A blood glucose detection system based on multi-wavelength PPG sensor fusion

Abstract

The present application is applicable to the technical field of blood glucose detection, and provides a blood glucose detection system based on multi-wavelength PPG sensor fusion, including: a signal acquisition device and a processor, the signal acquisition device includes a light source for generating multiple lights with different central wavelengths, a transmission receiving module, a first reflection receiving module, a second reflection receiving module, a first temperature and humidity acquisition module, a second temperature and humidity acquisition module, and a third temperature and humidity acquisition module; the processor is used to extract temperature and humidity signals, demographic information of the object to which the finger belongs, and feature information of PPG signals corresponding to multiple lights, and fuse the extracted feature information, input the fusion result into a blood glucose detection model for prediction, and obtain a blood glucose detection result. The present application can improve the accuracy of blood glucose detection.

Description

Technical Field

[0001] The present application belongs to the technical field of blood glucose detection, and in particular, relates to a blood glucose detection system based on multi-wavelength PPG sensor fusion. Background Art

[0002] Non-invasive blood glucose testing technology provides a painless solution for diabetic patients to monitor blood glucose levels through non-direct blood contact such as skin. Currently, most studies focus on using a single type of sensor to collect biological signals, and establish models through feature extraction and analysis to predict blood glucose concentrations. These methods have shown high accuracy in controlled environments such as laboratories, but their accuracy and reliability are greatly reduced when faced with complex and changing conditions in the real world. The main reason is that single sensor technology is difficult to effectively filter out interference from environmental noise and physiological conditions, such as changes in skin temperature, differences in sweat composition, etc. These factors will affect the detection signal, thereby reducing the signal-to-noise ratio and affecting the accuracy of blood glucose measurement.

[0003] Most of the traditional non-invasive blood glucose detection methods based on photoplethysmography (PPG) signals rely only on single-wavelength or dual-wavelength PPG signals and are studied by analyzing the characteristics of PPG morphology. Since the absorption rate of glucose molecules under different spectra is significantly different, the existing technology ignores this point, resulting in low accuracy of blood glucose detection. Summary of the invention

[0004] The embodiment of the present application provides a blood glucose detection system based on multi-wavelength PPG sensor fusion, which can solve the problem of low accuracy of blood glucose detection.

[0005] The embodiment of the present application provides a blood glucose detection system based on multi-wavelength PPG sensor fusion, including:

[0006] A signal acquisition device, the signal acquisition device includes a light source for generating a plurality of light with different central wavelengths, a transmission receiving module, a first reflection receiving module, a second reflection receiving module, a first temperature and humidity acquisition module for acquiring a first temperature and humidity signal of an environment where the signal acquisition device is located, a second temperature and humidity acquisition module for acquiring a second temperature and humidity signal at a finger pulp position, and a third temperature and humidity acquisition module for acquiring a third temperature and humidity signal at a finger back position, the first reflection receiving module is used to receive a light signal after light is reflected by the finger, and convert the received light signal into a first PPG signal, the second reflection receiving module is used to receive a light signal after light is reflected by the finger, and convert the received light signal into a second PPG signal, and the transmission receiving module is used to receive a light signal after light is transmitted through the finger, and convert the received light signal into a third PPG signal;

[0007] The processor, the control end of the light source, the signal transmission end of the transmission receiving module, the signal transmission end of the first reflection receiving module, the signal transmission end of the second reflection receiving module, the signal transmission end of the first temperature and humidity acquisition module, the signal transmission end of the second temperature and humidity acquisition module, and the signal transmission end of the third temperature and humidity acquisition module are all electrically connected to the processor. The processor is used to extract the first temperature and humidity signal, the second temperature and humidity signal, the third temperature and humidity signal, the demographic information of the object to which the finger belongs, and the feature information of the first PPG signal, the second PPG signal and the third PPG signal corresponding to multiple lights, and fuse the extracted feature information, and input the fusion result into the blood glucose detection model for prediction to obtain the blood glucose detection result.

[0008] The above solution of the present application has the following beneficial effects:

[0009] In an embodiment of the present application, a signal acquisition device generates a plurality of lights with different central wavelengths, and collects PPG signals corresponding to light signals after the plurality of lights are reflected by the fingers, PPG signals corresponding to light signals after the plurality of lights are transmitted by the fingers, temperature and humidity signals of the finger environment, temperature and humidity signals at the finger pulp position, and temperature and humidity signals at the finger back position, and then a processor extracts characteristic information of the PPG signal, temperature and humidity signal, and demographic information, and fuses the extracted characteristic information, and finally inputs the fusion result into a blood glucose detection model for prediction to obtain a blood glucose detection result. Among them, since the present application can more accurately distinguish the light absorption characteristics of glucose molecules and other biological molecules by analyzing light signals of multiple wavelengths, it helps to improve the accuracy of blood glucose detection. At the same time, since the processor fuses the three modal data of temperature and humidity signals, PPG signals, and demographic information as input to realize blood glucose prediction, it helps to further improve the accuracy of blood glucose detection.

[0010] Other beneficial effects of the present application will be described in detail in the subsequent specific implementation section. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0012] Figure 1 A schematic diagram of the structure of a blood sugar detection system provided in one embodiment of the present application;

[0013] Figure 2 Schematic diagram of the absorption and transmission spectra of blood samples in the near-infrared region;

[0014] Figure 3A schematic diagram of the structure of a signal acquisition device provided in one embodiment of the present application;

[0015] Figure 4 A circuit schematic diagram of each sensor provided in an embodiment of the present application;

[0016] Figure 5 A schematic diagram of the structure of a photoelectric receiving sensor provided in one embodiment of the present application;

[0017] Figure 6 A clock diagram of a photoelectric receiving sensor provided in one embodiment of the present application;

[0018] Figure 7 is a schematic diagram of a PPG signal;

[0019] Figure 8a A schematic diagram of PPG signals corresponding to each wavelength provided in an embodiment of the present application;

[0020] Figure 8b A schematic diagram of temperature data collected by a temperature and humidity sensor according to an embodiment of the present application;

[0021] Figure 8c A schematic diagram of humidity data collected by a temperature and humidity sensor provided in an embodiment of the present application;

[0022] Fig. 9 A clock diagram of a temperature and humidity sensor provided in an embodiment of the present application;

[0023] Fig.10 A schematic diagram of segmenting a PPG signal provided in an embodiment of the present application;

[0024] Fig.11 A data processing diagram of a blood glucose detection system provided in one embodiment of the present application;

[0025] Fig.12 Construct a schematic diagram for the data in an example;

[0026] Fig.13 A schematic diagram of the distribution of blood glucose levels in a data set in an example;

[0027] Fig.14 This is a Parkes error grid analysis diagram in an example;

[0028] Fig.15 This is a Bland-Altman analysis diagram in an example;

[0029] Fig.16 is a performance analysis diagram of different PPG combinations in an example;

[0030] Fig.17 The figure is a comparison chart of model performance indicators with and without meal intervals in an example. DETAILED DESCRIPTION

[0031] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0032] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, wholes, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or combinations thereof.

