Body fat prediction system and method based on multi-band impedance signals

Abstract

The invention discloses a body fat prediction system and method based on a multi-band impedance signal, and relates to the technical field of big data analys.The method comprises the steps that impedance data of different segments under different frequency bands, the body fat amount, the lean body weight and the human body weight are collected; integrity verification is carried out on the impedance data, missing values are filled with mean values, abnormal values are removed and corrected, and meanwhile timestamp alignment is carried out; extracting features to construct an original impedance vector, and mapping the original impedance vector to a high-dimensional space through a random matrix to generate a random mapping function; constructing a time sequence regression model, taking the embedding dimension as an input layer, taking the body fat amount and the lean body mass as an output layer, and using mean square error calibration and back propagation updating; and through a mean square error weighting evaluation model, outputting a performance standard result. The system comprises a data acquisition module, a data preprocessing module, a random distribution embedding module, a time sequence model training module and a display module. The method can adapt to impedance characteristics of different crowds, is suitable for portable terminal or smart home body measurement, and can be used at high frequency in daily life.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of big data analysis, and particularly relates to a body fat prediction system and method based on multi-frequency impedance signals. BACKGROUND

[0002] Dual-energy X-ray absorptiometry is a technology for accurately measuring body fat content, lean tissue mass and bone density by using two different energy X-rays to penetrate the human body and according to the difference in X-ray absorption rate of different tissues. The principle is based on the difference in attenuation of high and low energy X-rays in different tissues, and each tissue proportion is inversely deduced through a mathematical model. Although DXA is widely recognized as the gold standard for body composition measurement, its disadvantages include: expensive equipment, large size, site requirements, existence of trace radiation, and unsuitability for on-site portable use or home health management.

[0003] Bioelectrical impedance analysis technology is a non-invasive technology for estimating body composition such as body water, fat and lean tissue by measuring the impedance of the human body to weak alternating current, including impedance amplitude and phase angle, and establishing a human body model. The technology is non-invasive, fast and low in cost, and is suitable for clinical and home use, but in actual use, its precision is limited compared to the DXA method, it is easily disturbed, its modeling method relies more on experience, and underfitting may occur. Moreover, the model established is usually a linear model established according to the impedance values of the limbs, torso and whole body at different frequencies, which is difficult to meet the characteristics of the nonlinear model of the human body, especially in extreme cases such as overweight and underweight, the robustness is poor.

[0004] Random distribution embedded prediction model is a phase space reconstruction and random sampling technology, which can analyze the nonlinear signal of multi-dimensional human impedance and predict the body fat and lean tissue content under the chaotic model of the human body, solving the key technical problems of traditional BIA modeling methods that are difficult to handle nonlinear, high-dimensional and individual differences, so as to realize more accurate and stable prediction of human body fat and lean tissue in complex, disturbed and individual difference significant impedance signals. SUMMARY

[0005] The present application aims to provide a body fat prediction system and method based on multi-frequency impedance signals to solve the problems in the prior art.

[0006] To achieve the above-mentioned purpose, the present application provides the following technical scheme: a body fat prediction method based on multi-frequency impedance signals, the body fat prediction method comprising the following steps: Step S1, collecting impedance data at different frequencies and DXA measured human composition data; Step S2, performing integrity check and outlier check on the collected impedance data at different frequencies; Step S3, extracting data features of the checked impedance data and constructing a random mapping function; Step S4, constructing a time series regression model and training the time series regression model with the collected impedance data; Step S5, evaluating the performance of the time series regression model according to the DXA measured human composition data and the output result of the time series regression model, and taking the output result of the time series regression model that meets the performance standard as the prediction result.

[0007] Further, the specific steps of step S1 are as follows: Step S1-1, collecting impedance data of different segments at different frequencies and labeling time stamps, the different frequencies including 1 kHz, 5 kHz, 50 kHz, 250 kHz, 500 kHz, the different segments including left upper limb, right lower limb, trunk, left lower limb and right lower limb, extracting DXA human composition data from the system database and labeling time stamps, the DXA human composition data including body fat mass and lean body mass, collecting human body weight and recording as w; Step S1-2, marking the impedance data collected under a certain human body weight as , wherein , 1 corresponds to left upper limb, 2 corresponds to right lower limb, 3 corresponds to trunk, 4 corresponds to left lower limb, and 5 corresponds to right lower limb, and the input form of impedance value is constructed as: ; In the above formula, refers to the impedance value of the i-th segment at the j-th frequency; Step S1-3, taking the DXA human composition data under the same human body weight as the output result y, and the output form is: .

