Raman spectrum technology-based method for rapidly detecting types and contents of bacteria in gas

Abstract

The invention discloses a method for rapidly detecting the type and content of bacteria in gas based on a Raman spectrum technology, and relates to the technical field of microbiological detection.The method comprises the steps that a gas sample is collected, dried and dewatered, and the sampling volume is determined; the method comprises the following steps: pre-filtering a gas sample, guiding the pre-filtered gas sample to a low-background filter membrane to enrich bacteria, and irradiating the enriched bacteria on the low-background filter membrane by adopting near-infrared laser to generate a scattered Raman signal; raman signals are collected, and only light in a bacteria Raman characteristic peak interval is allowed to pass through in cooperation with a narrow-band optical filter; performing feature extraction and pattern recognition on the Raman signals, classifying to judge the types of bacteria, comparing the intensity of the Raman signals with a calibration curve of bacteria with known concentration, and calculating the concentration of the bacteria in the gas in combination with the sampling volume. Through exclusive gas sampling pretreatment and surface enhanced Raman scattering and fusion of representative Raman peak intensity, rapid bacteria detection with strong interference resistance, stability, reliability, low detection limit, high quantitative accuracy and cross-batch comparability is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of microorganism detection, in particular to a method for rapid detection of bacterial species and content in gas based on Raman spectroscopy. BACKGROUND

[0002] In the field of microorganism detection, rapid and accurate identification of bacterial species and quantity is crucial for preventing infection and ensuring food safety. Traditional bacterial detection methods (such as culture method and PCR technology) have obvious limitations: the culture method takes 24-72 hours, which is too time-consuming; the PCR technology relies on professional equipment and is prone to false positives due to contamination.

[0003] Raman spectroscopy has become a research hotspot for rapid detection because it can achieve non-contact detection through molecular vibration characteristics. However, existing Raman spectroscopy detection faces two major problems: first, the Raman signal of low-concentration bacteria (such as trace bacteria in the environment) is extremely weak, making direct detection prone to missed detection; second, there is a lack of efficient bacterial enrichment and detection integration scheme, resulting in a cumbersome detection process and low efficiency, which makes it difficult to meet real-time monitoring needs.

[0004] The closest prior art to the present application is a single bacterial detection method based on Raman spectroscopy using DMD and microfluidics. This method uses a laser, DMD, microscope, CCD camera, microfluidic chip, and Raman spectroscopy detection and analysis platform. Through sample preparation, injection into a microfluidic chip (to construct a static flow field to limit cells), 633nm laser excitation of Raman signal, beam expansion and filter processing (to separate imaging and spectral analysis light paths), DMD control of light spot to capture cells and collect Raman spectra, and signal preprocessing and comparison with known database to determine the presence of microorganisms, species, and metabolic state. This solution is limited in application scenarios, only suitable for liquid samples, and relies on the precise coordination of microfluidic chips and DMD optical tweezers, resulting in high equipment complexity. It is difficult to directly apply to gas sample detection, and there is no signal enhancement design, which may limit the detection sensitivity of low-abundance bacteria. SUMMARY

[0005] Based on the shortcomings of the existing technology described above, the present application aims to provide a method for rapid detection of bacterial species and content in gas based on Raman spectroscopy to solve the above technical problems.

[0006] To achieve the above-mentioned purpose, the present application provides the following technical solution: a method for rapid detection of bacterial species and content in gas based on Raman spectroscopy, comprising: A flow-controllable device is used to collect gas samples, and a drying agent or low-temperature condensation device is used to dry and remove water. A honeycomb or grid structure airflow rectifying device is used to make the gas flow field entering the sampling system uniform and stable. The sampling time and sampling flow rate are recorded to determine the sampling volume. The gas sample was pre-filtered by coarse filtration using a polyester fiber membrane and then by medium filtration using a glass fiber membrane. The pre-filtered gas sample was directed to a low-background filter membrane to enrich bacteria, and a nano-reinforcement substrate was modified on the surface of the low-background filter membrane to form a surface-enhanced Raman scattering hotspot to enhance the Raman signal of the bacteria. Near-infrared laser is used, which is expanded to a predetermined spot diameter by a lens group and then irradiates the enriched bacteria on the low background filter membrane to generate scattered Raman signals. Raman signals scattered by a low-background filter are collected using avalanche diodes or photomultiplier tubes, and narrowband filters are used to allow light from the characteristic Raman peaks of bacteria to pass through only. Feature extraction and pattern recognition were performed on the Raman signal. Principal component analysis was used to extract distinguishing features based on a known microbial Raman spectrum database, and support vector machine was used for classification to determine the bacterial species. The bacterial concentration in the gas was calculated by comparing the Raman signal intensity with the calibration curve of known bacterial concentrations and combining the sampling volume.

[0007] The present invention is further configured such that the flow controllable device controls the sampling flow rate through a precision valve or pump body, and calculates the sampling volume based on the sampling flow rate and sampling time.

[0008] The present invention is further configured such that the polyester fiber membrane and the glass fiber membrane have a porous structure; the pore size of the polyester fiber membrane is larger than that of the glass fiber membrane, and the polyester fiber membrane is used to trap large particulate impurities larger than the pore size of the polyester fiber membrane; the glass fiber membrane is used to trap small oil droplets and dust agglomerates larger than the pore size of the glass fiber membrane.

[0009] The present invention is further configured such that the low background filter membrane is a quartz fiber filter membrane or a polytetrafluoroethylene filter membrane, and a nano-reinforced substrate is modified on the surface of the low background filter membrane to amplify the Raman signal of bacteria by utilizing the surface plasmon resonance effect.