[0033] It should also be understood that the term “and / or” used in the specification and appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0034] As used in the specification and appended claims of this application, the term "if" can be interpreted as "when" or "uponce" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "uponce it is determined" or "in response to determining" or "uponce [described condition or event] is detected" or "in response to detecting [described condition or event]", depending on the context.

[0035] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0036] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0037] In view of the low accuracy of current blood sugar detection, the embodiment of the present application provides a blood sugar detection system based on multi-wavelength PPG sensor fusion, which generates multiple lights with different central wavelengths through a signal acquisition device, and collects PPG signals corresponding to light signals after multiple lights are reflected by fingers, PPG signals corresponding to light signals after multiple lights are transmitted by fingers, temperature and humidity signals of the finger environment, temperature and humidity signals at the finger pulp position, and temperature and humidity signals at the finger back position, and then extracts the characteristic information of PPG signals, temperature and humidity signals, and demographic information through a processor, and fuses the extracted characteristic information, and finally inputs the fusion result into a blood sugar detection model for prediction to obtain a blood sugar detection result. Among them, since the present application can more accurately distinguish the light absorption characteristics of glucose molecules and other biological molecules by analyzing light signals of multiple wavelengths, it helps to improve the accuracy of blood sugar detection. At the same time, since the processor fuses three modal data of temperature and humidity signals, PPG signals, and demographic information as input to realize blood sugar prediction, it helps to further improve the accuracy of blood sugar detection.

[0038] The relevant principles involved in this application are first explained below.

[0039] In human arterial blood, different substances have different absorption rates for different wavelengths of infrared light. According to Lambert-Beer's law, the absorbance of a composite substance can be expressed as follows:

[0040]

[0041] represents the total absorbance, is the initial light intensity, is the intensity of light passing through the absorbing medium, the variable represents the wavelength of the incident light, and the average path length of the photon propagation is express, Represents the optical properties and constant attenuation coefficient of superficial tissues such as muscle and bone, represents the molar absorptivity of the absorbing substance, Indicates concentration.

[0042] Therefore, the light intensity passing through the absorbing medium is can be restated as:

[0043]

[0044] Assuming that the absorption rate in tissues other than arterial blood is constant, the intensity of the transmitted light passing through the examination area is It can be expressed as:

[0045]

[0046] here, and It represents the molecular absorption coefficient matrix and concentration matrix of the absorbing substance that remain stable in a short time, and and Represents the molecular absorption coefficient matrix and concentration matrix of the variable species.

[0047] As arteries pulsate, the average light absorption path length in human tissue changes. Assuming the length The increment is , then the change in transmitted light intensity can be expressed as This assumption assumes that the absorption of light by human tissue (except arterial blood) remains constant. Therefore, the intensity of transmitted light can be expressed as:

[0048]

[0049] Dividing the two equations, we can get the following formula:

[0050]

[0051] Given the DC signal Significantly larger than the AC signal ,equation can be simplified to:

[0052]

[0053] If there is different light absorbing materials, and considering the incident light range from arrive , the above formula can be expanded as follows:

[0054]

[0055] Therefore, in theory, the wavelength can be determined , where irrelevant substances (analytes The molar absorption coefficient of the analyte (except for ) approaches zero. The relationship between the received light intensity and the analyte concentration is as follows:

[0056]

[0057] represents the analyte concentration, express The DC component of the transmitted light intensity at wavelength, express The AC component of the transmitted light intensity at wavelength, Represents analyte exist Molar absorptivity at wavelength.

[0058] Therefore, as long as the AC and DC values ​​of the PPG signal under multiple wavelengths are obtained, the concentration of the analyte can be theoretically determined in a data-driven manner. prediction.

[0059] In the embodiment of the present application, the above analyte concentration refers to the concentration of glucose.

[0060] The blood glucose detection system based on multi-wavelength PPG sensor fusion provided by this application is exemplified below in combination with specific embodiments.

[0061] like Figure 1 As shown, the blood glucose detection system based on multi-wavelength PPG sensor fusion provided by the embodiment of the present application includes a signal acquisition device and a processor; the signal acquisition device includes a light source for generating light of multiple different central wavelengths, a transmission receiving module, a first reflection receiving module, a second reflection receiving module, a first temperature and humidity acquisition module for acquiring a first temperature and humidity signal of the environment where the signal acquisition device is located, a second temperature and humidity acquisition module for acquiring a second temperature and humidity signal at the finger pulp position, and a third temperature and humidity acquisition module for acquiring a third temperature and humidity signal at the finger back position; the control end of the light source, the signal transmission end of the transmission receiving module, the signal transmission end of the first reflection receiving module, the signal transmission end of the second reflection receiving module, the signal transmission end of the first temperature and humidity acquisition module, the signal transmission end of the second temperature and humidity acquisition module, and the signal transmission end of the third temperature and humidity acquisition module are all electrically connected to the processor.

[0062] The first reflection receiving module is used to receive the light signal after the light is reflected by the finger, and convert the received light signal into a first PPG signal, the second reflection receiving module is used to receive the light signal after the light is reflected by the finger, and convert the received light signal into a second PPG signal, and the transmission receiving module is used to receive the light signal after the light is transmitted by the finger, and convert the received light signal into a third PPG signal. In some optional examples, the first reflection receiving module and the second reflection receiving module are respectively located at the light source to collect PPG signals in different directions to improve the accuracy of detection.

[0063] It is understood that the light source can generate light of different central wavelengths at a certain frequency. Since most of the human blood is composed of water, such as Figure 2As shown in the figure, water molecules show strong absorbance in the near-infrared spectrum, especially outside the 1400 nm wavelength, where the transmittance is less than one percent. This characteristic prevents near-infrared light in this range from completely penetrating human tissue (the transmission receiving module cannot capture such weak signals). However, light in this band can be reflected back from the surface of the skin through diffuse reflection and capture blood information. Therefore, the optical module of the device is designed with a transmission receiving module, a first reflection receiving module, and a second reflection receiving module for light in different bands.

