Intelligent early warning method and system for falling of old people in nursing home scene

Abstract

The invention discloses an intelligent early warning method and system for falling in a nursing home scene, and relates to the technical field of medical health, and the method comprises the steps: obtaining the static and dynamic data of an old person, carrying out the preprocessing and feature extraction of a data set, and forming a feature data set; classifying the feature data according to the data type, constructing a classification feature data set, and establishing an old man falling prediction model through a logic regression model, a linear combination of classification features and a Sigmoid function; and carrying out processing and feature extraction by using a pre-constructed old people tumble data set, and inputting the tumble data set into the prediction model for training to minimize a loss function value so as to obtain an optimized tumble prediction model. And finally, inputting the classification feature data into the optimized model, and outputting an old man falling prediction result.

Description

Technical Field

[0001] The present invention relates to the field of safety protection technology, and in particular to an intelligent early warning method and system for elderly people falling in a nursing home setting. Background Art

[0002] Faced with the increasingly serious problem of aging, the traditional family-based elderly care model can no longer meet the growing demand for elderly care. More and more elderly people, especially empty-nest elderly people and elderly people living alone, choose to enter nursing homes and other places to seek more professional care services. This phenomenon is common all over the world. According to the current situation in China, nursing homes, as places where the elderly live and care together, are becoming increasingly important. The degenerative lesions of bones and joints caused by aging and the loss of a large amount of muscle have made the elderly a high-risk group for falls. According to the World Health Organization and related studies, falls have become one of the main causes of death from injuries among the elderly. The consequences of falls include soft tissue contusions, fractures, and intracranial hemorrhage, which seriously affect the quality of life and life expectancy of the elderly. In the nursing home scenario, how to effectively prevent and reduce the occurrence of falls among the elderly and protect their life safety and physical health has become a focus of social attention.

[0003] The public patent document CN114676956A "Elderly Fall Risk Warning System Based on Multi-Dimensional Data Fusion" combines radar, wearable devices, depth of field cameras, etc. from a system perspective, and places third-party service providers, institutional management users, and the elderly in the entire system. Based on the analysis of the elderly's body information, institutional administrators can view the elderly's historical data information and implement fall warnings.

[0004] However, the limitation of the above-mentioned public patent is that it only focuses on real-time dynamic data information, while ignoring the equally important individual basic health information of the elderly and the comprehensive impact of environmental factors on the risk of falls. Summary of the invention

[0005] In order to overcome the shortcomings of the above-mentioned prior art that fails to screen out reversible factors according to the specific causes of falls and formulate personalized prevention strategies accordingly, the main purpose of the present invention is to provide an intelligent early warning method and system for falls of the elderly in a nursing home scenario for the elderly living in nursing homes.

[0006] To achieve the above purpose, the present invention adopts the following technical solution, an intelligent early warning method for elderly people falling in a nursing home scenario, comprising the following steps:

[0007] Obtain a known elderly person's fall data set, process it and extract features, and obtain fall classification feature data, wherein the known elderly person's fall data set includes: known personal static data, known personal dynamic data, and known environmental dynamic data;

[0008] A fall prediction model for the elderly was constructed based on the logistic regression model, the linear combination of the fall classification feature data, and the Sigmoid function;

[0009] Input the fall feature classification data into the elderly fall prediction model, train the weights in the elderly fall prediction model to minimize the value of the loss function, and obtain an optimized elderly fall prediction model;

[0010] Obtaining personal static data, personal dynamic data and environmental dynamic data of the elderly to form a data set; preprocessing the data set and performing feature extraction to obtain the feature data set; classifying the feature data set according to the data type to obtain the classified feature data set;

[0011] The classification feature data set is input into the optimized elderly fall prediction model to obtain the elderly fall prediction result, and the elderly fall prediction result is compared with the preset risk threshold range. If it is less than the preset risk threshold range, no alarm is issued; if it is within the preset risk threshold range, a warning reminder is issued and the detection rate is increased; if it is greater than the preset risk threshold range, an alarm is issued.

[0012] The method of constructing an elderly fall prediction model based on a logistic regression model, a linear combination of categorical data, and a Sigmoid function includes the following steps:

[0013] Assign an initial weight to each fall classification feature data and set an initial bias item;

[0014] Determine the structure of the logistic regression model, that is, the linear combination of features plus the Sigmoid function, to obtain the elderly falls prediction model.

