Bacterial real-time detection method and system

Abstract

The application provides a bacterial real-time detection method and system, which can be applied to the technical field of microorganisms. The bacterial real-time detection method comprises the following steps: collecting a plurality of Raman spectra of a sample at a preset period within a preset time length to obtain a time spectrum image matrix, wherein the time spectrum image matrix comprises a plurality of Raman spectra arranged in a time sequence; the time spectrum image matrix comprises spectral information and time dynamic information; performing low-rank reconstruction on the time spectrum image matrix to obtain a pretreated spectrum image; and performing classification on the pretreated spectrum image by using a trained two-dimensional convolutional neural network to obtain a bacterial detection result of the sample.

Description

Technical Field

[0001] At least one embodiment of the present invention relates to the field of microbial detection, and more specifically to a method and system for real-time bacterial detection. Background Technology

[0002] Raman optical tweezers combine Raman spectroscopy with optical tweezers to enable label-free, culture-independent, real-time, precise capture and non-contact separation of single-celled bacteria, and can be used for transient molecular identification. Raman optical tweezers offer advantages in non-invasiveness, minimal sample preparation, and direct single-cell detection.

[0003] However, in terms of real-time performance and practicality, the real-time bacterial detection technology based on Raman optical tweezers suffers from technical problems such as signal instability and severe noise interference caused by long-term exposure, which affects detection efficiency and identification accuracy. Summary of the Invention

[0004] In view of the above problems, the present invention provides a method and system for real-time detection of bacteria based on time-correlated Raman spectroscopy to improve detection efficiency and identification accuracy.

[0005] According to a first aspect of the present invention, a method for real-time bacterial detection is provided, the method comprising: acquiring multiple Raman spectra of a sample at a preset period within a preset time duration to obtain a time-spectral image matrix, the time-spectral image matrix comprising the multiple Raman spectra arranged in order of acquisition time; wherein the time-spectral image matrix contains spectral information and temporal dynamic information; performing low-rank reconstruction on the time-spectral image matrix to obtain a preprocessed spectral image; and classifying the preprocessed spectral image using a trained two-dimensional convolutional neural network to obtain the bacterial detection result of the sample.

[0006] According to an embodiment of the present invention, the acquisition of multiple Raman spectra of a sample within a preset time period and at a preset cycle includes: using the objective lens of a laser confocal Raman system to continuously acquire Raman scattering generated by laser excitation of the sample through the laser confocal Raman system within the preset time period; and using a spectrometer connected to the objective lens to integrate the Raman scattering for the preset time period to obtain the multiple Raman spectra.

[0007] According to an embodiment of the present invention, the above-mentioned low-rank reconstruction of the time-spectral image matrix to obtain a preprocessed spectral image includes: using alternating least squares method to perform low-rank approximation on the time-spectral image matrix to obtain the above-mentioned preprocessed spectral image.

[0008] According to an embodiment of the present invention, the method of low-rank approximation of the time-spectral image matrix using alternating least squares to obtain the preprocessed spectral image includes: constructing a solution matrix; in the i-th iteration of the alternating least squares method, determining the i-th residual matrix based on the i-th solution matrix and the time-spectral image matrix; determining the i-th search direction and the i-th scaling factor according to the i-th residual matrix; multiplying the i-th search direction and the i-th scaling factor to obtain the i-th adjusted search direction; linearly combining the i-th solution matrix and the i-th adjusted search direction to obtain the (i+1)-th solution matrix; performing the (i+1)-th iteration when the norm of the (i+1)-th solution matrix and the i-th solution matrix is ​​less than or equal to a preset threshold; and determining the (i+1)-th solution matrix as the preprocessed spectral image when the norm of the (i+1)-th solution matrix and the i-th solution matrix is ​​greater than the preset threshold; wherein i is a natural number.

[0009] According to an embodiment of the present invention, the above-mentioned classification of the preprocessed spectral image using a trained two-dimensional convolutional neural network to obtain the bacterial detection result of the sample includes: using the feature extraction module of the trained two-dimensional convolutional neural network to extract features from the preprocessed spectral image to obtain a feature map of the preprocessed spectral image; and using the feature classification module of the trained two-dimensional convolutional neural network to classify the feature map to obtain the bacterial detection result of the sample.