[0064] Specifically, in some embodiments of the present application, the multiple different center wavelengths include: 665nm, 905nm, 1200nm, 1300nm, 1460nm and 1550nm. When each light is irradiated on the finger, the transmission receiving module, the first reflection receiving module and the second reflection receiving module will collect the PPG signal corresponding to the light, so as to accurately analyze the light absorption characteristics of glucose molecules and other biological molecules based on the light signals of multiple wavelengths, thereby improving the accuracy of blood sugar detection.

[0065] The above-mentioned processor is used to extract feature information of the first temperature and humidity signal, the second temperature and humidity signal, the third temperature and humidity signal, the demographic information of the object to which the finger belongs, the first PPG signal, the second PPG signal and the third PPG signal corresponding to multiple lights, and fuse the extracted feature information, input the fusion result into the blood glucose detection model for prediction, and obtain the blood glucose detection result.

[0066] The above demographic information can be the age, gender, height, weight, etc. of the subject (i.e. the object to which the finger belongs). It is worth mentioning that the processor integrates three modal data of temperature and humidity signals, PPG signals and demographic information as input to realize blood sugar prediction, which helps to improve the accuracy of blood sugar detection.

[0067] In some embodiments of the present application, Figure 3 As shown, the signal acquisition device further includes: a first clamp 100, a second clamp 200, and an elastic component 700 located between the first clamp 100 and the second clamp 200, the top of the elastic component 700 is connected to one end of the first clamp 100, and the bottom of the elastic component 700 is connected to one end of the second clamp 200. When performing blood sugar testing, the finger 300 is located between the first clamp 100 and the second clamp 200.

[0068] Among them, the transmission receiving module 500 and the third temperature and humidity collection module 600 are both arranged on the side of the first clamp 100 facing the second clamp 200, and the first temperature and humidity collection module 400 is arranged on the side of the first clamp 100 facing away from the second clamp 200; the light source 900, the first reflection receiving module 800, the second reflection receiving module 910 and the second temperature and humidity collection module 920 are all arranged on the side of the second clamp 200 facing the first clamp 100, and the light source 900 is located between the first reflection receiving module 800 and the second reflection receiving module 910. Figure 3 The dotted line with an arrow in the middle represents the transmission direction of light, and the curves with arrows near the first temperature and humidity collection module 400, the second temperature and humidity collection module 920, and the third temperature and humidity collection module 600 represent external thermal radiation, which is received by the temperature and humidity collection modules at the corresponding positions.

[0069] When the elastic component is in a contracted state, the third temperature and humidity acquisition module contacts the back of the finger between the first clamp and the second clamp, and the second temperature and humidity acquisition module contacts the finger pulp. It is understandable that in actual application, when blood sugar testing is required, the subject can place the finger between the first clamp and the second clamp, and press the first clamp to put the elastic component in a contracted state, the third temperature and humidity acquisition module contacts the back of the finger, and the second temperature and humidity acquisition module contacts the finger pulp, light enters from the finger pulp, and the transmission receiving module, the first reflection receiving module, and the second reflection receiving module all have a certain gap with the surface of the finger; after completing the test, release the first clamp to put the elastic component in an extended state.

[0070] As an optional example, the above-mentioned elastic component can be a spring, the transmission receiving module can be a transmission receiving sensor (i.e., a transmission photoelectric receiving sensor), the first reflection receiving module and the second reflection receiving module can both be reflection receiving sensors (i.e., a reflection photoelectric receiving sensor), and the first temperature and humidity acquisition module, the second temperature and humidity acquisition module and the third temperature and humidity acquisition module can all be temperature and humidity sensors.

[0071] In order to facilitate the light source to generate light with multiple different central wavelengths of 665nm, 905nm, 1200nm, 1300nm, 1460nm and 1550nm, the above-mentioned light source includes six light-emitting LEDs, and the six light-emitting LEDs are used to generate light with central wavelengths of 665nm, 905nm, 1200nm, 1300nm, 1460nm and 1550nm respectively.

[0072] In some embodiments of the present application, the above-mentioned sensors can all be driven by an STM32 chip, collect optical signals at a sampling rate of 50 Hz, collect temperature and humidity signals at a sampling rate of 5 Hz, output data through a universal serial bus (USB) interface, and support data communication with multiple systems such as Windows and Android.

[0073] like Figures 4 to 5 As shown, in actual application, the six light-emitting LEDs of the light source are as follows Figure 4 As shown in the diode D1 (for generating light with a central wavelength of 1200nm), the diode D2 (for generating light with a central wavelength of 1300nm), the diode D3 (for generating light with a central wavelength of 1460nm), the diode D4 (for generating light with a central wavelength of 1550nm), the diode D5 (for generating light with a central wavelength of 665nm) and the diode D6 (for generating light with a central wavelength of 905nm), these light-emitting LEDs can be switched by a multi-channel switch (such as Figure 4 The switches S1, S2, S3, S4, S5 and S6 in the LED are opened and closed independently and driven by a cross-current controller to ensure the uniform brightness of each LED. The above-mentioned photoelectric receiving sensors are composed of two photodiodes DD1 (with a spectral response range of 400 to 1100 nm) and two photodiodes DD2 (with a spectral response range of 1000 to 1800 nm) (where, Figure 5 The circle in the figure is photodiode DD1, and the square is photodiode DD2. Figure 4 Only one photodiode DD1 and one photodiode DD2 are schematically shown in the figure. The processor sends control signals to the cross-flow controller, switch S1, switch S2, switch S3, switch S4, switch S5 and switch S6 through the micro-control unit to control the diodes D1, D2, D3, D4, D5 and D6 to light up in sequence. The photodiode DD1 is read when the diodes D1, D2, D3 and D4 are lit. The photodiode DD2 is read when the diodes D5 and D6 are lit. The data read by the photodiodes DD1 and DD2 are processed by the amplifier and transmitted to the micro-control unit, and then transmitted to the above-mentioned processor by the micro-control unit. At the same time, the above-mentioned first temperature and humidity acquisition module (such as Figure 4 The temperature and humidity sensor A in the second temperature and humidity acquisition module (such as Figure 4 The temperature and humidity sensor B) and the third temperature and humidity acquisition module (such as Figure 4The temperature and humidity sensor C) in the device collects the temperature and humidity signals and outputs the collected temperature and humidity signals to the micro control unit, which then transmits the collected temperature and humidity signals to the processor. It is understandable that the micro control unit can transmit relevant data signals to the processor via a USB interface. Figure 4 The GND in it indicates the ground terminal, and VCC indicates the power supply.

