Exhaled breath recognition system and method based on multi-channel absorption spectroscopy and machine learning

Abstract

This invention discloses an exhaled breath identification system and method based on multi-channel absorption spectroscopy and machine learning, relating to the fields of gas detection, absorption spectral analysis, and machine learning recognition. The system includes a sampling module, an electrode electric field preprocessing module, a multi-channel absorption spectroscopy detection module, a multi-modal sampling state constraint module, a data processing module, and a machine learning recognition module. The multi-channel detection module employs a slit grating beam splitting branch, a fiber optic spectral branch, and a non-dispersive infrared branch for complementary advantages in specific wavelengths. Electrode electric field preprocessing enhances the spectral response of weakly absorbing components. Multi-modal constraints use both visual and thermal imaging recognition to determine the sampling object, blowing position, and exhalation continuity. Data is processed by a machine learning model to identify single-component, multi-component mixed gases, and their concentrations. This invention improves the stability of weak absorption feature extraction and the accuracy of multi-component identification in complex exhaled breath samples, making it suitable for non-invasive health monitoring and metabolic state analysis scenarios.

Description

Technical Field

[0001] This invention relates to the fields of gas detection, absorption spectroscopy analysis, and machine learning recognition technology, specifically to an exhaled breath identification system and method based on multi-channel absorption spectroscopy and machine learning. Background Technology

[0002] Exhaled breath testing is a non-invasive, convenient, and repeatable method for analyzing physiological states. During energy metabolism, lipid metabolism, glucose metabolism, and microbial metabolism, the human body produces various small-molecule metabolites. These components can be transported to the alveoli via the bloodstream and expelled through gas exchange during exhalation. Therefore, exhaled breath is not only waste gas but can also reflect, to some extent, changes in human metabolic activity, inflammatory responses, oxidative stress, and organ function.

[0003] Compared to blood, urine, and histopathological tests, exhaled breath testing offers advantages such as non-invasive sampling, a natural sampling process, low testing burden, and suitability for continuous monitoring. It holds promise for applications in primary care screening, chronic disease management, family health monitoring, and disease risk assessment. Typical volatile components such as acetone, ammonia, methane, ethanol, and methanol are correlated with human metabolic state, liver and kidney function, alcohol metabolism, and gut microbiota activity; therefore, stable detection and accurate identification of these gaseous components are of great significance.

[0004] However, actual exhaled breath samples are characterized by high humidity, strong background, multi-component mixing, dynamic changes, and significant individual differences. Background components such as water vapor and carbon dioxide can easily cause enhanced light path scattering, baseline fluctuations, and spectral line overlap; the absorption response of low-concentration target gases is weak, and the absorption characteristics of different volatile gases may also overlap; at the same time, expiratory flow rate, blowing distance, blowing position, vital capacity, and environmental conditions all affect the consistency of test results. Therefore, weak absorption features are difficult to extract stably from complex exhaled breath samples, multi-component gases are difficult to identify accurately, and sampling conditions are challenging.

[0005] The difficulty in standardizing the exhaled breath test is a significant issue limiting its practical application.

[0006] Existing methods for exhaled breath detection mainly include gas-sensitive detection of functional materials, ion mobility spectrometry (IMS), gas chromatography-mass spectrometry (GC-MS), and optical detection. Functional material detection methods are structurally simple, but generally suffer from insufficient selectivity, long response recovery times, and limited resistance to humidity interference. Ion mobility spectrometry and GC-MS offer high sensitivity, but their high equipment cost, size, and operational complexity make them unsuitable for portable and low-cost detection. Optical detection offers advantages such as non-contact operation, fast response, and rich information content; however, a single optical detection channel typically only acquires gas information in a limited wavelength band or dimension, making it difficult to simultaneously achieve broad-spectrum response, background reference, continuous spectral acquisition, and weak absorption feature extraction.