[0010] The present invention is further configured to divide the detection area of ​​the low background filter membrane into multiple detection grids; Short-exposure multi-frame acquisition was performed on each detection grid under different excitation powers, and median fusion was performed on the multi-frame signals under the same power to obtain the response relationship between power and peak intensity. Linear fitting is performed on the low-power range of the response relationship, and the linearity score is calculated. The noise intensity is calculated in the baseline featureless interval to obtain the peak intensity to noise ratio, and the median is set as the signal-to-noise ratio score. The equivalent hotspot consistency factor is obtained by taking the weighted geometric average of the linearity score and the signal-to-noise ratio score. Based on the equivalent hotspot consistency factor, the Raman intensity of each detection grid is uniformly balanced, and the representative Raman peak intensity is obtained by weighted fusion.

[0011] The present invention is further configured such that the calculation logic for the linearity score includes: Select multiple excitation powers below a preset power threshold to construct a corresponding data sequence of power and peak intensity; Perform least-squares fitting on the data sequence to obtain the fitted straight line; The goodness of fit between the actual data and the fitted line is calculated, and the goodness of fit is used as the linearity score to characterize the linearity of the response of the detection grid in the low power range.

[0012] The present invention is further configured such that the calculation logic for the representative Raman peak intensity includes: The Raman intensity of each detection grid is normalized according to the corresponding equivalent hotspot consistency factor to obtain the balanced Raman intensity. The equivalent hotspot consistency factor of each detection grid is normalized to obtain the weighting coefficient of each detection grid. The balanced Raman intensity is multiplied by the weighting coefficient, and the results of all detection grids are summed to obtain the representative Raman peak intensity, which is used for bacterial species identification and concentration calculation.

[0013] The present invention is further configured to perform baseline subtraction, smoothing and wavenumber axis alignment on the acquired Raman signal, and normalize the intensity. Principal component analysis was performed using the processed Raman signal as input to extract distinguishing features and form feature vectors. Representative Raman peak intensities were incorporated into the feature vectors as independent features, and the incorporated feature vectors were standardized. A support vector machine classification model is trained based on a known microbial Raman spectroscopy database, and the feature vectors of the test samples are classified to output the bacterial species and their confidence scores. When the confidence score is lower than a preset threshold, a verification prompt is given. Under known concentration conditions, a mapping relationship between representative Raman peak intensity and bacterial concentration is established. The mapping relationship is obtained through linear or nonlinear regression. The representative Raman peak intensity of the sample to be tested is substituted into the mapping relationship and converted in combination with the sampling volume to obtain the bacterial concentration. The regression residual and the prediction interval are checked, and an over-range warning is given when the range exceeds the effective range.

[0014] The present invention is further configured to select samples with reliable annotations from the database and generate a training feature set according to the preprocessing and feature construction process consistent with the test samples; A support vector machine classification model is trained using the training feature set. The kernel function type, kernel function parameters, and penalty coefficient are determined through cross-validation. The classification performance is evaluated using the validation set. The decision output of the classification model is subjected to probability calibration, the classification decision value is mapped to confidence level, and the confidence level threshold is determined based on the validation data; Input the feature vector of the sample to be tested into the classification model, and output the bacterial species and their corresponding confidence scores; When the confidence level is lower than the confidence level threshold, a verification prompt is given and the sample is marked as needing verification.

[0015] The present invention is further configured to obtain multiple bacterial gas samples of different concentrations under known concentration conditions, obtain the corresponding representative Raman peak intensities, establish the mapping relationship between the representative Raman peak intensities and bacterial concentrations using linear regression, and determine the model form, parameters, effective range and working interval through validation data. To obtain the representative Raman peak intensity of the sample to be tested, substitute the representative Raman peak intensity into the mapping relationship to obtain the bacterial quantity or volume fraction corresponding to the sampling volume, and convert it according to the sampling volume to obtain the bacterial concentration per cubic meter. The conversion results are verified, the prediction interval is calculated and the regression residual is evaluated. When the intensity of the representative Raman peak or the converted concentration exceeds the effective range, or the width of the prediction interval and the regression residual exceed the preset threshold, an over-range prompt or a retest prompt is given, and the result is marked as needing to be retested.

[0016] This invention provides a rapid detection method for bacterial species and content in gas based on Raman spectroscopy. The method involves collecting gas samples using a flow-controlled device and drying them with a desiccant or low-temperature condenser. A honeycomb or grid-structured airflow rectifying device ensures a uniform and stable gas flow field entering the sampling system. Sampling time and flow rate are recorded to determine the sampling volume. The gas sample is pre-filtered using a coarse filtration stage with a polyester fiber membrane and a medium filtration stage with a glass fiber membrane. The pre-filtered gas sample is then directed to a low-background filter membrane to enrich bacteria. A nano-reinforcement substrate is modified on the surface of the low-background filter membrane to form surface-enhanced Raman scattering hotspots, enhancing bacterial Raman scattering. Raman signal generation: Near-infrared laser light is used, expanded to a predetermined spot diameter by a lens group, to irradiate enriched bacteria on a low-background filter membrane, generating scattered Raman signals. Avalanche diodes or photomultiplier tubes are used to collect the Raman signals scattered by the low-background filter membrane, and a narrow-band filter is used to allow only light within the characteristic Raman peak range of the bacteria to pass through. Feature extraction and pattern recognition are performed on the Raman signals. Principal component analysis is used to extract distinguishing features based on a known microbial Raman spectral database, and support vector machine is used for classification to determine bacterial species. The bacterial concentration in the gas is calculated by comparing the Raman signal intensity with a calibration curve of known bacterial concentrations and combining this with the sampling volume. The beneficial effects include: 1. Dedicated to gas sample detection, breaking through the limitations of existing technologies: It adopts a "controllable flow device + rectification device" to achieve quantitative collection of gas samples, and accurately calculates the sampling volume by controlling the flow rate and sampling time; it adds a drying and dehydration stage to remove water vapor in the gas, avoiding condensation on the filter membrane and interference of water peaks on the Raman signal; it designs a multi-stage pre-filtration (PET fiber membrane coarse filtration + glass fiber membrane medium filtration) to specifically remove impurities such as food debris and oil particles in the gas, solving the problem of complex impurity types in gas samples.