[0074] Specifically, the processor can drive six light-emitting LEDs through a microcontroller unit and a cross-current controller. Each light-emitting LED lights up for 2ms in turn according to the rules, and the six light-emitting LEDs light up for a total of 12ms. At the same time, the processor controls the photodiode to collect photoelectric signals at a sampling rate of 50Hz. Specifically, the photodiode reads 0.5ms data values ​​in sequence according to the lighting order of the corresponding wavelength light-emitting LEDs. Photodiode DD1 reads when diode D1, diode D2, diode D3, and diode D4 are lit, and photodiode DD2 reads when diode D5 and diode D6 are lit. After completing the lighting and reading operations of the six wavelengths, the system sets an 8ms delay for data packaging, and then performs the next round of lighting and reading operations. Among them, the clock diagram is as follows: Figure 6 It should be noted that the diode D5, the diode D6, the diode D1, the diode D2 and the transmission receiving module constitute a transmission module. When the diode D5, the diode D6, the diode D1 and the diode D2 are turned on, light is emitted from the fingertips and received at the back of the finger, which is used to observe the deep optical information of the fingertips; the diode D3, the diode D4, the first reflection receiving module and the second reflection receiving module constitute a reflection module. When the diode D3 and the diode D4 are turned on, light is emitted from the fingertips and received at the fingertips, which is used to observe the shallow optical information of the fingertips.

[0075] When the heart pumps blood, blood enters peripheral tissues (such as fingers) through arteries, causing the blood volume in the blood vessels to increase. This change causes more light to be absorbed in the light propagation path, and the intensity of the reflected or transmitted light detected by the photoelectric receiving sensor decreases. Conversely, when the heart is in diastole, the blood volume in the blood vessels decreases, the amount of light absorbed decreases, and the light intensity detected by the photoelectric receiving sensor increases. Since the light absorption in human tissues such as skin and bones remains relatively constant, the periodic changes in blood vessel closure synchronized with the cardiac cycle and the subsequent changes in blood vessel volume are the main reasons for the signal changes. The photodiode in the PPG sensor (i.e., photoelectric receiving sensor) receives the reflected or transmitted light signal and converts it into an electrical signal (i.e., PPG signal), the intensity of which is directly related to the volume change of blood in the blood vessels.

[0076] like Figure 7As shown, the PPG signal consists of two components: a direct current (DC) component and an alternating current (AC) component. The DC component corresponds to the light signal transmitted or reflected by the tissue, which depends on the tissue structure and the average volume of arterial and venous blood. Typically, the DC component changes gradually with breathing and activity, while the AC component superimposed on the DC component oscillates according to the blood volume fluctuations between the systolic and diastolic phases of the cardiac cycle. The fundamental frequency of the AC component is determined by the heart rate (HR). This blood glucose detection system is equipped with 6 different central wavelengths of light, and the visualization results of the signal are shown in Figure 8a As shown, it should be noted that the PPG signal of each light in the figure is: the first PPG signal, the second PPG signal and the third PPG signal corresponding to the light are weighted averaged to obtain an average PPG signal.

[0077] It is worth mentioning that, since skin temperature and humidity will affect the light flux of the PPG signal to a certain extent, this application designs and integrates three temperature and humidity sensors (the sensor model can be SHT30-DIS), which are placed in different positions to record the temperature and humidity changes of the fingertips, finger backs and the current measurement environment. Among them, SHT30-DIS is a universal independent component that can synchronously collect temperature and humidity signals with an error range of ±2.0%RH, ±0.48°C, where RH is the unit of relative humidity.

[0078] Specifically, the processor can control each temperature and humidity sensor to collect temperature and humidity signals at a sampling rate of 5 Hz through the microcontroller unit. The temperature data collected by the three temperature and humidity sensors (i.e., temperature and humidity sensor S1, temperature and humidity sensor S2, and temperature and humidity sensor S3) are as follows: Figure 8b As shown in Figure 2, the collected humidity data is as follows: Figure 8c Specifically, each temperature and humidity sensor lights up for 10us in turn according to the rules, and the three sensors light up for a total of 30us. After completing a round of operations, the system sets a 200ms delay for data packaging, so that the collected data can be transmitted to the processor through the microcontroller unit, and then the next round of operations can be carried out. The clock diagram is shown in Fig. 9 shown.

[0079] In practical applications, the microcontroller unit can be responsible for processing the raw signals from the photodiode and the temperature and humidity sensor. For the optical signal (i.e. the signal from the photodiode), it is continuously collected at a sampling rate of 50Hz, and is initially amplified and noise filtered through a high-fidelity amplifier. The temperature and humidity signal is sampled at a frequency of 5Hz and processed by a digital filter to eliminate environmental interference and the noise of the sensor itself.

[0080] The collected digital signals are formatted in hexadecimal inside the microcontroller unit and packaged into standardized data packets. Each data packet contains a timestamp, optical signal data, temperature and humidity signal data, and a checksum.

[0081] The data packets are transmitted to external devices (such as the above-mentioned processor) through the built-in USB interface. The USB interface not only supports high-speed data transmission, but also can power the components in the blood glucose detection system that need power, ensuring the stability and reliability of the blood glucose detection system during long-term operation. In addition, the blood glucose detection system also supports Bluetooth wireless transmission module for real-time data monitoring on mobile devices.

[0082] The following is an exemplary description of the process in which the processor makes predictions based on the above data.

[0083] For the convenience of description, the above-mentioned PPG signals are collectively referred to as optical signals, and the signals output by each temperature and humidity sensor are collectively referred to as thermodynamic signals.

[0084] To facilitate data processing, after receiving the optical signal, thermodynamic signal and demographic information (which can be input through the host computer), the processor can pre-process them and then use the pre-processed data for subsequent feature extraction, fusion and prediction.

[0085] Optical signal: The acquisition wavelength is The photoplethysmographic data of , M is the number of wavelengths (i.e., central wavelength), For the wavelength, for The corresponding PPG signal.

[0086] Thermodynamic signals: Synchronous acquisition of different measurement locations The temperature and humidity signals are recorded as and .

[0087] Demographic information: The age, gender, height, weight, etc. of the subjects, recorded as .

[0088] The preprocessing process of each signal is as follows:

[0089] For optical signals, a Butterworth filter can be used to remove high-frequency noise and obtain .

[0090]

[0091] in, is the cutoff frequency set empirically, is a Butterworth low-pass filter.

[0092] Then, The PPG morphological signal after baseline drift removal and normalization is obtained by normalizing and reducing low-frequency noise (the band with the most obvious morphological features), which is recorded as .