[0008] Further, the specific steps of step S2 are as follows: Step S2-1, performing a preset integrity check on the impedance data, marking the missing impedance data as , if there is , the mean value is used for filling, and the calculation formula of the mean value is: ; Where N is the total number of impedance values; Step S2-2, performing an abnormality check on the impedance data, and performing discrimination and elimination through a check formula, the check formula is: ; Wherein, is a certain impedance data, is the mean value of the whole system, is the standard deviation of the system, if If 3, it is determined as an abnormality and is rejected and corrected by the method of step S2-1; Step S2-3, sorting according to time stamp descending order, and aligning with DXA body composition data time stamp.

[0009] Further, the specific steps of step S3 are as follows: Step S3-1, extracting all impedance data, extracting impedance data collected at the same time stamp to construct an original impedance feature vector, denoted as X, the original impedance feature vector is: ; Wherein refers to the impedance data of 5 segments at 5 frequencies; Step S3-2, according to the original impedance feature vector constructed according to the current time t, traversing m original impedance feature vectors forward to construct an embedding vector, the expression of the embedding vector is: ; Wherein, Delay step, the value of the value is autocorrelation function algorithm; Step S3-3, using a random matrix to map the 25-dimensional data constructed by the impedance values of 5 segments at 5 frequencies of all sample populations in the data set to a high-dimensional embedding space, constructing a random mapping function Y, the random mapping function is: ; Wherein, is a nonlinear activation function, b is a bias vector, and W is a random value with normal distribution.

[0010] Further, the specific steps of step S4 are as follows: Step S4-1, taking the embedding dimension d as the number of input layer nodes, and constructing the number of output layer nodes according to the number of output data, the output data including body fat and lean body mass; Step S4-2, calibrating the time series regression model by a loss function, the loss function being mean square error, specifically: ; In the above formula, L represents the mean square error as the loss function, N represents the total number of samples, refers to the true DXA body composition data of the i-th sample, refers to the output of the time series regression model of the i-th sample, represents the Euclidean square distance between the true value and the predicted value, refers to the true value of the i-th sample in the k-th dimension; Step S4-3, updating the time series model according to the preset back propagation update rule until the difference between the time series model output value and the DXA human body composition data reaches the minimum.

[0011] Further, the specific steps of step S5 are as follows: Step S5-1, inputting the real-time measured impedance data and the body weight into the time series regression model to obtain time series regression model output data, and evaluating the time series regression model in combination with the DXA human body composition data; Step S5-2, the evaluation formula for the body fat mass is: ; Wherein, represents the true value of the body fat mass, and the numerical value herein is derived from the actual measurement value of the DXA human body composition data, represents the output value predicted by the time series regression model for the i th sample; Step S5-3, the evaluation formula for the lean body mass is: ; Wherein, represents the true value of the lean body mass, and the numerical value herein is derived from the actual measurement value of the DXA human body composition data, represents the output value predicted by the time series regression model for the i th sample; Step S5-4, performing weighted calculation on and to obtain the final performance evaluation of the time series regression model, and if the final performance evaluation exceeds the preset threshold, the output result of the time series regression model is taken as the prediction result.

[0012] Further, the body fat prediction system comprises a data acquisition module, a data preprocessing module, a random distribution embedding module, a time series model training module and a display module; The data acquisition module is used to acquire impedance data at different frequencies, DXA measured human body composition data and body weight; the data preprocessing module is used to perform integrity check and outlier check on the acquired impedance data at different frequencies; the random distribution embedding module is used to extract data features of the checked impedance data and construct a random mapping function; the time series model training module is used to construct a time series regression model and train the model using the acquired impedance data; the display module is used to evaluate the performance of the time series regression model according to the DXA data and the time series regression model output, and output the prediction result of the qualified time series regression model; The output end of the data acquisition module is electrically connected with the input end of a data preprocessing module; the output end of the data preprocessing module is electrically connected with the input end of a random distribution embedding module; the output end of the random distribution embedding module is electrically connected with the input end of a time series model training module; and the output end of the time series model training module is electrically connected with the input end of a display module.

