Human body composition parameter self-adaptive calibration method and system based on intelligent body fat scale

Abstract

The invention relates to the technical field of intelligent body fat measurement, and provides a body composition parameter adaptive calibration method and system based on an intelligent body fat scale. The method comprises the following steps: acquiring real-time weight data and real-time bioelectrical impedance signals of a user and real-time temperature and humidity data of a current environment; based on a preset bioelectrical impedance model, in combination with the real-time weight data, the real-time bioelectrical impedance signal and the real-time temperature and humidity data, initial body composition parameters are calculated and obtained; performing feature extraction on the calibration training data set according to a preset adaptive calibration algorithm to construct a real-time calibration model according to the adaptive calibration algorithm; and inputting the currently collected real-time weight data, the real-time bioelectrical impedance signal and the real-time environmental parameters to a real-time calibration model, outputting calibration correction parameters, and performing adaptive calibration on the initial human body composition parameters according to the calibration correction parameters to obtain target human body composition parameters. According to the invention, a systematic solution is provided for accurate measurement of the intelligent body fat scale.

Description

Technical Field

[0001] The present application relates to the technical field of intelligent body fat measurement, and in particular to a method and system for adaptively calibrating human body composition parameters based on an intelligent body fat scale. Background Art

[0002] Smart body fat scales, which use bioelectrical impedance (BIA) to measure body composition (such as body fat percentage and muscle mass), have been widely used. This method utilizes the conductivity of body fluids, applies a weak current to measure the impedance signal, and then uses mathematical models to estimate body composition parameters. However, existing technologies have the following significant drawbacks:

[0003] Inadequate correction for environmental factors: Traditional methods rely solely on fixed bioelectrical impedance models and fail to fully consider the impact of real-time environmental parameters (such as temperature and humidity) on measurement results. For example, temperature changes can cause fluctuations in body fluid ion concentrations, while humidity changes can affect skin surface resistance, both of which can introduce measurement bias. However, existing technologies lack dynamic environmental compensation mechanisms.

[0004] Inefficient use of historical data: Existing calibration methods often rely on static models or fixed-cycle calibration, failing to differentiate between the temporal relevance of historical data and user habits. For example, calibration data closer to the current moment typically better reflects the user's current physical condition, but traditional methods don't assign higher weight to it. Furthermore, the differences in calibration needs between high- and low-frequency user scenarios are ignored, resulting in the calibration model's inability to adapt to actual usage scenarios.

[0005] The calibration algorithm lacks adaptability: In existing technologies, the calibration of body composition parameters mostly relies on fixed formulas or simple linear regression, which makes it difficult to capture the complex nonlinear relationship between weight fluctuations, environmental parameters and body composition. Especially during long-term use, the cumulative errors caused by changes in individual physical condition or equipment aging cannot be effectively corrected.

[0006] Due to the above problems, the measurement accuracy of existing smart body fat scales decreases significantly when used in complex environments (such as seasonal changes, humidity differences) or for long periods of time, and cannot meet users' needs for real-time and accurate monitoring of human body composition parameters. Summary of the Invention

[0007] This application provides a method and system for adaptive calibration of human body composition parameters based on an intelligent body fat scale, aiming to solve the problems that the measurement accuracy of existing intelligent body fat scales decreases significantly when used in complex environments (such as seasonal changes, humidity differences) or for long-term use, and cannot meet users' needs for real-time and accurate monitoring of human body composition parameters.

[0008] In a first aspect, an embodiment of the present application provides a method for adaptively calibrating human body composition parameters based on an intelligent body fat scale, which is applied to an intelligent body fat scale; the method comprises:

[0009] Obtain the user's real-time weight data, real-time bioelectrical impedance signal and real-time temperature and humidity data of the current environment;

[0010] Based on a preset bioelectrical impedance model, the real-time weight data, real-time bioelectrical impedance signal, and real-time temperature and humidity data are combined to calculate and obtain initial body composition parameters, wherein the initial body composition parameters include body fat percentage, muscle mass, and body water percentage; and based on historical weight data, historical bioelectrical impedance signal, historical environmental parameters, and preset standard body composition parameters within a historical calibration period, a calibration training data set is formed;

[0011] Performing feature extraction on the calibration training data set according to a preset adaptive calibration algorithm, the adaptive calibration algorithm including a dynamic weight allocation model based on time series, the dynamic weight allocation model automatically adjusting the weight coefficient of historical data according to the time distance between the calibration training data set and the current calibration moment, and correcting the weight coefficient in combination with the user's historical usage frequency to construct a real-time calibration model according to the adaptive calibration algorithm;

[0012] The currently collected real-time weight data, real-time bioelectrical impedance signal, and real-time environmental parameters are input into the real-time calibration model, and calibration correction parameters are output to adaptively calibrate the initial body composition parameters according to the calibration correction parameters to obtain target body composition parameters.

[0013] In some embodiments, the bioelectrical impedance model has a built-in basic calculation formula, which is constructed based on the conductive properties of human body fluids, wherein real-time temperature data is used to correct the influence of body fluid ion concentration on impedance, and real-time humidity data is used to correct the environmental correlation of skin surface resistance; the preset bioelectrical impedance model is combined with the real-time weight data, real-time bioelectrical impedance signal and real-time temperature and humidity data to calculate and obtain initial human body composition parameters, including: dividing the human body composition calculation interval according to the real-time weight data through the bioelectrical impedance model, combining the frequency value of the corrected bioelectrical impedance signal and the temperature and humidity data, and estimating the initial values ​​of body fat percentage, muscle mass and body water percentage through hierarchical calculation logic.

[0014] In some embodiments, the historical calibration period covers a preset time range, and the historical weight data, historical bioelectrical impedance signals, and historical environmental parameters are stored in order of collection time and associated with the user's unique identifier; the preset standard body composition parameters are obtained through an external calibration reference device, and the external calibration reference device includes a medical body composition analyzer; the historical weight data, historical bioelectrical impedance signals, historical environmental parameters and preset standard body composition parameters within the historical calibration period are used to form a calibration training data set, including: aligning the historical weight data, historical bioelectrical impedance signals, historical environmental parameters under the same user identifier with the standard body composition parameters at the corresponding moment by timestamp to generate a structured calibration training data set.

[0015] In some embodiments, the feature extraction of the calibration training data set is performed according to a preset adaptive calibration algorithm, including: performing noise filtering on the bioelectrical impedance signal in the calibration training data set, and extracting signal feature values ​​using a sliding window method; normalizing the environmental parameters to generate a joint influencing factor of temperature and humidity; combining the changing trend of historical weight data to extract time series features such as weight fluctuation frequency and amplitude, and forming a multidimensional feature vector set containing signal features, environmental features and weight features.

[0016] In some embodiments, the weight coefficient of historical data is automatically adjusted according to the distance between the time of the calibration training data set and the current calibration moment, including: establishing a time decay function, giving a higher initial weight to historical data closer to the current calibration moment, and reducing the weight of historical data farther away according to a preset decay rate; the decay rate of the time decay function is dynamically adjusted according to the user's historical calibration frequency. When the user's continuous calibration interval is less than a preset threshold, the decay rate is automatically reduced to retain the reference value of historical data over a longer period.

[0017] In some embodiments, the weight coefficient is corrected in combination with the user's historical usage frequency, including: counting the number of times the user uses the body fat scale in a historical period to generate a usage frequency index; when the usage frequency index is higher than a preset threshold, an additional gain is added to the weight coefficient of the historical data used for the last three times, and the gain amplitude is positively correlated with the usage frequency index; when the usage frequency index is lower than the preset threshold, the weight coefficient of the historical data calibrated for the first time is retained to avoid excessive attenuation of historical data in low-frequency usage scenarios.

