A rapid detection method for microplastics based on image recognition

Abstract

This invention relates to the field of microplastic detection technology and discloses a rapid microplastic detection method based on image recognition. The method involves acquiring three-dimensional time-domain fluorescence lifetime data under multi-wavelength excitation using a tuned pulsed laser source and time-correlated single photon counting (TCSPC) technology. The fluorescence decay signal is then processed using a dynamic harmonic Fourier transform, and the harmonic order is adjusted based on the signal-to-noise ratio to optimize the frequency domain characteristics of the data. The fluorescence decay signal is then mapped into quantum state space, and the intrinsic parameters of the fluorescence lifetime are extracted using a variational quantum algorithm. The tensor factor combination is further optimized by combining a high-order tensor decomposition model and a quantum optimization algorithm. Finally, the fluorescence lifetime intrinsic parameters are reconstructed based on the matrix product state and the optimal tensor factor combination, and a eigenvector is extracted for use in microplastic classification and detection. By combining quantum computing with classical algorithms, this method improves the accuracy and efficiency of microplastic classification.

Description

Technical Field

[0001] The present invention relates to the technical field of microplastic detection, and in particular to a rapid detection method for microplastics based on image recognition. Background Art

[0002] Microplastic pollution has become a global environmental problem, with a wide range of sources, including industrial production, degradation of plastic waste, and textile wear. Microplastics are not only widely present in the ocean, soil, and air, but may also enter the human body through the food chain, posing a potential threat to the ecosystem and human health. Therefore, accurate and efficient detection of microplastic types is of great significance for environmental monitoring and pollution control. At present, many environmental protection organizations and scientific research institutions around the world have carried out microplastic monitoring research and promoted the establishment of standardized detection methods to deal with microplastic pollution in different environmental media. However, due to the wide variety of microplastics, wide particle size distribution, and different degrees of environmental aging, their detection still faces great technical challenges.

[0003] Currently, common microplastic detection methods mainly include spectral analysis and chromatography techniques. Among them, spectral analysis methods such as Raman spectroscopy and Fourier transform infrared spectroscopy (FTIR) are widely used to identify microplastic components and can provide information on the chemical composition of microplastics. However, spectral technology has low sensitivity when detecting tiny particles and is significantly affected by fluorescence interference, making it difficult to distinguish mixed microplastics in complex environmental samples. In addition, chromatographic techniques (such as pyrolysis gas chromatography-mass spectrometry, Py-GC-MS) heat and crack microplastics and analyze their decomposition products to infer their polymer type. However, this method usually requires complex sample pretreatment and a long analysis time, making it difficult to meet the needs of high-throughput detection.

[0004] Therefore, there is an urgent need to invent a rapid and accurate detection technology for microplastics to solve the limitations of existing microplastic detection methods in terms of sensitivity, anti-interference ability, sample pre-treatment complexity and detection efficiency, which are difficult to meet the needs of rapid, accurate and high-throughput detection. Summary of the Invention

[0005] In view of this, the present invention proposes a rapid detection method for microplastics based on image recognition, aiming to solve the problem that existing microplastic detection methods have limitations in sensitivity, anti-interference ability, sample pre-treatment complexity and detection efficiency, and are difficult to meet detection needs.

[0006] The present invention proposes a rapid detection method for microplastics based on image recognition, comprising:

[0007] The sample to be detected is excited by a tuned pulse laser source, and spatially resolved fluorescence decay signals are collected by time-correlated single photon counting to construct three-dimensional time-domain fluorescence lifetime data of multi-wavelength excitation.

[0008] Perform dynamic harmonic Fourier transform on the three-dimensional time-domain fluorescence lifetime data to obtain the time-domain fluorescence decay curve, and adjust the harmonic order according to the signal-to-noise ratio;

[0009] The time-domain fluorescence decay curve is mapped to the quantum state space, and the intrinsic parameters of the fluorescence lifetime are extracted based on the variational quantum algorithm;

[0010] Establish a high-order tensor decomposition model under multiple excitation wavelengths and use quantum optimization algorithms to determine the optimal tensor factor combination;

[0011] Based on the combination of matrix product state and optimal tensor factor, the fluorescence lifetime intrinsic parameters are reconstructed, and the fluorescence lifetime eigenvector is extracted. The type of microplastics in the sample to be detected is determined according to the fluorescence lifetime eigenvector.

[0012] Furthermore, when collecting spatially resolved fluorescence decay signals based on time-correlated single photon counting and constructing three-dimensional time-domain fluorescence lifetime data of multi-wavelength excitation, the following steps are included:

[0013] Record the arrival time delay of a single photon relative to the excitation pulse and construct a time-resolved fluorescence decay signal based on the arrival time delay;

[0014] Correcting the fluorescence decay signal based on a fluorescence lifetime physical model, and adjusting the signal-to-noise ratio when correcting the fluorescence decay signal using the fluorescence lifetime physical model based on a signal accumulation and noise suppression algorithm;

[0015] Correlate the time-resolved fluorescence decay signal with the spatial coordinates of the sample to establish two-dimensional spatial distribution information;

[0016] Based on the spectral dimension information obtained by multi-wavelength excitation, a three-dimensional correlation data set is formed with the spatiotemporal signal, and the three-dimensional correlation data set is determined as three-dimensional time-domain fluorescence lifetime data.

[0017] Furthermore, when correcting the fluorescence decay signal based on the fluorescence lifetime physical model, it includes:

[0018] A time response function when obtaining fluorescence based on a preset fluorescence sample measurement;

[0019] The fluorescence decay signal is corrected based on the iterative deconvolution algorithm and the time response function;

[0020] A multi-exponential fit is performed based on the corrected fluorescence decay signal, and whether the fluorescence decay signal is successfully corrected is determined based on the relationship between the fitting result and the configured preset fitting result, where:

[0021] If the fitting result is consistent with the preset fitting result, it is determined whether the fluorescence decay signal is calibrated successfully;

[0022] If the fitting result is inconsistent with the preset fitting result, the iterative deconvolution algorithm is adjusted until the fitting result is consistent with the preset fitting result.