[0013] Further, the data acquisition module comprises an impedance data acquisition unit and a DXA data acquisition unit. The impedance data acquisition unit is used for acquiring impedance data of the left upper limb, the right lower limb, the trunk, the left lower limb and the right lower limb at specified frequencies of 1 kHz, 5 kHz, 50 kHz, 250 kHz and 500 kHz; and the DXA data acquisition unit is used for extracting DXA human body composition data containing body fat and lean body mass from a system database and acquiring human body weight. The data preprocessing module comprises an integrity checking unit and an outlier checking unit. The integrity checking unit is used for performing preset integrity checking on the impedance data and using mean value filling for missing impedance data; and the outlier checking unit is used for judging whether the impedance data is abnormal through a formula, eliminating abnormal data and correcting by using a mean value filling method.

[0014] Further, the random distribution embedding module comprises an impedance feature extraction unit and a random mapping construction unit. The impedance feature extraction unit is used for extracting all impedance data, constructing an original impedance feature vector and traversing a specified number of vectors to construct an embedding vector; and the random mapping construction unit is used for mapping 25-dimensional impedance data input to a high-dimensional embedding space by using a random matrix and constructing a random mapping function containing a nonlinear activation function.

[0015] Further, the time series model training module comprises a time series regression model construction unit and a model parameter updating unit. The time series regression model construction unit is used for constructing a time series regression model by taking the embedding dimension as the number of input layer nodes and the number of output data as the number of output layer nodes; and the model parameter updating unit is used for calibrating the model by using a mean square error loss function and updating the model according to a back propagation update rule until the difference between the output and the DXA data is minimum. The display module comprises a model performance evaluation unit and a prediction result output unit. The model performance evaluation unit is used for calculating the MSE of body fat and lean body mass in combination with the DXA data, and then obtaining the final performance evaluation of the model by weighted calculation; and the prediction result output unit is used for taking the output result of the time series regression model as the prediction result when the final performance evaluation of the model exceeds a preset threshold.

[0016] Compared with the prior art, the present application has the following beneficial effects: 1. This invention can be run by using an impedance sensor and a calculation model. Based on bioelectrical impedance, it is non-invasive, radiation-free, and harmless to the human body. It can be used safely in daily or high-frequency use and is suitable for portable terminals or smart home body measurement systems.

[0017] 2. This invention extracts deeper electrophysiological features of human body composition through a random distribution embedding algorithm, which can more accurately reflect the distribution of fat and muscle at different frequencies compared with the traditional BIA model, and improves the fitting ability of body fat and lean body mass.