[0007] Furthermore, different spectral detection structures exhibit varying recognition performance across different wavelength bands. Slit grating spectroscopic detection structures can effectively acquire spectral shape variations and broadband response information in the 400nm to 800nm ​​band, making them suitable for extracting gas category differences within the visible to near-infrared short-wavelength range. Fiber optic spectral detection structures, on the other hand, can obtain more stable continuous spectral data in the 800nm ​​to 1200nm band, making them suitable for extracting weak absorption and statistical features in the near-infrared range. Therefore, how to collaboratively detect based on the advantageous wavelengths of different detection branches and combine this with machine learning methods to achieve the identification of complex exhaled breath samples is a pressing technical problem that needs to be solved in this field. Summary of the Invention

[0008] To address the shortcomings of existing technologies, this invention provides an exhaled breath identification system and method based on multi-channel absorption spectroscopy and machine learning. This system can improve the stability of weak absorption feature extraction in complex exhaled breath samples, enhance the accuracy of multi-component gas identification, and reduce the impact of sampling state fluctuations on detection results.

[0009] To achieve the above objectives, the present invention provides the following technical solution:

[0010] A gas identification system based on multi-channel absorption spectroscopy and machine learning includes an exhaled air sampling module, a multi-channel absorption spectroscopy detection module, an electrode electric field preprocessing module, a multi-modal sampling state constraint module, a data processing module, and a machine learning identification module. The exhaled air sampling module collects the exhaled air sample to be detected and introduces it into the detection area. The multi-channel absorption spectroscopy detection module acquires the multi-dimensional spectral response information of the exhaled air sample, including a slit grating spectroscopic detection branch, a non-dispersive infrared detection branch, and a fiber optic spectral detection branch.

[0011] Specifically, the slit grating spectroscopic detection branch includes a light source, a slit, a grating, and an image receiving unit. It acquires the spectroscopic response information of exhaled breath samples in the 400nm to 800nm ​​wavelength range and extracts broadband spectral shape variation features within this range. The non-dispersive infrared detection branch includes an infrared light source, a sample cavity, and an infrared detector. It acquires a carbon dioxide background reference signal and assists in determining the validity of the exhalation process. The fiber optic spectral detection branch includes a combined light source, an optical fiber coupling structure, and a miniature spectrometer. It acquires continuous spectral data of exhaled breath samples in the 800nm ​​to 1200nm wavelength range and extracts the spectral features required for identifying weakly absorbing gas components and predicting their concentration within this range.

[0012] Furthermore, the slit grating spectroscopic detection branch and the fiber optic spectral detection branch complement each other in terms of detection bands. Specifically, the slit grating spectroscopic detection branch exhibits better spectral response acquisition and gas category differentiation capabilities in the 400nm to 800nm ​​band, making it suitable for extracting broad-spectrum variation features in the visible to near-infrared short-wavelength range. The fiber optic spectral detection branch, on the other hand, offers better continuous spectral acquisition stability and feature recognition in the 800nm ​​to 1200nm band, making it suitable for extracting weak absorption features in the near-infrared long-wavelength range. By fusing the advantageous bands of these two branches, the comprehensive identification capability of different target gases in exhaled breath samples can be improved.

[0013] The electrode electric field pretreatment module is located at the front end of the main detection area. By forming an electric field region between two electrode plates, the exhaled gas sample is subjected to an electric field before entering the spectral detection area, thereby enhancing the observability of the spectral response of some typical gas components. Preferably, the electrode electric field pretreatment module consists of two parallel electrode plates with a spacing of 5 mm and an applied voltage of 600 V.

[0014] The multimodal sampling state constraint module includes a visual recognition unit and a thermal imaging recognition unit. The visual recognition unit is a common vision camera used for identifying the sampling object and determining the exhalation position. The thermal imaging recognition unit is a thermal imaging camera used to acquire changes in the temperature field of the exhaled air plume and determine whether the exhalation process is continuous and effective.