[0015] The method of inputting the fall feature classification data into the elderly fall prediction model, training the weights in the elderly fall prediction model, minimizing the value of the loss function, and obtaining an optimized elderly fall prediction model comprises the following steps:

[0016] Calculate the linear combination of features and weights: z = w1x1 + w2x2 + ... + w n *x n +b; and convert the output of the linear combination into a probability prediction by applying the Sigmoid function: h(x) = 1 / (1+e^(-z));

[0017] The cross entropy loss function is used to calculate the loss between the predicted probability and the actual label: L(y,h(x))=-y*log(h(x))-(1-y)*log(1-h(x)), where y is the actual label;

[0018] By calculating the gradient of the loss function for each weight and bias; updating the weights and biases of the elderly fall prediction model parameters according to the gradient descent algorithm, Where α is the learning rate, and the loss function is minimized to obtain an optimized elderly fall prediction model.

[0019] The individual static information data includes numerical data and categorical data. The numerical data includes age, mental health status, Barthel index, triglyceride index, and fasting blood sugar; the categorical data includes the type of disease, medication history, fall history, sleep disorder, and mental cognitive disorder.

[0020] The personal dynamic data includes image data and non-image data, wherein the image data is facial change data; the non-image data includes: acceleration of joint movements, accumulated exercise time, heart rate, oxygen saturation, and blood pressure.

[0021] The environmental dynamic data includes light data, humidity data and obstacle detection data.

[0022] The preprocessing of the data set and feature extraction to obtain the feature data set includes the following steps:

[0023] Identifying and processing missing data, detecting and processing outliers in the data set, and obtaining a cleaned data set;

[0024] Performing heat coding on the categorical data and standardizing the numerical data in the cleaned data set to obtain a converted data set;

[0025] The converted data set is used for selecting features related to fall prediction using correlation analysis and information gain method, and the features are screened using stepwise regression analysis to obtain a feature data set.

[0026] The preset risk threshold range is:

[0027] The risk value is less than 0.3, which is the low risk threshold;

[0028] The risk value is between 0.3 and 0.6, which is the medium risk threshold;

[0029] The risk value is greater than 0.6, which is the high risk threshold.

[0030] The intelligent early warning system for elderly falls in nursing homes includes:

[0031] The data acquisition module is used to acquire the static data and dynamic data of the elderly, personal static data, personal dynamic data and environmental dynamic data, to form a data set;

[0032] The data processing module preprocesses the data set and extracts features to obtain a feature data set. The feature data set is classified according to the data type to obtain a classified feature data set. Based on the classified feature data set, a linear combination of the classified data is obtained.

[0033] The model building module builds a fall prediction model for the elderly based on the logistic regression model, the linear combination of classification data, and the Sigmoid function; obtains a known fall data set of the elderly, processes it, and extracts features to obtain the fall classification feature data, inputs the fall feature classification data into the fall prediction model for the elderly, trains the weights in the fall prediction model for the elderly, minimizes the value of the loss function, and obtains an optimized fall prediction model for the elderly;

[0034] The fall prediction module inputs the classification feature data set into the optimized elderly fall prediction model to obtain the elderly fall prediction results, and compares the elderly fall prediction results with the preset risk threshold range. If it is less than the preset risk threshold range, no alarm is issued. If it is within the preset risk threshold range, an early warning reminder is issued and the detection rate is increased. If it is greater than the preset risk threshold range, an alarm is issued.

[0035] Compared with the prior art, the beneficial effects of the present invention are:

[0036] Based on the personal static information and real-time dynamic characteristics of the elderly, fall risk thresholds are generated and risk stratification is achieved. This can more accurately predict the fall risk of the elderly based on their personal static information and real-time dynamic characteristics, and achieve early warning risk stratification accordingly.

[0037] A fall prediction model is constructed using a logistic regression model combined with a Sigmoid function to process and extract features from the elderly's personal static data, personal dynamic data, and environmental dynamic data. The cross entropy loss function and gradient descent algorithm are used to train and optimize the model. The risk threshold range is set, and the decision to issue an alarm is made based on the comparison between the prediction result and the threshold. The system can predict the risk of falls based on the elderly's real-time data and issue corresponding warnings. By analyzing the dynamic data of the elderly's personal and environment, the model can more accurately predict fall events; through the training and optimization process, the model can gradually improve the accuracy of the prediction. Timely warnings can reduce the occurrence of elderly fall accidents and improve the safety level of nursing homes. Continuous monitoring of the elderly's health status helps to adjust care and treatment plans in a timely manner. The intelligent early warning system can reduce dependence on manual monitoring, thereby reducing long-term care costs. The elderly and their families can enjoy a higher quality of life by reducing the risk of falls. Through the analysis of personal data, the system can provide more personalized early warning services to meet the needs of different elderly people.