[0017] 2. Ultra-high sensitivity for accurate detection of low-concentration gaseous bacteria: Introducing SERS (Surface Enhanced Raman Scattering) technology, by modifying the filter membrane surface with a nano-enhancing substrate, the Raman signal is amplified by 10⁻¹⁰ ohms using surface plasmon resonance. 6 -10 8 This breakthrough overcomes the bottleneck of weak signals at low concentrations; it employs avalanche diodes / photomultiplier tubes as signal acquisition devices, both of which have single-photon detection capabilities, enabling the capture of weak signals after SERS enhancement and achieving a detection limit of "identifiable single bacteria".

[0018] 3. Strong anti-interference capability, reducing the impact of background signals on detection results: Low background filter membranes are selected to reduce the Raman background of the filter membrane itself from the source; multi-stage pre-filtration is used to avoid the Raman signals of impurities from masking the characteristic peaks of bacteria; near-infrared lasers are used to reduce fluorescence interference from biological samples, and narrow-band filters are used to allow light in the bacterial characteristic peak range to pass through only, further filtering ambient light interference.

[0019] 4. Improved detection stability and reliability: The lens group increases the diameter of the laser beam, reduces the power density, minimizes light damage to bacteria and thermal deformation of the filter membrane, and ensures the stability of the bacterial structure; the rectifier optimizes the airflow, avoids uneven bacterial distribution caused by turbulence, ensures stable distribution of enriched bacteria on the filter membrane, and improves the consistency of signal excitation. 5. Suppressing Hotspot Inhomogeneity and Improving Repeatability, Quantitative Accuracy, and Cross-Batch Comparability: By using a consistency factor composed of linearity and signal-to-noise ratio to screen, gain balance, and weighted fusion of multi-grid, multi-power short-exposure data to obtain representative Raman peak intensities, the system bias caused by surface enhancement hotspot inhomogeneity can be effectively reduced. This improves repeatability and result consistency across grids and films, enhances quantitative accuracy and linearity, and reduces detection and quantitation limits. It also improves robustness to interferences such as power fluctuations, spot shifts, and local contamination. Furthermore, it serves as a robust engineering feature for species identification and as a core input for concentration conversion, facilitating the output of uncertainty and quality control criteria, and enabling comparability of results across batches and time periods.

[0020] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, specific embodiments of this application are given below. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings: Figure 1 This is a flowchart illustrating a rapid detection method for bacterial species and content in gas based on Raman spectroscopy, as an exemplary embodiment of the present invention. Detailed Implementation

[0022] The embodiments of the present invention will be described below with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are only for illustrating the present invention and not for limiting the scope of protection of the present invention.

[0023] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0024] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the invention.

[0025] First, it's important to understand that the principle of Raman spectroscopy is as follows: When a laser irradiates bacteria, some photons undergo inelastic collisions with biomolecules within the bacteria, such as proteins, nucleic acids, lipids, and carbohydrates. This causes a change in photon energy, which corresponds to molecular vibrations, thus producing a unique Raman spectrum. Different bacteria have different cell structures and biochemical compositions, and their Raman spectra's characteristic peaks (such as peak position, peak intensity, and peak shape) are also specific, much like "molecular fingerprints." By analyzing these characteristic peaks, it's possible to identify bacterial species, analyze their structure, and even assess their physiological state. Bacterial enrichment technology is a commonly used technique that utilizes the pore size of filter membranes to selectively retain bacteria. Its core principle is to select a filter membrane with a matching pore size (e.g., 0.22 μm or 0.45 μm) based on the typical size of the target bacteria (most bacteria have a diameter of 0.5-5 μm). When a sample containing bacteria (such as water, body fluids, or food extracts) passes through the filter membrane, the bacteria are trapped because they cannot penetrate the membrane pores, while small molecule impurities and solvents in the sample can pass through smoothly, thus achieving separation of bacteria from other components. In this way, bacteria that were originally dispersed in a large volume of sample are concentrated on the surface of the filter membrane, effectively increasing the bacterial concentration per unit volume. This solution is a process for detecting bacteria in gases, achieving bacterial concentration analysis and bacterial species detection through multi-stage synergy.

[0026] Rapid detection methods for bacterial species and concentration in gases based on Raman spectroscopy, such as Figure 1 As shown, it includes: Gas samples are collected using a flow controllable device and dried by a desiccant or a low-temperature condenser. A honeycomb or grid-structured airflow rectifier is used to ensure a uniform and stable gas flow field entering the sampling system. The sampling time and flow rate are recorded to determine the sampling volume. The gas sample was pre-filtered by coarse filtration using a polyester fiber membrane and then by medium filtration using a glass fiber membrane. The pre-filtered gas sample was directed to a low-background filter membrane to enrich bacteria, and a nano-reinforcement substrate was modified on the surface of the low-background filter membrane to form a surface-enhanced Raman scattering hotspot to enhance the Raman signal of the bacteria. Near-infrared laser is used, which is expanded to a predetermined spot diameter by a lens group and then irradiates the enriched bacteria on the low background filter membrane to generate scattered Raman signals. Raman signals scattered by a low-background filter are collected using avalanche diodes or photomultiplier tubes, and narrowband filters are used to allow light from the characteristic Raman peaks of bacteria to pass through only. Feature extraction and pattern recognition were performed on the Raman signal. Principal component analysis was used to extract distinguishing features based on a known microbial Raman spectrum database, and support vector machine was used for classification to determine the bacterial species. The bacterial concentration in the gas was calculated by comparing the Raman signal intensity with the calibration curve of known bacterial concentrations and combining the sampling volume.