[0018] In some embodiments, the real-time calibration model adopts a long short-term memory neural network architecture, the input layer receives a multi-dimensional feature vector, and the hidden layer learns the long-term dependency of human body composition parameters on environmental parameters and weight changes through a time gating mechanism; the real-time calibration model is constructed according to the adaptive calibration algorithm, including: using the error between the external calibration reference data and the model output as a reward signal, dynamically adjusting the connection weights of the neural network, so that the real-time calibration model can adaptively generate environmental compensation factors and weight fluctuation compensation factors.

[0019] In some embodiments, the input of currently collected real-time weight data, real-time bioelectrical impedance signal, and real-time environmental parameters to the real-time calibration model and the output of calibration correction parameters include: converting the real-time collected data into a feature vector isomorphic to the calibration training data set, and inputting the feature vector into the trained real-time calibration model; the real-time calibration model generates environmental compensation parameters and weight compensation parameters through forward propagation calculation, wherein the environmental compensation parameters correct for the interference of the current temperature and humidity on the bioelectrical impedance signal, and the weight compensation parameters compensate for the deviation in human body composition estimation based on the difference between the real-time weight and the historical average weight, so as to constitute the calibration correction parameters.

[0020] In some embodiments, the initial human body composition parameters are adaptively calibrated according to the calibration correction parameters to obtain target human body composition parameters, including: setting calibration correction rules for body fat percentage, muscle mass, and body water percentage respectively, and the calibration correction rules are designed with different correction logics according to the physical characteristics of the human body composition parameters, including: using multiplication correction for body fat percentage, adjusting the adipose tissue conductivity coefficient of the bioelectrical impedance model through environmental compensation parameters; using addition correction for muscle mass, compensating for weight estimation deviation through weight compensation parameters; forming the corrected target human body composition parameters.

[0021] In a second aspect, the present application provides a human body composition parameter adaptive calibration system based on an intelligent body fat scale, which is applied to an intelligent body fat scale; the system comprises:

[0022] A data acquisition unit is used to obtain the user's real-time weight data, real-time bioelectrical impedance signal and real-time temperature and humidity data of the current environment;

[0023] a composition calculation unit for calculating and obtaining initial body composition parameters based on a preset bioelectrical impedance model and in combination with the real-time weight data, the real-time bioelectrical impedance signal, and the real-time temperature and humidity data, the initial body composition parameters including body fat percentage, muscle mass, and body water percentage; and forming a calibration training data set based on historical weight data, historical bioelectrical impedance signals, historical environmental parameters, and preset standard body composition parameters within a historical calibration period;

[0024] a feature extraction unit, configured to extract features from the calibration training dataset according to a preset adaptive calibration algorithm, wherein the adaptive calibration algorithm includes a dynamic weight allocation model based on a time series, wherein the dynamic weight allocation model automatically adjusts a weight coefficient of historical data according to the time distance between the calibration training dataset and the current calibration moment, and modifies the weight coefficient based on the user's historical usage frequency, so as to construct a real-time calibration model according to the adaptive calibration algorithm;

[0025] The target acquisition unit is used to input the currently collected real-time weight data, real-time bioelectrical impedance signal, and real-time environmental parameters into the real-time calibration model, and output calibration correction parameters to adaptively calibrate the initial body composition parameters according to the calibration correction parameters to obtain target body composition parameters.

[0026] The embodiments of the present application provide a method and system for adaptive calibration of human composition parameters based on an intelligent body fat scale. The method deeply integrates real-time weight data, bioelectrical impedance signals, and environmental temperature and humidity data, and constructs a bioelectrical impedance model that is closer to the actual human body's conductive characteristics by dynamically correcting the environmental correlation of body fluid ion concentration and skin resistance. A dynamic weight distribution model based on time series is designed to automatically adjust the weight coefficient according to the distance of historical data from the current calibration moment, and to make corrections based on the user's usage frequency, so that recent data and high-frequency scene data have higher reference value, significantly improving the adaptability of the calibration model to individual changes in users. A real-time calibration model can be constructed using intelligent algorithms such as long short-term memory neural networks (LSTMs). By learning the long-term dependency between environmental parameters, weight changes, and human composition parameters, environmental compensation factors and weight fluctuation compensation factors are automatically generated to achieve nonlinear dynamic calibration of the initial measurement results.

[0027] By combining real-time temperature and humidity data to correct the bioelectrical impedance model, the interference of environmental factors on the conductive properties of body fluids and skin resistance can be effectively reduced, and the measurement errors of parameters such as body fat percentage and muscle mass can be reduced compared with traditional methods; through the dynamic weight distribution of time decay function and usage frequency correction, it is ensured that the calibration model gives priority to the use of recent valid data and high-frequency scenario data, avoiding excessive attenuation of historical data during low-frequency use or interference from invalid data during high-frequency use, and improving the model's real-time response ability to individual changes of users; through the real-time calibration model based on LSTM neural network, it can capture the complex nonlinear relationship between human body composition parameters and the environment and weight, and automatically generate targeted compensation parameters, which solves the error accumulation problem during long-term use of traditional fixed algorithms and achieves continuous optimization of measurement accuracy.

[0028] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0030] Figure 1 This is a schematic flow chart of the steps of a method for adaptively calibrating human body composition parameters based on an intelligent body fat scale provided in one embodiment of the present application;

[0031] Figure 2 This is a schematic diagram of the structure of the smart body fat scale provided in one embodiment of the present application;

[0032] Figure 3 This is a schematic block diagram of a human body composition parameter adaptive calibration system based on an intelligent body fat scale provided in one embodiment of the present application;

[0033] Figure 4 This is a schematic block diagram of the structure of the smart body fat scale provided in one embodiment of the present application.

[0034] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. DETAILED DESCRIPTION

[0035] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0036] The flowcharts shown in the accompanying drawings are for illustrative purposes only and do not necessarily include all contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may be decomposed, combined, or partially merged, so the actual execution order may vary depending on the actual situation.

[0037] It should be understood that, in order to clearly describe the technical solutions of the embodiments of the present invention, in the embodiments of the present invention, terms such as "first" and "second" are used to distinguish between identical or similar items having substantially the same functions and effects. Those skilled in the art will understand that terms such as "first" and "second" do not limit the quantity or order of execution, and that terms such as "first" and "second" do not necessarily define differences.

[0038] It should be understood that the terms used in this specification are only for the purpose of describing specific embodiments and are not intended to limit the present application. As used in this specification and the appended claims, the singular forms "a", "an", and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0039] It will also be understood that the term "and / or" as used in this specification and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0040] The following describes some embodiments of the present application in detail with reference to the accompanying drawings. In the absence of conflict, the following embodiments and features therein may be combined with each other.

[0041] Smart body fat scales, which use bioelectrical impedance (BIA) to measure body composition (such as body fat percentage and muscle mass), have been widely used. This method utilizes the conductivity of body fluids, applies a weak current to measure the impedance signal, and then uses mathematical models to estimate body composition parameters. However, existing technologies have the following significant drawbacks:

[0042] Inadequate correction for environmental factors: Traditional methods rely solely on fixed bioelectrical impedance models and fail to fully consider the impact of real-time environmental parameters (such as temperature and humidity) on measurement results. For example, temperature changes can cause fluctuations in body fluid ion concentrations, while humidity changes can affect skin surface resistance, both of which can introduce measurement bias. However, existing technologies lack dynamic environmental compensation mechanisms.

[0043] Inefficient use of historical data: Existing calibration methods often rely on static models or fixed-cycle calibration, failing to differentiate between the temporal relevance of historical data and user habits. For example, calibration data closer to the current moment typically better reflects the user's current physical condition, but traditional methods don't assign higher weight to it. Furthermore, the differences in calibration needs between high- and low-frequency user scenarios are ignored, resulting in the calibration model's inability to adapt to actual usage scenarios.