[0023] Furthermore, when the signal-to-noise ratio of the fluorescence decay signal is corrected by adjusting the fluorescence lifetime physical model based on the signal accumulation and noise suppression algorithm, the method includes:

[0024] The fluorescence photon events of multiple excitation cycles are accumulated for a preset number of times, the decay curve is optimized based on the maximum likelihood estimation method, and abnormal photon events are identified and eliminated based on time correlation;

[0025] Obtain the decay curves of each excitation after eliminating abnormal photon events, and superimpose the decay curves of each excitation based on the time channel alignment method;

[0026] Obtain the real-time signal-to-noise ratio of the cumulative attenuation signal, and determine the fitting strategy based on the relationship between the real-time signal-to-noise ratio and the configured preset signal-to-noise ratio, where:

[0027] When the real-time signal-to-noise ratio is higher than the preset signal-to-noise ratio, the real-time signal-to-noise ratio of the cumulative attenuation signal is determined to be a high signal-to-noise ratio, and the fitting strategy is determined to be free fitting based on a double exponential model;

[0028] When the real-time signal-to-noise ratio is consistent with the preset signal-to-noise ratio, the real-time signal-to-noise ratio of the cumulative attenuation signal is determined to be a medium signal-to-noise ratio, and the fitting strategy is determined to be fixed long-life component parameters and adjust the short-life component and amplitude;

[0029] When the real-time signal-to-noise ratio is lower than the preset signal-to-noise ratio, the real-time signal-to-noise ratio of the cumulative attenuation signal is determined to be a low signal-to-noise ratio, and the fitting strategy is determined to be smoothing the spatial neighborhood data, and the fitting strategy is re-determined based on the relationship between the processed real-time signal-to-noise ratio and the preset signal-to-noise ratio.

[0030] Furthermore, performing a dynamic harmonic Fourier transform on the three-dimensional time-domain fluorescence lifetime data to obtain a time-domain fluorescence decay curve and adjusting the harmonic order according to the signal-to-noise ratio includes:

[0031] The signal-to-noise ratio of the fluorescence decay signal is calculated based on power spectrum analysis, and the spectrum information content is evaluated based on spectral entropy to determine the validity of the fluorescence decay signal, where:

[0032] Determining whether the fluorescence decay signal is valid based on a relationship between the spectral entropy and a preset spectral entropy, wherein if the spectral entropy is less than the preset spectral entropy, determining that the fluorescence decay signal is valid;

[0033] Determine the harmonic order of fast Fourier transform according to the signal-to-noise ratio of the effective fluorescence decay signal;

[0034] The effective fluorescence decay signal is converted into time-frequency domain based on fast Fourier transform to obtain the frequency characteristics of fluorescence lifetime.

[0035] Furthermore, according to the signal-to-noise ratio of the effective fluorescence decay signal, the harmonic order of the fast Fourier transform is determined, including:

[0036] A preset signal-to-noise ratio is pre-configured, and the harmonic order of the fast Fourier transform is determined based on the relationship between the real-time signal-to-noise ratio of the effective fluorescence decay signal and the preset signal-to-noise ratio:

[0037] When the real-time signal-to-noise ratio is lower than or equal to the preset signal-to-noise ratio, determining that the harmonic order of the fast Fourier transform is a low-order harmonic;

[0038] When the real-time signal-to-noise ratio is greater than the preset signal-to-noise ratio, the harmonic order of the fast Fourier transform is determined to be a high-order harmonic;

[0039] Among them, low-order harmonics are (n≥3) and high-order harmonics are (n=1).

[0040] Furthermore, when mapping the time-domain fluorescence decay curve to the quantum state space and extracting the intrinsic parameters of the fluorescence lifetime based on the variational quantum algorithm, the following steps are involved:

[0041] The time-domain fluorescence decay signal is converted into a quantum state by an amplitude encoding method, and the quantum state is prepared into a quantum state corresponding to the fluorescence decay curve based on a quantum circuit;

[0042] Iteratively adjust the parameters of quantum circuits based on classical optimizers and quantum computing;

[0043] Based on quantum superposition state and quantum parallel computing, the target fluorescence lifetime parameters are determined.

[0044] Furthermore, a high-order tensor decomposition model under multiple excitation wavelengths is established, and a quantum optimization algorithm is used to determine the optimal tensor factor combination, including:

[0045] Fluorescence data from multiple excitation wavelengths are constructed as a high-order tensor. Based on tensor decomposition technology, the high-order tensor is decomposed into multiple low-order factor matrices and core tensors to extract relevant features between dimensions.

[0046] Optimize high-order tensor decomposition problems based on quantum optimization algorithms, and determine the optimal factor matrix combination based on the superposition and parallelism of quantum computing;

[0047] The quantum optimization algorithm minimizes the loss function and determines the optimal tensor decomposition combination of the tensor factor matrix.

[0048] Furthermore, the fluorescence lifetime intrinsic parameters are reconstructed based on the matrix product state and the optimal tensor factor combination, and the fluorescence lifetime eigenvector is extracted. When the type of microplastics in the sample to be detected is determined according to the fluorescence lifetime eigenvector, it includes:

[0049] Fit the fluorescence decay curve to a biexponential model to determine the intrinsic parameters of the fluorescence lifetime, where the intrinsic parameters include decay time constants τ1, τ2 and their corresponding amplitudes A1, A2;

[0050] The intrinsic parameters of the fluorescence lifetime are fitted based on the least squares method, and the decay parameters are extracted, and the extracted decay parameters are combined into a fluorescence lifetime characteristic vector;

[0051] The extracted fluorescence lifetime feature vector is subjected to mean-variance processing, and the processed fluorescence lifetime feature vector is substituted into the pre-established microplastic classification dataset to obtain the similarity between the fluorescence lifetime feature vector and the feature vector of each microplastic classification in the microplastic classification dataset;

[0052] The similarities between the fluorescence lifetime characteristic vector and the characteristic vector of each microplastic classification are ranked in reverse order, and the microplastic classification corresponding to the first-ranked similarity is determined as the microplastic type of the sample to be tested.