[0027] Specifically, the sample was in a volume of 0.5 m³. 3 The sample gas is generated and collected within a controlled cavity, with a relative humidity ≤30%; the sampling flow rate is 2 L / min, the sampling time is 15 min, and the corresponding sampling volume is 30 L. The sample gas is sequentially passed through a 10 μm PET fiber membrane for coarse filtration and a 5 μm glass fiber membrane for medium filtration, and finally enriched on a 0.22 μm quartz fiber membrane (alternative: polytetrafluoroethylene membrane). The excitation wavelength is 785 nm, and the beam is expanded to approximately 8 mm spot diameter by a lens group; power settings are 5% / 10% / 15% / 20% / 25%, with 5 short exposures per setting, and each frame is 80 ms. The filter membrane detection area is divided into a 3×3 grid (9 grids); the target peak window half-width Δ = 5 cm. -1 Baseline range 1800–1900 cm - ¹. Linearity score threshold ≥ 0.90; reference signal-to-noise ratio (SNR) = 12; consistency factor admission threshold ≥ 0.75 (0.50–0.75 weighted). Total time for a single measurement and processing ≈ 7–12 min. During gas sample collection, a flow control device controls the sampling flow rate through a precision valve or pump, calculating the sampling volume based on the sampling flow rate and sampling time to ensure the accuracy of bacterial concentration calculation. The rectifier uses a honeycomb or grid-structured airflow rectifier to ensure a uniform and stable gas flow field entering the sampling system, avoiding uneven bacterial distribution caused by turbulence and reducing deviations in subsequent enrichment stages. Moisture in the gas is removed using a desiccant or low-temperature condensation device, reducing the gas humidity to below 30%. This prevents water vapor from condensing on the filter membrane surface, thus avoiding bacterial aggregation, and also eliminates the interference of water molecule Raman characteristic peaks on bacterial characteristic peaks.

[0028] The invention is further configured such that both the polyester fiber membrane and the glass fiber membrane have a porous structure; the pore size of the polyester fiber membrane is larger than that of the glass fiber membrane, and the polyester fiber membrane is used to trap large particulate impurities larger than the pore size of the polyester fiber membrane; the glass fiber membrane is used to trap small oil droplets and dust agglomerates larger than the pore size of the glass fiber membrane. Specifically, the purpose of pre-filtration is to specifically remove impurities in the gas that may interfere with detection, protect subsequent filter membranes, and reduce signal noise. The first stage (coarse filtration) uses a polyester (PET) fiber membrane, which has the characteristics of being resistant to oil and having high mechanical strength (resistant to airflow impact). The 10μm pore size design can accurately trap large particulate impurities >10μm (such as food scraps, hair, and dust clumps). This step can prevent large particles from directly clogging the subsequent fine-pore filter membrane and extend the service life of the filter membrane. The second stage (medium filtration) uses a glass fiber membrane with a 5μm pore size, which can trap small oil droplets (such as volatile oils from meat / dairy products in a refrigerator) and dust agglomerates of 5-20μm. The porous structure of the glass fiber membrane combines adsorption and air permeability, which can efficiently capture oil (oil easily adheres to the glass fiber surface) without affecting the passage of bacteria (0.5-5μm), thus clearing obstacles for subsequent bacterial enrichment. The two-stage filtration forms a "gradient interception", removing impurities in layers from large to small, minimizing the interference of impurities on the Raman signal (such as the Raman peak of oil may mask the characteristic peak of bacteria).

[0029] The purpose of bacterial enrichment via filter membrane is to efficiently enrich low concentrations of bacteria while reducing background interference. Combined with SERS technology to enhance the signal, this invention further specifies that the low-background filter membrane is a quartz fiber filter membrane or a polytetrafluoroethylene (PTFE) filter membrane, with a nano-reinforcement substrate modified on the surface of the low-background filter membrane to amplify the Raman signal of the bacteria using surface plasmon resonance. Specifically, in the filter membrane selection logic, quartz fiber filter membranes are preferred because they have extremely low Raman background, only between 400-1800 cm⁻¹. -1 (The main range of bacterial characteristic peaks) contains a weak Si-O bond peak (approximately 1080 cm⁻¹). - ¹), and this peak position is far from the core characteristic peak of bacteria, avoiding background signals from masking the target signal; the 0.22μm pore size can retain 100% of all bacteria, ensuring no missed detection; PTFE filter membrane (alternative): excellent hydrophobic properties (contact angle >110°), which can reduce water vapor residue on the filter membrane surface and avoid bacteria from being affected by water vapor agglomeration; although the 0.45μm pore size has a slightly lower rejection rate for extremely small bacteria than the quartz fiber membrane, it is suitable for scenarios with higher requirements for hydrophobicity. By modifying the filter membrane surface with a nano-reinforced substrate, the Raman signal of bacteria is amplified by the surface plasmon resonance effect (enhancement factor can reach 10). 6 -10 8 This addresses the issue of low bacterial concentration in the refrigerator (<100 CFU / m³). 3 This addresses the weak signal issue caused by [unclear], achieving ultra-high sensitivity where "a single bacterium can be identified".

[0030] When exciting Raman signals, it is crucial to stabilize the Raman signal of bacteria while minimizing damage to both bacteria and equipment. 785nm near-infrared laser selection: Compared to 633nm visible light lasers, 785nm lasers cause less light damage to biological samples (lower energy, reducing photochemical effects) and offer superior penetration in gas-filter membrane systems, effectively exciting Raman scattering from bacteria enriched on the filter membrane surface. Beam expansion: A convex lens group expands the laser beam diameter from the initial 1-2mm to 5-10mm, reducing the beam power density (power density = power / spot area) and preventing high-energy laser ablation of bacteria (leading to structural damage) or damage to the filter membrane (e.g., PTFE membranes are easily deformed at high temperatures). Simultaneously, beam expansion improves the spatial uniformity of the beam, ensuring consistent bacterial excitation within the laser irradiation area and enhancing signal stability.