[0038] In summary, the above method combines advanced machine learning technology and multi-source data to achieve intelligent prediction and early warning of the risk of falls for the elderly, thereby ensuring the safety of the elderly while also improving the operational efficiency and quality of nursing services in nursing homes. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The drawings described herein are used to provide further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute improper limitations on the present application.

[0040] Figure 1 It is a flowchart of the intelligent early warning system for falls of the present invention;

[0041] Figure 2 This is the process of pushing early warning information based on the fusion of individual information and real-time vital signs of the elderly;

[0042] Figure 3 It is a schematic diagram of the process structure of the present invention;

[0043] Figure 4 It is an R2 value table of the fall warning risk prediction model constructed based on SEM in the present invention. DETAILED DESCRIPTION

[0044] According to the World Health Organization and related research, falls have become one of the main causes of injury-related deaths among the elderly. The consequences of falls include soft tissue contusions, fractures, and intracranial hemorrhage, which seriously affect the quality of life and life expectancy of the elderly. In nursing homes, how to effectively prevent and reduce the occurrence of falls among the elderly and protect their life safety and physical health has become a focus of social concern.

[0045] Among the existing publicly available technologies, the focus of fall warning for the elderly is only on real-time dynamic data information, while ignoring the equally important individual basic health information of the elderly and the comprehensive impact of environmental factors on the risk of falls.

[0046] In addition, existing patents fail to screen out reversible factors based on the specific causes of falls, formulate personalized prevention strategies based on them, and effectively push the strategies to target users, providing guidance and reference for fall prevention and early intervention.

[0047] In order to overcome the shortcomings of the above-mentioned prior art that the reversible factors cannot be screened out according to the specific causes of falls, and personalized prevention strategies are formulated accordingly, the main purpose of the present invention is to provide an intelligent early warning method and system for elderly people falling in nursing homes. To achieve the above purpose, the present invention adopts the following technical scheme, and the intelligent early warning method for elderly people falling in nursing homes includes the following steps:

[0048] Obtain a known elderly person's fall data set, process it and extract features, and obtain fall classification feature data, wherein the known elderly person's fall data set includes: known personal static data, known personal dynamic data, and known environmental dynamic data;

[0049] A fall prediction model for the elderly was constructed based on the logistic regression model, the linear combination of the fall classification feature data, and the Sigmoid function;

[0050] Input the fall feature classification data into the elderly fall prediction model, train the weights in the elderly fall prediction model to minimize the value of the loss function, and obtain an optimized elderly fall prediction model;

[0051] Obtaining personal static data, personal dynamic data and environmental dynamic data of the elderly to form a data set; preprocessing the data set and performing feature extraction to obtain the feature data set; classifying the feature data set according to the data type to obtain the classified feature data set;

[0052] The classification feature data set is input into the optimized elderly fall prediction model to obtain the elderly fall prediction result, and the elderly fall prediction result is compared with the preset risk threshold range. If it is less than the preset risk threshold range, no alarm is issued; if it is within the preset risk threshold range, a warning reminder is issued and the detection frequency is increased; if it is greater than the preset risk threshold range, an alarm is issued.

[0053] The present invention will be further described below in conjunction with the accompanying drawings and implementation modes.

[0054] Example

[0055] See also Figure 1-Figure 4 ,The intelligent early warning method for elderly falling in the nursing home scenario includes the following steps:

[0056] Obtain a known data set of falls of the elderly, process it and extract its features to obtain the feature data of fall classification;

[0057] A fall prediction model for the elderly was constructed based on the logistic regression model, the linear combination of the fall classification feature data, and the Sigmoid function;

[0058] Input the fall feature classification data into the elderly fall prediction model, train the weights in the elderly fall prediction model to minimize the value of the loss function, and obtain an optimized elderly fall prediction model;

[0059] Collect data sets, including personal static data, personal dynamic data and environmental dynamic data.

[0060] Specifically, static data refers to individual static information data of the elderly, including:

[0061] 1) Age: A key factor that leads to falls. As people age, their muscle loss accelerates and their body coordination ability declines, which can easily lead to falls.