[0015] The data processing module is used to preprocess the data obtained from each detection branch, including background correction, absorbance conversion, feature band extraction, statistical feature calculation, and multi-channel feature fusion. The absorbance conversion is based on the calculation of absorbance changes according to the light intensity variation, and its formula is: A = -lg(I / I0), where A is the absorbance, I is the sample light intensity, and I0 is the background light intensity. The data processing module concatenates, filters, or reduces the dimensionality of the features extracted by different detection branches within their respective advantageous bands to form a machine learning input feature vector.

[0016] The machine learning identification module is used to establish a mapping relationship between spectral features and gas categories, gas concentrations, or multi-component compositions. The machine learning model includes one or more of principal component analysis (PCA), K-means clustering, linear discriminant analysis (LDA), support vector machines (SVM), and support vector regression (SVR). Specifically, the machine learning identification module uses an SVM model to map gas samples of acetone, ammonia, methane, ethanol, and methanol to different categories; it uses an SVR model to establish a nonlinear mapping relationship between spectral features and gas concentrations; it uses a multi-label identification method combining feature peak extraction and one-to-many SVM to determine whether a mixed sample contains one or more target gases; and it uses one or more of principal component analysis (PCA), K-means clustering, and LDA for dimensionality reduction, unsupervised clustering, and LDA effect analysis.

[0017] Furthermore, this invention also provides a gas identification method based on multi-channel absorption spectroscopy and machine learning, comprising the following steps:

[0018] Step S1: Collect exhaled air samples through the exhaled air sampling module, and pass the exhaled air samples through the electrode electric field pretreatment area;

[0019] Step S2: The visual recognition unit and thermal imaging recognition unit of the multimodal sampling state constraint module are used to identify the sampling object, determine whether the blowing position is in the preset area, and obtain the temperature distribution change of the exhaled air plume to determine whether the exhalation is continuous. Specifically, when the result of the thermal imaging recognition unit shows that the exhaled air plume continuously enters the detection area, and the result of the visual recognition unit shows that the blowing position meets the requirements, the current sampling is determined to be a valid sampling.

[0020] Step S3: Based on the judgment result of step S2, when the current sample is determined to be a valid sample, the multidimensional spectral response data of the exhaled breath sample is collected collaboratively through the slit grating spectroscopic detection branch, the non-dispersive infrared detection branch, and the fiber optic spectral detection branch in the multi-channel absorption spectral detection module; wherein, the slit grating spectroscopic detection branch acquires the spectroscopic image features and spectral shape change features in the 400nm to 800nm ​​band, the fiber optic spectral detection branch acquires the continuous spectral features, peak features, integral features, and statistical features in the 800nm ​​to 1200nm band, and the non-dispersive infrared detection branch acquires the carbon dioxide background reference features and exhaled breath validity judgment features;

[0021] Step S4: The data processing module performs background correction, absorbance conversion, feature band extraction, and statistical feature calculation on the acquired spectral data, and fuses the features of different detection branches to form a multi-channel spectral feature vector;

[0022] Step S5: Input the multi-channel spectral feature vector into the machine learning identification module. The machine learning model identifies the gas types, concentrations, or multi-component compositions in the exhaled breath sample and outputs the identification results. Specifically, a support vector machine model is used to identify single-component gas categories such as acetone, ammonia, methane, ethanol, and methanol; a support vector regression model is used to establish a non-linear relationship between spectral features and gas concentrations for concentration prediction; and a multi-label identification method combining feature peak extraction and one-to-many support vector machines is used to identify multi-component gases in mixed samples.

[0023] Compared with existing technologies, this invention provides an exhaled breath identification system and method based on multi-channel absorption spectroscopy and machine learning, which has the following beneficial effects:

[0024] (1) The present invention can simultaneously acquire wide-band spectral response, carbon dioxide background reference information and continuous spectral curve by combining slit grating spectral detection, non-dispersive infrared detection and fiber optic spectral detection, thereby improving the integrity of spectral information of complex exhaled breath samples.

[0025] (2) The present invention divides the work according to the advantageous bands of different detection branches, so that the slit grating spectral detection branch mainly undertakes the task of gas category identification in the 400nm to 800nm ​​band, and the fiber optic spectral detection branch mainly undertakes the task of weak absorption feature extraction and identification in the 800nm ​​to 1200nm band, thereby improving the effectiveness of multi-channel spectral features.