[0031] When collecting Raman signals, efficient acquisition of weak Raman scattering signals ensures the detectability of low-concentration bacteria. Avalanche diode (APD) / photomultiplier tube (PMT) selection: Both have single-photon detection capabilities, capable of capturing extremely weak Raman scattering photons, especially suitable for signals that remain weak even after SERS enhancement. APDs offer fast response times (nanosecond level), suitable for rapid scanning detection; PMTs offer higher gain (up to 10). 6 (Multiple times), suitable for signal acquisition in ultra-low temperature environments. Signal acquisition logic: By focusing the scattered light through the optical path, the light is converged onto the photosensitive surface of the detector. In conjunction with a narrow-band filter (which only allows light from the bacterial Raman characteristic peak range to pass through), ambient light interference is reduced, ensuring that the detector only receives the target signal and improving the signal-to-noise ratio.

[0032] The present invention is further configured to divide the detection area of ​​the low background filter membrane into multiple detection grids; specifically, nine grids of 3×3 are divided on the membrane, with indices g=1,…,9; For each detection grid, short-exposure multi-frame acquisition was performed at different excitation powers. Median fusion was then performed on the multi-frame signals at the same power to obtain the power-peak intensity response relationship. Specifically, five nominal power levels were set {5%, 10%, 15%, 20%, 25%}, with a single-frame exposure of 80ms. K=5 frames were acquired for each power level, with the target peak center wavenumber being v0 (determined according to the target bacterial library), and the peak window half-width Δ=5cm. -1 The baseline noise range is selected as 1800-1900 cm⁻¹, and at each power level... Down, For power range index, , ,Record Frame spectrum ,exist Calculate the peak height or integral for each frame to obtain... ; A linear fit is performed on the low-power range of the response relationship, and a linearity score is calculated; the present invention further specifies that the calculation logic for the linearity score includes: Multiple excitation powers below a preset power threshold are selected to construct a data sequence corresponding to power and peak intensity; specifically, the low power threshold is set to 15% of the nominal power; three low power levels are selected: 5%, 10%, and 15%, denoted as... Short exposures were used to capture images at each power level. For each frame, the median peak intensity of the target peak window is taken as the robust peak intensity for that frame. , , For power Corresponding robust peak strength; The least squares linear fitting model is obtained by performing a least squares fit on the data sequence. Slope With intercept (Ordinary least squares closed-form solution): , ,in, and This is the arithmetic mean of power and peak intensity. , , Calculate the fitted value , For the linear model, the first The predicted peak strength of the file; Calculate the goodness of fit between the actual data and the fitted straight line, using the goodness of fit as a linearity score to characterize the linearity of the detection grid's response in the low-power range; calculate the residual sum of squares and the total sum of squares: , Goodness of fit (coefficient of determination): ,Will Defined as linearity score: ; Noise intensity is calculated in the baseline featureless region to obtain the peak intensity to noise ratio, and the median is set as the signal-to-noise ratio score; specifically, for the same level... Frame median: Calculate the standard deviation in the baseline region. ; Calculate the signal-to-noise ratio of a single range ;Get the median Normalize the median signal-to-noise ratio. ; The equivalent hotspot consistency factor is obtained by taking a weighted geometric average of the linearity score and the signal-to-noise ratio score; specifically... , Weights can be calibrated and obtained; default weights are available. It is 0.7. It is 0.3; Based on the equivalent hotspot consistency factor, the Raman intensities of each detection grid are uniformly balanced, and the representative Raman peak intensity is obtained through weighted fusion. The present invention further specifies that the calculation logic for the representative Raman peak intensity includes: The Raman intensity of each detection grid is normalized according to the corresponding equivalent hotspot consistency factor to obtain the balanced Raman intensity. Specifically, to suppress the bias of low-quality grids and highlight the reliable contribution of high-quality grids, the peak intensity of each grid is normalized and balanced according to its consistency factor. ,in For the balancing index (preferred) (The default value is 0.5). A higher value enhances the quality of the mesh. A threshold can be set for obviously substandard meshes. (e.g., 0.5), when When necessary, remove or set to zero; The equivalent hotspot consistency factor of each detection grid is normalized to obtain the weighting coefficient of each detection grid; specifically, the consistency factor of each grid is normalized to obtain the weight. ; The balanced Raman intensity is multiplied by the weighting coefficients, and the results of all detection grids are summed to obtain the representative Raman peak intensity, which is used for bacterial species identification and concentration calculation; specifically, the balanced intensity is weighted and summed with the weights to obtain the representative Raman peak intensity. .

[0033] Furthermore, under the same conditions as Example 1, three paths were compared: ① no balancing or weighting; ② linearity screening only; ③ the proposed method (linearity + signal-to-noise ratio → consistency factor, γ=0.5 balancing, δ=1 weighting). The cross-grid coefficient of variation (CV) was approximately 24%, 15%, and 8–10%, respectively. After using the representative Raman peak intensity for the calibration curve, the quantitative linear region expanded by approximately 1.6 × 10⁻⁶, and the median low-end residual decreased by approximately 35%. Within the range of 80–120 CFU / m³, the detection rate increased from ~72% to ~93%.