[0062] 2) Anxiety or depression: Poor mental state can affect the overall health level and quality of life of the elderly. Anxiety or depression can cause individual distraction or decreased judgment, thereby inducing falls.

[0063] 3) Daily living ability (Barthel index): An indicator to assess an individual's ability to live independently in daily living activities. The lower the index, the higher the individual's dependence. Individuals with high dependence have lower self-care ability and a higher risk of falling.

[0064] 4) Number of comorbidities: The number of comorbidities refers to the number of diseases an individual suffers from. The elderly may suffer from multiple health problems at the same time, which may interact with each other, thereby increasing physical dysfunction and promoting the occurrence of falls.

[0065] 5) Taking sleeping pills and sedatives: The use of sleeping pills and sedatives will reduce the reaction speed of the nervous system and the sensitivity of the body, increasing the individual's risk of falling.

[0066] 6) Previous history of falls: Previous history of falls is an important indicator of increased risk of falls. If the elderly have fallen before, it indicates that they may have certain risk factors that lead to falls, such as muscle weakness and poor sense of balance, which may cause falls again.

[0067] 7) Sleep disorders: If the elderly have sleep problems, such as insomnia, short sleep time or poor sleep quality, it may affect their concentration, reaction time and body coordination, and increase the risk of falling.

[0068] 8) Mental cognitive disorder: Mental cognitive disorder refers to dementia or inattention, which are significantly associated with an increased risk of falls; and may affect the elderly's judgment, decision-making ability, and reaction speed to environmental changes, making individuals more prone to falls than normal people.

[0069] 9) Albumin <30g / L: Albumin levels below 30g / L may indicate malnutrition in the elderly. Malnutrition can lead to muscle weakness, decreased immunity and physical function decline, all of which are associated with an increased risk of falls.

[0070] 10) Triglycerides ≥ 2.3mmol / L: Triglyceride levels higher than 2.3mmol / L indicate that the elderly have too high blood lipid levels. Individuals with high blood lipid levels may cause cardiovascular diseases, such as coronary atherosclerotic heart disease, and affect heart function and blood circulation, which in turn directly affects the walking stability and endurance of the elderly. At the same time, high blood lipids may also affect the blood supply to the leg muscles, leading to muscle weakness; the above factors may increase the risk of falls in the elderly.

[0071] 11) Fasting blood sugar > 7.0mmol / L: Fasting blood sugar levels above 7.0mmol / L may indicate that the elderly have diabetes or poor blood sugar control. High blood sugar levels may cause peripheral nerve damage, vision problems, and even diabetic foot, all of which may increase the risk of falls.

[0072] Personal dynamic data refers to the real-time dynamic characteristics of the elderly in their daily activities, such as movement changes, expressions, physiological data, etc., including:

[0073] 1) Immediate movement changes (acceleration changes in joint movements): Monitoring sudden changes in joint acceleration can promptly capture possible precursors to falls, such as unstable standing or walking movements.

[0074] 2) Cumulative exercise time: When the elderly exercise for a long time or with high intensity, the body may bear a high burden. This burden will cause fatigue of muscles and joints, because muscles need to contract and relax for a long time to maintain exercise, and joints need to bear the weight of the body and the impact of exercise during exercise. Long-term or high-intensity exercise will increase the fatigue of muscles and joints, thereby reducing their stability.

[0075] 3) Expressions of pain or tension: Through facial expression recognition technology, monitoring expressions of pain or tension can serve as early indicators of discomfort or potential health problems. These expression characteristics may indicate an acute pathological state that causes body imbalance.

[0076] 4) Heart rate changes: Abnormal changes in heart rate are an important indicator of cardiovascular problems, which may cause a decrease in body balance or loss of consciousness, thereby increasing the risk of falling.

[0077] 5) Oxygen saturation: Low oxygen saturation may indicate impaired function of the respiratory or circulatory systems, which may cause dizziness or confusion and increase the risk of falls.

[0078] 6) Blood pressure changes: Large fluctuations in blood pressure can lead to fainting and directly cause falls.

[0079] Environmental dynamic data is collected from light and humidity sensors, as well as environmental image data obtained by indoor cameras, to identify and monitor obstacles on the ground such as furniture, debris or stagnant water, which may easily cause falls.

