module includes a convolutional neural network and a Gaussian mixture model, wherein the convolutional neural network is used as a supervised feature extraction operator to extract features from the collected data, and the Gaussian mixture model is used as an unsupervised feature extraction operator to extract features from the collected data. The sequence modeling module is used to perform sequence modeling on the features extracted by the convolutional neural network and the features extracted by the Gaussian mixture model. as well as The classification module is used to classify the output of the sequence modeling.

2. The system according to claim 1, wherein, The convolutional neural network is a single-layer convolutional neural network.

3. The system according to claim 1, wherein, The Gaussian mixture model includes five Gaussian models, each used to characterize one of the five human behavioral states.

4. The system according to claim 3, wherein, The five behavioral states are sleep, sedentary, light activity, walking, and exercise.

5. The system according to claim 1, wherein, The sequence modeling module includes a long short-term memory model or a gated loop unit.

6. The system according to claim 1, wherein, The classification module includes a logistic regression classifier.

7. A training method for a cognitive impairment early warning model, the training method comprising: Obtain information related to user behavior patterns; The information is used to extract features using a convolutional neural network and a Gaussian mixture model. The convolutional neural network is used as a supervised feature extraction operator to extract features from the information, and the Gaussian mixture model is used as an unsupervised feature extraction operator to extract features from the information. A recurrent neural network is used to perform sequence modeling on the features extracted by the convolutional neural network and the features extracted by the Gaussian mixture model to generate an initial warning model; The output of the initial early warning model is classified; The classification results and a predetermined threshold are input into the loss function to train the initial warning model until the loss function meets the convergence condition, thus obtaining the warning model.

8. The training method according to claim 7, wherein, The convolutional neural network used is a single-layer convolutional neural network.

9. The training method according to claim 7, wherein, The Gaussian mixture model used employs five Gaussian models to represent the five behavioral states.

10. The training method according to claim 9, wherein, The five behavioral states are sleep, sedentary, light activity, walking, and exercise.

11. The training method according to claim 7, wherein, After the step of extracting features from the acquired information using a convolutional neural network and a Gaussian mixture model, the method further includes: The features extracted by the convolutional neural network and the Gaussian mixture model used are input side by side into the recurrent neural network.

12. The training method according to claim 7, wherein, Sequence modeling using recurrent neural networks to extract features includes: Use gated recurrent units to perform sequence modeling on the extracted features; or The extracted features are sequence-modeled using a long short-term memory model.

13. The training method according to claim 7, wherein, The classification of the output of the initial warning model includes: The output of the initial warning model is classified into two categories using a logistic regression classifier.

14. The training method according to claim 7, wherein, The information is a multivariate time series.

15. The training method according to claim 14, wherein, The multivariate time series includes displacement information in three directions and illumination information.

16. A method for early warning of cognitive impairment, comprising: Determine the target information; Based on the target information and the early warning model obtained by the training method as described in any one of claims 7-15, determine the label corresponding to the target information; as well as Early warning operations are performed based on the identified labels.

17. The method according to claim 16, wherein, The steps for determining the tag corresponding to the target information include: Feature extraction is performed on the target information; The extracted features are input into the early warning model, and the output of the early warning model is input into the classifier; and The classifier classifies the output of the early warning model to obtain the label corresponding to the output.

18. A cognitive impairment early warning device, the device comprising: Memory, which stores instructions that a computer can execute; as well as A processor for executing computer-executable instructions stored in the memory to implement the method as described in any one of claims 7-15.

19. A computer storage medium comprising computer-executable instructions, which, when executed by a processor, cause the method of any one of claims 7-15 to be implemented.

Images