[0010] According to an embodiment of the present invention, the trained two-dimensional convolutional neural network is obtained by the following operation: using multiple Raman spectra continuously acquired for a sample containing predetermined bacteria, the two-dimensional convolutional neural network is trained to obtain the trained two-dimensional convolutional neural network.

[0011] According to an embodiment of the present invention, the preset period is 0.1 seconds.

[0012] A second aspect of the present invention provides a real-time bacterial detection system, comprising: a data acquisition device adapted to acquire multiple Raman spectra of a sample at a preset period within a preset time duration to obtain a time-spectral image matrix, wherein the time-spectral image matrix includes the multiple Raman spectra arranged in chronological order of acquisition time; wherein the time-spectral image matrix contains spectral information and temporal dynamic information; a processing device adapted to perform low-rank reconstruction on the time-spectral image matrix to obtain a preprocessed spectral image; and to classify the preprocessed spectral image using a trained two-dimensional convolutional neural network to obtain bacterial detection results of the sample, wherein the two-dimensional convolutional neural network is trained to extract spectral features and temporal dynamic features.

[0013] According to an embodiment of the present invention, the acquisition device includes: an objective lens of a laser confocal Raman system, adapted to continuously acquire Raman scattering generated by laser excitation of the sample through the laser confocal Raman system within the preset time period; and a spectrometer connected to the objective lens, adapted to integrate the Raman scattering for the preset period to obtain the plurality of Raman spectra.

[0014] According to an embodiment of the present invention, the above-described processing apparatus is also suitable for using alternating least squares method to perform low-rank approximation on the above-described time-spectral image matrix to obtain the above-described preprocessed spectral image.

[0015] According to embodiments of the present invention, Raman scattering is continuously acquired within a preset time period, and integration is performed at a preset period, effectively mitigating spectral instability caused by motion and reducing interference. Acquiring multiple short exposures instead of a single continuous measurement effectively captures bacterial movement within each acquisition frame, preserving spectral characteristics while extracting valuable temporal information. Low-rank reconstruction of the temporal spectral image matrix obtained in chronological order of acquisition improves the signal-to-noise ratio. For the proposed new framework, the rate of temporal change is used as a key discriminative feature for single-cell detection. A two-dimensional convolutional neural network can simultaneously extract spectral features and temporal dynamic features, processing information in both the spectral and temporal dimensions. Compared to analyzing static spectra, this adds a discriminative mode based on time-related spectral change information, improving identification accuracy. Attached Figure Description

[0016] The above-mentioned contents, as well as other objects, features and advantages of the present invention, will become clearer from the following description of embodiments of the present invention with reference to the accompanying drawings.

[0017] Figure 1 A flowchart of a real-time bacterial detection method according to an embodiment of the present invention is shown.

[0018] Figure 2 A block diagram of a real-time bacterial detection system according to an embodiment of the present invention is shown. Detailed Implementation

[0019] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the invention. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the invention for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concept of the invention.

[0020] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the invention. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0021] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0022] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0023] Bacterial contamination poses a significant threat to public health, food safety, and environmental sustainability. Globally, more than 600 million foodborne illnesses are caused by bacterial contamination each year. Pathogens such as Salmonella, Escherichia coli, and Staphylococcus aureus not only cause economic losses but can also lead to life-threatening infections. To address the threat of bacterial contamination, the development of real-time and on-site bacterial detection systems is crucial. These systems can intervene promptly to stop transmission and disasters, support precision medicine, and reduce antibiotic resistance caused by drug abuse.

[0024] Bacterial detection methods in related technologies, including culture-based assays, PCR (polymerase chain reaction), and immunological assays, are increasingly showing limitations in rapid detection. While culture-based assays are cost-effective and specific, they require incubation periods of several days, failing to meet the demands for rapid detection. PCR, through DNA (deoxyribonucleic acid) amplification, offers high sensitivity, but its cumbersome procedures and specialized equipment hinder its widespread adoption in environments with limited infrastructure. Immunological assays provide relatively rapid results through antigen-antibody interactions, but still require optimized incubation conditions and multiple washing steps, introducing delays even in optimized laboratory environments. Furthermore, these methods generally suffer from stringent reagent storage conditions and reliance on centralized infrastructure, limiting their application in dynamic, real-time detection.