[0044] The calibration algorithm lacks adaptability: In existing technologies, the calibration of body composition parameters mostly relies on fixed formulas or simple linear regression, which makes it difficult to capture the complex nonlinear relationship between weight fluctuations, environmental parameters and body composition. Especially during long-term use, the cumulative errors caused by changes in individual physical condition or equipment aging cannot be effectively corrected.

[0045] Due to the above problems, the measurement accuracy of existing smart body fat scales decreases significantly when used in complex environments (such as seasonal changes, humidity differences) or for long periods of time, and cannot meet users' needs for real-time and accurate monitoring of human body composition parameters.

[0046] To resolve the above, please refer to Figure 1 The embodiment of the present application provides a method for adaptive calibration of human body composition parameters based on an intelligent body fat scale, which is applied to Figure 2 At the same time, it should be noted that each piece of information involved in the method provided in this application is extracted with the authorization of the relevant user and in compliance with relevant regulations, and will not infringe on the user's privacy.

[0047] The provided method for adaptively calibrating human body composition parameters based on an intelligent body fat scale includes steps S101 to S104. Detailed descriptions are as follows:

[0048] Step S101: Acquire the user's real-time weight data, real-time bioelectrical impedance signal, and real-time temperature and humidity data of the current environment.

[0049] Specifically, this step uses the built-in sensors of the smart body fat scale to collect three types of key data in real time: User real-time weight data: The user's weight when standing is measured through a pressure sensor, which serves as the basic parameter for body composition calculation. Real-time bioelectrical impedance signal: A weak alternating current (within a safe range, usually ≤50kHz, 10-100μA) is applied to the human body through electrodes to measure the body's electrical impedance (including resistance R and capacitance Xc), reflecting the conductivity of body fluid distribution and body composition. Real-time temperature and humidity data: The temperature (unit: ℃) and relative humidity (unit: %RH) of the current measurement environment are collected through environmental sensors (such as the temperature and humidity sensor module) for subsequent environmental factor compensation.

[0050] The weight sensor uses a high-precision pressure strain gauge integrated into the bottom of the scale, supporting dynamic filtering to eliminate interference from standing shaking. The bioelectrical impedance measurement module includes a constant current source circuit, a signal amplification circuit, and an analog-to-digital converter (ADC). The electrodes are usually designed as metal contact points on the surface of the scale, forming a current loop when the user stands. The temperature and humidity sensors are integrated inside or on the edge of the scale to avoid user obstruction and ensure real-time data (sampling frequency ≥ 1 time / second). The three types of data are synchronously collected by an embedded microcontroller (such as an MCU), with timestamps accurate to the millisecond level, ensuring the temporal and spatial consistency of the data in subsequent calculations.

[0051] Incorporating environmental parameters (temperature and humidity) into the real-time measurement system overcomes the flaw of traditional methods that ignore environmental influences and provides data support for subsequent dynamic compensation. Through hardware synchronization acquisition, parameter mismatches caused by data delays are avoided, ensuring that the initial body composition parameters calculated later are closer to the actual state.

[0052] Step S102. Based on a preset bioelectrical impedance model, combined with the real-time weight data, real-time bioelectrical impedance signal, and real-time temperature and humidity data, initial body composition parameters are calculated and obtained. The initial body composition parameters include body fat percentage, muscle mass, and body water percentage. A calibration training data set is formed based on historical weight data, historical bioelectrical impedance signal, historical environmental parameters, and preset standard body composition parameters within a historical calibration period.

[0053] Specifically, the initial body composition parameters are calculated based on a preset bioelectrical impedance model (such as the classic height, weight, and electrical impedance correlation formula, or an improved multivariate regression model), combined with real-time weight (m), electrical impedance (Z), temperature (T), and humidity (H) to calculate the initial body fat percentage (BF%), muscle mass (MM), and body water percentage (TBW%). For example, the model may be formalized as follows:

[0054] BF%=f1(m,Z,T,H),MM=f2(m,Z,T,H),TBW%=f3(m,Z,T,H); among them, temperature affects the mobility of body fluid ions, and humidity affects the skin contact resistance, both of which are embedded in the model as variables.

[0055] The calibration training dataset is constructed by collecting historical data from the past calibration period (e.g., the last 30 days), including historical weight data (mhist), historical electrical impedance signals (Zhist), historical environmental parameters (Thist, Hhist); and preset standard body composition parameters (such as body fat percentage and muscle mass from the user's recent physical examination report, or reference values ​​calibrated by professional equipment (such as DXA), recorded as BF%std, MMstd, TBW%std). Each data record is called a sample (mhist, Zhist, Thist, Hhist, BF%std, MMstd, TBW%std), forming the training dataset for supervised learning.

[0056] Standard parameters are acquired through user input: manual entry of physical examination data or measurements from third-party devices via the companion app. Automatic calibration mode periodically prompts users to calibrate using professional equipment (such as gym fitness monitors) and automatically synchronizes standard values. Data storage utilizes a time series database (such as InfluxDB) to store historical data, indexed by timestamp, to facilitate calculation of time distances during subsequent dynamic weight allocation.

[0057] Introducing temperature and humidity variables into the initial calculation quantifies the environmental impact on electrical impedance for the first time, compared to traditional fixed models, and reduces the inherent bias of static models. By pairing historical data with standard values, a labeled dataset is constructed for calibration, providing training samples for subsequent adaptive models and addressing the lack of dynamic calibration data found in traditional methods.

[0058] Step S103. Feature extraction is performed on the calibration training data set according to a preset adaptive calibration algorithm. The adaptive calibration algorithm includes a dynamic weight allocation model based on time series. The dynamic weight allocation model automatically adjusts the weight coefficient of historical data according to the distance between the calibration training data set and the current calibration moment, and corrects the weight coefficient in combination with the user's historical usage frequency to construct a real-time calibration model according to the adaptive calibration algorithm.

[0059] Specifically, feature extraction is performed by normalizing the input features (weight, electrical impedance, temperature, humidity) and output labels (standard body composition parameters) in the calibration training dataset, and extracting time-related features (such as the time difference Δt between the data collection moment and the current calibration moment) and user usage frequency features (such as the number of measurements f in the past 7 days).

[0060] The dynamic weight allocation model includes: Time decay weight: an exponential decay function is used to give higher weight to recent data. The formula is: The closer the distance is to the current state (the smaller Δt), the higher the weight, reflecting the time-varying nature of the user's physical condition. Frequency correction adjusts the weight based on the user's historical usage frequency ff. Frequent users (e.g., those measuring daily) receive a higher weight for recent data (e.g., multiplied by a correction factor kf > 1), while infrequent users (e.g., those measuring weekly) receive a more gradual weight reduction to avoid interference from occasional data.

[0061] The adaptive calibration algorithm uses a machine learning model (such as gradient boosting tree GBDT, deep neural network DNN) or a time series model (such as LSTM), takes weighted training data as input, and constructs a nonlinear mapping relationship: (BF%cal,MMcal,TBW%cal)=g(m,Z,T,H;θ,w); where θ is the model parameter and w is the dynamic weight. By minimizing the weighted mean square error Train models to capture the complex nonlinear relationships among body weight, environmental parameters, and body composition.

[0062] Dynamic weight adjustment recalculates Δt and f for all historical data based on the current time before each calibration, updating weights in real time to avoid the lag of static weights. Considering the embedded computing resource limitations of smart body fat scales, model compression techniques (such as pruning and quantization) are used to ensure real-time inference efficiency.

[0063] Dynamic weighting prioritizes recent data, addressing the inefficiency of traditional methods that treat historical data equally, making the model more tailored to the user's current physical condition. Frequency-corrected weighting distinguishes between high-frequency (requiring fine-grained calibration) and low-frequency (requiring noise reduction) scenarios, improving the generalization of the calibrated model and avoiding overfitting to incidental data. This method overcomes the limitations of traditional linear regression and can capture the complex interactions between temperature, humidity, and electrical impedance (for example, humidity affects skin resistance, which indirectly affects body water content), enhancing error correction capabilities during long-term use.