[0034] The present invention is further configured to perform baseline subtraction, smoothing, and wavenumber axis alignment on the acquired Raman signal, and normalize the intensity. Specifically, baseline subtraction uses an adaptive baseline algorithm to eliminate slowly varying backgrounds, with parameters fixed to remain consistent within a single experimental period; smoothing employs polynomial smoothing (with a fixed window and order) to suppress high-frequency noise while maintaining peak shape; wavenumber axis alignment uses a stable intrinsic parameter peak or consensus peak to correct for minor wavenumber drift, controlling the deviation of the main peak position before and after correction within an allowable threshold; intensity normalization involves uniform scaling of the entire spectrum based on the total area or the maximum peak value to ensure comparability of intensities between different samples. Principal component analysis is performed using the processed Raman signal as input to extract distinguishing features and form feature vectors. The intensity of representative Raman peaks is incorporated into the feature vectors as independent features, and the incorporated feature vectors are standardized uniformly. Specifically, the preprocessed full spectrum is used as input to extract several principal components with a cumulative variance explanation rate of not less than a preset threshold (e.g., reaching 90% explanation rate), resulting in a principal component score vector. The intensity of representative Raman peaks is incorporated into the above vectors as independent engineering features. If necessary, a reference peak ratio feature (the ratio of representative peak intensity to reference peak intensity) is added to improve robustness. The entire set of features after splicing is standardized in the same way, and the standardized parameters used in the inference stage are saved. A support vector machine (SVM) classification model is trained based on a known microbial Raman spectroscopy database. The model classifies the feature vectors of test samples, outputting bacterial species and their confidence scores. A verification prompt is given when the confidence score falls below a preset threshold. Specifically, the training set is used with samples from the known microbial Raman spectroscopy database, and training features are generated following a preprocessing and feature construction process identical to that used for the test samples. SVM is used for training, with the kernel type and penalty parameters determined through cross-validation. Accuracy, recall, and macro-average metrics are evaluated using an independent validation set. The classification output is probabilistically calibrated, mapping decision values ​​to confidence scores. Feature vectors are generated for the test samples, input into the SVM, and the bacterial species and corresponding confidence scores are output. When the confidence score falls below a preset threshold, a "verification required" message is displayed. Under known concentration conditions, a mapping relationship between representative Raman peak intensities and bacterial concentrations is established, obtained through linear or nonlinear regression. The representative Raman peak intensities of the sample to be tested are substituted into the mapping relationship and converted using the sampling volume to obtain the bacterial concentration. The regression residuals and prediction intervals are checked, and an out-of-range warning is given when the concentration exceeds the effective range. Specifically, bacterial aerosols are generated under known concentration conditions (covering low, medium, and high concentrations, at least five points), and the corresponding representative Raman peak intensities are obtained following the same procedure as for the sample to be tested. A linear or nonlinear regression model is established with the representative Raman peak intensities as the independent variable and the corresponding known concentration as the dependent variable. The model form and parameters are determined through hold-out or cross-validation, and the effective range and working interval are recorded. The representative Raman peak intensities of the sample to be tested are substituted into the established mapping relationship to obtain the bacterial quantity or volume fraction corresponding to the sampling volume. Unit conversion is performed using the sampling volume to output the bacterial concentration per cubic meter.

[0035] The invention further comprises selecting reliably labeled samples from a database and generating a training feature set according to a preprocessing and feature construction process consistent with that of the test samples. Specifically, based on a known microbial Raman spectroscopy database (containing standard spectra of thousands of bacteria, such as Escherichia coli and Staphylococcus aureus), as a "benchmark" for species identification, a standard library is established by selecting multiple representative bacteria (including at least Escherichia coli, Staphylococcus aureus, Pseudomonas aeruginosa, and Bacillus subtilis). The standard library contains 4 types of bacteria; each type has ≥3 batches of independent culture; each batch has ≥30 spectra; and the total number of spectra is ≥360. The entire spectrum is normalized after baseline subtraction, smoothing, and wavenumber axis alignment; principal components are extracted using PCA, with a cumulative variance explanation rate ≥90%; representative Raman peak intensities are incorporated as independent features and uniformly standardized. SVM (RBF kernel) is used, with stratified 5-fold cross-validation and probability calibration. Without the inclusion of representative peak intensities, the macro-average accuracy is ≈92%; after inclusion, it is ≈96%, and the recall rate for low signal-noise samples is improved by ≈5–7 pct. With a confidence threshold of 0.85, the rejection rate is approximately 6%, and the pass rate upon review is approximately 82%; the spectral range is Raman frequency shift of 400–1800 cm⁻¹. -1 A metal nano-reinforcement layer was modified on a quartz fiber filter membrane; excitation was performed at 785 nm; APD or PMT detection was conducted; sampling and SERS were performed according to the aforementioned steps and representative Raman peak intensities were obtained. A support vector machine (SVM) classification model is trained using the training feature set. The kernel function type, kernel function parameters, and penalty coefficient are determined through cross-validation, and the classification performance is evaluated using the validation set. Specifically, for the SVM, a radial basis function kernel is preferred. Within the training set, a logarithmic grid of penalty coefficient and kernel width is searched using hierarchical cross-validation to select the optimal combination for validation performance. If necessary, a simple early stopping strategy is used to prevent overfitting. The decision output of the classification model is probabilistically calibrated to map the classification decision value to a confidence level, and the confidence level threshold is determined based on the validation data. Specifically, the classification output is probabilistically calibrated on the validation set (e.g., using common calibration methods) to reduce overconfidence or conservative estimation. The confidence level threshold is determined by combining the accuracy, recall and false alarm cost of the validation set, and the corresponding performance metrics and confusion matrix are recorded. Input the feature vector of the sample to be tested into the classification model, and output the bacterial species and their corresponding confidence scores; When the confidence level is lower than the confidence level threshold, a verification prompt is given and the sample is marked as needing verification.