[0093]

[0094] in, is the cutoff frequency set empirically, represents normalization processing, Represents a Butterworth high-pass filter.

[0095] For thermodynamic signals, the temperature and humidity signals can be smoothed in moving intervals to obtain and .

[0096]

[0097]

[0098] in, Indicates moving range smoothing.

[0099] For demographic information, interpolation is performed on demographic data to obtain .

[0100]

[0101] in, Indicates data interpolation processing.

[0102] The following is an exemplary description of the process of feature extraction, fusion and prediction performed by the processor.

[0103] In some embodiments of the present application, the above-mentioned characteristic information includes temperature and humidity characteristic information, demographic characteristic information, and PPG characteristic information corresponding to multiple lights. Accordingly, the specific implementation method of the processor extracting the characteristic information of the first temperature and humidity signal, the second temperature and humidity signal, the third temperature and humidity signal, the demographic information of the object to which the finger belongs, and the first PPG signal, the second PPG signal, and the third PPG signal corresponding to multiple lights includes the following steps:

[0104] Step 1.1: The processor inputs the first temperature and humidity signal, the second temperature and humidity signal, and the third temperature and humidity signal into a feature extraction network for processing to obtain temperature and humidity feature information.

[0105] Specifically, the processor can convert the temperature signals at three locations into and humidity signal The signals are concatenated into a 6-channel signal and then upsampled to 50 Hz to match the input size of the PPG signal. The upsampled signal is then input into a feature extraction network (such as the ResNet-1D backbone network) to obtain temperature and humidity feature information. .

[0106] In step 1.2, the processor inputs the demographic information of the object to which the finger belongs into the feature embedding layer for processing to obtain demographic feature information.

[0107] Specifically, the processor can use the preprocessed As input, it is input into the feature embedding layer to obtain demographic feature information .

[0108]

[0109] in, Represents the feature embedding layer.

[0110] In step 1.3, the processor performs weighted averaging processing on the first PPG signal, the second PPG signal and the third PPG signal corresponding to each light to obtain an average PPG signal, and inputs the average PPG signal into the feature extraction network for processing to obtain the PPG feature information corresponding to the light.

[0111] Specifically, for each type of light, the first PPG signal, the second PPG signal and the third PPG signal corresponding to the light can be weighted averaged to obtain an average PPG signal; then the average PPG signal is input into a feature extraction network (such as a ResNet-1D backbone network) for processing to obtain the PPG feature information corresponding to the light. It can be understood that the number of channels of the ResNet-1D backbone network can be set to the number of wavelengths. , in order to obtain the PPG characteristic information corresponding to each light .

[0112] To improve the detection accuracy, the above feature information also includes optical absorbance feature information, so that the system can not only use the PPG signal, but also directly calculate the concentration of glucose in the blood based on the spectral absorption characteristics, thereby improving the specificity and sensitivity of the model for glucose detection. Accordingly, the processing is also used to perform the following absorbance field feature extraction steps:

[0113] Step 1.4, for each light, use the peak detection algorithm to determine the peak position of the average PPG signal corresponding to the light, and use the peak position to perform waveform segmentation on the average PPG signal corresponding to the light to obtain a segmented waveform. Each peak in the segmented waveform corresponds to a heartbeat cycle.

[0114] Specifically, for (the band with the most obvious morphological characteristics), the average PPG signal corresponding to this light can be segmented by the peak detection algorithm to obtain the peak position ; Then for each light, use the peak position The average PPG signal is segmented to obtain the segmented waveform, which is recorded as For example, the segmented waveform is as follows: Fig.10 As shown (the figure only shows the waveform corresponding to one type of light), Fig.10 A represents the peak position, B represents the valley position, and C represents a heartbeat cycle. A heartbeat cycle refers to the process between the start of two heartbeats.

[0115] Step 1.5, for each light, calculate the DC value and AC value of the average PPG signal corresponding to the light in each heartbeat cycle, and filter from all DC values ​​in order from large to small DC value, for The DC value and AC value of the heartbeat cycle corresponding to the DC value are spliced ​​to obtain the optical absorbance characteristic information corresponding to the light; the DC value is used to represent the minimum value of the voltage within the heartbeat cycle, and the AC value is used to represent the fluctuation amplitude of the voltage within the heartbeat cycle.

[0116] Specifically, The average PPG signal at wavelength Optical absorbance characteristic information of a heartbeat cycle for: , The average PPG signal at wavelength DC value of a heartbeat cycle for: , The average PPG signal at wavelength AC value of a heartbeat cycle for: , Indicates The minimum value of the voltage (i.e. the average PPG signal) within a heartbeat cycle, Indicates the The maximum value of the voltage during a heartbeat cycle.

[0117] Average PPG signal at all wavelengths The result of concatenating the AC and DC values ​​of a heartbeat cycle for: , Indicates splicing processing.

[0118] Due to individual heart rate differences, the same acquisition window may contain different heartbeat times. To ensure the consistency of the model network input structure, the extraction The heartbeat cycle with the largest AC value is The wavelength is obtained by splicing the DC value and AC value within a heartbeat cycle. Optical absorbance characteristic information of light .

[0119] Step 1.6, concatenate the optical absorbance characteristic information corresponding to all lights to obtain the optical absorbance characteristic information.

[0120] Specifically, for all The optical absorbance characteristic information DF obtained by splicing is:

[0121]

[0122] Indicates splicing processing.

[0123] In some embodiments of the present application, after obtaining all feature information, the processor may fuse the extracted feature information in the following manner:

[0124] The processor uses a fusion model to process the temperature and humidity characteristic information, demographic characteristic information, optical absorbance characteristic information and PPG characteristic information corresponding to various lights respectively, and splices all the processed features to obtain a fusion result.

[0125] The above fusion model includes a cross attention module and a global average pooling module connected in sequence. Exemplarily, the structure of the cross attention module can adopt a commonly used cross attention module, and the structure of the global average pooling module can adopt a commonly used global average pooling module.

[0126] Specifically, the temperature and humidity feature information can be input into the cross attention module for cross attention processing, and the processing result can be input into the global average pooling module for global average pooling processing to obtain the final features of temperature and humidity; similarly, other feature information can also be processed in the same way to obtain the corresponding final features. Other feature information includes demographic feature information, optical absorbance feature information, and PPG feature information corresponding to various lights.

[0127] After obtaining the final features of all feature information, the processor performs splicing processing on all features to obtain a fusion result. Specifically, all features can be spliced ​​through a common feature splicing module.