[0064] Step S104: Input the currently collected real-time weight data, real-time bioelectrical impedance signal, and real-time environmental parameters into the real-time calibration model, and output the calibration correction parameters to adaptively calibrate the initial body composition parameters according to the calibration correction parameters to obtain the target body composition parameters.

[0065] Specifically, the real-time data (mreal, Zreal, Treal, Hreal) collected in step S101 is input into the real-time calibration model constructed in step S103, and calibration correction parameters (such as body fat percentage correction coefficient δBF and muscle mass offset ΔMM) are output to adjust the initial parameters calculated in step S102:

[0066] BF%target=BF% initial+δBF,MMtarget=MMinitial+ΔMM; the correction logic can be based on the absolute deviation or relative proportion of the model output, depending on the loss function design during training.

[0067] The microcontroller calls trained model parameters, performs forward propagation calculations on real-time data, and outputs correction values, keeping processing latency within 50ms to ensure a guaranteed user experience. It supports scheduled calibration (such as weekly automatic calibration) and event-triggered calibration (such as forced calibration when user measurement data fluctuations exceed a threshold for three consecutive times), ensuring real-time updates of model parameters.

[0068] A real-time calibration model compensates for measurement deviations caused by environmental changes (such as seasonal temperature fluctuations and indoor humidity differences) and device aging, addressing the long-term accuracy degradation of traditional methods. The model is trained based on individual historical data rather than a universal fixed formula, adapting to variations in body composition (e.g., muscle building and fat loss) for different users, improving personalized monitoring accuracy.

[0069] Steps S101-S104 collect temperature and humidity data in real time and embed them into a calibration model to quantify the environmental impact on electrical impedance, breaking the limitations of traditional methods that rely on fixed models and ignore the environment. Dynamic weighting enables on-demand allocation of historical data, prioritizing recent data and high-frequency scenario data. This ensures that calibration is more tailored to the current state and reduces interference from invalid data. A nonlinear model and dynamic update mechanism effectively address complex errors associated with long-term use, such as weight fluctuations and device aging. This creates a closed loop of "data acquisition-model training-real-time correction," ensuring consistent and robust measurement accuracy.

[0070] In some embodiments, the bioelectrical impedance model has a built-in basic calculation formula, which is constructed based on the conductive properties of human body fluids, wherein real-time temperature data is used to correct the influence of body fluid ion concentration on impedance, and real-time humidity data is used to correct the environmental correlation of skin surface resistance; the preset bioelectrical impedance model is combined with the real-time weight data, real-time bioelectrical impedance signal and real-time temperature and humidity data to calculate and obtain initial human body composition parameters, including: dividing the human body composition calculation interval according to the real-time weight data through the bioelectrical impedance model, combining the frequency value of the corrected bioelectrical impedance signal and the temperature and humidity data, and estimating the initial values ​​of body fat percentage, muscle mass and body water percentage through hierarchical calculation logic.

[0071] This embodiment focuses on the hierarchical calculation logic of the bioelectrical impedance model, and corrects the basic calculation formula through real-time temperature and humidity data to solve the interference of environmental factors on electrical impedance measurement.

[0072] Based on real-time weight data, users are categorized into different body types (e.g., underweight, normal, overweight). Different basic formulas are used for each type to match physiological differences in body composition. As temperature increases, ion mobility increases and electrical impedance decreases, a mapping relationship between temperature and ionic conductivity is established using the Arrhenius equation. High humidity enhances hydration of the skin's stratum corneum, reducing contact resistance, leading to an empirical correction formula for humidity and skin resistance. Body fat percentage, muscle mass, and body water percentage are estimated step by step using the hierarchical relationship of "conductive body fluids → lean body mass → adipose tissue." For example, body water percentage is first calculated using corrected electrical impedance, and then muscle mass and body fat percentage are derived. The World Health Organization (WHO) BMI standard is used for body stratification (e.g., BMI < 18.5 is considered underweight, 18.5 ≤ BMI < 24 is considered normal), with multiple built-in sub-models corresponding to different body types.

[0073] Temperature and humidity correction formulas include: Ionic conductivity correction: σ(T)=σ0*e Ea / (kT) , where σ0 is the conductivity at reference temperature, Ea is the activation energy, and T is the absolute temperature; skin resistance correction: , α is the humidity sensitivity coefficient, which is calibrated through experiments.

[0074] The stratified calculation steps include calculating total body fluid volume (TBW) based on body weight and corrected electrical impedance; estimating muscle mass based on TBW and the high water content of muscle tissue; and calculating fat mass by the difference between body weight and lean body mass (muscle + bone, etc.) to deduce body fat percentage.

[0075] The effects of temperature and humidity on electrical impedance are embedded in the basic calculation formula, addressing the flaw of traditional models that ignore dynamic environmental changes and making the initial parameter calculation more accurate to actual physiological conditions. Formula parameters are adaptively divided into weight ranges, avoiding body shape bias caused by traditional calculation methods. For example, overweight people have a higher proportion of fat tissue, requiring adjustment of the fat conductivity coefficient to improve basic calculation accuracy.

[0076] In some embodiments, the historical calibration period covers a preset time range, and the historical weight data, historical bioelectrical impedance signals, and historical environmental parameters are stored in order of collection time and associated with the user's unique identifier; the preset standard body composition parameters are obtained through an external calibration reference device, and the external calibration reference device includes a medical body composition analyzer; the historical weight data, historical bioelectrical impedance signals, historical environmental parameters and preset standard body composition parameters within the historical calibration period are used to form a calibration training data set, including: aligning the historical weight data, historical bioelectrical impedance signals, historical environmental parameters under the same user identifier with the standard body composition parameters at the corresponding moment by timestamp to generate a structured calibration training data set.

[0077] This embodiment standardizes the construction process of the calibration training dataset to ensure the spatiotemporal consistency of historical data and standard parameters. The core includes: Historical data organization: Historical data is stored according to a preset duration (such as 90 days). Each data item contains the collection timestamp, user ID, weight, electrical impedance, temperature and humidity; Standard parameter acquisition: High-precision human body composition parameters are obtained through external medical equipment (such as DXA dual-energy X-ray absorptiometry analyzer and BodPod body composition analyzer) as labels for supervised learning; Data alignment mechanism: Based on the timestamp, the historical measurement data of the same user is matched with the most recent external calibration data to ensure a one-to-one correspondence between the "measurement time-standard value" of the calibration sample.

[0078] The app reminds users to calibrate their medical devices regularly (e.g., monthly), manually entering or scanning a QR code to synchronize standard data. The app supports Bluetooth connection with third-party body measurement devices to automatically obtain standard parameters such as body fat percentage and muscle mass at the time of calibration. Data storage: A relational database table is used with fields including user_id, timestamp, weight, impedance, temp, humi, bf_std, mm_std, and tbw_std. Uniquely indexed by user_id and timestamp ensures duplicates and traceability. A time tolerance of ±10 minutes is allowed, matching the most recent external calibration data to the corresponding measurement record (e.g., if a user uses a body fat scale at 10:00 AM and measures on a calibrator at 10:05 AM, this is considered a valid pairing).

[0079] Standard parameters obtained through medical equipment offer greater credibility, resolving the issue of traditional calibration relying on user self-reporting or low-precision equipment, and improving the quality of supervisory signals for training data. Timestamp alignment avoids logical errors such as matching outdated standard values ​​with current data (e.g., calibrating current measurements with physical examination data from six months ago), ensuring that training samples reflect the physical condition at the same moment and enhancing the accuracy of model fitting.