[0036] This invention is further configured to: acquire bacterial gas samples of different concentrations under known concentration conditions, obtain corresponding representative Raman peak intensities, establish a mapping relationship between representative Raman peak intensities and bacterial concentrations using linear regression, and determine the model form, parameters, effective range, and working interval through validation data; specifically, the bacterial strains include at least one or more of *Escherichia coli*, *Staphylococcus aureus*, *Pseudomonas aeruginosa*, and *Bacillus subtilis* for quantification; prepare five or more levels (e.g., near-low, low, medium, high, and upper limit) under known concentration conditions to cover the expected application range; each level has at least three independent samples; generate bacterial aerosols of the target concentration in a closed or ventilated controlled cavity using a nebulizer or standard generator, and begin collection after stabilization; sample and test the same batch of aerosols using plate counting or an equivalent authoritative method as a reference value for that level of concentration; repeat sampling is performed if necessary to confirm batch-to-batch consistency; the sampling order of different concentration levels is randomly arranged to avoid deviations caused by system drift; blank collection is performed after each sampling to monitor the baseline; To obtain the representative Raman peak intensity of the sample to be tested, substitute the representative Raman peak intensity into the mapping relationship to obtain the bacterial quantity or volume fraction corresponding to the sampling volume. Based on the sampling volume, calculate the bacterial concentration per cubic meter. Specifically, acquire the spectrum according to the aforementioned multi-grid, multi-power short exposure strategy, and read the peak intensity of each grid at the same operating power. Calculate the linearity score and signal-to-noise ratio score to obtain the equivalent hotspot consistency factor. Grids that do not meet the standards are removed or weighted down. Perform consistency balancing and weighted fusion on the peak intensities of each grid to obtain the representative Raman peak intensity. Simultaneously save the weighted variance as a reference for subsequent uncertainty. Use the same operating power, peak window, preprocessing parameters, and sampling protocol as the calibration and subsequent tests. Pair the representative Raman peak intensity corresponding to each concentration level with the reference concentration to form a calibration dataset. Use linear regression to establish the mapping relationship between intensity and concentration. Use hold-out or cross-validation to evaluate the goodness of fit and bias. Based on the fitting residual, prediction interval, signal-to-noise ratio, and monotonicity, give the effective range and recommended operating interval. Record the corresponding decision threshold and acceptance criteria. The conversion results are verified, the prediction interval is calculated, and the regression residual is evaluated. When the representative Raman peak intensity or the converted concentration exceeds the effective range, or the prediction interval width and regression residual exceed the preset threshold, an out-of-range warning or retest warning is given, and the result is marked as requiring verification. Specifically, the representative Raman peak intensity of the sample to be tested is obtained according to the same procedure as the calibration, and the sampling volume is recorded. The representative Raman peak intensity is substituted into the established linear mapping relationship to obtain the bacterial quantity or volume fraction corresponding to the sampling volume. The concentration is then converted to per cubic meter based on the sampling volume. Degree; confirm that the working power, peak window, preprocessing and weighting parameters are consistent with those in the calibration stage; if inconsistent, refuse calculation and prompt to resample according to the calibration parameters; calculate the prediction interval and evaluate the residual based on the error distribution of the calibration model; when the prediction interval width or residual exceeds the preset threshold, it is judged as a suspicious result; if the representative Raman peak intensity or converted concentration falls outside the effective range, directly give an out-of-range prompt; for suspicious or out-of-range samples, give retest suggestions (such as extending the sampling time, increasing the number of effective grids, checking humidity and pre-filter blockage, and verifying laser power and alignment).

[0037] Compared with existing technologies, this invention, based on multi-grid and multi-power short-exposure acquisition, introduces a low-power region linearity score to constrain response predictability and a signal-to-noise ratio score to constrain spectral quality. The weighted geometric average of these two scores constitutes a monotonic evaluation of hotspot quality. Under the guidance of this evaluation, gain balancing and weighted fusion are implemented to form representative Raman peak intensities. Examples show that removing any step significantly degrades cross-grid variation, low-concentration detection rate, classification robustness, and quantitative residuals. Using the complete process reduces the cross-grid variation coefficient from approximately 24% to approximately 8% to 10%, and improves the detection rate and expands the quantitative linear region in the low-concentration range. Therefore, each step is a necessary technical means to solve the problems of hotspot heterogeneity and spectral quality fluctuations in gaseous scenarios, producing quantifiable and verifiable technical effects.

[0038] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A rapid detection method for bacterial species and content in gas based on Raman spectroscopy, characterized in that, include: Gas samples are collected using a flow controllable device and dried by a desiccant or a low-temperature condenser. A honeycomb or grid-structured airflow rectifier is used to ensure a uniform and stable gas flow field entering the sampling system. The sampling time and flow rate are recorded to determine the sampling volume. The gas sample was pre-filtered by coarse filtration using a polyester fiber membrane and then by medium filtration using a glass fiber membrane. The pre-filtered gas sample was directed to a low-background filter membrane to enrich bacteria, and a nano-reinforcement substrate was modified on the surface of the low-background filter membrane to form a surface-enhanced Raman scattering hotspot to enhance the Raman signal of the bacteria. Near-infrared laser is used, which is expanded to a predetermined spot diameter by a lens group and then used to irradiate the enriched bacteria on the low background filter membrane to generate scattered Raman signals. Raman signals scattered by a low-background filter are collected using avalanche diodes or photomultiplier tubes, and narrowband filters are used to allow light from the characteristic Raman peaks of bacteria to pass through only. Feature extraction and pattern recognition were performed on the Raman signal. Principal component analysis was used to extract distinguishing features based on a known microbial Raman spectrum database, and support vector machine was used for classification to determine the bacterial species. The bacterial concentration in the gas was calculated by comparing the Raman signal intensity with the calibration curve of known bacterial concentrations and combining the sampling volume.