[0025] Raman optical tweezers combine Raman spectroscopy with optical tweezers to achieve label-free, culture-independent, real-time, precise capture and non-contact separation of single-celled bacteria, enabling instantaneous molecular identification. Raman optical tweezers offer advantages in non-invasiveness, minimal sample preparation, and direct single-cell detection. However, this detection technology faces three major technical bottlenecks in terms of real-time performance and practicality:

[0026] First, the Raman signal itself is very weak. The Raman signal from a single bacterium is inherently weak, requiring a longer integration time to obtain sufficient signal intensity. This longer integration time leads to unstable spectral measurements, directly affecting detection accuracy, especially for single-celled bacterial detection that requires continuous signal acquisition. This makes it unsuitable for real-time, rapid detection scenarios, limiting the practical application of Raman technology.

[0027] Second, interference from bacterial dynamic behavior. Using longer integration times to improve signal quality, typically 3 to 20 seconds or even longer, can introduce additional noise during the long integration process. During this extended integration period, bacteria may exhibit dynamic behaviors such as motility, morphological changes, and metabolic fluctuations, thus affecting detection reliability.

[0028] Third, laser-induced thermal effects. In Raman optical tweezers detection, commonly used for single-cell detection in microfluidics, laser-induced thermal effects complicate the detection process. To minimize bacterial damage, low laser power is typically used, further extending the integration time. The laser-induced thermal effect also causes a gradual increase in local temperature, exacerbating changes in cell motility and physiological state, leading to Raman spectral instability and reducing the accuracy of bacterial identification.

[0029] Integration time is the exposure duration for which the spectrometer "continuously collects photons," essentially "superimposing and amplifying" the originally weak Raman signal: the longer the time, the more photons are accumulated, and the stronger the signal. However, just as handheld cameras are affected by camera shake during long exposures, if the Raman signal is integrated for too long, noise such as bacterial movement, laser power fluctuations, and environmental vibrations will also be "recorded," leading to blurred or distorted spectra, thus affecting detection efficiency and identification accuracy.

[0030] In view of this, the present invention provides a method and system for real-time bacterial detection. By using time-correlated Raman spectroscopy to shorten the integration time during the acquisition of Raman scattering signals and introducing time-varying information, single-cell bacteria are identified based on the spectral and temporal dynamic characteristics of the acquired Raman spectra, thereby improving detection efficiency and identification accuracy and realizing real-time detection of single-cell bacteria.

[0031] Figure 1 A flowchart of a real-time bacterial detection method according to an embodiment of the present invention is shown.

[0032] like Figure 1As shown, the real-time bacterial detection method includes operations S110 to S130.

[0033] In operation S110, multiple Raman spectra of the sample are acquired at a preset period within a preset duration to obtain a time-spectral image matrix. The time-spectral image matrix includes multiple Raman spectra arranged in the order of acquisition time. The time-spectral image matrix contains spectral information and time dynamic information.

[0034] In operation S120, low-rank reconstruction is performed on the time-spectral image matrix to obtain a preprocessed spectral image.

[0035] In operation S130, the preprocessed spectral image is classified using a trained two-dimensional convolutional neural network to obtain the bacterial detection results of the sample.

[0036] Based on spectral and temporal dynamic information, a two-dimensional convolutional neural network is trained to extract spectral and temporal dynamic features.

[0037] According to embodiments of the present invention, Raman scattering is continuously acquired within a preset time period, and integration is performed at a preset period, effectively mitigating spectral instability caused by motion and reducing interference. Acquiring multiple short exposures instead of a single continuous measurement effectively captures bacterial movement within each acquisition frame, preserving spectral characteristics while extracting valuable temporal information. Low-rank reconstruction of the temporal spectral image matrix obtained in chronological order of acquisition improves the signal-to-noise ratio. For the proposed new framework, the rate of temporal change is used as a key discriminative feature for single-cell detection. A two-dimensional convolutional neural network can simultaneously extract spectral features and temporal dynamic features, processing information in both the spectral and temporal dimensions. Compared to analyzing static spectra, this adds a discriminative mode based on time-related spectral change information, improving identification accuracy.

