Fall prediction method and system based on deep learning

Abstract

The invention discloses a tumble prediction method and system based on deep learning, and belongs to the technical field of tumble prediction. The tumble prediction method based on deep learning comprises the following steps: constructing and training a CNN prediction model; obtaining motion data of a human body, and generating an intermediate prediction probability value based on the CNN prediction model and the motion data; basic data of the human body are obtained, a final prediction probability value is generated based on the basic data and the middle prediction probability value, and the basic data comprise the age and the historical falling frequency. The accuracy of the prediction result is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of fall prediction, and in particular, relates to a fall prediction method and system based on deep learning. Background Art

[0002] With the aging of the population, falls have become one of the major health risks faced by the elderly and those with limited mobility. Falls may not only cause physical injuries such as fractures and joint dislocations, but also lead to serious consequences. Therefore, developing a technology that can effectively predict and prevent falls is of great significance for improving the quality of life of the elderly and reducing the medical burden. At present, there are many fall detection technologies on the market. However, these technologies usually only predict falls based on the movement state of the human body, ignoring the differences between individuals, resulting in inaccurate and unreliable prediction results. Summary of the invention

[0003] The purpose of the present invention is to overcome the deficiencies of the prior art and provide a fall prediction method and system based on deep learning.

[0004] The objective of the present invention is achieved through the following technical solutions:

[0005] The first aspect of this embodiment discloses a fall prediction method based on deep learning, comprising:

[0006] Build and train a CNN prediction model;

[0007] Obtain human motion data and generate intermediate prediction probability values ​​based on the CNN prediction model and motion data;

[0008] Basic data of a human body is acquired, and a final predicted probability value is generated based on the basic data and the intermediate predicted probability value, wherein the basic data includes age and the number of historical falls.

[0009] Furthermore, the motion data includes acceleration data, angular velocity data and magnetometer data of the human body.

[0010] Further, generating a final predicted probability value based on the basic data and the intermediate predicted probability value includes:

[0011] The final predicted probability value is calculated based on the basic data and the intermediate predicted probability value. The calculation formula for the final predicted probability value is:

[0012]

[0013] Where R represents the final predicted probability value, R0 represents the intermediate predicted probability value, e represents the base of the natural logarithm, A represents age, and the value of A is greater than or equal to 60, H represents the number of historical falls, and β1 and β2 represent coefficients.

[0014] Furthermore, the number of historical falls is automatically updated when the final predicted probability value is greater than a preset value.

[0015] Furthermore, an alarm is issued when the final predicted probability value is greater than a preset value.

[0016] Furthermore, a CNN prediction model is constructed and trained, including:

[0017] Constructing a CNN prediction model, wherein the CNN prediction model includes an input layer, a convolution layer, a pooling layer, a fully connected layer, and an output layer, wherein the convolution layer is a one-dimensional convolution layer, the pooling layer adopts a maximum pooling layer, and the output layer adopts a sigmoid activation function to output the probability of falling;

[0018] Acquire raw data, wherein the raw data includes samples of multiple fall events and multiple normal activities;

[0019] Preprocess the raw data;

[0020] Divide the preprocessed raw data into training set and validation set;

[0021] Input the data of the training set into the CNN prediction model for training;

[0022] The CNN prediction model is validated using the validation set data, and the hyperparameters are adjusted to optimize the model performance.

[0023] Furthermore, preprocessing the raw data includes: removing noise data and invalid data in the raw data; dividing the raw data into time windows of fixed length, each window containing acceleration, angular velocity and magnetometer data within a period of time; and normalizing the raw data.

[0024] Furthermore, the binary cross entropy loss function is used as the loss function when training the CNN prediction model.

[0025] A second aspect of this embodiment discloses a fall prediction system based on deep learning, including:

[0026] Model building module, used to build and train CNN prediction models;

[0027] The first prediction module is used to obtain the motion data of the human body and generate an intermediate prediction probability value based on the CNN prediction model and the motion data;

[0028] The second prediction module is used to obtain basic data of the human body and generate a final prediction probability value based on the basic data and the intermediate prediction probability value, wherein the basic data includes age and the number of historical falls.