[0026] (3) The present invention introduces an electrode electric field pretreatment module at the front end of the detection channel, which can enhance the spectral response of weakly absorbing gas components and make the characteristics of low-concentration target gases easier to extract.

[0027] (4) The present invention constrains the sampling object, blowing position and exhalation continuity by visual recognition and thermal imaging recognition, thereby reducing the impact of sampling posture and exhalation state fluctuations on the detection results.

[0028] (5) The present invention classifies, regresses and identifies multi-channel fusion features through machine learning models, which can adapt to the problem of overlapping multi-component responses and nonlinear feature mapping in complex exhaled breath samples. Attached Figure Description

[0029] Figure 1 Design goals and research ideas for integrated multi-channel spectrometers;

[0030] Figure 2 Schematic diagram of the design objectives and technical positioning of the main detection channel;

[0031] Figure 3 To address the pain points and design goals of the auxiliary channel. Detailed Implementation

[0032] 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.

[0033] Unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of the invention. It should also be understood that, for ease of description, the dimensions of the various parts shown in the drawings are not drawn to actual scale. Techniques, methods, and devices known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and devices should be considered part of the specification. In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may include different values. It should be noted that similar reference numerals and letters in the following figures denote similar items; therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.

[0034] In the description of this application, it should be understood that the terms "center", "longitudinal", "lateral", "upper", "lower", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only used to facilitate the description of the present invention and to simplify the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting the scope of protection of the present invention.

[0035] Example 1

[0036] like Figures 1-3 As shown, this invention proposes an exhaled breath identification system and method based on multi-channel absorption spectroscopy and machine learning. Figure 1 It demonstrates the overall architecture that integrates and complements system integration design, experimental design, and data processing. Figure 2 The structure of three complementary paths—the slit grating beam splitting branch, the non-dispersive infrared detection branch, and the miniature fiber optic spectral branch—is demonstrated in the main detection channel. Figure 3 The working mechanism of the electrode electric field auxiliary branch and the vision-thermal imaging recognition auxiliary branch is demonstrated. This system integrates the main / auxiliary detection branches into the overall enclosure structure, achieving system integration of sample import, optical / gas / electrical path coordination, and simultaneous acquisition and processing of multi-source data. The specific operation of this system includes the following steps:

[0037] Step S1: Collect an exhaled air sample and pass the exhaled air sample through the electrode electric field pretreatment area.

[0038] Specifically, the exhaled air sampling module includes an air inlet, a gas buffer area, a detection chamber, and an exhaust port. The subject blows an exhaled air sample into the air inlet. After buffering, the exhaled air sample enters the detection chamber to reduce the impact of airflow fluctuations on the spectral signal. The exhaled air sample can be a real exhaled air sample or a simulated exhaled air sample composed of background gas, humidity components, and a target gas.

[0039] The electrode electric field preprocessing module consists of two parallel electrode plates, with the exhaled breath sample passing between them. An electric field region is formed between the plates by applying a voltage. Preferably, the distance between the two parallel electrode plates is 5 mm, and the applied voltage is 600 V. After the exhaled breath sample passes through the electric field region, the spectral integral response of some typical gas components is enhanced, thereby improving the stability of subsequent spectral identification. These typical gas components include, but are not limited to, acetone, ammonia, methane, ethanol, methanol, and carbon dioxide.

[0040] Step S2: Use visual recognition and thermal imaging to determine whether the sampling object, blowing position and exhalation continuity meet the detection conditions.

[0041] Specifically, the multimodal sampling state constraint module includes a visual recognition unit (ordinary visual camera) and a thermal imaging recognition unit (thermal imaging camera). The visual recognition unit is used to identify the sampling object and determine whether its blowing position is within a preset area; the thermal imaging recognition unit is used to acquire the temperature distribution changes of the exhaled air plume. When the thermal imaging result shows that the exhaled air plume continuously enters the detection area, and the visual recognition result shows that the blowing position meets the requirements, the system determines that the sampling is valid. That is, when the result of the thermal imaging recognition unit shows that the exhaled air plume continuously enters the detection area, and the result of the visual recognition unit shows that the blowing position meets the requirements, the current sampling is determined to be valid. This step can reduce data fluctuations caused by personnel position deviation, inconsistent sampling posture, and discontinuous exhalation.