[0018] 3. This invention, through a large amount of DXA ground truth training data, can automatically learn individualized, non-linear mapping relationships, adaptively learn the relationship between impedance characteristics and body fat and muscle in different populations, support cross-group, integrated modeling, and significantly improve the model's fitting ability and individual accuracy. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating a body fat prediction method based on multi-band impedance signals according to the present invention. Figure 2 This is a schematic diagram of the structure of a body fat prediction system based on multi-band impedance signals according to the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Example 1: As Figure 1 As shown, the present invention provides a technical solution: a body fat prediction method based on multi-band impedance signals. The body fat prediction method includes the following steps: Step S1: Collect impedance data at different frequencies and DXA measurement of human body composition data; The specific steps of step S1 are as follows: Step S1-1: Collect impedance data of different segments at different frequencies and mark them with timestamps. The different frequencies include 1kHz, 5kHz, 50kHz, 250kHz, and 500kHz. The different segments include the left upper limb, right lower limb, trunk, left lower limb, and right lower limb. Extract DXA body composition data from the system database and mark it with timestamps. The DXA body composition data includes body fat mass and lean body mass. Collect body weight and record it as w. Step S1-2: Mark the impedance data collected at a certain body weight as... wherein 1 corresponds to the left upper limb, 2 corresponds to the right lower limb, 3 corresponds to the trunk, 4 corresponds to the left lower limb, 5 corresponds to the right lower limb, and the input form of the impedance value is: ; In the above formula, refers to the impedance value of the ith segment at the jth frequency; Step S1-3, the DXA body composition data under the same body weight is taken as the output result y, and the output form is: ; Step S2, the impedance data collected at different frequencies is subjected to integrity check and outlier check; The specific steps of the step S2 are as follows: Step S2-1, the impedance data is subjected to a preset integrity check, and the missing impedance data is recorded as If there is , the mean value is used for filling, and the calculation formula of the mean value is: ; wherein N is the total number of impedance values; Step S2-2, the impedance data is subjected to an outlier check, and is discriminated and removed through a check formula, and the check formula is: ; wherein is a certain impedance data, is the mean value of the entire system, is the standard deviation of the system, if > 3, it is determined to be abnormal and is removed and corrected by the method of step S2-1; Step S2-3, according to the descending order of the time stamp, the time stamp is aligned with the DXA body composition data; Step S3, the data features of the checked impedance data are extracted and a random mapping function is constructed; The specific steps of the step S3 are as follows: Step S3-1, all impedance data is extracted, the impedance data collected at the same time stamp is extracted to construct an original impedance feature vector, denoted as X, and the original impedance feature vector is: ; wherein refers to the impedance data of 5 segments at 5 frequencies; Step S3-2, according to the original impedance feature vector constructed according to the current time t, m original impedance feature vectors are traversed forward to construct an embedding vector, and the expression of the embedding vector is: ; wherein, delay step, the value of the method is autocorrelation function algorithm; Step S3-3, input the 25-dimensional data constructed by using the random matrix to map the 5 frequency impedance values of the 5 segments of all sample populations in the data set to the high-dimensional embedding space, and construct a random mapping function Y, the random mapping function is: ; wherein, is a nonlinear activation function, b is a bias vector, and W is a random value of normal distribution; Step S4, constructing a time series regression model and training the time series regression model using the collected impedance data; The specific steps of the step S4 are as follows: Step S4-1, taking the embedding dimension d as the input layer node number, and taking the output data number as the output layer node number, the output data including body fat mass and lean body mass, and constructing a time series regression model; Step S4-2, calibrating the time series regression model through a loss function, the loss function being mean square error, specifically: ; In the above formula, L represents the mean square error as the loss function, N represents the total number of samples, refers to the true DXA human body composition data of the i-th sample, refers to the output of the time series regression model of the i-th sample, represents the Euclidean square distance between the true value and the predicted value, refers to the true value of the i-th sample in the k-th dimension; Step S4-3, updating the time series model according to the preset back propagation update rule until the difference between the output value of the time series model and the DXA human body composition data reaches the minimum; Step S5, evaluating the performance of the time series regression model according to the DXA measured human body composition data and the output result of the time series regression model, and taking the output result of the time series regression model whose performance meets the standard as the prediction result; The specific steps of the step S5 are as follows: Step S5-1, inputting the real-time measured impedance data and human body weight into the time series regression model to obtain the time series regression model output data, and evaluating the time series regression model in combination with the DXA human body composition data; Step S5-2, the evaluation formula for body fat mass is: ; wherein, refers to the true value of the body fat mass, and the value is derived from the actual measured value of the DXA body composition data, refers to the output value predicted by the time series regression model for the i-th sample; Step S5-3, the evaluation formula for lean body mass is: ; wherein, refers to the true value of the lean body mass, and the value is derived from the actual measured value of the DXA body composition data, refers to the output value predicted by the time series regression model for the i-th sample; Step S5-4, the and are weighted to obtain the final performance evaluation of the time series regression model, and if the final performance evaluation exceeds a preset threshold, the output result of the time series regression model is taken as the prediction result. and

[0022] For example: Step 1: The user collects a high-throughput 5-segment multi-frequency human impedance value dataset through a data acquisition module as follows:

[0023] which contains five-segment impedance values of the human body at 1 kHz, 5 kHz, 50 kHz, 250 kHz and 500 kHz, and body fat rate and lean body mass values measured by the DXA method; Step 2: The collected human impedance original values are compensated for missing values, screened for abnormal values, and labeled and aligned through a data preprocessing module, and the processed 5-segment multi-frequency human impedance value dataset is as follows:

[0024] Step 3: According to the above preprocessed human impedance dataset, impedance sequences and high-dimensional space mappings are constructed, and the delay step of the dataset is calculated by the autocorrelation function method is 3, and the embedding dimension is 3; Step 4: The time series regression model is set and the parameters are confirmed: the network structure is confirmed, including confirming that the input layer node number (embedding dimension) is 3, which can be greater than the input dimension to expand the feature space, the structure information and dynamic characteristics of the input data are maintained through random mapping, and new discriminant features are generated for the subsequent time series regression model. Other parameter settings: the number of hidden layers is 2, and the number of output layer nodes is 2. At this point, the model of the application has been trained and can predict the body fat rate and lean body mass value by inputting a new multi-frequency 5-segment human impedance dataset; Step 5: Input the human body weight as 63.1 kg, and input the new multi-frequency 5-segment human impedance value as follows:​

[0025] The random distribution embedding prediction model method predicts that the body fat percentage of this person at this weight is 21.9% and the lean body mass is 49.3 kg.

[0026] Example 2, as Figure 2 As shown, the present invention provides a body fat prediction system based on multi-band impedance signals. The body fat prediction system includes a data acquisition module, a data preprocessing module, a random distribution embedding module, a time series model training module, and a display module. The data acquisition module is used to collect impedance data at different frequencies, DXA measurement of human body composition data, and human weight; the data preprocessing module is used to perform integrity verification and outlier verification on the collected impedance data at different frequencies; the random distribution embedding module is used to extract data features of the verified impedance data and construct a random mapping function; the time series model training module is used to construct a time series regression model and train the model using the collected impedance data; the display module is used to output an evaluation of the time series regression model's performance based on the DXA data and the time series regression model, and output the prediction results of the qualified time series regression model. The output of the data acquisition module is electrically connected to the input of the data preprocessing module; the output of the data preprocessing module is electrically connected to the input of the random distribution embedding module; the output of the random distribution embedding module is electrically connected to the input of the time series model training module; and the output of the time series model training module is electrically connected to the input of the display module.

[0027] The data acquisition module includes an impedance data acquisition unit and a DXA data acquisition unit; The impedance data acquisition unit is used to acquire impedance data of the left upper limb, right lower limb, trunk, left lower limb, and right lower limb at specified frequencies of 1kHz, 5kHz, 50kHz, 250kHz, and 500kHz; the DXA data acquisition unit is used to extract DXA body composition data containing body fat mass and lean body mass from the system database, and to acquire body weight. The data preprocessing module includes an integrity verification unit and an outlier verification unit; The integrity verification unit is used to perform a preset integrity verification on the impedance data and fill missing impedance data with the mean; the outlier verification unit is used to determine whether the impedance data is abnormal by formula, remove abnormal data and correct it by the mean filling method.

[0028] The random distribution embedding module includes an impedance feature extraction unit and a random mapping construction unit; The impedance feature extraction unit is configured to extract all impedance data, construct an original impedance feature vector, and traverse a specified number of vectors to construct an embedding vector; and the random mapping construction unit is configured to map 25-dimensional impedance data input to a high-dimensional embedding space using a random matrix, and construct a random mapping function with a nonlinear activation function.

[0029] The time series model training module comprises a time series regression model construction unit and a model parameter updating unit. The time series regression model construction unit is configured to construct a time series regression model by taking the embedding dimension as the input layer node number and the output data number as the output layer node number; and the model parameter updating unit is configured to calibrate the model by using a mean square error loss function, and update the model according to a back propagation update rule until the output difference from DXA data is minimized. The display module comprises a model performance evaluation unit and a prediction result output unit. The model performance evaluation unit is configured to calculate the MSE of body fat mass and lean body mass in combination with DXA data, and then obtain the final performance evaluation of the model by weighted calculation; and the prediction result output unit is configured to output the prediction result of the time series regression model when the final performance evaluation of the model exceeds a preset threshold.

[0030] It is apparent for those skilled in the art that the present application is not limited to the details of the above exemplary embodiments, and can be implemented in other specific forms without departing from the spirit or essential characteristics of the present application. Therefore, the embodiments should be regarded as exemplary and non-limiting, and the scope of the present application is defined by the appended claims rather than the above description, and all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be included in the present application. Any reference signs in the claims should not be regarded as limiting the claims to which they relate.