[0080] Obtain personal static data, personal dynamic data and environmental dynamic data to form a data set; preprocess the data set and perform feature extraction to obtain a feature data set; classify the feature data set according to the data type to obtain a classified feature data set;

[0081] Among them, for the acquisition of data sets:

[0082] 1) User static data collection

[0083] A. When the elderly check in, the staff of the institution collects and records all the static data information of the elderly.

[0084] B. Data update frequency: The update frequency of user static information data depends on the individual's medical condition, but all static information is extracted based on the latest medical records.

[0085] 2) User dynamic data collection

[0086] A. Use wearable devices to collect dynamic information data of the elderly in real time, such as immediate action changes, accumulated exercise time and other indicators. The real-time dynamic data results are uploaded to the server with a frequency set to once per minute. The specific data content is as follows:

[0087] Motion data: Use wearable devices to obtain and identify the real-time motion patterns of the elderly (accelerometer and gyroscope data).

[0088] Physiological data: Obtain real-time vital signs data such as heart rate and blood pressure through wearable devices.

[0089] Expression data: Collect the elderly’s facial expression data through cameras or specific sensors.

[0090] Positioning data: Upload the user's current location information data in real time through wearable devices.

[0091] Integrate the above dynamic and static data into the central database, and ensure the consistency and accuracy of the data through user identity association.

[0092] 3) Environmental dynamic data collection

[0093] A. Light and humidity information: The environmental data (light brightness value, humidity value) collected in real time by the light, temperature and humidity integrated sensor is uploaded to the monitoring station via the serial port and then uploaded to the server in real time.

[0094] B. Obstacle detection video signal: The environmental video signal collected by the camera is uploaded to the server in real time via the wireless network.

[0095] The data set is preprocessed and features are extracted to obtain a feature data set, including the following steps:

[0096] Identifying and processing missing data, detecting and processing outliers in the data set, and obtaining a cleaned data set;

[0097] Performing heat coding on the categorical data and standardizing on the numerical data in the cleaned data to obtain a converted data set;

[0098] The converted data set is used for selecting features related to fall prediction using correlation analysis and information gain method, and the features are screened using stepwise regression analysis to obtain a feature data set.

[0099] Data classification, identifying the type of each data to determine the appropriate preprocessing method.

[0100] 1) Static data includes:

[0101] A. Numerical data: age, Barthel index, albumin, triglyceride, fasting blood glucose, and air humidity.

[0102] B. Categorical data: anxiety or depression (yes / no), taking sedatives (yes / no), previous history of falls (yes / no), sleep disorders (yes / no), mental cognitive disorders (yes / no).

[0103] C. Count data: number of comorbidities (treated as numerical data).

[0104] 2) Dynamic data includes:

[0105] A. Continuous data: immediate movement changes (changes in acceleration of joint movements), heart rate changes, and blood pressure changes.

[0106] B. Time data: accumulated exercise time.

[0107] C. Categorical data: expression of pain or tension (converted to numerical type for processing).

[0108] D. Proportional data: oxygen saturation.

[0109] For data processing

[0110] 1) Static data preprocessing, including personal background information and health status indicators, which generally do not change over time. Numerical data (age, Barthel index, albumin, triglycerides, fasting blood sugar) are standardized using the Z-score method. Categorical data (anxiety or depression, taking sedatives, previous history of falls, sleep disorders, mental cognitive disorders) are encoded using one-hot encoding to convert categorical data into binary format. In the supplement of missing values, the mean or median is used for numerical data, and the mode is used for categorical data.

[0111] 2) Dynamic data involves behavioral or physiological indicators that change over time. Continuous data (immediate changes in movement, heart rate, and blood pressure) can still be normalized using the same Z-score method as static data. Time data (accumulated exercise time) uses Min-Max normalization to scale the data to the range of [0,1]. Categorical data (expressions of pain or tension) uses binary data.

[0112] Feature extraction for datasets

[0113] Using the central database, we can directly obtain the elderly’s personal static information such as age, fall history, and case data, as well as some dynamic information such as blood pressure, heart rate, and oxygen saturation. In addition, the elderly’s joint acceleration, cumulative movement time, and dynamic facial phenotype need to be feature extracted. The specific steps are as follows:

[0114] 1) Acceleration of joint movements & accumulated movement time

[0115] 1. Sensor data collection: Use the accelerometer and gyroscope in the wearable device to collect the linear acceleration and angular velocity of the joints in different directions.

[0116] 2. Preprocessing: Preprocess the collected raw sensor data and use a Butterworth low-pass filter to remove high-frequency noise. The filter cutoff frequency should be set according to the motion frequency range, usually 1-5Hz.