[0038] The following is a specific example to help understand the inventive concept of this application.

[0039] Bacterial suspensions were prepared as samples. Five representative foodborne bacterial strains were selected from the BeNa Culture Collection (BNCC): Escherichia coli (BNCC186347), Campylobacter jejuni (BNCC360736), Listeria monocytogenes (BNCC185986), Staphylococcus aureus (BNCC376051), and Salmonella typhimurium (BNCC365311).

[0040] Campylobacter jejuni was cultured in a sterile liquid medium, such as Mueller-Hinton broth (Genview, Tianjin Dingguo Biotechnology Co. Ltd.), under microaerophilic conditions with 10% carbon dioxide concentration. The culture was then soaked at 37°C with continuous stirring at 175 revolutions per minute for 16 hours.

[0041] The other four bacterial strains were cultured under aerobic conditions in trypsin-soy broth (Haibo Biotechnology, Tianjin Dingguo Biotechnology Co., Ltd.) at the same temperature for a slightly shorter time, approximately 12 hours. After incubation, the bacterial cultures were harvested, and the cell density was adjusted to approximately 10⁸ colonies per milliliter. To remove any residual growth medium, the samples were centrifuged at 15,000 times gravity for 3 minutes. The supernatant was discarded, and the resulting bacterial pellet was washed twice with sterile phosphate-buffered saline (PFS) to remove any residual culture components. Finally, a clean bacterial suspension was prepared by resuspending the washed pellet in PFS for subsequent Raman spectroscopy analysis.

[0042] According to an embodiment of the present invention, the objective lens of a laser confocal Raman system is used to continuously acquire Raman scattering generated by laser excitation of a sample within a preset time period. The Raman scattering is then integrated using a spectrometer connected to the objective lens at preset time intervals to obtain multiple Raman spectra.

[0043] In one embodiment, the laser confocal Raman system can be a diode laser confocal Raman system with an output wavelength of 671 nm. The laser excites the bacterial suspension, and individual bacterial cells in the suspension produce Raman scattering. The spectrometer can be a spectrometer with a diffraction grating of 1200 lines / mm. Individual bacterial cells are visualized using an inverted optical microscope equipped with a 60x objective. Raman scattering from individual bacterial cells can be acquired using an objective lens with a numerical aperture (NA) of 0.85 and a working distance of 0.2 mm. The bacterial suspension is injected into a microfluidic channel made of polydimethylsiloxane, allowing each bacterial cell to be presented individually. A focused laser with a spot size of approximately 1 μm and an output power of 20 mW is used in the microfluidic channel to apply a light-trapping force.

[0044] In one embodiment, microfluidic channels can be used to continuously detect single bacterial cells, and also at 400... ~2200 Raman displacement with 1 The spectral resolution captures Raman scattering.

[0045] According to embodiments of the present invention, real-time detection of bacterial Raman scattering can be achieved through continuous integration, thereby enabling simultaneous analysis of single-cell compositional fingerprints and temporal evolution patterns. The inherent temporal coherence of continuous Raman spectroscopy is utilized to stabilize spectral characteristics, thereby reducing interference from bacterial dynamic behavior and laser-induced thermal effects.

[0046] According to an embodiment of the present invention, the preset period can be 0.1 seconds. Integrating over an exposure duration of 0.1 seconds can capture transient molecular vibration information.

[0047] In one embodiment, the preset duration can be 10 seconds, and the preset period can be 0.1 seconds. Integrating over a 10-second acquisition duration with an exposure time of 0.1 seconds, 100 Raman spectra can be continuously acquired.

[0048] Arranging 100 Raman spectra in chronological order of acquisition yields a time-spectral image matrix. Time-spectral image matrix The low-rank structure arises from the temporal redundancy generated by stable cellular structures, and from the combination of spatial spectral separability reflected by the different molecular fingerprints of bacteria and substrates.