[0029] The beneficial effects of the present invention are:

[0030] (1) The present invention predicts falls based on the body's motion state and the body's age and number of historical falls, taking into account the differences between different people and improving the accuracy of the prediction results;

[0031] (2) The present invention combines human acceleration, angular velocity and magnetometer data to predict falls, which can effectively distinguish between fall events and some daily activities (such as turning around, bending over, etc.), thereby improving the accuracy of fall prediction results. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 A flow chart of the fall prediction method in the present invention;

[0033] Figure 2 The figure is a block diagram of a fall prediction system in the present invention. DETAILED DESCRIPTION

[0034] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0035] See also Figure 1 to Figure 2 , the present invention provides a fall prediction method and system based on deep learning:

[0036] The first aspect of this embodiment discloses a fall prediction method based on deep learning, such as Figure 1 As shown, the fall prediction method includes steps S100 to S300.

[0037] Step S100: Build and train a CNN prediction model.

[0038] In one embodiment, building and training a CNN prediction model includes: building a CNN prediction model; obtaining raw data; preprocessing the raw data; dividing the preprocessed raw data into a training set and a validation set; inputting the data of the training set into the CNN prediction model for training; using the data of the validation set to validate the CNN prediction model, and adjusting hyperparameters (such as learning rate, batch size, etc.) to optimize model performance.

[0039] The CNN prediction model includes an input layer, a convolutional layer, a pooling layer, a fully connected layer and an output layer.

[0040] The input layer is used for data input.

[0041] The convolution layer is used to extract features from the input data. The convolution layer uses a one-dimensional convolution layer.

[0042] The pooling layer is used to reduce the dimension of the data, reduce the amount of computation, and extract more important features.

[0043] The pooling layer uses the maximum pooling layer, which can extract the most significant features in the feature map, such as the maximum value in the acceleration signal, thereby improving the accuracy of the prediction results.

[0044] There can be multiple convolutional layers and pooling layers to gradually reduce the dimension of the data and extract higher-level features.

[0045] The activation function of the fully connected layer uses the ReLU activation function, and other functions such as the Dropout function are used for regularization before the last layer.

[0046] The output layer consists of a single neuron, and the output layer uses a sigmoid activation function to output the probability of falling.

[0047] The raw data includes samples of multiple fall events and various normal activities.

[0048] When the original data is divided into a training set and a validation set, the ratio of the training set to the validation set can be determined according to the specific situation.

[0049] In one embodiment, preprocessing the raw data includes: removing noise data and invalid data in the raw data; dividing the raw data into time windows of fixed length, each window containing acceleration, angular velocity and magnetometer data within a period of time; and normalizing the raw data.

[0050] In one embodiment, a binary cross entropy loss function is used as a loss function when training a CNN prediction model.

[0051] Step S200: Obtain human motion data, and generate an intermediate prediction probability value based on the CNN prediction model and the motion data.

[0052] Specifically, the motion data of the human body is obtained, and then the motion data is input into the CNN prediction model. The CNN prediction model outputs the probability value of whether a person falls according to the input motion data, which is the intermediate prediction probability value.

[0053] In one embodiment, the motion data includes acceleration data, angular velocity data and magnetometer data of the human body.

[0054] The acceleration data, angular velocity data and magnetometer data are collected by the wearable device. After the user wears the wearable device, the wearable device collects the acceleration data, angular velocity data and magnetometer data of the user.

[0055] Acceleration data and angular velocity data provide dynamic and kinematic information of human movement, and magnetometer data provides directional information. Combining acceleration data, angular velocity data and magnetometer data for fall prediction can perform more accurate posture calculation, thereby more accurately distinguishing fall events from daily activities, reducing false alarms and missed alarms, and improving the accuracy of fall prediction.

[0056] In one embodiment, a wearable device located at the waist or ankle of the human body collects acceleration data, angular velocity data, and magnetometer data of the human body, thereby reducing the amount of data while ensuring the accuracy of the prediction results.

[0057] Step S300: Obtain basic data of a human body, and generate a final predicted probability value based on the basic data and the intermediate predicted probability value, wherein the basic data includes age and the number of historical falls.

[0058] Based on the prediction using the CNN prediction model, this embodiment further calculates the age of the human body and the number of historical falls, etc., to generate the final prediction probability value of the fall prediction, thereby realizing personalized prediction of each person's fall risk and improving the accuracy and reliability of the prediction results.

[0059] In one embodiment, generating a final prediction probability value based on basic data and intermediate prediction probability values ​​includes: calculating the final prediction probability value according to the basic data and the intermediate prediction probability values, and the calculation formula of the final prediction probability value is:

[0060]

[0061] Wherein, R represents the final predicted probability value, R0 represents the intermediate predicted probability value, e represents the base of the natural logarithm, A represents age, the value of A is greater than or equal to 60, H represents the number of historical falls, β1 and β2 represent coefficients. In other embodiments, the minimum value of age may also be other values, and the minimum value of age is generally greater than or equal to 18.

