A marine water body microplastic particle monitoring system and method thereof

Abstract

The application relates to the technical field of marine environment monitoring, and discloses a marine water body micro-plastic particle monitoring system and a method thereof, wherein the method comprises the following steps: acquiring a three-mode synchronous signal data stream; generating an event window marker sequence based on forward scattering pulse analysis; performing sparse deconvolution separation on photoacoustic signals in a multi-particle overlapping event window; extracting particle physical parameters; taking the number of particles as a component number hard constraint, converting a volume proportion weight vector into an abundance coefficient constraint, performing constraint non-negative matrix decomposition on overlapping stimulated Raman spectra; performing reconstruction verification; merging identification results and counting concentrations.

Description

Technical Field

[0001] This invention relates to the field of marine environmental monitoring technology, specifically to a marine water microplastic particle monitoring system and method. Background Technology

[0002] Monitoring microplastic pollution in marine waters is an important part of marine environmental protection. In existing technologies, online flow cytometry detection methods typically employ a combined photoacoustic and stimulated Raman scattering approach to extract physical parameters and identify the material of microplastic particles in the sampled water. This method assumes that only a single particle is within the laser irradiation area at any given time.

[0003] However, in high-concentration marine areas such as nearshore sewage outfalls and estuaries, particle concentrations can reach tens of thousands per liter, with multiple particles frequently passing through the detection area simultaneously. In such cases, photoacoustic signals undergo temporal aliasing, resulting in the superposition of multiple polymer spectra in stimulated Raman spectroscopy. While photoacoustic signal demixing can separate particle physical parameters, photoacoustic features lack the ability to distinguish the chemical composition of polymers, making material identification impossible on their own. Stimulated Raman spectroscopy demixing faces unknown component numbers and a vast endmember combination space when dealing with overlapping spectra; sparse regularization alone cannot guarantee the uniqueness of the solution, leading to false material determinations when component estimation is incorrect. The independent processing of the two modes each presents information gaps, resulting in decreased accuracy in material identification and distorted concentration statistics in high-concentration marine monitoring. Summary of the Invention

[0004] This invention provides a marine microplastic particle monitoring system and method, which solves the technical problem in related technologies where multiple particles simultaneously enter the detection area, causing the photoacoustic signal and Raman spectrum to overlap, making it impossible to independently extract particle size and identify the material of each particle.

[0005] This invention discloses a method for monitoring microplastic particles in marine water, comprising: acquiring photoacoustic signals, stimulated Raman loss signals, and forward scattered light pulse signals generated by pulsed laser irradiation of particle groups in a marine water sampling stream, and performing synchronization alignment of the three signals according to a unified clock; Peak detection is performed on the forward-scattered light pulse signal, and the event window is marked as a single-particle event window or a multi-particle overlapping event window based on the comparison between the time interval of adjacent pulses and the duration threshold. For the photoacoustic signals within the multi-particle overlapping event window, sparse deconvolution is performed using the scattering event timestamp as a priori constraint to obtain the estimated separation photoacoustic amplitude and separation delay of each particle. The equivalent particle size of each particle is extracted based on the separated photoacoustic amplitude estimate, and the volume proportion weight vector of each particle is calculated. The number of particles derived from photoacoustic analysis is used as a hard constraint on the components of non-negative matrix decomposition. The volume proportion weight vector is converted by Raman scattering cross section and used as the initial value of the abundance coefficient and the relaxation range constraint. Constrained non-negative matrix decomposition is performed on the overlapping stimulated Raman spectrum to obtain the abundance coefficient of each component and the candidate material identifier.

[0006] Furthermore, the duration threshold is obtained based on the statistical characteristics of the photoacoustic pulses collected in a single-particle event window, and the average of the durations of multiple single-particle photoacoustic pulses is taken as the duration threshold; wherein, when the difference in arrival time between two adjacent scattering pulses is less than the duration threshold, the photoacoustic response of the previous particle has not yet decayed and the photoacoustic response of the next particle has already overlapped, and this window is marked as a multi-particle overlapping event window.