[0117] 3. Joint acceleration feature extraction

[0118] 3.1 Calculation of joint angles:

[0119] The joint angles are represented by quaternions or Euler angles, and the real-time angles of the joints in three-dimensional space are calculated by fusing accelerometer and gyroscope data, for example, through Kalman filtering or complementary filtering.

[0120] 3.2 Integration of angular velocity:

[0121] The angular velocity data collected by the sensor is integrated once to obtain the change of joint angle over time, that is, the angular position.

[0122] The integral of the angular velocity is calculated using the following formula:

[0123] θ(t)=θ(t-1)+ω(t)*Δt

[0124] Among them, θ(t) represents the angle at time t, ω(t) represents the angular velocity, and Δt is the time interval.

[0125] 3.3 Calculation of linear acceleration:

[0126] The linear acceleration of a joint can be calculated by the second-order derivative of the joint angle with time

[0127] 4. Motion Cumulative Time Feature Extraction

[0128] 4.1 Status Detection:

[0129] By real-time monitoring of the acceleration or angular velocity data of the joint, it is determined whether it exceeds the preset motion threshold S(t). If the acceleration or angular velocity is greater than the threshold, S(t) = 1, otherwise S(t) = 0).

[0130] 4.2 Time accumulation:

[0131] All time periods marked as motion states are accumulated to obtain the cumulative motion time:

[0132]

[0133] Among them, T 运动 represents the total accumulated time of movement, T is the total observation time, and Δt is the time interval.

[0134] 2) Dynamic facial expressions of the elderly

[0135] For the dynamic facial expressions of elderly users, this system will analyze facial video signals through a convolutional neural network model to identify the user's painful, nervous and other uncomfortable expressions. The specific processing process is as follows:

[0136] A. Data preprocessing Video acquisition: Extract frames from the video. To balance the amount of computation and real-time performance, 5 frames per second are used.

[0137] Training data source: Use image data containing various facial expressions, which comes from the public dataset CK+:

[0138] Face detection and alignment: Use OpenCV's Haar cascade classifier face detection algorithm to detect the face area in each frame. Align the detected faces to make the position and scale of the faces consistent in the image.

[0139] B. Construct a convolutional neural network (CNN) model using a convolutional neural network architecture, as shown below:

[0140] Figure 1 Schematic diagram of convolutional neural network architecture

[0141] C. Training Model

[0142] 1) Use the cross entropy loss function and Adam optimizer for model training:

[0143]

[0144] N is the number of samples and K is the number of categories.

[0145] y i,k , is the actual label (using one-hot encoding) that the i-th sample belongs to the k-th class.

[0146] is the probability that the model predicts that the i-th sample belongs to the k-th class

[0147] D. Model Deployment

[0148] After training is completed, a real-time processing pipeline is built to collect video data using a camera and perform face detection, alignment, feature extraction, and classification.

[0149] A fall prediction model for the elderly is constructed based on the logistic regression model, the linear combination of the fall classification feature data, and the Sigmoid function, as follows:

[0150] Assign an initial weight to each fall classification feature data and set an initial bias item;

[0151] Determine the structure of the logistic regression model, that is, the linear combination of features plus the Sigmoid function, to obtain the elderly falls prediction model.

[0152] Fall prediction model training for the elderly

[0153] N is the number of elderly fall cases in the data set. p is the label (0,1) of whether the elderly falls in the ith case. i is the probability of predicting that the i-th case is a fall:

[0154] p i =σ(X i β+b)

[0155] σ is the Sigmoid function, which maps the linear output of the model to the (0,1) interval so that it can represent probability.

[0156] Adjust β and b through gradient descent to minimize the value of the loss function. The weight gradient is calculated as follows:

[0157]

[0158] The bias gradient is calculated as follows

[0159]

[0160] After the above training process, the optimal weight is obtained for calculating the probability of falling risk.

[0161] After obtaining the probability calculation of the risk of falling for the elderly, this probability value is integrated into an intelligent reminder system that can automatically notify the elderly and their caregivers.

[0162] Low risk threshold: p<0.3, no special action required.

[0163] Medium risk threshold: 0.3≤p<0.6 will send a reminder and it is recommended to increase the monitoring frequency.

[0164] High risk threshold: p ≥ 0.6 immediately sends an alert to healthcare professionals or family members, who may need urgent preventive measures.