[0042] Step S3: Collect multidimensional spectral response data of exhaled breath samples using a multi-channel absorption spectroscopy detection module.

[0043] Specifically, the multi-channel absorption spectroscopy detection module is arranged around the detection cavity, including a slit grating spectroscopic detection branch, a non-dispersive infrared detection branch, and a fiber optic spectroscopy detection branch. These three types of detection branches use changes in light intensity absorption as a unified characterization basis, acquiring absorption spectral information in different dimensions. Combined with... Figure 2As shown, the slit grating spectral detection branch includes a light source, a slit, a grating, and an image receiving unit; the non-dispersive infrared detection branch includes an infrared light source, a sample cavity, and an infrared detector; and the fiber optic spectral detection branch includes a combined light source, an optical fiber coupling structure, and a miniature spectrometer.

[0044] The slit-grating spectroscopic detection branch includes a light source, a slit, a grating, and an image receiving unit. The light beam emitted by the light source passes through the exhaled breath sample and is then split by the slit and grating. The image receiving unit acquires the split image, thus obtaining broadband response information. This branch can obtain significant spectral shape changes in the 400nm to 800nm ​​band, making it suitable for extracting features such as color distribution, grayscale variations, band intensity differences, and overall spectral shape changes from broadband spectroscopic images. Therefore, it exhibits good gas category identification performance within this band.

[0045] The nondispersive infrared detection branch includes an infrared light source, a sample chamber, and an infrared detector. Light of a specific wavelength emitted by the infrared light source passes through the exhaled breath sample and is received by the infrared detector. The change in received light intensity yields a carbon dioxide background reference signal. This signal can be used to help determine whether a valid exhalation process occurred and provides a reference for subsequent spectral data correction.

[0046] The fiber optic spectral detection branch comprises a combined light source, an optical fiber coupling structure, and a miniature spectrometer. Light emitted from the combined light source passes through the exhaled breath sample and is coupled into the optical fiber, where the miniature spectrometer acquires a continuous spectral curve. This branch can obtain more stable continuous spectral data in the 800nm ​​to 1200nm band, making it suitable for extracting near-infrared weak absorption peaks, integral areas, peak intensities, mean, variance, and other statistical characteristics. Therefore, it exhibits good gas identification and concentration modeling performance within this band.

[0047] Step S4: Perform background correction, absorbance conversion, feature band extraction, and statistical feature calculation on the acquired multi-channel spectral data.

[0048] Specifically, the data processing module first collects blank background data and exhaled breath sample data, and then calculates absorbance based on changes in light intensity. The absorbance calculation formula is:

[0049] A = -lg(I / I0)

[0050] Where A is absorbance, I is sample light intensity, and I0 is background light intensity.

[0051] Subsequently, the data processing module extracts features from the data of different detection branches. For the slit grating spectroscopic detection branch, the focus is on extracting spectroscopic image features and spectral shape variation features in the 400nm to 800nm ​​band; for the fiber optic spectral detection branch, the focus is on extracting continuous spectral features, peak features, integral features, and statistical features in the 800nm ​​to 1200nm band; for the nondispersive infrared detection branch, the focus is on extracting carbon dioxide background reference features and exhalation effectiveness judgment features.

[0052] Step S5: The slit grating spectral characteristics in the 400nm to 800nm ​​band, the fiber spectral characteristics in the 800nm ​​to 1200nm band, and the non-dispersive infrared background reference characteristics are spliced, filtered, or reduced in dimensionality to form a machine learning input feature vector.