Claims

1. A method for predicting body fat based on multi-band impedance signals, characterized in that: The body fat prediction method includes the following steps: Step S1: Collect impedance data of different segments at different frequencies and DXA measurement of human body composition data; Step S2: Perform integrity verification and outlier verification on the impedance data collected at different frequencies; Step S3: Extract the data features of the verified impedance data and construct a random mapping function; Step S4: Construct a time series regression model and train the time series regression model using the collected impedance data; Step S5: Evaluate the performance of the time series regression model based on the DXA measurement of human body composition data and the output of the time series regression model, and use the output of the time series regression model that meets the performance standard as the prediction result.

2. The body fat prediction method based on multi-band impedance signals according to claim 1, characterized in that: The specific steps of step S1 are as follows: Step S1-1: Collect impedance data of different segments at different frequencies and mark them with timestamps. The different frequencies include 1kHz, 5kHz, 50kHz, 250kHz, and 500kHz. The different segments include the left upper limb, right lower limb, trunk, left lower limb, and right lower limb. Extract DXA body composition data from the system database and mark it with timestamps. The DXA body composition data includes body fat mass and lean body mass. Collect body weight and record it as w. Step S1-2: Mark the impedance data collected at a certain body weight as... ,in 1 corresponds to the left upper limb, 2 to the right lower limb, 3 to the trunk, 4 to the left lower limb, and 5 to the right lower limb. The input format for constructing the impedance value is as follows: ； In the above formula, refers to the impedance value of the i-th segment at the j-th frequency; Steps S1-3: Use the DXA body composition data for the same body weight as the output result y, and output it in the following format: 。 3. The body fat prediction method based on multi-band impedance signals according to claim 2, characterized in that: The specific steps of step S2 are as follows: Step S2-1: Perform a preset integrity check on the impedance data, and record any missing impedance data as... If it exists Then the mean is used to fill the gap, and the formula for calculating the mean is: ； Where N is the total number of impedance values; Step S2-2: Perform anomaly verification on the impedance data, and identify and eliminate anomalies using a verification formula, which is: ； in, For a certain impedance data, The mean of the entire system. It is the standard deviation of the system. If the value is greater than 3, it is considered abnormal and removed, and then corrected using the method in step S2-1. Step S2-3: Sort the data in descending order of timestamps and align the timestamps with the DXA human body composition data.

4. The body fat prediction method based on multi-band impedance signals according to claim 3, characterized in that: The specific steps of step S3 are as follows: Step S3-1: Extract all impedance data. Extract impedance data collected at the same timestamp to construct an original impedance feature vector, denoted as X. The original impedance feature vector is: ； in This refers to the impedance data for five segments at five frequencies; Step S3-2: Based on the original impedance feature vector constructed at the current time t, traverse forward m original impedance feature vectors to construct an embedding vector. The expression for the embedding vector is: ； in, The delay step size is determined using the autocorrelation function algorithm. Step S3-3: Using a random matrix, the 25-dimensional data input, constructed from the five frequency impedance values ​​of the five segments of all sample populations in the dataset, is mapped to a high-dimensional embedding space, thus constructing a random mapping function Y. The random mapping function is as follows: ； in, is a non-linear activation function, b is a bias vector, and W is a random value from a normal distribution.

5. The body fat prediction method based on multi-band impedance signals according to claim 4, characterized in that: The specific steps of step S4 are as follows: Step S4-1: Use the embedding dimension d as the number of input layer nodes, and construct the number of output layer nodes based on the number of output data, where the output data includes body fat mass and lean body mass, and construct a time-series regression model. Step S4-2: Calibrate the time series regression model using a loss function, where the loss function is the mean squared error, specifically: ； In the above formula, L represents the mean squared error as the loss function, and N represents the total number of samples. This refers to the actual DXA human body composition data of the i-th sample. This refers to the output of the time-series regression model for the i-th sample. This represents the Euclidean squared distance between the actual and predicted values. This refers to the true value of the i-th sample in the k-th dimension; Step S4-3: Update the time series model according to the preset backpropagation update rules until the difference between the output value of the time series model and the DXA human body composition data reaches the minimum.