[0049] According to an embodiment of the present invention, the alternating least squares method is used to perform low-rank approximation on the time-spectral image matrix to obtain a preprocessed spectral image.

[0050] According to embodiments of the present invention, an alternating least squares method is used to iteratively refine the singular value distribution. The iterative process achieves a balance between noise removal efficiency and preservation of true signal transformation by dynamically adjusting parameters. Low-rank constrained time processing mitigates the inherent signal instability in long-exposure protocols, thereby effectively separating transient noise from true spectral information.

[0051] According to an embodiment of the present invention, a solution matrix is ​​constructed. At the beginning of the iteration, the solution matrix can be set. It is zero.

[0052] In the alternating least squares method In the nth iteration operation, based on the nth Solution matrix and time-spectral image matrix Determine the first Each residual matrix According to the first Each residual matrix Determine the first Search directions and the Each proportional factor Optimize the first step size Search directions This minimizes the residual error in the interval [0,1] to determine the optimal first... Each proportional factor The first Search directions and the Each proportional factor Multiply to get the first... One adjusted search direction The first Solution matrix With the One adjusted search direction By performing a linear combination, we obtain the (i+1)th solution matrix. .

[0053] When the Solution matrix With the Solution matrix The norm is less than or equal to a preset threshold. Execute the first This is the second iteration operation.

[0054] When the Solution matrix With the Solution matrix The norm is greater than a preset threshold Determine the first The solution matrix is ​​a preprocessed spectral image, where, It is a natural number.

[0055] In one embodiment, the number of iterations can be set to 20. In low signal-to-noise ratio (SNR) environments, the number of iterations can be set to 30 to 50. The preset threshold can be set to 0.001 to 0.01. Smaller preset thresholds generally prioritize noise removal efficiency to improve the SNR. However, overly strict constraints can suppress changes in the true signal and introduce temporal aliasing errors. It is necessary to weigh the selection of the number of iterations and the preset threshold according to experimental conditions to achieve a balance between noise removal efficiency and preservation of the true signal transformation.

[0056] According to an embodiment of the present invention, a feature extraction module of a trained two-dimensional convolutional neural network is used to extract features from a preprocessed spectral image to obtain a feature map of the preprocessed spectral image. Furthermore, a feature classification module of the trained two-dimensional convolutional neural network is used to classify the feature map to obtain the bacterial detection results of the sample.

[0057] In one embodiment, a 2D convolutional neural network may include 7 convolutional blocks and 3 pooling layers. Each of the first three convolutional blocks includes a convolutional layer, a batch normalization layer, and an activation layer. The convolutional kernels of the convolutional layers can be set to 3×5, with the number of channels progressively increasing from 32 to 128, capturing multi-scale features. The batch normalization layer plays a crucial role in stabilizing training, accelerating convergence, and mitigating the vanishing / exploding gradient problem. Activation layers, such as ReLU activation layers, introduce non-linear transformations, enabling the 2D convolutional neural network to learn more complex feature representations. Each of the last four convolutional blocks includes a convolutional layer and an activation layer (e.g., using ReLU activation), extracting hierarchical features. The number of channels progressively increases from 128 to 2048, enhancing feature representation capabilities. The 3 pooling layers can be placed after the 2nd, 5th, and 7th convolutional blocks, respectively. The convolutional kernels of the pooling layers can be set to 2×2 with a stride of 2, reducing parameter complexity and controlling feature map size.

[0058] In one embodiment, the 2D convolutional neural network may further include three fully connected layers. After feature extraction, the 2D convolutional neural network can enter the fully connected layers through three consecutive dimensionality reduction stages. For example, the first fully connected layer contains 1024 neurons, the second fully connected layer contains 512 neurons, and the third fully connected layer contains 256 neurons. Each fully connected layer includes an activation layer, such as using the ReLU activation function, to non-linearly transform high-dimensional features, enhancing the 2D convolutional neural network's ability to express complex patterns. Adding a dropout layer after the first fully connected layer, such as a Dropout (regularization) layer with a dropout probability of 50%, can alleviate overfitting and noise interference.