[0053] Furthermore, when pre-establishing a microplastic classification dataset, it includes:

[0054] Obtain the fluorescence lifetime decay parameters of each microplastic, and establish a microplastic data correlation formula based on the fluorescence lifetime decay parameters;

[0055] The distance metric between each microplastic data association formula is obtained based on the Euclidean distance, and a distance matrix is ​​established based on the distance metric. The microplastic data association formulas are iteratively clustered based on the distance matrix.

[0056] According to the clustering results, the characteristic vectors of each microplastic type were determined, and a microplastic classification dataset was established based on the characteristic vectors of each microplastic type.

[0057] Compared with existing technologies, the present invention offers the following advantages: By combining tuned pulsed laser excitation and time-correlated single photon counting (TSPC), spatially resolved fluorescence decay signal acquisition is achieved, and three-dimensional time-domain fluorescence lifetime data from multi-wavelength excitation is constructed, improving the spectral information richness and data integrity of microplastic detection. Secondly, through dynamic harmonic Fourier transform (DHFT), the harmonic order can be adaptively adjusted, enhancing signal processing robustness and improving the resolution and noise immunity of the fluorescence decay curve. Furthermore, quantum computing technology is utilized to efficiently extract fluorescence lifetime eigenvalues ​​using a variational quantum algorithm. High-order tensor decomposition and quantum optimization algorithms are combined to optimize the feature extraction process, effectively reducing computational complexity and improving the accuracy of data dimensionality reduction and feature extraction. Finally, fluorescence lifetime features are reconstructed using the matrix product state (MPS) method and combined with fluorescence lifetime eigenvectors for microplastic classification, ensuring classification accuracy and generalizability. Overall, this method enables rapid and accurate detection and identification of microplastic types in complex environmental samples, improving detection efficiency while mitigating the limitations of traditional methods in data processing, noise interference, and computational complexity. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiment below. The accompanying drawings are for illustration purposes only and are not to be considered as limiting the present invention. The same reference symbols are used throughout the drawings to represent the same components. In the drawings:

[0059] Figure 1 A flowchart of a rapid detection method for microplastics based on image recognition provided by an embodiment of the present invention;

[0060] Figure 2 Optical images of microplastics provided by embodiments of the present invention;

[0061] Figure 3 This is a scan of the excitation and emission spectra of microplastics provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0062] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art. It should be noted that, unless there is a conflict, the embodiments of the present disclosure and the features in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.

[0063] like Figure 1-Figure 3As shown, in some embodiments of the present application, this embodiment provides a rapid detection method for microplastics based on image recognition, comprising:

[0064] Step S100 : Excite the sample to be detected based on a tuned pulse laser source, collect spatially resolved fluorescence decay signals based on time-correlated single photon counting, and construct three-dimensional time-domain fluorescence lifetime data of multi-wavelength excitation.

[0065] Specifically, when collecting spatially resolved fluorescence decay signals based on the time-correlated single-photon counting method and constructing three-dimensional time-domain fluorescence lifetime data for multi-wavelength excitation, it includes: recording the arrival time delay of a single photon relative to the excitation pulse, and constructing a time-resolved fluorescence decay signal based on the arrival time delay; correcting the fluorescence decay signal based on the fluorescence lifetime physical model, and adjusting the signal-to-noise ratio when correcting the fluorescence decay signal based on the fluorescence lifetime physical model based on the signal accumulation and noise suppression algorithm; correlating the time-resolved fluorescence decay signal with the spatial coordinates of the sample to establish two-dimensional spatial distribution information; forming a three-dimensional correlation data set with the spatiotemporal signal based on the spectral dimension information obtained by multi-wavelength excitation, and determining the three-dimensional correlation data set as three-dimensional time-domain fluorescence lifetime data.

[0066] Specifically, when correcting the fluorescence decay signal based on the fluorescence lifetime physical model, it includes: obtaining the time response function when the fluorescence is collected based on the preset fluorescence sample measurement; correcting the fluorescence decay signal based on the iterative deconvolution algorithm and the time response function; performing multi-exponential fitting based on the corrected fluorescence decay signal, and determining whether the fluorescence decay signal is successfully corrected based on the relationship between the fitting result and the configured preset fitting result, wherein: if the fitting result is consistent with the preset fitting result, it is determined whether the fluorescence decay signal is successfully corrected; if the fitting result is inconsistent with the preset fitting result, the iterative deconvolution algorithm is adjusted until the fitting result is consistent with the preset fitting result.

[0067] Specifically, when the signal-to-noise ratio of the fluorescence decay signal is corrected by adjusting the fluorescence lifetime physical model based on the signal accumulation and noise suppression algorithm, the method includes: accumulating a preset number of fluorescence photon events of multiple excitation cycles, optimizing the decay curve based on the maximum likelihood estimation method, and identifying and eliminating abnormal photon events based on time correlation; obtaining the decay curve of each excitation after eliminating the abnormal photon event, and superimposing the decay curve of each excitation based on the time channel alignment method; obtaining the real-time signal-to-noise ratio of the accumulated decay signal, and determining the fitting strategy based on the relationship between the real-time signal-to-noise ratio and the configured preset signal-to-noise ratio, wherein: when the real-time signal-to-noise ratio When it is higher than the preset signal-noise ratio, the real-time signal-noise ratio of the cumulative attenuation signal is determined to be a high signal-noise ratio, and the fitting strategy is determined to be free fitting based on the double exponential model; when the real-time signal-noise ratio is consistent with the preset signal-noise ratio, the real-time signal-noise ratio of the cumulative attenuation signal is determined to be a medium signal-noise ratio, and the fitting strategy is determined to fix the long-life component parameters and adjust the short-life component and amplitude; when the real-time signal-noise ratio is lower than the preset signal-noise ratio, the real-time signal-noise ratio of the cumulative attenuation signal is determined to be a low signal-noise ratio, and the fitting strategy is determined to be smoothing of the spatial neighborhood data, and the fitting strategy is re-determined based on the relationship between the processed real-time signal-noise ratio and the preset signal-noise ratio.