6. The body fat prediction method based on multi-band impedance signals according to claim 5, characterized in that: The specific steps of step S5 are as follows: Step S5-1: Input the real-time measured impedance data and human body weight into the time series regression model to obtain the output data of the time series regression model, and evaluate the time series regression model in combination with DXA human body composition data. Step S5-2, The formula for assessing body fat percentage is: ； in, This refers to the actual value of body fat percentage, which is derived from actual measurements using DXA body composition data. This refers to the output value predicted by the current time series regression model for the i-th sample; Step S5-3: The formula for assessing lean body mass is as follows: ； This refers to the true value of lean body mass, which is the actual measured value from DXA body composition data. This refers to the output value predicted by the current time series regression model for the i-th sample; Step S5-4, combine and Weighted calculations are performed to obtain the final performance evaluation of the time series regression model. If the final performance evaluation exceeds the preset threshold, the output of the time series regression model is used as the prediction result.

7. A body fat prediction system based on multi-band impedance signals, applied to the body fat prediction method based on multi-band impedance signals as described in any one of claims 1-6, characterized in that: The body fat prediction system includes a data acquisition module, a data preprocessing module, a random distribution embedding module, a time series model training module, and a display module; The data acquisition module is used to acquire impedance data at different frequencies, DXA measurement of human body composition data, and human body weight; the data preprocessing module is used to perform integrity verification and outlier verification on the acquired impedance data at different frequencies. The random distribution embedding module is used to extract data features from the verified impedance data and construct a random mapping function; the time series model training module is used to construct a time series regression model and train the model using the collected impedance data; the display module is used to output an evaluation of the performance of the time series regression model based on the DXA data and the time series regression model, and output the prediction results of the qualified time series regression model. The output of the data acquisition module is electrically connected to the input of the data preprocessing module; the output of the data preprocessing module is electrically connected to the input of the random distribution embedding module; the output of the random distribution embedding module is electrically connected to the input of the time series model training module; and the output of the time series model training module is electrically connected to the input of the display module.

8. A body fat prediction system based on multi-band impedance signals according to claim 7, characterized in that: The data acquisition module includes an impedance data acquisition unit and a DXA data acquisition unit; The impedance data acquisition unit is used to acquire impedance data of the left upper limb, right lower limb, trunk, left lower limb, and right lower limb at specified frequencies of 1kHz, 5kHz, 50kHz, 250kHz, and 500kHz; the DXA data acquisition unit is used to extract DXA body composition data containing body fat mass and lean body mass from the system database, and to acquire body weight. The data preprocessing module includes an integrity verification unit and an outlier verification unit; The integrity verification unit is used to perform a preset integrity verification on the impedance data and fill missing impedance data with the mean; the outlier verification unit is used to determine whether the impedance data is abnormal by formula, remove abnormal data and correct it by the mean filling method.

9. A body fat prediction system based on multi-band impedance signals according to claim 7, characterized in that: The random distribution embedding module includes an impedance feature extraction unit and a random mapping construction unit; The impedance feature extraction unit is used to extract all impedance data, construct the original impedance feature vector, and traverse forward a specified number of vectors to construct the embedding vector; the random mapping construction unit is used to map the 25-dimensional impedance data input to a high-dimensional embedding space using a random matrix, and construct a random mapping function containing a nonlinear activation function.

10. A body fat prediction system based on multi-band impedance signals according to claim 7, characterized in that: The time series model training module includes a time series regression model construction unit and a model parameter update unit; The time-series regression model construction unit is used to construct a time-series regression model by taking the embedding dimension as the number of input layer nodes and the number of output data as the number of output layer nodes; the model parameter update unit is used to calibrate the model through the mean squared error loss function and update the model according to the backpropagation update rule until the difference between the output and the DXA data is minimized. The display module includes a model performance evaluation unit and a prediction result output unit; The model performance evaluation unit is used to calculate the MSE of body fat mass and lean body mass by combining DXA data, and then obtain the final performance evaluation of the model through weighted calculation; the prediction result output unit is used to take the output result of the time series regression model as the prediction result when the final performance evaluation of the model exceeds a preset threshold.

Images