[0080] In some embodiments, the feature extraction of the calibration training data set is performed according to a preset adaptive calibration algorithm, including: performing noise filtering on the bioelectrical impedance signal in the calibration training data set, and extracting signal feature values ​​using a sliding window method; normalizing the environmental parameters to generate a joint influencing factor of temperature and humidity; combining the changing trend of historical weight data to extract time series features such as weight fluctuation frequency and amplitude, and forming a multidimensional feature vector set containing signal features, environmental features and weight features.

[0081] This embodiment optimizes the feature engineering of calibration training data and enhances the richness of model input through multidimensional feature extraction, including: electrical impedance signal processing: using a sliding window (e.g., 5 measurement cycles) to perform median filtering or wavelet denoising on the raw electrical impedance signal to extract statistical features such as mean, variance, and peak value; environmental feature generation: after normalizing temperature and humidity (e.g., to [-1, 1]), calculating joint influencing factors (e.g., temperature-humidity product terms, comprehensive environmental indicators after principal component analysis); weight time series features: analyzing the time series of historical weight data to extract dynamic features such as fluctuation frequency (e.g., the number of weight changes per week) and fluctuation amplitude (the difference between the current weight and the 7-day average).

[0082] Signal filtering is achieved by setting the sliding window size to 5 seconds (corresponding to 5 sampling times at 1 Hz). The mean μZ and standard deviation σZ of the electrical impedance signal within the window are calculated each time as signal stability characteristics. A Butterworth low-pass filter is used to remove high-frequency noise (cutoff frequency 1 kHz) and retain the effective signal components of the bioelectrical impedance (usually <50 kHz).

[0083] Environmental characteristics normalization: Temperature normalization: ,in (covering the normal use environment); humidity normalization is similar, ranging from 0% to 100%, generating Hnorm, and the combined factor is Tnorm*Hnorm.

[0084] Weight characteristic calculation: Fluctuation frequency: Count the percentage of days with weight changes exceeding 1% in the past 30 days; Fluctuation amplitude: The absolute difference between the current weight and the 30-day average weight, divided by the 30-day average weight to obtain the relative amplitude.

[0085] Signal filtering reduces random interference during measurement (such as impedance fluctuations caused by slight user movement), extracts more stable eigenvalues, and prevents noisy data from misleading model training. It also generates a temperature-humidity joint factor and weight time series features to capture interactions between environmental parameters (such as how high temperature and humidity synergistically reduce skin resistance) and weight trends (e.g., rapid weight loss may be accompanied by water loss), providing richer input dimensions for subsequent nonlinear models.

[0086] In some embodiments, the weight coefficient of historical data is automatically adjusted according to the distance between the time of the calibration training data set and the current calibration moment, including: establishing a time decay function, giving a higher initial weight to historical data closer to the current calibration moment, and reducing the weight of historical data farther away according to a preset decay rate; the decay rate of the time decay function is dynamically adjusted according to the user's historical calibration frequency. When the user's continuous calibration interval is less than a preset threshold, the decay rate is automatically reduced to retain the reference value of historical data over a longer period.

[0087] This embodiment improves the time weight allocation mechanism and realizes adaptive screening of historical data through dynamic decay rate, including: Time decay function: using exponential decay model to give higher weight to recent data, the basic formula is , where Δt is the number of days from the current data; dynamic adjustment of the decay rate: λ is automatically adjusted based on the user's historical calibration frequency (such as the number of calibrations in the past 30 days). For users with high-frequency calibration (such as weekly calibration), λ is reduced (decay becomes slower) to retain the reference value of longer-term data; for users with low-frequency calibration, λ is increased (decay becomes faster) to focus on recent data.

[0088] Attenuation rate calculation rule: Define the calibration frequency f = the number of calibrations in the past 30 days / 30. When f ≥ 0.3 (high frequency), λ = 0.1; when 0.1 < f < 0.3 (medium frequency), λ = 0.3; when f ≤ 0.1 (low frequency), λ = 0.5. Before each calibration model is built, recalculate Δt and the corresponding weights for all historical data according to the current time to ensure that the weights change dynamically according to the user's usage habits (for example, when the user changes from low frequency to high frequency usage, the attenuation rate is automatically reduced). If the user calibrates for the first time (no historical data), the weights of all data are equal to avoid algorithm failure.

[0089] The physical condition of high-frequency users (such as fitness enthusiasts) changes relatively fast. Retaining long-term data can capture trend changes; the recent data of low-frequency users (such as occasional measurements) is more representative. Dynamically adjusting can avoid "outdated data" dominating the calibration and improve the model generalization ability in different usage scenarios. Through the dynamic attenuation rate, when there are systematic biases caused by device aging (such as the sensor sensitivity decreasing year by year), the long-term accumulated calibration data can still provide effective correction signals, avoiding information loss caused by the traditional fixed attenuation rate.

[0090] In some embodiments, the combination of the user's historical usage frequency to correct the weight coefficient includes: counting the number of times the user uses the body fat scale in the historical period to generate a usage frequency index; when the usage frequency index is higher than the preset threshold, an additional gain is applied to the weight coefficients of the historical data of the last three usages, and the gain amplitude is positively correlated with the usage frequency index; when the usage frequency index is lower than the preset threshold, the weight coefficient of the historical data of the first calibration is retained to avoid excessive attenuation of historical data in low-frequency usage scenarios.

[0091] This embodiment further corrects the weight coefficient in combination with the user's usage frequency, distinguishing the "fine calibration" requirement in the high-frequency scenario and the "noise reduction requirement" in the low-frequency scenario, including: Usage frequency index: Count the number of measurements of the user in the historical period (such as the past 7 days) and normalize it to an index findex in the 0-1 interval; High-frequency scenario gain: When findex > 0.5, multiply the weights of the data of the last 3 measurements by an additional gain factor g = 1 + 0.2 * findex to strengthen the influence of recent high-frequency data; Low-frequency scenario retention: When findex ≤ 0.5, retain the weight of the first calibration data (without attenuation) to avoid calibration model deviation caused by data sparsity in the low-frequency scenario.

[0092] Frequency index calculation: The number of measurements over the past seven days, c, is used, and the frequency index, findex, is equal to c / 7 (maximum value 1, corresponding to daily measurements). Weight correction logic: High-frequency gain: For the three most recent data records, the weight is updated to W' = w * g, where g increases linearly with findex (e.g., g = 1.2 for daily measurements). Low-frequency retention includes: The weight of the first calibration data (the record with the earliest timestamp) is set to a fixed value of 1, which is unaffected by time decay, ensuring that initial calibration data from low-frequency users is not filtered. Data storage enhancements include adding a "Measurement Sequence Number" field to historical data records to identify whether it is the user's first measurement, the last three measurements, etc., making it easier to quickly locate samples requiring correction.

[0093] Recent data from high-frequency users is more likely to reflect current physical status (e.g., daily changes during a weight loss period). Additional gain ensures that the model prioritizes this "high-value data," reducing interference from occasional fluctuations and enabling fine-grained calibration. Low-frequency users have less measurement data, so retaining the initial calibration data (typically collected when the user's status is stable) serves as a baseline to avoid over-reliance on noisy data due to data sparsity and improve the stability of calibration results.

[0094] In some embodiments, the real-time calibration model adopts a long short-term memory neural network architecture, the input layer receives a multi-dimensional feature vector, and the hidden layer learns the long-term dependency of human body composition parameters on environmental parameters and weight changes through a time gating mechanism; the real-time calibration model is constructed according to the adaptive calibration algorithm, including: using the error between the external calibration reference data and the model output as a reward signal, dynamically adjusting the connection weights of the neural network, so that the real-time calibration model can adaptively generate environmental compensation factors and weight fluctuation compensation factors.