[0165] These rules are encoded in an intelligent system that continuously collects data from home cameras and smart bracelets, running predictive models to update the assessment of fall risk. Alerts and reminders can be sent through multiple communication channels such as smartphone app push, SMS or emergency call system. To enhance the user experience, the system provides a user-friendly interface that allows the elderly and caregivers to easily view risk assessment and alert information, and a feedback mechanism to provide feedback on the relevance and accuracy of the alert, helping to optimize the performance of the predictive model.

[0166] The system allows users (seniors, medical staff, or family members) to rate each alert, and the feedback can be a simple binary response (such as "relevant / not relevant" or "accurate / inaccurate"). The collected feedback data can be represented as a binary variable:

[0167] If the user believes that the i-th alert is accurate;

[0168] If the user believes that the i-th alert is inaccurate;

[0169] Use proportional estimation to measure the accuracy of the alarm. Based on the feedback data, adjust the alarm threshold or other parameters of the system to optimize the accuracy. For example, if the accuracy is lower than a certain target value (such as 75%), it may be necessary to adjust the parameters in the risk assessment model or improve the feature processing method. Based on the feedback collected, use the optimization algorithm to adjust the model parameters:

[0170] β,b=argmin β,b -logL(β,b|X,f)

[0171] Find a set of parameters (weight β and bias b) so that given the input feature X and user feedback f, the probability error between the model prediction and the actual observed data is minimized.

[0172] And comprehensive assessment of fall risk

[0173] 1) Using the pre-trained model obtained in the model training step, calculate the user's fall risk value.

[0174] 2) Based on the fall risk value obtained by the model and the risk threshold range, users are stratified according to their fall risk.

[0175] 3) After building the fall risk prediction model and completing the training, the present invention further applies R 2 (R-squared) value is used to identify the key factors that affect the risk of falls in the elderly. 2 The R value is an important indicator for measuring the explanatory power of the model. It can effectively evaluate the contribution of each predictor variable to the risk of falls, thereby determining the key influencing factors for each elderly person at the individual level. The R value of each predictor variable was calculated by the structural equation model. 2 Value. 2 The specific calculation process of the value is as follows:

[0176] A. Data standardization: Standardize the collected static data (such as medical records) and dynamic data (such as physical signs, movements, expressions, etc.) to eliminate the impact of different dimensions. The formula is as follows:

[0177]

[0178] Among them, represents the standardized data, represents the original observation value, and is the mean and standard deviation of variable j respectively.

[0179] B. Construction and fitting of structural equation model: Based on the fall risk prediction model, a structural equation model including various observed variables such as age, previous fall history, psychological state, etc. and the explained variable, i.e., fall risk, was constructed. Specifically, a hybrid method was used, i.e., reflective model and formative model.

[0180] C. Calculate R 2 Value: Calculate the R value of each observed variable on the risk of falls through the SEM model. 2 The value is as follows:

[0181]

[0182] SStot is the total sum of squares, SS res is the residual sum of squares.

[0183] This application can realize the intelligent identification of abnormal environmental conditions and the push of early warning information. The designed system can monitor the environmental conditions in the nursing home in real time. Once an abnormal environmental condition that may cause a fall is identified, the system will immediately push fall warning information to users in the target area or all users in the nursing home, thereby promptly reminding the elderly and caregivers to pay attention to potential risks and take effective measures to prevent the occurrence of falls.

[0184] It should be noted that, in the present invention, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0185] The above embodiments are merely examples of the present invention and do not limit the protection scope of the present invention. All designs that are the same or similar to the present invention fall within the protection scope of the present invention.

Claims

1. An intelligent early warning method for elderly people falling in nursing homes, characterized by: The following steps are involved: Obtain a known elderly person's fall data set, process it and extract features, and obtain fall classification feature data, wherein the known elderly person's fall data set includes: known personal static data, known personal dynamic data, and known environmental dynamic data; A fall prediction model for the elderly was constructed based on the logistic regression model, the linear combination of the fall classification feature data, and the Sigmoid function; Input the fall feature classification data into the elderly fall prediction model, train the weights in the elderly fall prediction model to minimize the value of the loss function, and obtain an optimized elderly fall prediction model; Obtaining personal static data, personal dynamic data and environmental dynamic data of the elderly to form a data set; preprocessing the data set and performing feature extraction to obtain the feature data set; classifying the feature data set according to the data type to obtain the classified feature data set; The classification feature data set is input into the optimized elderly fall prediction model to obtain the elderly fall prediction result, and the elderly fall prediction result is compared with the preset risk threshold range. If it is less than the preset risk threshold range, no alarm is issued; if it is within the preset risk threshold range, a warning reminder is issued and the detection rate is increased; if it is greater than the preset risk threshold range, an alarm is issued.