[0059] In one embodiment, the two-dimensional convolutional neural network may also include a classifier such as a Softmax classifier (multi-class classifier), which outputs the probability prediction results of multi-class classification to obtain the bacterial detection results of the sample.

[0060] According to embodiments of the present invention, an optimized two-dimensional convolutional neural network can reduce computational load and improve inference speed for analyzing preprocessed spectral images, capturing the spatiotemporal correlation and molecular-specific patterns of multiple Raman spectra. By separating temporal dynamics from signal attenuation, the limitations of traditional Raman techniques are overcome, enabling robust detection of bacterial features under dynamic conditions.

[0061] According to an embodiment of the present invention, the trained two-dimensional convolutional neural network is obtained by the following operation: training the two-dimensional convolutional neural network using multiple Raman spectra continuously acquired for a sample containing predetermined bacteria, thereby obtaining the trained two-dimensional convolutional neural network.

[0062] The beneficial effects of the time-correlated Raman spectroscopy-based real-time bacterial detection method of this invention were verified through experiments.

[0063] Clean bacterial suspensions were prepared targeting the main pathogens of foodborne illnesses: Escherichia coli, Campylobacter jejuni, Listeria monocytogenes, Staphylococcus aureus, and Salmonella typhimurium. A total of 2500 single-cell Raman spectra were collected using a diode laser confocal Raman system with a wavelength of 671 nm integrated with a microfluidic device.

[0064] Classification was performed within a preset timeframe of 10 seconds using both conventional Raman spectroscopy and time-correlated Raman spectroscopy. Conventional Raman spectroscopy employed continuous acquisition, with each Raman spectrum having a complete integration time of 10 seconds. Time-correlated Raman spectroscopy collected multiple Raman spectra at 0.1-second sampling intervals within the 10-second preset timeframe, and then sorted them temporally to obtain a time-spectral image matrix. A convolutional neural network was used to classify the Raman spectra obtained using conventional Raman spectroscopy. A two-dimensional convolutional neural network was used to classify the time-spectral image matrix obtained using time-correlated Raman spectroscopy.

[0065] Using 1500 Raman spectra as the training set and 1000 Raman spectra as the test set, the overall classification accuracy of five bacteria using traditional Raman spectroscopy reached 93.0% within a 10-second integration time. The time-correlated Raman spectroscopy technique significantly improved the overall classification accuracy to 97.7%. Experimental results demonstrate that the time-correlated Raman spectroscopy technique is significantly superior to the traditional Raman spectroscopy technique.

[0066] The time-correlated Raman spectroscopy technique also improved the identification accuracy of each bacterial species, reaching 98.5% for Escherichia coli, 97.0% for Chlamydia jejuni, 98.0% for Listeria monocytogenes, 96.5% for Staphylococcus aureus, and 97.0% for Salmonella typhimurium.

[0067] The real-time bacterial detection method based on time-correlated Raman spectroscopy in this invention effectively mitigates motion-induced spectral instability and improves the signal-to-noise ratio by collecting instantaneous continuous Raman spectra and extracting time-correlated features. By acquiring multiple short exposures instead of a single continuous measurement, time-correlated Raman spectroscopy effectively captures bacterial motion within each acquisition frame, preserving spectral characteristics while extracting valuable temporal information. Combining a two-dimensional convolutional neural network enhances the robustness of bacterial identification by capturing spectral and temporal dynamics, achieving continuous performance improvements across different integration times. This surpasses the accuracy bottleneck observed in traditional Raman spectroscopy techniques. The real-time bacterial detection method based on time-correlated Raman spectroscopy can rapidly and reliably classify bacteria in food safety monitoring, clinical pathogen detection, and environmental monitoring. It can be widely applied in fields such as microbial detection, food safety, and medical diagnostics, providing an efficient and reliable solution for single-cell bacterial detection.

[0068] The time-correlated Raman spectroscopy-based real-time bacterial detection method of this invention for single-cell bacterial analysis overcomes the limitations of traditional Raman spectroscopy and paves the way for high-precision, label-free bacterial detection in dynamic environments. Future research will focus on further optimizing this method, extending its applicability to a wider range of bacterial species, and integrating it into portable Raman detection platforms for field-deployable biosensing applications.