[0095] This embodiment uses a long short-term memory (LSTM) neural network to build a real-time calibration model, utilizes a time-gating mechanism to capture long-term dependencies, and optimizes model parameters through reinforcement learning. Specifically, the model architecture design includes: the input layer receives multidimensional feature vectors (signal features, environmental features, and weight features), the hidden layer contains LSTM units, and the output layer generates environmental compensation factors and weight compensation factors; the time-gating mechanism selectively retains historical information through input gates, forget gates, and output gates, such as remembering the impact of long-term environmental changes (seasonal temperature trends) on body water content; and reward signal-driven: the error between external calibration reference data and model output (such as the root mean square error (RMSE)) is used as a reward signal, and network weights are dynamically adjusted through backpropagation to enable the model to adaptively generate compensation factors.

[0096] LSTM parameter configuration: Input layer dimension: consistent with the feature vector dimension (e.g., 10 dimensions); hidden layer: 128 LSTM units, with a dropout rate of 0.2 to prevent overfitting; output layer: 2 neurons, corresponding to the environmental compensation factor (temperature and humidity comprehensive correction value) and the weight compensation factor (weight fluctuation correction value).

[0097] Reward signal calculation: After each external calibration, the RMSE between the model output and the standard value is calculated. If the RMSE is less than a preset threshold (e.g., body fat percentage ±1%), the reward value is +1; otherwise, the reward value is -1. Gating parameters are adjusted using a reinforcement learning algorithm (e.g., Q-learning). Model deployment optimization: On embedded devices, a lightweight LSTM framework (e.g., TensorFlow Lite) is used, and quantization technology (converting 32-bit floating point to 8-bit fixed point) is used to reduce computing power.

[0098] The LSTM gating mechanism can capture the cumulative impact of environmental parameters (such as seasonal temperature changes) and weight trends (such as sustained weight gain over six months) on body composition over several months, addressing the inability of traditional linear models to handle temporal dependencies. Through dynamic optimization of reward signals, the model can autonomously learn from individual differences among users (for example, athletes' muscle mass is more sensitive to weight changes than the average person), generating more precise compensation factors and improving calibration accuracy over long-term use.

[0099] In some embodiments, the input of currently collected real-time weight data, real-time bioelectrical impedance signal, and real-time environmental parameters to the real-time calibration model and the output of calibration correction parameters include: converting the real-time collected data into a feature vector isomorphic to the calibration training data set, and inputting the feature vector into the trained real-time calibration model; the real-time calibration model generates environmental compensation parameters and weight compensation parameters through forward propagation calculation, wherein the environmental compensation parameters correct for the interference of the current temperature and humidity on the bioelectrical impedance signal, and the weight compensation parameters compensate for the deviation in human body composition estimation based on the difference between the real-time weight and the historical average weight, so as to constitute the calibration correction parameters.

[0100] This embodiment standardizes the processing flow of the real-time data input model, ensures that the data format is consistent with the training data set, and clarifies the physical meaning of the compensation parameters, including: data format conversion: the real-time collected weight, electrical impedance, temperature and humidity are extracted according to the feature extraction method of Example 3 to generate a multidimensional feature vector isomorphic to the training set (such as containing the signal mean, temperature normalization value, weight fluctuation amplitude, etc.); compensation parameter generation: environmental compensation parameters: output skin resistance correction value, ion concentration correction coefficient, etc. based on the interference of current temperature and humidity on electrical impedance; weight compensation parameters: output weight estimation deviation, fat mass adjustment factor, etc. based on the difference between real-time weight and historical average weight.

[0101] Feature vector generation process: The real-time electrical impedance signal is filtered using a 5-second sliding window to calculate the mean and standard deviation; the temperature and humidity are normalized and the joint influencing factor is calculated; the difference and relative amplitude between the real-time weight and the average weight of the past 30 days are calculated as the weight fluctuation feature.

[0102] Forward propagation calculation: Input the feature vector into the trained LSTM model (Example 6), and generate the environmental compensation parameter (dimension 1×3, corresponding to the environmental correction values ​​of body fat percentage, muscle mass, and body water percentage) and weight compensation parameter (dimension 1×3, corresponding to the weight-related correction values ​​of each component) through hidden layer calculation.

[0103] Physical meaning of compensation parameters: The skin resistance correction value in the environmental compensation parameter directly acts on the contact resistance term of the bioelectrical impedance model; the lean body mass deviation in the weight compensation parameter is used to adjust the basic formula coefficient for muscle mass calculation.

[0104] Isomorphic feature vectors ensure compatibility between real-time data and training data, avoiding model inference errors caused by inconsistent formats and improving system robustness. Clearly distinguishing the impact of environmental factors and weight changes on body composition makes compensation parameters more physically interpretable (for example, high temperatures increase body water conductivity, and the model outputs negative environmental compensation values ​​to reduce body fat percentage estimation bias), facilitating the design and verification of subsequent calibration rules.

[0105] In some embodiments, the initial human body composition parameters are adaptively calibrated according to the calibration correction parameters to obtain target human body composition parameters, including: setting calibration correction rules for body fat percentage, muscle mass, and body water percentage respectively, and the calibration correction rules are designed with different correction logics according to the physical characteristics of the human body composition parameters, including: using multiplication correction for body fat percentage, adjusting the adipose tissue conductivity coefficient of the bioelectrical impedance model through environmental compensation parameters; using addition correction for muscle mass, compensating for weight estimation deviation through weight compensation parameters; forming the corrected target human body composition parameters.

[0106] This embodiment designs differentiated calibration correction rules based on the physical properties of human body composition to improve the scientific nature of the correction logic, including: multiplication correction for body fat percentage: based on the characteristic that the conductivity of adipose tissue is much lower than that of lean body mass, the conductivity coefficient of adipose tissue is adjusted through environmental compensation parameters (such as σfat=σfat0*(1+δenv)) to indirectly correct the estimated body fat percentage; addition correction for muscle mass: since muscle mass is directly related to the lean body mass in body weight, the deviation in weight estimation is directly compensated through weight compensation parameters (such as Δmm=k*weight fluctuation amplitude, where k is the ratio coefficient of muscle to lean body mass); combined with the environment (temperature affects body fluid distribution) and body weight (changes in lean body mass affect body water), a mixed correction of multiplication and addition is adopted.

[0107] See also Figure 3 As shown, Figure 3 2 is a schematic diagram of the structure of a human body composition parameter adaptive calibration system 200 based on an intelligent body fat scale, provided in an embodiment of the present application. This human body composition parameter adaptive calibration system 200 based on an intelligent body fat scale is used to execute the steps of the human body composition parameter adaptive calibration method based on an intelligent body fat scale, as described in the above embodiments. This human body composition parameter adaptive calibration system 200 based on an intelligent body fat scale can be a single server or a server cluster, or it can be a terminal, such as a handheld terminal, a laptop computer, a wearable device, or a robot.

[0108] like Figure 3 As shown, the human body composition parameter adaptive calibration system 200 based on the intelligent body fat scale includes:

[0109] The data acquisition unit 201 is used to acquire the user's real-time weight data, real-time bioelectrical impedance signal and real-time temperature and humidity data of the current environment;

[0110] a composition calculation unit 202 for calculating and obtaining initial body composition parameters based on a preset bioelectrical impedance model and in combination with the real-time weight data, the real-time bioelectrical impedance signal, and the real-time temperature and humidity data, wherein the initial body composition parameters include body fat percentage, muscle mass, and body water percentage; and forming a calibration training data set based on historical weight data, historical bioelectrical impedance signals, historical environmental parameters, and preset standard body composition parameters within a historical calibration period;

[0111] A feature extraction unit 203 is configured to extract features from the calibration training dataset according to a preset adaptive calibration algorithm, wherein the adaptive calibration algorithm includes a dynamic weight allocation model based on a time series, wherein the dynamic weight allocation model automatically adjusts the weight coefficient of historical data according to the time distance between the calibration training dataset and the current calibration moment, and modifies the weight coefficient based on the user's historical usage frequency to construct a real-time calibration model according to the adaptive calibration algorithm;

[0112] The target acquisition unit 204 is used to input the currently collected real-time weight data, real-time bioelectrical impedance signal, and real-time environmental parameters into the real-time calibration model, and output calibration correction parameters to adaptively calibrate the initial body composition parameters according to the calibration correction parameters to obtain target body composition parameters.