[0128] It is understandable that the processor inputs the obtained fusion result into the blood glucose detection model for prediction to obtain the blood glucose detection result. Exemplarily, the above blood glucose detection model can be a fully connected layer, and correspondingly, the blood glucose detection result for:

[0129]

[0130] in, represents the fully connected layer, Indicates the fusion result.

[0131] It can be seen that Fig.11 As shown in the figure, by extracting the absorbance field features of the PPG signal, Before the filter is selected The DC and AC values ​​are used to obtain the optical absorbance characteristic information (i.e. Fig.11 The feature embedding of the output of the feature extraction part in the domain is performed, and the feature extraction network (such as the ResNet-1D backbone network) is used to extract the features of the PPG signal and the temperature and humidity signal to obtain the PPG feature information (i.e. Fig.11 The feature embedding of the PPG output by the feature extraction network), temperature and humidity feature information (i.e. Fig.11 The feature embedding layer is used to process the demographic information and obtain the demographic characteristic information (i.e. Fig.11 The feature embedding corresponding to the demographic information in the dataset is then sequentially input into the cross attention module, the global average pooling module, the feature concatenation module, and the fully connected layer for processing to obtain the blood glucose value.

[0132] The blood glucose detection system of the present application is exemplified below with reference to specific examples.

[0133] In this example, a small-scale exploratory study was conducted in a community using the blood glucose monitoring system described above. This study was approved by the hospital ethics committee (ethics approval number: 22117).

[0134] In order to understand the blood sugar fluctuations of the subjects, this study used the oral glucose tolerance test (OGTT) method to collect blood sugar data of the subjects. Biochemical blood sugar measurements were performed on an empty stomach, 30 minutes after a random meal, 1 hour after a meal, 1.5 hours after a meal, and 2 hours after a meal, and non-invasive device signals were collected. The experiment also recorded demographic information (age, gender, height, weight, etc.) and clinical diagnosis results for statistical analysis (see Fig.12 ).

[0135] The study collected data from 301 participants and obtained 27,290 sensor data. After preliminary screening, 254 participants (including 52 diabetic patients) were retained, containing 26,063 valid samples. After preprocessing, the data of the 254 participants were divided into three different groups in a ratio of 3:1:1 for training, validation, and test sets, respectively, to ensure that there was no overlap between the participants in each group (see Fig.12 ). Fig.12 The dataset construction process is presented, showing the steps from collecting raw data using multiple sensors to creating a raw dataset containing 301 subjects, ultimately resulting in 27,290 measurements. Fig.12 The process of data filtering and data modeling is also shown. The filtered dataset is divided into training set, validation set and test set in a ratio of 3:1:1. The demographic characteristics of the study participants are shown in Table 1. The distribution of blood glucose levels in the dataset is shown in Fig.13 As shown, Fig.13 The horizontal axis represents the blood glucose level in mg / dL, and the vertical axis represents the amount.

[0136] Table 1 Demographic characteristics of study participants

[0137]

[0138] To evaluate the performance of the blood glucose prediction model, this study used R2 score, root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) as the main evaluation indicators. In addition, ISO 15197:2013 (ISO 15197:2013 is a revised standard of "In vitro diagnostic test systems-General technical requirements for blood glucose monitoring systems") and Parkes error grid (PEG) were introduced to further comprehensively evaluate the accuracy and clinical acceptability of blood glucose meter measurements. The ISO standard evaluates the percentage of measured values ​​with a relative error within 15% (blood glucose concentration>100mg / dL) or within 15mg / dL (blood glucose concentration≤100mg / dL) compared with the standard value. On the other hand, PEG involves dividing the measured values ​​of the blood glucose meter to observe the risk level of using predicted blood glucose values ​​for clinical treatment decisions.

[0139] Table 2 gives the detailed performance results on the test set based on 5-fold cross validation. Fig.14 The visualization results of the best-performing PEG analysis are given. It can be seen that 99.47% of the measurement points belong to areas with no clinical risk (Area A: 88.23%) or low clinical risk (Area B: 11.24%), and 0.53% of the measurement points fall into the remaining partitions (Regions C, D, and E are values ​​for assigning risks based on professional knowledge, and each region represents a different level of clinical importance). 81.2% of the measurement point errors are within the ISO acceptable range, indicating that most of the measurement results are within the clinically acceptable range and are close to the accuracy of commercially available blood glucose meters. Fig.14 The horizontal axis is the calibrated blood glucose value, the unit is mg / dL, and the vertical axis is the predicted blood glucose value, the unit is mg / dL.

[0140] Table 2 Overall performance of 5-fold cross validation on the test set

[0141]

[0142] In addition, we performed a comparative analysis using the Bland-Altman method. Fig.15 As shown, the vast majority of points fall within the 95% confidence interval (upper limit), and the mean difference line is close to 0, indicating that there is no obvious systematic bias between the two measurement methods. This strongly supports the potential of this method and device for home blood glucose monitoring. Fig.15 The horizontal axis represents the average of the reference value and the predicted value in mg / dL, and the vertical axis represents the difference (reference value - predicted value) in mg / dL.

[0143] In order to verify the importance of the domain features proposed in this paper, the following experiments were designed. Initially, various regression methods were used without using a deep learning feature extraction network. The experimental results are shown in Table 3, where most of the classic regressors achieved good performance. Among them, MLP performed the best, with RMSE, MAE, ISO accuracy, and PEG A area indicators reaching 27.09%, 19.97%, 65.9%, and 79.3%, respectively. In addition, this ablation study tested the comparison between models with and without domain features DF, as shown in Table 3. It can be clearly seen that without using domain features, the performance of the model in RMSE, MAE, ISO accuracy, and PEG A area is 24.16%, 15.1%, 79.1%, and 86.71%, respectively, which is about 2% lower than the model using domain feature features. This proves the importance of domain features in the task of non-invasive blood glucose detection. In Table 3, XGBoost is extreme gradient boosting tree, RF is random forest, MLP is multilayer perceptron, Ours w / o DF is the network framework proposed in this application (without using domain features), Ours w / i DF is the network framework proposed in this application (using domain features), ISO accuracy is the percentage of measurements with a relative error within 15% (blood glucose concentration>100 mg / dL) or within 15 mg / dL (blood glucose concentration≤100 mg / dL), PEG A area is the proportion of PEG A area, and PEG A+B area is the sum of the proportions of PEG A area and B area.