[0053] Specifically, the slit grating spectroscopic detection branch and the fiber optic spectral detection branch do not simply repeat the acquisition of the same spectral information, but rather divide their functions according to the advantageous wavelength bands of their respective detection structures. The slit grating spectroscopic detection branch is mainly used for gas identification in the 400nm to 800nm ​​wavelength band, while the fiber optic spectral detection branch is mainly used for gas identification in the 800nm ​​to 1200nm wavelength band. The data processing module fuses the features extracted by the two detection branches in their respective advantageous wavelength bands to form a more complete multi-channel spectral feature vector.

[0054] Step S6: Use a machine learning model to identify the types, concentrations, or multi-component compositions of gases in the exhaled breath sample.

[0055] Specifically, the machine learning identification module includes a single-component gas identification submodule, a concentration prediction submodule, and a multi-component gas identification submodule. The machine learning model includes one or more of principal component analysis, K-means clustering, linear discriminant analysis, support vector machine, and support vector regression. The single-component gas identification submodule uses a support vector machine model to map gas samples such as acetone, ammonia, methane, ethanol, and methanol to different categories. The concentration prediction submodule uses a support vector regression model to establish a non-linear relationship between spectral features and gas concentration. The multi-component gas identification submodule employs a multi-label identification method combining feature peak extraction and one-to-many support vector machines to determine whether a mixed sample contains one or more target gases.

[0056] Furthermore, principal component analysis is used to reduce the dimensionality of high-dimensional spectral features and observe sample distribution; K-means clustering is used to analyze the natural clustering of different gas samples under unsupervised conditions; linear discriminant analysis is used to analyze the linear discriminant effect; support vector machine is used to handle nonlinear classification problems caused by weak feature overlap of different gases; and support vector regression is used to establish a nonlinear mapping relationship between spectral features and gas concentration.

[0057] Through the above steps, the present invention can acquire multi-band, multi-dimensional spectral information in complex exhaled gas samples, and improve the reliability of target gas identification through plate electric field enhancement, sampling state constraints, and machine learning identification.

[0058] Example 2

[0059] This embodiment further illustrates the band-based collaborative detection method.

[0060] In this embodiment, the slit grating spectroscopic detection branch is used to acquire spectroscopic response data in the 400nm to 1200nm wavelength range, with a focus on identifying spectral data in the 400-800nm ​​band. Since the slit grating structure in this band can effectively reflect spectral shape changes caused by different gas samples, features such as brightness distribution, color changes, grayscale gradients, spectral band intensity changes, and overall spectral shape changes in the spectroscopic image can be extracted and used as important inputs for gas category identification.

[0061] The fiber optic spectral detection branch is used to acquire continuous spectral data in the 400nm to 1200nm band, with a focus on acquiring infrared spectral data in the 800-1200nm range. Since the fiber optic spectral structure in this band yields more stable continuous spectral curves, features such as absorption peak position, peak intensity, spectral band integral area, mean, variance, peak width, and absorption rate of change can be extracted and used as important inputs for identifying weakly absorbing gas components and predicting their concentration.

[0062] The non-dispersive infrared detection branch is used to acquire the carbon dioxide background reference signal. This signal can not only be used to determine whether the exhalation process is effective, but also participate as a background feature in subsequent spectral feature fusion. As a result, the system can form multi-source fusion features of 400nm to 800nm, 800nm ​​to 1200nm and carbon dioxide background reference information, improving the accuracy and stability of exhaled breath identification.

[0063] Example 3

[0064] This embodiment further illustrates the pretreatment method for the plate electric field.

[0065] In this embodiment, the electrode electric field pretreatment module is located before the exhaled air enters the main detection chamber. Two electrode plates are arranged in parallel, and the exhaled air sample passes between the two electrode plates. By applying a voltage to the electrode plates, an electric field region is formed between the two electrode plates, so that the exhaled air sample is subjected to the electric field before entering the detection area.

[0066] Preferably, the electrode spacing is 5 mm and the applied voltage is 600 V. Under these conditions, the spectral integral responses of typical gas components such as acetone, ammonia, ethanol, methanol, and carbon dioxide can all be enhanced to varying degrees. This preprocessing method can improve the observability of the spectral response of weakly absorbing gas components, making subsequent feature extraction and machine learning recognition more stable.