[0113] It should be noted that those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the human body composition parameter adaptive calibration system based on the smart body fat scale and each module described above can refer to the corresponding contents in the various embodiments of the human body composition parameter adaptive calibration method based on the smart body fat scale, and will not be repeated here.

[0114] The above-mentioned method for adaptive calibration of human body composition parameters based on intelligent body fat scale can be implemented in the form of a computer program. The computer program can be used in Figure 3 Run on the device shown.

[0115] See also Figure 4 , Figure 4 This is a schematic block diagram of the structure of the smart body fat scale provided by an embodiment of the present application. The smart body fat scale includes a processor, a memory, and a network interface connected via a device bus, wherein the memory may include a storage medium and an internal memory.

[0116] The storage medium can store an operating device and a computer program. The computer program includes program instructions, which, when executed, can cause the processor to execute any one of the methods for adaptively calibrating human body composition parameters based on the intelligent body fat scale.

[0117] The processor is used to provide computing and control capabilities to support the operation of the entire smart body fat scale.

[0118] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor, the processor can execute any one of the methods for adaptively calibrating human body composition parameters based on the intelligent body fat scale.

[0119] The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art will understand that Figure 4 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the terminal to which the solution of the present application is applied. The specific smart body fat scale may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0120] It should be understood that the processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0121] In one embodiment, the processor is configured to execute a computer program stored in the memory to implement the following steps:

[0122] Obtain the user's real-time weight data, real-time bioelectrical impedance signal and real-time temperature and humidity data of the current environment;

[0123] Based on a preset bioelectrical impedance model, the real-time weight data, real-time bioelectrical impedance signal, and real-time temperature and humidity data are combined to calculate and obtain initial body composition parameters, wherein the initial body composition parameters include body fat percentage, muscle mass, and body water percentage; and based on historical weight data, historical bioelectrical impedance signal, historical environmental parameters, and preset standard body composition parameters within a historical calibration period, a calibration training data set is formed;

[0124] Performing feature extraction on the calibration training data set according to a preset adaptive calibration algorithm, the adaptive calibration algorithm including a dynamic weight allocation model based on time series, the dynamic weight allocation model automatically adjusting the weight coefficient of historical data according to the time distance between the calibration training data set and the current calibration moment, and correcting the weight coefficient in combination with the user's historical usage frequency to construct a real-time calibration model according to the adaptive calibration algorithm;

[0125] The currently collected real-time weight data, real-time bioelectrical impedance signal, and real-time environmental parameters are input into the real-time calibration model, and calibration correction parameters are output to adaptively calibrate the initial body composition parameters according to the calibration correction parameters to obtain target body composition parameters.

[0126] In some embodiments, the bioelectrical impedance model has a built-in basic calculation formula, which is constructed based on the conductive properties of human body fluids, wherein real-time temperature data is used to correct the influence of body fluid ion concentration on impedance, and real-time humidity data is used to correct the environmental correlation of skin surface resistance; the preset bioelectrical impedance model is combined with the real-time weight data, real-time bioelectrical impedance signal and real-time temperature and humidity data to calculate and obtain initial human body composition parameters, including: dividing the human body composition calculation interval according to the real-time weight data through the bioelectrical impedance model, combining the frequency value of the corrected bioelectrical impedance signal and the temperature and humidity data, and estimating the initial values ​​of body fat percentage, muscle mass and body water percentage through hierarchical calculation logic.

[0127] In some embodiments, the historical calibration period covers a preset time range, and the historical weight data, historical bioelectrical impedance signals, and historical environmental parameters are stored in order of collection time and associated with the user's unique identifier; the preset standard body composition parameters are obtained through an external calibration reference device, and the external calibration reference device includes a medical body composition analyzer; the historical weight data, historical bioelectrical impedance signals, historical environmental parameters and preset standard body composition parameters within the historical calibration period are used to form a calibration training data set, including: aligning the historical weight data, historical bioelectrical impedance signals, historical environmental parameters under the same user identifier with the standard body composition parameters at the corresponding moment by timestamp to generate a structured calibration training data set.

[0128] In some embodiments, the feature extraction of the calibration training data set is performed according to a preset adaptive calibration algorithm, including: performing noise filtering on the bioelectrical impedance signal in the calibration training data set, and extracting signal feature values ​​using a sliding window method; normalizing the environmental parameters to generate a joint influencing factor of temperature and humidity; combining the changing trend of historical weight data to extract time series features such as weight fluctuation frequency and amplitude, and forming a multidimensional feature vector set containing signal features, environmental features and weight features.

[0129] In some embodiments, the weight coefficient of historical data is automatically adjusted according to the distance between the time of the calibration training data set and the current calibration moment, including: establishing a time decay function, giving a higher initial weight to historical data closer to the current calibration moment, and reducing the weight of historical data farther away according to a preset decay rate; the decay rate of the time decay function is dynamically adjusted according to the user's historical calibration frequency. When the user's continuous calibration interval is less than a preset threshold, the decay rate is automatically reduced to retain the reference value of historical data over a longer period.

[0130] In some embodiments, the weight coefficient is corrected in combination with the user's historical usage frequency, including: counting the number of times the user uses the body fat scale in a historical period to generate a usage frequency index; when the usage frequency index is higher than a preset threshold, an additional gain is added to the weight coefficient of the historical data used for the last three times, and the gain amplitude is positively correlated with the usage frequency index; when the usage frequency index is lower than the preset threshold, the weight coefficient of the historical data calibrated for the first time is retained to avoid excessive attenuation of historical data in low-frequency usage scenarios.

[0131] In some embodiments, the real-time calibration model adopts a long short-term memory neural network architecture, the input layer receives a multi-dimensional feature vector, and the hidden layer learns the long-term dependency of human body composition parameters on environmental parameters and weight changes through a time gating mechanism; the real-time calibration model is constructed according to the adaptive calibration algorithm, including: using the error between the external calibration reference data and the model output as a reward signal, dynamically adjusting the connection weights of the neural network, so that the real-time calibration model can adaptively generate environmental compensation factors and weight fluctuation compensation factors.

[0132] In some embodiments, the input of currently collected real-time weight data, real-time bioelectrical impedance signal, and real-time environmental parameters to the real-time calibration model and the output of calibration correction parameters include: converting the real-time collected data into a feature vector isomorphic to the calibration training data set, and inputting the feature vector into the trained real-time calibration model; the real-time calibration model generates environmental compensation parameters and weight compensation parameters through forward propagation calculation, wherein the environmental compensation parameters correct for the interference of the current temperature and humidity on the bioelectrical impedance signal, and the weight compensation parameters compensate for the deviation in human body composition estimation based on the difference between the real-time weight and the historical average weight, so as to constitute the calibration correction parameters.

[0133] In some embodiments, the initial human body composition parameters are adaptively calibrated according to the calibration correction parameters to obtain target human body composition parameters, including: setting calibration correction rules for body fat percentage, muscle mass, and body water percentage respectively, and the calibration correction rules are designed with different correction logics according to the physical characteristics of the human body composition parameters, including: using multiplication correction for body fat percentage, adjusting the adipose tissue conductivity coefficient of the bioelectrical impedance model through environmental compensation parameters; using addition correction for muscle mass, compensating for weight estimation deviation through weight compensation parameters; forming the corrected target human body composition parameters.

[0134] The present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the processor implements the steps of the method for adaptive calibration of human body composition parameters based on an intelligent body fat scale as provided in any embodiment of the present application.