[0144] Table 3. Performance analysis of domain features of different regressors

[0145]

[0146] In order to study the impact of different input combinations on the prediction performance of the model, the following experiments were designed in this study. Initially, the experiment used all six wavelengths of PPG. Subsequently, the impact of additional inputs such as temperature, humidity, and demographic information on the model was studied. The experimental results are shown in Table 4. It is obvious that excluding any modality input will lead to varying degrees of performance degradation. The model performs the worst when only PPG is used. This is because the study included hundreds of test subjects, among whom individual differences in skin color, bone structure, and measurement environment can significantly affect the amplitude of the PPG signal. In Table 4, w / o TH&Demo is a model that excludes temperature, humidity, and demographic information, w / o TH is a model that excludes temperature and humidity, w / o Demo is a model that excludes demographic information, and Ours is a model that includes all input combinations.

[0147] Integrating temperature, humidity, and demographic information significantly improves performance. This is because the combined model mitigates the impact of environmental and individual differences on predictions and enhances the model's understanding of the individual's current physiological state. This also strengthens the feature representation of the PPG signal to a certain extent, thereby improving the accuracy of blood glucose prediction.

[0148] Table 4 Performance analysis of different input combinations

[0149]

[0150] To further elucidate the impact of incorporating multiple wavelengths on the model's predictive power, the study analyzed PPG signal combinations of six different optical wavelengths, which were aggregated into 63 different configurations (single wavelength 6 combinations, dual wavelength 15 combinations, triple wavelength 20 combinations, quad wavelength 15 combinations, five wavelength 6 combinations, six wavelength 1 combination). Violin plots were used to describe the distribution and probability density of the data to illustrate how the model performance evolves with the addition of each new wavelength. Fig.16 As shown, the performance of the model continues to improve as the number of PPG channels increases, reaching a peak when all six wavelengths are combined. This enhancement can be attributed to two main factors. First, the change in wavelength, especially wavelengths above 1000 nm, significantly increases the penetration depth of light, enabling the capture of detailed changes in the skin, such as changes in vascular elasticity caused by elevated glucose levels. Second, the integration of multiple wavelengths supports a more comprehensive absorption spectrum of glucose molecules, allowing the model to capture absorption data over a wider range, thereby improving the accuracy of predictions. Fig.16 The horizontal axis is the channel combination, and the vertical axis is the ISO accuracy (%).

[0151] As mentioned earlier, many studies on non-invasive blood glucose monitoring have incorporated dietary information (such as food logs) into the model input, which significantly improved the model performance. However, this approach is difficult to implement in real-world settings. In order to study the impact of dietary information on model performance, we also added meal intervals (i.e., the time from the last meal) to the data collection and conducted a comparative experiment. The experimental results are shown in Figure 2. Fig.17 shown. Fig.17 The horizontal axis is the evaluation index and the vertical axis is the value.

[0152] It can be seen that incorporating meal interval data into the model greatly enhanced various performance indicators. RMSE and MAE values ​​were significantly reduced, with RMSE decreasing from 23.24 to 18.64 and MAE decreasing from 14.21 to 11.32, indicating more accurate prediction of blood glucose levels. ISO accuracy also significantly improved from 81.2% to 87.0%, indicating better compliance with the standard accuracy threshold. PEG analysis showed that the values ​​in the safer areas (A zone and A+B zone combination) were significantly enhanced, with A zone values ​​increasing from 88.23% to 92.65% and A+B zone values ​​increasing from 99.47% to 99.79%. These results indicate that incorporating meal intervals into the model has substantial benefits for its performance.

[0153] Unlike similar studies that rely on detailed food diaries as input, using meal intervals provides a more convenient and less intrusive way to incorporate meal-related information. Maintaining a food diary accurately and consistently is a cumbersome task for users. In contrast, recording meal intervals is simpler but still significantly improves the model's performance. This approach enhances the usability and practicality of the system in real-world applications, making it more user-friendly while maintaining high prediction accuracy. The model's strong performance even without meal information emphasizes its versatility, making it a reliable tool for continuous glucose monitoring.

[0154] To verify the adaptability of the proposed network, a comparative analysis of different backbone network architectures is performed and used as feature encoders in the network. The evaluation is performed on multiple backbone network architectures, which are known for their effectiveness in processing time series and multi-dimensional data, and are specialized for enhancing feature extraction capabilities in different operational settings.

[0155] Table 5 Performance analysis of different backbone networks

[0156]

[0157] As shown in Table 5, the performance analysis of different backbone networks shows that all tested networks performed well. In comparison, the ResNet-1D model (ResNet-1D is a ResNet model built using a one-dimensional residual block) showed the strongest performance. It was followed by the Convolutional Neural Network-Long Short-Term Memory Network (CNN-LSTM) and the Temporal Convolutional Network (TCN), which shows their robustness in processing the complex patterns of glucose level data. In contrast, the performance of WaveNet (WaveNet is an autoregressive model) and InceptionTime (InceptionTime is an integrated model consisting of 5 deep learning classifiers with the same structure) lagged behind, showing the different effectiveness of different neural network architectures in this specific application.

[0158] This variability emphasizes the adaptability of the framework, highlighting that while most network structures achieve good performance, some are significantly more effective depending on the specifics of the task. The superior performance of the ResNet-1D model suggests that it is particularly well suited for sequential data tasks in medical applications, such as blood glucose level prediction, where understanding long-term dependencies is critical. These insights will be critical for improving model selection in future research to improve the accuracy and reliability of non-invasive blood glucose monitoring systems.

[0159] Assessing the performance of different studies in the field of blood glucose level estimation can be challenging due to differences in evaluation criteria and dataset sample sizes.

[0160] To provide a meaningful comparison, we selected studies with comparable sample sizes and performance indicators, as shown in the performance comparison with existing studies in Table 6. In Table 6, CEG A area is the proportion of A area in the Clarke error grid.

[0161] Table 6 Performance comparison with existing studies

[0162]

[0163] Moajjem et al. collected single-channel PPG, temperature, and biopotential signals and carefully recorded the participants' diet logs. They used the MMG-Net method to tailor personalized modeling for individual subjects. In contrast, our framework significantly outperforms the CEG region A index without the need for a food log, and its calibration-free model provides higher versatility.

[0164] Chen et al. used a single-channel PPG sensor to collect signals, but only collected fasting blood glucose signals. They used a multi-view neural network framework for modeling. In contrast, our framework achieves comparable prediction performance using only one-fifth of the parameters. In addition, our dataset covers a wider range of blood glucose, including multiple sensor readings from fasting to 2 hours after a meal, which is more practical and more challenging than simply predicting fasting blood glucose values.