[0069] Figure 2 A block diagram of a real-time bacterial detection system according to an embodiment of the present invention is shown.

[0070] like Figure 2 As shown, the real-time bacterial detection system 200 includes a collection device 210 and a processing device 220.

[0071] According to an embodiment of the present invention, the acquisition device 210 is suitable for acquiring multiple Raman spectra of a sample at a preset period within a preset time duration to obtain a time-spectral image matrix. The time-spectral image matrix includes multiple Raman spectra arranged in order of acquisition time, wherein the time-spectral image matrix contains spectral information and temporal dynamic information. The processing device 220 is suitable for performing low-rank reconstruction on the time-spectral image matrix to obtain a preprocessed spectral image. The preprocessed spectral image is then classified using a trained two-dimensional convolutional neural network to obtain the bacterial detection result of the sample, wherein the two-dimensional convolutional neural network is trained to extract spectral features and temporal dynamic features.

[0072] According to an embodiment of the present invention, the acquisition device 210 includes: an objective lens of a laser confocal Raman system and a spectrometer connected to the objective lens. The objective lens of the laser confocal Raman system is suitable for continuously acquiring Raman scattering generated by a laser-excited sample through the laser confocal Raman system within a preset time period. The spectrometer connected to the objective lens is suitable for integrating the Raman scattering at preset periodic times to obtain multiple Raman spectra.

[0073] According to an embodiment of the present invention, the processing device 220 is also suitable for using alternating least squares to perform low-rank approximation on the time-spectral image matrix to obtain a preprocessed spectral image.

[0074] According to an embodiment of the present invention, the processing device 220 is also suitable for using a feature extraction module of a trained two-dimensional convolutional neural network to extract features from a preprocessed spectral image, thereby obtaining a feature map of the preprocessed spectral image. Furthermore, the processing device 220 is also suitable for using a feature classification module of a trained two-dimensional convolutional neural network to classify the feature map, thereby obtaining the bacterial detection results of the sample.

[0075] Those skilled in the art will understand that the features described in the various embodiments of the present invention can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in the present invention. In particular, the features described in the various embodiments of the present invention can be combined and / or combined in various ways without departing from the spirit and teachings of the present invention. All such combinations and / or combinations fall within the scope of the present invention.

[0076] The embodiments of the present invention have been described above. However, these embodiments are merely illustrative and not intended to limit the scope of the invention. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. Various substitutions and modifications can be made by those skilled in the art without departing from the scope of the invention, and all such substitutions and modifications should fall within the scope of the invention.

Claims

1. A method for real-time detection of bacteria, characterized in that, The real-time bacterial detection method includes: Multiple Raman spectra of a sample are acquired at a preset period within a preset time duration to obtain a time-spectral image matrix, wherein the time-spectral image matrix includes the multiple Raman spectra arranged in the order of acquisition time; wherein the time-spectral image matrix contains spectral information and temporal dynamic information; Low-rank reconstruction is performed on the time-spectral image matrix to obtain a preprocessed spectral image; and The preprocessed spectral image is classified using a trained two-dimensional convolutional neural network to obtain the bacterial detection results of the sample; The step of performing low-rank reconstruction on the time-spectral image matrix to obtain a preprocessed spectral image includes: The alternating least squares method is used to perform a low-rank approximation on the time-spectral image matrix to obtain the preprocessed spectral image. The step of using alternating least squares to perform low-rank approximation on the time-spectral image matrix to obtain the preprocessed spectral image includes: Construct the solution matrix; In the i-th iteration of the alternating least squares method Based on the i-th solution matrix and the time-spectral image matrix, determine the i-th residual matrix; Based on the i-th residual matrix, determine the i-th search direction and the i-th scaling factor; Multiply the i-th search direction by the i-th scaling factor to obtain the i-th adjusted search direction; The i-th solution matrix is ​​linearly combined with the i-th adjusted search direction to obtain the (i+1)-th solution matrix; If the norm of the (i+1)th solution matrix and the i-th solution matrix is ​​less than or equal to a preset threshold, the (i+1)th iteration operation is performed. When the norm of the (i+1)th solution matrix and the ith solution matrix is ​​greater than the preset threshold, the (i+1)th solution matrix is ​​determined to be the preprocessed spectral image; Where i is a natural number.