[0062] In one embodiment, the number of historical falls is automatically updated when the final predicted probability value is greater than a preset value, that is, the number of historical falls is increased by 1 when the final predicted probability value is greater than the preset value. By automatically updating the number of historical falls when the final predicted probability value is greater than the preset value, the timeliness of updating the historical fall data is improved, and the reliability of the prediction results is ensured.

[0063] Generally, the number of historical falls can be updated manually while the number of historical falls is automatically updated.

[0064] In one embodiment, an alarm is triggered when the final predicted probability value is greater than a preset value.

[0065] There are many ways to alarm, such as sound alarm, flash alarm, and sending alarm information to a preset terminal (such as a mobile phone) for remote alarm.

[0066] A second aspect of this embodiment discloses a fall prediction system based on deep learning, such as Figure 2 As shown, the fall prediction system includes a model building module, a first prediction module and a second prediction module.

[0067] The model building module is used to build and train the CNN prediction model.

[0068] The first prediction module is used to obtain the motion data of the human body and generate an intermediate prediction probability value based on the CNN prediction model and the motion data.

[0069] The second prediction module is used to obtain basic data of the human body and generate a final prediction probability value based on the basic data and the intermediate prediction probability value, wherein the basic data includes age and the number of historical falls.

[0070] It should be noted that those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system or device described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0071] The above is only a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the form disclosed herein, and should not be regarded as excluding other embodiments, but can be used in various other combinations, modifications and environments, and can be modified within the scope of the concept described herein through the above teachings or the technology or knowledge of the relevant field. The changes and modifications made by those skilled in the art shall not deviate from the spirit and scope of the present invention, and shall be within the scope of protection of the claims attached to the present invention.

Claims

1. A fall prediction method based on deep learning, characterized in that: include: Build and train a CNN prediction model; Obtain human motion data and generate intermediate prediction probability values ​​based on the CNN prediction model and motion data; Basic data of a human body is acquired, and a final predicted probability value is generated based on the basic data and the intermediate predicted probability value, wherein the basic data includes age and the number of historical falls.

2. The fall prediction method based on deep learning according to claim 1, characterized in that: The motion data includes acceleration data, angular velocity data and magnetometer data of the human body.

3. The fall prediction method based on deep learning according to claim 1, characterized in that: The final predicted probability value is generated based on the basic data and the intermediate predicted probability value, including: The final predicted probability value is calculated based on the basic data and the intermediate predicted probability value. The calculation formula for the final predicted probability value is: Where R represents the final predicted probability value, R0 represents the intermediate predicted probability value, e represents the base of the natural logarithm, A represents age, and the value of A is greater than or equal to 60, H represents the number of historical falls, and β1 and β2 represent coefficients.

4. The fall prediction method based on deep learning according to claim 1, characterized in that: The historical number of falls is automatically updated when the final predicted probability value is greater than the preset value.

5. The fall prediction method based on deep learning according to claim 1, characterized in that: An alarm will be issued when the final predicted probability value is greater than the preset value.

6. The fall prediction method based on deep learning according to claim 1, characterized in that: Build and train a CNN prediction model, including: Constructing a CNN prediction model, wherein the CNN prediction model includes an input layer, a convolution layer, a pooling layer, a fully connected layer, and an output layer, wherein the convolution layer is a one-dimensional convolution layer, the pooling layer adopts a maximum pooling layer, and the output layer adopts a sigmoid activation function to output the probability of falling; Acquire raw data, wherein the raw data includes samples of multiple fall events and multiple normal activities; Preprocess the raw data; Divide the preprocessed raw data into training set and validation set; Input the data of the training set into the CNN prediction model for training; The CNN prediction model is validated using the validation set data, and the hyperparameters are adjusted to optimize the model performance.

7. The fall prediction method based on deep learning according to claim 6, characterized in that: The preprocessing of the raw data includes: removing noise data and invalid data in the raw data; dividing the raw data into time windows of fixed length, each window containing acceleration, angular velocity and magnetometer data within a period of time; and normalizing the raw data.

8. The fall prediction method based on deep learning according to claim 6, characterized in that: The binary cross entropy loss function is used as the loss function when training the CNN prediction model.

9. A fall prediction system based on deep learning, characterized in that: include: Model building module, used to build and train CNN prediction models; The first prediction module is used to obtain the motion data of the human body and generate an intermediate prediction probability value based on the CNN prediction model and the motion data; The second prediction module is used to obtain basic data of the human body and generate a final prediction probability value based on the basic data and the intermediate prediction probability value, wherein the basic data includes age and the number of historical falls.

Images