[0165] In summary, the present application generates a plurality of lights with different central wavelengths through a signal acquisition device, and collects PPG signals corresponding to light signals after the plurality of lights are reflected by the fingers, PPG signals corresponding to light signals after the plurality of lights are transmitted by the fingers, temperature and humidity signals of the finger environment, temperature and humidity signals at the finger pulp position, and temperature and humidity signals at the finger back position, and then extracts the characteristic information of the PPG signal, temperature and humidity signal, and demographic information through a processor, and fuses the extracted characteristic information, and finally inputs the fusion result into a blood glucose detection model for prediction to obtain a blood glucose detection result. Among them, since the present application can more accurately distinguish the light absorption characteristics of glucose molecules and other biological molecules by analyzing light signals of multiple wavelengths, it helps to improve the accuracy of blood glucose detection. At the same time, since the processor fuses the three modal data of temperature and humidity signals, PPG signals, and demographic information as input to achieve blood glucose prediction, it helps to further improve the accuracy of blood glucose detection.

[0166] The above is a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles described in the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A blood glucose detection system based on multi-wavelength PPG sensor fusion, characterized in that: include: A signal acquisition device, the signal acquisition device comprising a light source for generating a plurality of light with different central wavelengths, a transmission receiving module, a first reflection receiving module, a second reflection receiving module, a first temperature and humidity acquisition module for acquiring a first temperature and humidity signal of an environment where the signal acquisition device is located, a second temperature and humidity acquisition module for acquiring a second temperature and humidity signal at a finger pulp position, and a third temperature and humidity acquisition module for acquiring a third temperature and humidity signal at a finger back position, the first reflection receiving module being used to receive a light signal after the light is reflected by the finger, and convert the received light signal into a first PPG signal, the second reflection receiving module being used to receive a light signal after the light is reflected by the finger, and convert the received light signal into a second PPG signal, and the transmission receiving module being used to receive a light signal after the light is transmitted by the finger, and convert the received light signal into a third PPG signal; the plurality of different central wavelengths comprising: 665nm, 905nm, 1200nm, 1300nm, 1460nm and 1550nm; a processor, wherein the control end of the light source, the signal transmission end of the transmission receiving module, the signal transmission end of the first reflection receiving module, the signal transmission end of the second reflection receiving module, the signal transmission end of the first temperature and humidity acquisition module, the signal transmission end of the second temperature and humidity acquisition module, and the signal transmission end of the third temperature and humidity acquisition module are all electrically connected to the processor, and the processor is used to extract the first temperature and humidity signal, the second temperature and humidity signal, the third temperature and humidity signal, the demographic information of the object to which the finger belongs, and the feature information of the first PPG signal, the second PPG signal, and the third PPG signal corresponding to multiple lights, and fuse the extracted feature information, and input the fusion result into the blood glucose detection model for prediction to obtain the blood glucose detection result; The characteristic information includes temperature and humidity characteristic information, demographic characteristic information, and PPG characteristic information corresponding to various lights; The processor extracts feature information of the first temperature and humidity signal, the second temperature and humidity signal, the third temperature and humidity signal, demographic information of the object to which the finger belongs, and the first PPG signal, the second PPG signal, and the third PPG signal corresponding to multiple lights, including: The processor inputs the first temperature and humidity signal, the second temperature and humidity signal, and the third temperature and humidity signal into a feature extraction network for processing to obtain temperature and humidity feature information; The processor inputs the demographic information of the object to which the finger belongs into the feature embedding layer for processing to obtain demographic feature information; The processor performs weighted average processing on the first PPG signal, the second PPG signal and the third PPG signal corresponding to each light to obtain an average PPG signal, and inputs the average PPG signal into a feature extraction network for processing to obtain PPG feature information corresponding to the light; The characteristic information also includes optical absorbance characteristic information, and the processing is further used to perform the following steps: For each type of light, a peak detection algorithm is used to determine the peak position of the average PPG signal corresponding to the light, and the peak position is used to perform waveform segmentation on the average PPG signal corresponding to the light to obtain a segmented waveform; each peak in the segmented waveform corresponds to a heartbeat cycle; For each type of light, the DC value and AC value of the average PPG signal corresponding to the light in each heartbeat cycle are calculated respectively, and K DC values ​​are selected from all DC values ​​in order from large to small, and the DC value and AC value of the heartbeat cycle corresponding to the K DC values ​​are spliced ​​to obtain the optical absorbance characteristic information corresponding to the light; the DC value is used to represent the minimum value of the voltage within the heartbeat cycle, and the AC value is used to represent the fluctuation amplitude of the voltage within the heartbeat cycle; The optical absorbance characteristic information corresponding to all lights is spliced ​​to obtain the optical absorbance characteristic information.

2. The blood sugar detection system according to claim 1, characterized in that: The processor fuses the extracted feature information, including: The processor uses a fusion model to process temperature and humidity feature information, demographic feature information, optical absorbance feature information and PPG feature information corresponding to multiple lights respectively, and splices all processed features to obtain a fusion result; the fusion model includes a cross-attention module and a global average pooling module connected in sequence.

3. The blood sugar detection system according to claim 1, characterized in that: The signal acquisition device further includes: a first clamp, a second clamp, and an elastic component located between the first clamp and the second clamp, wherein the top end of the elastic component is connected to one end of the first clamp, and the bottom end of the elastic component is connected to one end of the second clamp; The transmission receiving module and the third temperature and humidity acquisition module are both arranged on a side of the first clamp facing the second clamp, and the first temperature and humidity acquisition module is arranged on a side of the first clamp facing away from the second clamp; The light source, the first reflection receiving module, the second reflection receiving module and the second temperature and humidity acquisition module are all arranged on a side of the second clamp facing the first clamp; When the elastic component is in a contracted state, the third temperature and humidity collection module contacts the back of the finger located between the first clamp and the second clamp, and the second temperature and humidity collection module contacts the web of the finger.

4. The blood sugar detection system according to claim 3, characterized in that: The light source is located between the first reflection receiving module and the second reflection receiving module.

5. The blood sugar detection system according to claim 3, characterized in that: The elastic component is a spring.

6. The blood sugar detection system according to claim 3, characterized in that: The transmission receiving module is a transmission receiving sensor, the first reflection receiving module and the second reflection receiving module are both reflection receiving sensors, and the first temperature and humidity acquisition module, the second temperature and humidity acquisition module and the third temperature and humidity acquisition module are all temperature and humidity sensors.

7. The blood sugar detection system according to claim 6, characterized in that: The light source includes six light-emitting LEDs, and the six light-emitting LEDs are used to generate light with central wavelengths of 665nm, 905nm, 1200nm, 1300nm, 1460nm and 1550nm respectively.

Images