[0068] As can be understood, in fluorescence lifetime measurement, TCSPC technology constructs a time-resolved fluorescence decay curve by recording the arrival time delay of a single photon relative to the excitation pulse. Because fluorescence decay is a probabilistic and statistical process, TCSPC can accumulate the statistical distribution of photon arrival times across multiple excitation-fluorescence events, generating a high-resolution time-decay curve. By correlating the time-resolved fluorescence decay signal with the spatial coordinates of the sample, spatial distribution information of the fluorescence lifetime can be obtained. Combined with spectral data from multi-wavelength excitation, a three-dimensional (spatial, temporal, and wavelength) fluorescence lifetime dataset is generated, providing complete fluorescence dynamics information for subsequent signal processing and classification analysis. Secondly, because the measurement system (such as the detector and optical components) introduces temporal response errors, the directly acquired fluorescence decay signal often includes the influence of the system response function. Therefore, a fluorescence lifetime physics model is employed to perform signal correction to improve the accuracy of fluorescence decay parameters. Specifically, the temporal response function of a pre-determined fluorescence sample is first measured, and an iterative deconvolution algorithm is used to remove the influence of the system response, ensuring that the fluorescence signal more closely matches the sample's true decay behavior. Subsequently, the corrected decay signal is analyzed using a multi-exponential fitting method and compared with a pre-determined standard fitting result to ensure the effectiveness of the correction. If the fitting result has a large deviation, the deconvolution algorithm parameters are adjusted until the correction reaches the expected accuracy. In addition, the fluorescence measurement process is affected by the statistical noise of photon detection, and the signal noise of a single measurement is high, so it is necessary to improve the signal-to-noise ratio through signal accumulation technology. Specifically, the fluorescence photon events of multiple excitation cycles are accumulated, and the maximum likelihood estimation method is used to optimize the fluorescence decay curve. At the same time, the time correlation is used to identify and eliminate abnormal photon events to reduce the interference of data noise on the fitting results. Subsequently, the fluorescence decay curves of each excitation are superimposed by time channel alignment to further improve the signal quality and make the fluorescence decay curve more statistically stable. At the same time, since the signal-to-noise ratio under different measurement conditions may vary, it is necessary to dynamically adjust the fitting strategy according to the real-time signal-to-noise ratio to ensure the extraction accuracy of the fluorescence lifetime parameters. When the real-time signal-to-noise ratio is high, the fitting strategy uses a bi-exponential model for free fitting to obtain the most detailed fluorescence lifetime information. When the signal-to-noise ratio is at a moderate level, a strategy of fixing the long-lifetime component parameters and adjusting the short-lifetime component and amplitude is adopted to reduce free parameters while maintaining good fitting accuracy. When the signal-to-noise ratio is low, the spatial neighborhood data is first smoothed, and then the fitting strategy is dynamically adjusted based on the processed signal-to-noise ratio to improve data stability and reduce noise interference. Finally, through the above signal acquisition, correction, noise suppression, and fitting optimization process, the final fluorescence lifetime data obtained includes the time dimension (fluorescence decay characteristics), the spatial dimension (the distribution of fluorescence lifetime on the sample surface), and the wavelength dimension (spectral information under multi-wavelength excitation), constructing a complete three-dimensional time-domain fluorescence lifetime dataset.

[0069] As can be seen, by exciting the sample with a tuned pulsed laser source and combining it with time-correlated single photon counting (TCSPC), the arrival time of individual photons relative to the excitation pulse can be precisely recorded, thereby obtaining fluorescence decay curves with high temporal resolution. Furthermore, spatially resolved data acquisition methods allow the time-resolved fluorescence signal to be correlated with the spatial coordinates of the sample, thereby constructing two-dimensional spatial distribution information and accurately mapping the fluorescence lifetime parameters in the spatial dimension, providing more fine-grained spatial information for microplastic detection. Secondly, by presetting the time response function of the fluorescence sample measurement system and correcting it using an iterative deconvolution algorithm, system response distortion can be effectively compensated, ensuring that the fluorescence lifetime measurement results more closely resemble the sample's true decay behavior. Furthermore, a multi-exponential fitting method is used to further optimize the fluorescence lifetime calculation. Comparing the fitted results with standard fitting ensures the accuracy of the signal correction, thereby improving data reliability. Furthermore, by accumulating fluorescence photon events over multiple excitation cycles and optimizing the decay curve using maximum likelihood estimation, the statistical stability of the signal is enhanced. Furthermore, using temporal correlation to identify and eliminate anomalous photon events helps reduce errors caused by scattering or detector dark noise, thereby improving the quality of fluorescence lifetime data. Furthermore, an adaptive fitting strategy based on real-time signal-to-noise ratio (SNR) optimizes fluorescence lifetime parameter extraction under varying measurement conditions. When the SNR is high, the system uses a bi-exponential model for free fitting to obtain the most complete fluorescence dynamics. When the SNR is moderate, the long-lifetime component parameters are fixed and the short-lifetime component and amplitude are adjusted to reduce free parameters and improve fitting stability. Under low SNR conditions, spatial neighborhood data smoothing is performed before dynamically adjusting the fitting strategy to minimize noise interference. This dynamic fitting strategy ensures high accuracy and robustness in fluorescence lifetime extraction. Finally, spectral dimension information is acquired through multi-wavelength excitation and combined with the spatiotemporal signal to form a three-dimensional correlation dataset, achieving a multidimensional representation of fluorescence lifetime data. This three-dimensional time-domain fluorescence lifetime dataset not only captures the temporal characteristics of fluorescence decay but also incorporates spatial distribution and spectral information, providing richer characteristic information for microplastic classification.

[0070] Step S200 : performing a dynamic harmonic Fourier transform on the three-dimensional time-domain fluorescence lifetime data to obtain a time-domain fluorescence decay curve, and adjusting the harmonic order according to the signal-to-noise ratio.

[0071] Specifically, a dynamic harmonic Fourier transform is performed on three-dimensional time-domain fluorescence lifetime data to obtain a time-domain fluorescence decay curve, and the harmonic order is adjusted according to the signal-to-noise ratio, including: calculating the signal-to-noise ratio of the fluorescence decay signal based on power spectrum analysis, and evaluating the amount of spectrum information based on spectral entropy to determine the validity of the fluorescence decay signal, wherein: determining whether the fluorescence decay signal is valid based on the relationship between the spectral entropy and the configured preset spectral entropy, wherein if the spectral entropy is less than the preset spectral entropy, the fluorescence decay signal is determined to be valid; determining the harmonic order of the fast Fourier transform based on the signal-to-noise ratio of the valid fluorescence decay signal; and performing time-frequency domain conversion on the valid fluorescence decay signal based on the fast Fourier transform to obtain the frequency characteristics of the fluorescence lifetime.