2. The rapid detection method for bacterial species and content in gas based on Raman spectroscopy according to claim 1, characterized in that, The flow controllable device controls the sampling flow rate through a precision valve or pump, and calculates the sampling volume based on the sampling flow rate and sampling time.

3. The rapid detection method for bacterial species and content in gas based on Raman spectroscopy according to claim 1, characterized in that, Polyester fiber membranes and glass fiber membranes have porous structures; the pore size of polyester fiber membranes is larger than that of glass fiber membranes. Polyester fiber membranes are used to trap large particulate impurities larger than the pore size of polyester fiber membranes; glass fiber membranes are used to trap small oil droplets and dust agglomerates larger than the pore size of glass fiber membranes.

4. The rapid detection method for bacterial species and content in gas based on Raman spectroscopy according to claim 1, characterized in that, The low background filter membrane is a quartz fiber filter membrane or a polytetrafluoroethylene filter membrane. A nano-reinforced substrate is modified on the surface of the low background filter membrane to amplify the Raman signal of bacteria by utilizing the surface plasmon resonance effect.

5. The rapid detection method for bacterial species and content in gas based on Raman spectroscopy according to claim 1, characterized in that, Multiple detection grids are divided in the detection area of ​​the low-background filter membrane; Short-exposure multi-frame acquisition was performed on each detection grid under different excitation powers, and median fusion was performed on the multi-frame signals under the same power to obtain the response relationship between power and peak intensity. Linear fitting is performed on the low-power range of the response relationship, and the linearity score is calculated. The noise intensity is calculated in the baseline featureless interval to obtain the peak intensity to noise ratio, and the median is set as the signal-to-noise ratio score. The equivalent hotspot consistency factor is obtained by taking the weighted geometric average of the linearity score and the signal-to-noise ratio score. Based on the equivalent hotspot consistency factor, the Raman intensity of each detection grid is uniformly balanced, and the representative Raman peak intensity is obtained by weighted fusion.

6. The rapid detection method for bacterial species and content in gas based on Raman spectroscopy according to claim 5, characterized in that, The calculation logic for the linearity score includes: Select multiple excitation powers below a preset power threshold to construct a data sequence corresponding to power and peak intensity; Perform least-squares fitting on the data sequence to obtain the fitted straight line; The goodness of fit between the actual data and the fitted line is calculated, and the goodness of fit is used as the linearity score to characterize the linearity of the response of the detection grid in the low power range.

7. The rapid detection method for bacterial species and content in gas based on Raman spectroscopy according to claim 5, characterized in that, The calculation logic for the intensity of a representative Raman peak includes: The Raman intensity of each detection grid is normalized according to the corresponding equivalent hotspot consistency factor to obtain the balanced Raman intensity. The equivalent hotspot consistency factor of each detection grid is normalized to obtain the weighting coefficient of each detection grid. The balanced Raman intensity is multiplied by the weighting coefficient, and the results of all detection grids are summed to obtain the representative Raman peak intensity, which is used for bacterial species identification and concentration calculation.

8. The rapid detection method for bacterial species and content in gas based on Raman spectroscopy according to claim 5, characterized in that, The acquired Raman signals were subjected to baseline subtraction, smoothing, and wavenumber axis alignment, and the intensity was normalized. Principal component analysis was performed using the processed Raman signal as input to extract distinguishing features and form feature vectors. Representative Raman peak intensities were incorporated into the feature vectors as independent features, and the incorporated feature vectors were standardized. A support vector machine classification model is trained based on a known microbial Raman spectroscopy database, and the feature vectors of the test samples are classified to output the bacterial species and their confidence scores. When the confidence score is lower than a preset threshold, a verification prompt is given. Under known concentration conditions, a mapping relationship between representative Raman peak intensity and bacterial concentration is established. The mapping relationship is obtained through linear or nonlinear regression. The representative Raman peak intensity of the sample to be tested is substituted into the mapping relationship and converted in combination with the sampling volume to obtain the bacterial concentration. The regression residual and the prediction interval are checked, and an over-range warning is given when the range exceeds the effective range.

9. The rapid detection method for bacterial species and content in gas based on Raman spectroscopy according to claim 8, characterized in that, Select samples with reliable annotations from the database and generate a training feature set according to the same preprocessing and feature construction process as the samples to be tested; A support vector machine classification model is trained using the training feature set. The kernel function type, kernel function parameters, and penalty coefficient are determined through cross-validation. The classification performance is evaluated using the validation set. The decision output of the classification model is subjected to probability calibration, the classification decision value is mapped to confidence level, and the confidence level threshold is determined based on the validation data; Input the feature vector of the sample to be tested into the classification model, and output the bacterial species and their corresponding confidence scores; When the confidence level is lower than the confidence level threshold, a verification prompt is given and the sample is marked as needing verification.

10. The rapid detection method for bacterial species and content in gas based on Raman spectroscopy according to claim 8, characterized in that, Under known concentration conditions, multiple bacterial gas samples of different concentrations were obtained, and the corresponding representative Raman peak intensities were acquired. Linear regression was used to establish the mapping relationship between the representative Raman peak intensities and bacterial concentrations. The model form, parameters, effective range, and working interval were determined through validation data. To obtain the representative Raman peak intensity of the sample to be tested, substitute the representative Raman peak intensity into the mapping relationship to obtain the bacterial quantity or volume fraction corresponding to the sampling volume, and convert it according to the sampling volume to obtain the bacterial concentration per cubic meter. The conversion results are verified, the prediction interval is calculated and the regression residual is evaluated. When the intensity of the representative Raman peak or the converted concentration exceeds the effective range, or the width of the prediction interval and the regression residual exceed the preset threshold, an over-range prompt or a retest prompt is given, and the result is marked as needing to be retested.

Images