2. The intelligent early warning method for elderly people falling in a nursing home as claimed in claim 1 is characterized in that: The method of constructing an elderly fall prediction model based on a logistic regression model, a linear combination of categorical data, and a Sigmoid function includes the following steps: Assign an initial weight to each fall classification feature data and set an initial bias item; Determine the structure of the logistic regression model, that is, the linear combination of features plus the Sigmoid function, to obtain the elderly falls prediction model.

3. The intelligent early warning method for elderly people falling in a nursing home as claimed in claim 1 is characterized in that: The step of inputting the fall feature classification data into the elderly fall prediction model, training the weights in the elderly fall prediction model to minimize the value of the loss function, and obtaining an optimized elderly fall prediction model comprises the following steps: Calculate the linear combination of features and weights: z = w1x1 + w2x2 + ... + w n *x n +b; and convert the output of the linear combination into a probability prediction by applying the Sigmoid function: h(x) = 1 / (1+e^(-z)); The cross entropy loss function is used to calculate the loss between the predicted probability and the actual label: L(y,h(x))=-y*log(h(x))-(1-y)*log(1-h(x)), where y is the actual label; By calculating the gradient of the loss function for each weight and bias; updating the weights and biases of the elderly fall prediction model parameters according to the gradient descent algorithm, Where α is the learning rate, and the loss function is minimized to obtain an optimized elderly fall prediction model.

4. The intelligent early warning method for elderly people falling in a nursing home scenario as claimed in claim 1 is characterized in that: The individual static information data of the elderly includes numerical data and categorical data. The numerical data includes age, mental health status, Barthel index, triglyceride index, and fasting blood sugar; the categorical data includes the type of disease, medication history, fall history, sleep disorder, and mental cognitive disorder. The personal dynamic data includes image data and non-image data, wherein the image data is facial change data; the non-image data includes: acceleration of joint movements, accumulated exercise time, heart rate, oxygen saturation, and blood pressure. The environmental dynamic data includes light data, humidity data and obstacle detection data.

5. The intelligent early warning method for elderly people falling in a nursing home as claimed in claim 4 is characterized in that: The preprocessing of the data set and feature extraction to obtain the feature data set includes the following steps: Identifying and processing missing data, detecting and processing outliers in the data set, and obtaining a cleaned data set; Performing heat coding on the categorical data and standardizing the numerical data in the cleaned data set to obtain a converted data set; The converted data set is used for selecting features related to fall prediction using correlation analysis and information gain method, and the features are screened using stepwise regression analysis to obtain a feature data set.

6. The intelligent early warning method for elderly people falling in a nursing home as claimed in claim 1 is characterized in that: The preset risk threshold range is: The risk value is less than 0.3, which is the low risk threshold; The risk value is between 0.3 and 0.6, which is the medium risk threshold; The risk value is greater than 0.6, which is the high risk threshold.

7. The intelligent early warning system for elderly people falling in nursing homes is characterized by: include: The data acquisition module is used to acquire the static data and dynamic data of the elderly, including personal static data, personal dynamic data and environmental dynamic data, to form a data set; The data processing module preprocesses the data set and performs feature extraction to obtain a feature data set, and classifies the feature data set according to the data type to obtain a classified feature data set; According to the classification feature data set, the linear combination of classification data is obtained. The model building module builds a fall prediction model for the elderly based on the logistic regression model, the linear combination of classification data, and the Sigmoid function; obtains a known fall data set of the elderly, processes it, and extracts features to obtain the fall classification feature data, inputs the fall feature classification data into the fall prediction model for the elderly, trains the weights in the fall prediction model for the elderly, minimizes the value of the loss function, and obtains an optimized fall prediction model for the elderly; The fall prediction module inputs the classification feature data set into the optimized elderly fall prediction model to obtain the elderly fall prediction results, and compares the elderly fall prediction results with the preset risk threshold range. If it is less than the preset risk threshold range, no alarm is issued. If it is within the preset risk threshold range, an early warning reminder is issued and the detection rate is increased. If it is greater than the preset risk threshold range, an alarm is issued.

Images