[0072] Specifically, when determining the harmonic order of the fast Fourier transform based on the signal-to-noise ratio of the effective fluorescence attenuation signal, it includes: pre-configuring a preset signal-to-noise ratio, and determining the harmonic order of the fast Fourier transform based on the relationship between the real-time signal-to-noise ratio of the effective fluorescence attenuation signal and the preset signal-to-noise ratio: when the real-time signal-to-noise ratio is lower than or equal to the preset signal-to-noise ratio, the harmonic order of the fast Fourier transform is determined to be a low-order harmonic; when the real-time signal-to-noise ratio is greater than the preset signal-to-noise ratio, the harmonic order of the fast Fourier transform is determined to be a high-order harmonic; wherein, the low-order harmonic is (n≥3) and the high-order harmonic is (n=1).

[0073] As can be understood, power spectrum analysis calculates the signal-to-noise ratio (SNR) of the fluorescence decay signal, and spectral entropy is used to assess the signal's spectral information content. Spectral entropy, as a measure of a signal's spectral complexity, effectively reflects signal validity. When the spectral entropy is below a preset threshold, it indicates a high level of spectral information and is therefore considered valid. This method provides a reliable method for screening valid signals, ensuring data quality and accuracy in subsequent processing. Secondly, based on the SNR of the valid signal, the harmonic order of the Fourier transform can be dynamically adjusted. This dynamic adjustment strategy determines the harmonic order by comparing the preset SNR with the real-time SNR. For signals with low SNRs, low-order harmonics (n ≥ 3) are used to avoid errors introduced by higher-order harmonics. For signals with high SNRs, higher-order harmonics (n = 1) are selected to extract finer frequency features. This adaptive mechanism optimizes the Fourier transform process based on signal quality, enhancing the accuracy and validity of the analysis results. Furthermore, the valid signal is converted from the time domain to the frequency domain using a fast Fourier transform (FFT), thereby extracting the frequency characteristics of the fluorescence lifetime. The Fourier transform can decompose complex time-domain signals into different frequency components, which is crucial for analyzing the different physical characteristics of the fluorescence decay process. By extracting frequency features, subtle changes in fluorescence lifetime can be revealed, providing a more accurate basis for the characteristic identification of microplastics. Finally, by combining the signal-to-noise ratio and spectral entropy to dynamically adjust the harmonic order of the Fourier transform, not only the noise resistance in the data processing process is improved, but also the accuracy of signal analysis is enhanced. When the signal-to-noise ratio is low, the use of low-order harmonics can reduce the impact of noise; while signals with a high signal-to-noise ratio extract more detailed features through high-order harmonics, thereby effectively improving the time-frequency analysis capability of the fluorescence decay signal and ensuring that reliable analysis results can be obtained under different environments.

[0074] As can be seen, the effectiveness of fluorescence decay signals can be assessed by combining power spectrum analysis and spectral entropy. Spectral entropy, as a measure of signal complexity, can effectively distinguish valid signals from noise or invalid signals by comparing them against a preset spectral entropy threshold. This ensures that only valid signals with high spectral information are used in subsequent processing, avoiding errors caused by noise and improving signal reliability and processing accuracy. Secondly, the harmonic order of the fast Fourier transform (FFT) is dynamically adjusted based on the signal-to-noise ratio (SNR), thereby optimizing the time-frequency analysis of the fluorescence decay signal. When the SNR is low, using low-order harmonics effectively reduces errors introduced by high-order harmonics, ensuring signal analysis stability. For high SNR signals, using high-order harmonics allows for more detail extraction and enhanced frequency resolution. This dynamic adjustment mechanism enables signal processing to adapt to changes in signal quality, thereby improving analysis accuracy. Furthermore, by performing a fast Fourier transform (FFT) on the valid signal, the fluorescence decay signal is converted from the time domain to the frequency domain, extracting the frequency characteristics of the fluorescence lifetime. This process can better reveal the subtle features in the fluorescence decay process, thereby providing more accurate data support for the subsequent identification of microplastic types. Time-frequency domain conversion can provide more in-depth frequency information for efficient signal processing and analysis. Finally, through real-time monitoring and adjustment of the signal-to-noise ratio, not only the noise resistance of the signal is improved, but also the fitting strategy is optimized. In the case of low signal-to-noise ratio, low-order harmonics are used to reduce noise interference, while in the case of high signal-to-noise ratio, high-order harmonics are used to extract more detailed features, so that it can adapt to different signal environments and ensure stability and accuracy in complex samples. This strategy effectively improves the robustness of fluorescence lifetime analysis, ensuring a wider range of applications and higher reliability.

[0075] Step S300: Map the time-domain fluorescence decay curve to the quantum state space, and extract the fluorescence lifetime intrinsic parameters based on the variational quantum algorithm.

[0076] Specifically, when mapping the time-domain fluorescence decay curve to the quantum state space and extracting the fluorescence lifetime intrinsic parameters based on the variational quantum algorithm, it includes: converting the time-domain fluorescence decay signal into a quantum state through the amplitude encoding method, and preparing the quantum state as a quantum state corresponding to the fluorescence decay curve based on the quantum circuit; iteratively adjusting the parameters of the quantum circuit based on the classical optimizer and quantum computing; and determining the target fluorescence lifetime parameters based on the quantum superposition state and quantum parallel computing.