[0067] Example 4

[0068] This embodiment further illustrates the machine learning identification method.

[0069] In this embodiment, the data processing module first performs background correction and absorbance conversion on the multi-channel spectral data, and then extracts features from the slit grating spectroscopic detection branch, the fiber optic spectral detection branch, and the non-dispersive infrared detection branch, respectively. Subsequently, the features from different branches are spliced, filtered, or dimensionality-reduced to form a machine learning input feature vector.

[0070] For single-component gas identification tasks, a support vector machine (SVM) is used to establish a gas category classification model. For concentration prediction tasks, a support vector regression (SVM) is used to establish a nonlinear mapping model between spectral features and concentration. For multi-component mixed gas identification tasks, a multi-label identification method combining feature peak extraction and one-to-many SVM is used to determine whether the mixed gas sample contains one or more of acetone, ammonia, methane, ethanol, and methanol.

[0071] Through the aforementioned machine learning identification method, this invention can address the problems of overlapping multi-component responses, difficulty in directly interpreting weak absorption features, and significant nonlinear relationships between spectral features and concentration in complex exhaled breath samples.

[0072] This invention improves the stability of weak absorption feature extraction in complex exhaled breath samples and enhances the accuracy of single-component and multi-component gas identification through research and analysis of exhaled breath identification methods and systems based on multi-channel absorption spectroscopy and machine learning. It also strengthens the consistency of the sampling process through visual and thermal imaging recognition. This method can be used for the detection of simulated or real exhaled breath samples, with target gases including but not limited to acetone, ammonia, methane, ethanol, and methanol. It has application value in non-invasive health monitoring, metabolic state analysis, and exhaled breath-assisted detection.

[0073] It should be noted that in this application, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0074] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A gas identification system based on multi-channel absorption spectroscopy and machine learning, characterized in that, include: The exhaled air sampling module is used to collect an exhaled air sample to be detected and introduce the exhaled air sample into the detection area; An electrode electric field pretreatment module is located at the front end of the detection area and is used to apply an electric field to the exhaled air sample. A multi-channel absorption spectroscopy detection module is arranged around the detection area to acquire multi-dimensional spectral response information of the exhaled breath sample. It includes a slit grating spectroscopic detection branch, a non-dispersive infrared detection branch, and a fiber optic spectroscopy detection branch. The multimodal sampling state constraint module includes a visual recognition unit and a thermal imaging recognition unit, used to determine the sampling object, blowing position, and exhalation continuity; the data processing module is connected to the slit grating spectral detection branch, the non-dispersive infrared detection branch, and the fiber optic spectral detection branch, respectively, and is used to preprocess, extract features, and fuse multi-channel features on the acquired detection data; The machine learning identification module, connected to the data processing module, is used to identify single-component gases, multi-component mixed gases, and gas concentrations in exhaled breath samples using a machine learning model based on the data fused from the multi-channel features.

2. The gas identification system based on multi-channel absorption spectroscopy and machine learning according to claim 1, characterized in that, The electrode electric field pretreatment module consists of two parallel electrode plates with a spacing of 5 mm and an applied voltage of 600 V.

3. The gas identification system based on multi-channel absorption spectroscopy and machine learning according to claim 1, characterized in that, The slit grating spectroscopic detection branch includes a light source, a slit, a grating, and an image receiving unit, used to acquire the spectroscopic response information of the exhaled breath sample in the 400nm to 800nm ​​wavelength range; the non-dispersive infrared detection branch includes an infrared light source, a sample cavity, and an infrared detector, used to acquire a carbon dioxide background reference signal; the fiber optic spectral detection branch includes a combined light source, an optical fiber coupling structure, and a miniature spectrometer, used to acquire continuous spectral data of the exhaled breath sample in the 800nm ​​to 1200nm wavelength range.