2. The method for real-time bacterial detection according to claim 1, characterized in that, The acquisition of multiple Raman spectra of the sample within a preset time period at a preset cycle includes: Using the objective lens of a laser confocal Raman system, Raman scattering generated by laser excitation on the sample is continuously acquired within the preset time period; Using a spectrometer connected to the objective lens, the Raman scattering is integrated over a preset period to obtain the multiple Raman spectra.

3. The method for real-time bacterial detection according to claim 1, characterized in that, The step of classifying the preprocessed spectral image using a trained two-dimensional convolutional neural network to obtain the bacterial detection results of the sample includes: The feature extraction module of the trained two-dimensional convolutional neural network is used to extract features from the preprocessed spectral image to obtain the feature map of the preprocessed spectral image; The feature map is classified using the feature classification module of the trained two-dimensional convolutional neural network to obtain the bacterial detection results of the sample.

4. The method for real-time bacterial detection according to claim 1, characterized in that, The trained two-dimensional convolutional neural network is obtained through the following operations: A two-dimensional convolutional neural network is trained using multiple Raman spectra continuously acquired from samples containing predetermined bacteria, resulting in the trained two-dimensional convolutional neural network.

5. The method for real-time bacterial detection according to claim 1, characterized in that, The preset period is 0.1 seconds.

6. A real-time bacterial detection system, characterized in that, include: The acquisition device is suitable for acquiring multiple Raman spectra of a sample at a preset period within a preset time to obtain a time-spectral image matrix, wherein the time-spectral image matrix includes the multiple Raman spectra arranged in the order of acquisition time; wherein the time-spectral image matrix contains spectral information and temporal dynamic information; The processing device is suitable for performing low-rank reconstruction on the time-spectral image matrix to obtain a preprocessed spectral image; and for classifying the preprocessed spectral image using a trained two-dimensional convolutional neural network to obtain the bacterial detection result of the sample, wherein the two-dimensional convolutional neural network is trained to extract spectral features and temporal dynamic features. The step of performing low-rank reconstruction on the time-spectral image matrix to obtain a preprocessed spectral image includes: The alternating least squares method is used to perform a low-rank approximation on the time-spectral image matrix to obtain the preprocessed spectral image. The step of using alternating least squares to perform low-rank approximation on the time-spectral image matrix to obtain the preprocessed spectral image includes: Construct the solution matrix; In the i-th iteration of the alternating least squares method Based on the i-th solution matrix and the time-spectral image matrix, determine the i-th residual matrix; Based on the i-th residual matrix, determine the i-th search direction and the i-th scaling factor; Multiply the i-th search direction by the i-th scaling factor to obtain the i-th adjusted search direction; The i-th solution matrix is ​​linearly combined with the i-th adjusted search direction to obtain the (i+1)-th solution matrix; If the norm of the (i+1)th solution matrix and the i-th solution matrix is ​​less than or equal to a preset threshold, the (i+1)th iteration operation is performed. When the norm of the (i+1)th solution matrix and the ith solution matrix is ​​greater than the preset threshold, the (i+1)th solution matrix is ​​determined to be the preprocessed spectral image; Where i is a natural number.

7. The real-time bacterial detection system according to claim 6, characterized in that, The data acquisition device includes: The objective lens of the laser confocal Raman system is suitable for continuously acquiring Raman scattering generated by the laser excitation of the sample through the laser confocal Raman system within the preset time period; The spectrometer connected to the objective lens is adapted to integrate the Raman scattering over the duration of the preset period to obtain the multiple Raman spectra.

8. The real-time bacterial detection system according to claim 6, characterized in that, The processing device is also suitable for using the alternating least squares method to perform low-rank approximation on the time-spectral image matrix to obtain the preprocessed spectral image.

Images