[0077] As can be understood, the time-domain fluorescence decay signal is converted into a quantum state using amplitude encoding. This process uses quantum bits to represent the different amplitudes of the signal, and through the operation of a quantum circuit, the classical signal is mapped into quantum information. Amplitude encoding is a common quantum information encoding method that effectively converts continuous time-domain signals into discrete quantum states, thereby bridging the gap between traditional signals and quantum computing. Secondly, after the time-domain fluorescence decay signal is mapped into a quantum state, a variational quantum algorithm (VQA) is used to optimize the parameters of the quantum circuit based on a combination of a classical optimizer and quantum computing. By iteratively adjusting the parameters of the quantum circuit, the optimized quantum state is optimized to better correspond to the fluorescence decay curve. The classical optimizer guides the optimization process, while quantum computing provides the efficient ability to handle complex problems, enabling this iterative process to converge quickly and provide accurate fluorescence lifetime intrinsic parameters. Finally, during the execution of the variational quantum algorithm, the properties of quantum parallel computing and quantum superposition are exploited to accelerate the solution process. Quantum superposition allows multiple computational paths to be processed simultaneously, allowing multiple possible solutions to be explored in a short period of time. Through quantum parallelism, multiple possible solutions for fluorescence lifetime parameters can be processed simultaneously, greatly improving the computational efficiency and accuracy, and ultimately determining the most appropriate target fluorescence lifetime parameter.

[0078] Step S400: Establish a high-order tensor decomposition model under multiple excitation wavelengths, and use a quantum optimization algorithm to determine the optimal tensor factor combination.

[0079] Specifically, a high-order tensor decomposition model under multiple excitation wavelengths is established, and a quantum optimization algorithm is used to determine the optimal tensor factor combination, including: constructing the fluorescence data from multiple excitation wavelengths into a high-order tensor, and decomposing the high-order tensor into multiple low-order factor matrices and core tensors based on tensor decomposition technology, and extracting relevant features between dimensions; optimizing the high-order tensor decomposition problem based on the quantum optimization algorithm, and determining the optimal factor matrix combination based on the superposition and parallelism of quantum computing; minimizing the loss function using the quantum optimization algorithm to determine the optimal tensor decomposition combination of the tensor factor matrix.

[0080] As can be understood, fluorescence data from multiple excitation wavelengths is constructed as a high-order tensor. A high-order tensor is a data structure with multiple dimensions corresponding to different experimental conditions (such as wavelength, time, and space). Using tensor decomposition techniques, this high-order tensor is decomposed into multiple low-order factor matrices and a core tensor. Tensor decomposition extracts correlation features between dimensions and represents the originally complex multidimensional data in a more simplified and manageable form, making subsequent analysis and optimization more efficient. Next, the high-order tensor decomposition problem is optimized using a quantum optimization algorithm. By leveraging the superposition and parallelism of quantum computing, quantum optimization algorithms offer greater efficiency than classical optimization methods when processing large amounts of data and complex optimization problems. Quantum computing enables rapid exploration of combinations of factor matrices in the tensor decomposition, reducing computational time and improving solution accuracy. Quantum algorithms can simultaneously evaluate multiple potential solution paths, effectively finding the optimal tensor factor combination within the vast solution space. Finally, within the framework of the quantum optimization algorithm, a loss function is minimized to determine the optimal tensor factor matrix combination. The loss function measures the difference between the current factor matrix combination and the actual data. The optimization process minimizes this difference by adjusting the values ​​of the factor matrix to find the best tensor decomposition result. The parallelism and efficiency of the quantum optimization algorithm enable this process to quickly converge to the global optimal solution, ensuring that the extracted factor matrix accurately reflects the inherent structure and characteristics of the data.

[0081] Step S500: reconstruct the fluorescence lifetime intrinsic parameters based on the matrix product state and the optimal tensor factor combination, extract the fluorescence lifetime eigenvector, and determine the type of microplastics in the sample to be detected according to the fluorescence lifetime eigenvector.

[0082] Specifically, the fluorescence lifetime intrinsic parameters are reconstructed based on the combination of matrix product states and optimal tensor factors, and the fluorescence lifetime eigenvector is extracted. When the type of microplastics in the sample to be detected is determined according to the fluorescence lifetime eigenvector, it includes: fitting the fluorescence decay curve to a double exponential model to determine the intrinsic parameters of the fluorescence lifetime, where the intrinsic parameters include decay time constants τ1, τ2 and their corresponding amplitudes A1, A2; fitting the intrinsic parameters of the fluorescence lifetime based on the least squares method, extracting the decay parameters, and combining the extracted decay parameters into a fluorescence lifetime eigenvector; performing mean-variance processing on the extracted fluorescence lifetime eigenvector, and substituting the processed fluorescence lifetime eigenvector into a pre-established microplastic classification data set to obtain the similarity between the fluorescence lifetime eigenvector and the eigenvector of each microplastic classification in the microplastic classification data set; the similarity between the fluorescence lifetime eigenvector and the eigenvector of each microplastic classification is ranked in reverse order, and the microplastic classification corresponding to the first-ranked similarity is determined as the microplastic type of the sample to be detected.

[0083] Specifically, when pre-establishing a microplastic classification dataset, it includes: obtaining the fluorescence lifetime decay parameters of each microplastic, and establishing a microplastic data association formula based on the fluorescence lifetime decay parameters; obtaining the distance measurement between each microplastic data association formula based on the Euclidean distance, and establishing a distance matrix based on the distance measurement, and iteratively clustering the microplastic data association formulas based on the distance matrix; based on the clustering results, determining the characteristic vector of each microplastic type, and establishing a microplastic classification dataset based on the characteristic vector of each microplastic type.

[0084] As can be understood, by fitting the fluorescence decay curve to a biexponential model, the intrinsic parameters of the fluorescence lifetime are determined, including the decay time constants (τ1, τ2) and their corresponding amplitudes (A1, A2). These parameters reflect the kinetic characteristics of the sample's fluorescence decay process. Fitting these intrinsic parameters using the least squares method accurately extracts the decay parameters and generates fluorescence lifetime feature vectors. These feature vectors contain key information about the fluorescence decay process and provide fundamental data for microplastic classification. Secondly, the extracted fluorescence lifetime feature vectors are normalized using mean-variance processing to reduce noise and improve stability. The processed feature vectors are then compared with a pre-established microplastic classification dataset. By calculating the similarity between the fluorescence lifetime feature vectors and the feature vectors of each microplastic classification, the microplastic type of the sample can be effectively determined. The microplastic type with the highest similarity is the sample's classification result, which improves classification accuracy and efficiency. Ultimately, to accurately classify microplastics, a microplastic classification dataset must be established. This process involves obtaining fluorescence lifetime decay parameters for different microplastic types and establishing a correlation equation for the microplastic data. Next, the distances between the microplastic data associations were calculated using the Euclidean distance metric, and cluster analysis was performed using the distance matrix. Finally, based on the clustering results, the characteristic vectors for each microplastic type were determined, and a classification dataset was constructed. This classification dataset provided the basis for subsequent sample classification, ensuring the effectiveness and reliability of the classification algorithm.