4. The gas identification system based on multi-channel absorption spectroscopy and machine learning according to claim 1, characterized in that, The visual recognition unit is a common visual camera, used to identify the sampling object and determine whether the blowing position is within a preset area; the thermal imaging recognition unit is a thermal imaging camera, used to acquire the temperature distribution changes of the exhaled air plume.

5. A gas identification system based on multi-channel absorption spectroscopy and machine learning according to claim 1, characterized in that, The data processing module calculates absorbance based on changes in light intensity. The formula for calculating absorbance is: A = -lg(I / I0), where A is absorbance, I is sample light intensity, and I0 is background light intensity. The data processing module splices, filters, or reduces the dimensions of the features extracted by different detection branches in their respective bands to form a machine learning input feature vector.

6. The gas identification system based on multi-channel absorption spectroscopy and machine learning according to claim 1, characterized in that, The machine learning identification module is configured with: a support vector machine model for mapping acetone, ammonia, methane, ethanol, and methanol gas samples to different categories; a support vector regression model for establishing a nonlinear mapping relationship between spectral features and gas concentrations; a multi-label identification method combining feature peak extraction and one-to-many support vector machines for determining whether a mixed sample contains the target gas; and one or more of principal component analysis, K-means clustering, and linear discriminant analysis for dimensionality reduction, unsupervised clustering, and linear discriminant effect analysis.

7. A gas identification method based on multi-channel absorption spectroscopy and machine learning, applied to the system described in any one of claims 1 to 6, characterized in that, Includes the following steps: Step S1: Collect exhaled air samples through the exhaled air sampling module, and pass the exhaled air samples through the electrode electric field pretreatment area; Step S2: Through the visual recognition unit and thermal imaging recognition unit of the multimodal sampling state constraint module, identify the sampling object, determine whether the blowing position is in the preset area, and obtain the temperature distribution change of the exhaled plume to determine whether the exhalation is continuous. Step S3: Based on the judgment result of step S2, when the current sample is determined to be a valid sample, the multidimensional spectral response data of the exhaled breath sample is collected collaboratively through the slit grating spectroscopic detection branch, the non-dispersive infrared detection branch, and the fiber optic spectral detection branch in the multi-channel absorption spectral detection module. Step S4: The data processing module performs background correction, absorbance conversion, feature band extraction, and statistical feature calculation on the acquired spectral data, and fuses the features of different detection branches to form a multi-channel spectral feature vector; Step S5: Input the multi-channel spectral feature vector into the machine learning identification module, and use the machine learning model to identify the gas type, concentration or multi-component composition in the exhaled breath sample, and output the identification result.

8. The gas identification method based on multi-channel absorption spectroscopy and machine learning according to claim 7, characterized in that, In step S3, the slit grating spectroscopic detection branch is used to acquire spectroscopic image features and spectral shape change features in the 400nm to 800nm ​​band; the fiber optic spectral detection branch is used to acquire continuous spectral features, peak features, integral features and statistical features in the 800nm ​​to 1200nm band; and the non-dispersive infrared detection branch is used to acquire carbon dioxide background reference features and exhalation effectiveness judgment features.

9. A gas identification method based on multi-channel absorption spectroscopy and machine learning according to claim 7, characterized in that, In step S5, the identification process of the machine learning model includes: using a support vector machine model to identify single-component gas categories of acetone, ammonia, methane, ethanol, and methanol; using a support vector regression model to establish a nonlinear relationship between spectral features and gas concentration for concentration prediction; and using a multi-label identification method combining feature peak extraction and one-to-many support vector machines to identify multi-component gases in mixed samples.

10. A gas identification method based on multi-channel absorption spectroscopy and machine learning according to claim 7, characterized in that, In step S2, a standard vision camera is used to identify the sampling object and determine whether its blowing position is within a preset area; a thermal imaging camera is used to acquire the temperature distribution change of the exhaled air plume. When the thermal imaging result shows that the exhaled air plume continuously enters the detection area and the visual recognition result shows that the blowing position meets the requirements, the current sampling is determined to be a valid sampling.

Images