[0135] The computer-readable storage medium may be the internal storage unit of the smart body fat scale described in the aforementioned embodiment, such as the hard disk or memory of the smart body fat scale. The computer-readable storage medium may also be an external storage device of the smart body fat scale, such as a plug-in hard disk, a smart memory card (SMC), a secure digital (SD) card, a flash memory card, etc., equipped on the smart body fat scale.

[0136] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present application, and such modifications or substitutions should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A method for adaptive calibration of human body composition parameters based on an intelligent body fat scale, characterized in that: Applied to an intelligent body fat scale; the method comprises: Obtain the user's real-time weight data, real-time bioelectrical impedance signal and real-time temperature and humidity data of the current environment; Based on a preset bioelectrical impedance model, the real-time weight data, real-time bioelectrical impedance signal, and real-time temperature and humidity data are combined to calculate and obtain initial body composition parameters, wherein the initial body composition parameters include body fat percentage, muscle mass, and body water percentage; and based on historical weight data, historical bioelectrical impedance signal, historical environmental parameters, and preset standard body composition parameters within a historical calibration period, a calibration training data set is formed; Performing feature extraction on the calibration training data set according to a preset adaptive calibration algorithm, the adaptive calibration algorithm including a dynamic weight allocation model based on time series, the dynamic weight allocation model automatically adjusting the weight coefficient of historical data according to the time distance between the calibration training data set and the current calibration moment, and correcting the weight coefficient in combination with the user's historical usage frequency to construct a real-time calibration model according to the adaptive calibration algorithm; The currently collected real-time weight data, real-time bioelectrical impedance signal, and real-time environmental parameters are input into the real-time calibration model, and calibration correction parameters are output to adaptively calibrate the initial body composition parameters according to the calibration correction parameters to obtain target body composition parameters.

2. The method according to claim 1, characterized in that The bioelectrical impedance model has a built-in basic calculation formula, which is constructed based on the conductive properties of human body fluids. Real-time temperature data is used to correct the effect of body fluid ion concentration on electrical impedance, and real-time humidity data is used to correct the environmental dependence of skin surface resistance. The initial body composition parameters are calculated based on the preset bioelectrical impedance model and combined with the real-time weight data, real-time bioelectrical impedance signal and real-time temperature and humidity data, including: The bioelectrical impedance model is used to divide the human body composition calculation interval according to the real-time weight data. Combined with the frequency value of the corrected bioelectrical impedance signal and the temperature and humidity data, the initial values ​​of body fat percentage, muscle mass and body water percentage are estimated respectively through hierarchical calculation logic.

3. The method according to claim 1, characterized in that The historical calibration period covers a preset time range, and the historical weight data, historical bioelectrical impedance signals, and historical environmental parameters are stored in the order of collection time and associated with the user's unique identifier; the preset standard body composition parameters are obtained through an external calibration reference device, and the external calibration reference device includes a medical body composition analyzer; the calibration training data set is formed based on the historical weight data, historical bioelectrical impedance signals, historical environmental parameters, and preset standard body composition parameters within the historical calibration period, including: The historical weight data, historical bioelectrical impedance signals, and historical environmental parameters under the same user ID are timestamped and aligned with the standard body composition parameters at the corresponding moments to generate a structured calibration training dataset.

4. The method according to claim 1, wherein The feature extraction of the calibration training data set according to a preset adaptive calibration algorithm includes: The bioelectrical impedance signals in the calibration training dataset were subjected to noise filtering, and the signal eigenvalues ​​were extracted using the sliding window method. Normalize the environmental parameters to generate the combined impact factor of temperature and humidity; Combined with the changing trend of historical weight data, time series features such as weight fluctuation frequency and amplitude are extracted to form a multidimensional feature vector set containing signal features, environmental features and weight features.

5. The method according to claim 1, wherein The method of automatically adjusting the weight coefficient of historical data according to the distance between the calibration training data set and the current calibration time includes: A time decay function is established to assign a higher initial weight to historical data closer to the current calibration moment, and to reduce the weight of historical data farther away according to a preset decay rate. The decay rate of the time decay function is dynamically adjusted according to the user's historical calibration frequency. When the user's continuous calibration interval is less than a preset threshold, the decay rate is automatically reduced to retain the reference value of historical data over a longer period.

6. The method according to claim 1, characterized in that The modification of the weight coefficient based on the user's historical usage frequency includes: Count the number of times a user has used a body fat scale in a historical period and generate a usage frequency index; When the usage frequency index is higher than the preset threshold, the weight coefficient of the historical data used in the last three times is additionally gained, and the gain amplitude is positively correlated with the usage frequency index; when the usage frequency index is lower than the preset threshold, the weight coefficient of the historical data calibrated for the first time is retained to avoid excessive attenuation of historical data in low-frequency usage scenarios.

7. The method according to claim 1, characterized in that The real-time calibration model adopts a long short-term memory neural network architecture. The input layer receives a multi-dimensional feature vector, and the hidden layer learns the long-term dependency of body composition parameters with environmental parameters and weight changes through a time-gating mechanism. The method of constructing a real-time calibration model according to an adaptive calibration algorithm includes: Using the error between the external calibration reference data and the model output as a reward signal, the connection weights of the neural network are dynamically adjusted, enabling the real-time calibration model to adaptively generate environmental compensation factors and weight fluctuation compensation factors.

8. The method according to claim 1, characterized in that The input of the currently collected real-time weight data, real-time bioelectrical impedance signal, and real-time environmental parameters into the real-time calibration model and the output of the calibration correction parameters include: The real-time collected data is converted into a feature vector isomorphic to the calibration training data set and input into the trained real-time calibration model; the real-time calibration model generates environmental compensation parameters and weight compensation parameters through forward propagation calculation, wherein the environmental compensation parameters correct for the interference of the current temperature and humidity on the bioelectrical impedance signal, and the weight compensation parameters compensate for the deviation in human body composition estimation based on the difference between the real-time weight and the historical average weight, thereby forming the calibration correction parameters.

9. The method according to claim 1, characterized in that The adaptively calibrating the initial body composition parameters according to the calibration correction parameters to obtain target body composition parameters includes: Calibration correction rules are set for body fat percentage, muscle mass, and body water percentage. These correction rules use different correction logics based on the physical characteristics of human body composition parameters, including: multiplication correction for body fat percentage, adjusting the adipose tissue conductivity coefficient of the bioelectrical impedance model through environmental compensation parameters; addition correction for muscle mass, compensating for weight estimation deviations through weight compensation parameters; The corrected target body composition parameters are formed.

10. A human body composition parameter adaptive calibration system based on an intelligent body fat scale, characterized in that: Applied to smart body fat scale; the system includes: A data acquisition unit is used to obtain the user's real-time weight data, real-time bioelectrical impedance signal and real-time temperature and humidity data of the current environment; a composition calculation unit for calculating and obtaining initial body composition parameters based on a preset bioelectrical impedance model and in combination with the real-time weight data, the real-time bioelectrical impedance signal, and the real-time temperature and humidity data, the initial body composition parameters including body fat percentage, muscle mass, and body water percentage; and forming a calibration training data set based on historical weight data, historical bioelectrical impedance signals, historical environmental parameters, and preset standard body composition parameters within a historical calibration period; a feature extraction unit, configured to extract features from the calibration training dataset according to a preset adaptive calibration algorithm, wherein the adaptive calibration algorithm includes a dynamic weight allocation model based on a time series, wherein the dynamic weight allocation model automatically adjusts a weight coefficient of historical data according to the time distance between the calibration training dataset and the current calibration moment, and modifies the weight coefficient based on the user's historical usage frequency, so as to construct a real-time calibration model according to the adaptive calibration algorithm; The target acquisition unit is used to input the currently collected real-time weight data, real-time bioelectrical impedance signal, and real-time environmental parameters into the real-time calibration model, and output calibration correction parameters to adaptively calibrate the initial body composition parameters according to the calibration correction parameters to obtain target body composition parameters.

Images