[0085] As can be seen, by fitting the fluorescence decay curve to a biexponential model and extracting the decay time constant and amplitude, the intrinsic parameters of the fluorescence lifetime were accurately determined. These intrinsic parameters were converted into fluorescence lifetime feature vectors after mean-variance processing, which serve as key data for microplastic classification. This extraction and processing method effectively eliminates the influence of noise, enhances the classification system's ability to identify microplastic types, and ensures high accuracy and reliability of the classification results. Secondly, by pre-constructing a microplastic classification dataset and performing cluster analysis based on the Euclidean distance metric, feature vectors of different microplastic types can be extracted and organized. This process enables the microplastic classification model to quickly match the samples to be tested by comparing similarity with feature vectors of known microplastic types, significantly improving classification efficiency. In addition, iterative clustering helps to refine the distinctions between different microplastic types, reducing false positives and missed detections during the classification process. Finally, by calculating and ranking similarity between the fluorescence lifetime feature vectors and the microplastic classification dataset, this method achieves a fully automated microplastic classification process. This not only reduces the need for human intervention, but also enables the rapid identification of microplastic types in a large number of samples, greatly improving detection efficiency and accuracy, making it suitable for large-scale monitoring applications.

[0086] In the above-described embodiment, by combining tuned pulsed laser excitation and time-correlated single photon counting (SPC) technology, spatially resolved fluorescence decay signal acquisition is achieved, and three-dimensional time-domain fluorescence lifetime data with multi-wavelength excitation is constructed, improving the spectral information richness and data integrity of microplastic detection. Secondly, through dynamic harmonic Fourier transform, the harmonic order can be adaptively adjusted, enhancing the robustness of signal processing and improving the resolution and noise immunity of the fluorescence decay curve. Quantum computing technology is further utilized to efficiently extract fluorescence lifetime eigenvalues ​​using a variational quantum algorithm. High-order tensor decomposition and quantum optimization algorithms are combined to optimize the feature extraction process, effectively reducing computational complexity and improving the accuracy of data dimensionality reduction and feature extraction. Finally, fluorescence lifetime features are reconstructed based on the matrix product state (MPS) method and combined with fluorescence lifetime eigenvectors for microplastic classification, ensuring classification accuracy and generalizability. Overall, this method enables rapid and accurate detection and identification of microplastic types in complex environmental samples, improving detection efficiency while mitigating the limitations of traditional methods in data processing, noise interference, and computational complexity.

[0087] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0088] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0089] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0090] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0091] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.

Claims

1. A rapid detection method for microplastics based on image recognition, characterized in that: include: The sample to be detected is excited by a tuned pulse laser source, and spatially resolved fluorescence decay signals are collected by time-correlated single photon counting to construct three-dimensional time-domain fluorescence lifetime data of multi-wavelength excitation. Perform dynamic harmonic Fourier transform on the three-dimensional time-domain fluorescence lifetime data to obtain the time-domain fluorescence decay curve, and adjust the harmonic order according to the signal-to-noise ratio; The time-domain fluorescence decay curve is mapped to the quantum state space, and the intrinsic parameters of the fluorescence lifetime are extracted based on the variational quantum algorithm; Establish a high-order tensor decomposition model under multiple excitation wavelengths and use quantum optimization algorithms to determine the optimal tensor factor combination; Based on the combination of matrix product state and optimal tensor factor, the fluorescence lifetime intrinsic parameters are reconstructed, and the fluorescence lifetime eigenvector is extracted. The type of microplastics in the sample to be detected is determined according to the fluorescence lifetime eigenvector.

2. The rapid detection method for microplastics based on image recognition according to claim 1, characterized in that: When collecting spatially resolved fluorescence decay signals based on time-correlated single photon counting and constructing three-dimensional time-domain fluorescence lifetime data with multi-wavelength excitation, the following steps are included: Record the arrival time delay of a single photon relative to the excitation pulse and construct a time-resolved fluorescence decay signal based on the arrival time delay; Correcting the fluorescence decay signal based on a fluorescence lifetime physical model, and adjusting the signal-to-noise ratio when correcting the fluorescence decay signal using the fluorescence lifetime physical model based on a signal accumulation and noise suppression algorithm; Correlate the time-resolved fluorescence decay signal with the spatial coordinates of the sample to establish two-dimensional spatial distribution information; Based on the spectral dimension information obtained by multi-wavelength excitation, a three-dimensional correlation data set is formed with the spatiotemporal signal, and the three-dimensional correlation data set is determined as three-dimensional time-domain fluorescence lifetime data.

3. The rapid detection method for microplastics based on image recognition according to claim 2, characterized in that: Correction of fluorescence decay signals based on the fluorescence lifetime physical model includes: A time response function when obtaining fluorescence based on a preset fluorescence sample measurement; The fluorescence decay signal is corrected based on the iterative deconvolution algorithm and the time response function; A multi-exponential fit is performed based on the corrected fluorescence decay signal, and whether the fluorescence decay signal is successfully corrected is determined based on the relationship between the fitting result and the configured preset fitting result, where: If the fitting result is consistent with the preset fitting result, it is determined whether the fluorescence decay signal is calibrated successfully; If the fitting result is inconsistent with the preset fitting result, the iterative deconvolution algorithm is adjusted until the fitting result is consistent with the preset fitting result.

4. The rapid detection method for microplastics based on image recognition according to claim 3, characterized in that: When the signal-to-noise ratio of the fluorescence decay signal is corrected by adjusting the fluorescence lifetime physical model based on the signal accumulation and noise suppression algorithm, it includes: The fluorescence photon events of multiple excitation cycles are accumulated for a preset number of times, the decay curve is optimized based on the maximum likelihood estimation method, and abnormal photon events are identified and eliminated based on time correlation; Obtain the decay curves of each excitation after eliminating abnormal photon events, and superimpose the decay curves of each excitation based on the time channel alignment method; Obtain the real-time signal-to-noise ratio of the cumulative attenuation signal, and determine the fitting strategy based on the relationship between the real-time signal-to-noise ratio and the configured preset signal-to-noise ratio, where: When the real-time signal-to-noise ratio is higher than the preset signal-to-noise ratio, the real-time signal-to-noise ratio of the cumulative attenuation signal is determined to be a high signal-to-noise ratio, and the fitting strategy is determined to be free fitting based on a double exponential model; When the real-time signal-to-noise ratio is consistent with the preset signal-to-noise ratio, the real-time signal-to-noise ratio of the cumulative attenuation signal is determined to be a medium signal-to-noise ratio, and the fitting strategy is determined to be fixed long-life component parameters and adjust the short-life component and amplitude; When the real-time signal-to-noise ratio is lower than the preset signal-to-noise ratio, the real-time signal-to-noise ratio of the cumulative attenuation signal is determined to be a low signal-to-noise ratio, and the fitting strategy is determined to be smoothing the spatial neighborhood data, and the fitting strategy is re-determined based on the relationship between the processed real-time signal-to-noise ratio and the preset signal-to-noise ratio.

5. The rapid detection method for microplastics based on image recognition according to claim 1, characterized in that: Performing dynamic harmonic Fourier transform on three-dimensional time-domain fluorescence lifetime data to obtain a time-domain fluorescence decay curve and adjusting the harmonic order according to the signal-to-noise ratio includes: The signal-to-noise ratio of the fluorescence decay signal is calculated based on power spectrum analysis, and the spectrum information content is evaluated based on spectral entropy to determine the validity of the fluorescence decay signal, where: Determining whether the fluorescence decay signal is valid based on a relationship between the spectral entropy and a preset spectral entropy, wherein if the spectral entropy is less than the preset spectral entropy, determining that the fluorescence decay signal is valid; Determine the harmonic order of fast Fourier transform according to the signal-to-noise ratio of the effective fluorescence decay signal; The effective fluorescence decay signal is converted into time-frequency domain based on fast Fourier transform to obtain the frequency characteristics of fluorescence lifetime.

6. The rapid detection method for microplastics based on image recognition according to claim 5, characterized in that: When determining the harmonic order of the fast Fourier transform based on the signal-to-noise ratio of the effective fluorescence decay signal, the following are included: A preset signal-to-noise ratio is pre-configured, and the harmonic order of the fast Fourier transform is determined based on the relationship between the real-time signal-to-noise ratio of the effective fluorescence decay signal and the preset signal-to-noise ratio: When the real-time signal-to-noise ratio is lower than or equal to the preset signal-to-noise ratio, determining that the harmonic order of the fast Fourier transform is a low-order harmonic; When the real-time signal-to-noise ratio is greater than the preset signal-to-noise ratio, the harmonic order of the fast Fourier transform is determined to be a high-order harmonic; Among them, low-order harmonics are n≥3, and high-order harmonics are n=1.

7. The rapid detection method for microplastics based on image recognition according to claim 6, characterized in that: Mapping the time-domain fluorescence decay curve to the quantum state space and extracting the intrinsic parameters of the fluorescence lifetime based on the variational quantum algorithm includes: The time-domain fluorescence decay signal is converted into a quantum state by an amplitude encoding method, and the quantum state is prepared into a quantum state corresponding to the fluorescence decay curve based on a quantum circuit; Iteratively adjust the parameters of quantum circuits based on classical optimizers and quantum computing; Based on quantum superposition state and quantum parallel computing, the target fluorescence lifetime parameters are determined.

8. The rapid detection method for microplastics based on image recognition according to claim 1, characterized in that: Establishing a high-order tensor decomposition model under multiple excitation wavelengths and using a quantum optimization algorithm to determine the optimal tensor factor combination includes: Fluorescence data from multiple excitation wavelengths are constructed as a high-order tensor. Based on tensor decomposition technology, the high-order tensor is decomposed into multiple low-order factor matrices and core tensors to extract relevant features between dimensions. Optimize high-order tensor decomposition problems based on quantum optimization algorithms, and determine the optimal factor matrix combination based on the superposition and parallelism of quantum computing; The quantum optimization algorithm minimizes the loss function and determines the optimal tensor decomposition combination of the tensor factor matrix.

9. The rapid detection method for microplastics based on image recognition according to claim 1, characterized in that: Reconstructing the fluorescence lifetime intrinsic parameters based on the matrix product state and the optimal tensor factor combination, extracting the fluorescence lifetime eigenvector, and determining the type of microplastics in the sample to be detected based on the fluorescence lifetime eigenvector, including: Fit the fluorescence decay curve to a biexponential model to determine the intrinsic parameters of the fluorescence lifetime, where the intrinsic parameters include decay time constants τ1, τ2 and their corresponding amplitudes A1, A2; The intrinsic parameters of the fluorescence lifetime are fitted based on the least squares method, and the decay parameters are extracted, and the extracted decay parameters are combined into a fluorescence lifetime characteristic vector; The extracted fluorescence lifetime feature vector is subjected to mean-variance processing, and the processed fluorescence lifetime feature vector is substituted into the pre-established microplastic classification dataset to obtain the similarity between the fluorescence lifetime feature vector and the feature vector of each microplastic classification in the microplastic classification dataset; The similarities between the fluorescence lifetime characteristic vector and the characteristic vector of each microplastic classification are ranked in reverse order, and the microplastic classification corresponding to the first-ranked similarity is determined as the microplastic type of the sample to be tested.

10. The rapid detection method for microplastics based on image recognition according to claim 9, characterized in that: When pre-establishing a microplastic classification dataset, include: Obtain the fluorescence lifetime decay parameters of each microplastic, and establish a microplastic data correlation formula based on the fluorescence lifetime decay parameters; The distance metric between each microplastic data association formula is obtained based on the Euclidean distance, and a distance matrix is ​​established based on the distance metric. The microplastic data association formulas are iteratively clustered based on the distance matrix. According to the clustering results, the characteristic vectors of each microplastic type were determined, and a microplastic classification dataset was established based on the characteristic vectors of each microplastic type.

Images