[0007] Furthermore, the step of performing sparse deconvolution using the scattering event timestamp as a priori constraint includes: The aliased photoacoustic signal within the multi-particle overlapping event window is modeled as the linear superposition of multiple time-shifted photoacoustic impulse responses and the sum of noise, wherein each time-shifted photoacoustic impulse response is represented by a convolution kernel and corresponding amplitude coefficients and time delay parameters. The convolution kernel is obtained by averaging the waveforms of multiple photoacoustic pulse signals collected from a single particle event window; Using the timestamp of the scattering event as the initial estimate of each delay, the solution range of each delay is limited to a narrow interval near the corresponding timestamp, and the amplitude coefficient and precise delay of each particle are solved under the sparsity regularization constraint.

[0008] Furthermore, the calculation of the volume percentage weight vector for each particle includes: A short-time Fourier transform is performed on each separated photoacoustic pulse response, and the equivalent particle size of each particle is obtained through a forward mapping relationship based on the peak frequency of the photoacoustic spectrum. The volume percentage weight of each particle is determined by the ratio of the cube of the equivalent particle size to the sum of the cubes of the equivalent particle sizes of all particles in the window.

[0009] Furthermore, the volume percentage weight vector, after being converted using the Raman scattering cross section, is used as the initial value for the abundance coefficient, including: The volume percentage weight of each particle is multiplied by the Raman scattering cross section of the corresponding candidate polymer material, and then the product result is normalized to obtain the initial value of the abundance coefficient of each component. The Raman scattering cross section is assigned according to the cross section values ​​of each reference material in the standard polymer Raman spectroscopy library, and is updated synchronously with the endmember selection during the iterative process of constrained nonnegative matrix decomposition.

[0010] Furthermore, the constrained nonnegative matrix decomposition includes: Starting from the initial value of the abundance coefficient, and under the constraint that the number of components is fixed as the number of photoacoustic derived particles, the overlapping stimulated Raman spectra are optimized by alternating least squares iterative optimization using all reference spectra in the standard polymer Raman spectroscopy library as endmember matrices. In each iteration, fixed endmembers are selected to update the abundance coefficients, while non-negativity constraints and sparse regularization terms are applied, and the value of each abundance coefficient is restricted to an interval centered on the corresponding initial value and with a preset relaxation parameter as the radius. When the iteration converges, the candidate material identifiers of each component are obtained based on the index of the endmembers corresponding to the non-zero components in the abundance coefficients in the standard polymer Raman spectroscopy library.

[0011] Furthermore, after obtaining the candidate material identifier, the process also includes: The mixed spectrum is linearly reconstructed using the reference spectrum and abundance coefficients corresponding to the candidate material to generate the reconstructed spectrum; The ratio of the L2 norm of the difference between the reconstructed spectrum and the original overlapped stimulated Raman spectrum to the L2 norm of the original overlapped stimulated Raman spectrum is calculated and used as an indicator of reconstruction accuracy. When the reconstruction accuracy index is lower than a preset threshold, the candidate material identifier is confirmed as the material identification result of each particle.

[0012] Furthermore, when the reconstruction accuracy index exceeds a preset threshold, it also includes: Calculate the residual spectrum between the original overlapping stimulated Raman spectrum and the reconstructed spectrum; The residual spectra are matched twice with an extended spectral library containing the spectra of weathered and degraded polymers to identify material components not covered by the standard polymer Raman spectral library. After adding the newly identified materials to the candidate component list, the constrained nonnegative matrix decomposition is re-executed, the abundance coefficients are updated, and the reconstruction accuracy index is recalculated to generate the final material identification results for each overlapping particle.

[0013] Furthermore, it also includes: The photoacoustic signal and stimulated Raman spectrum within a single particle event window are extracted independently, and the material determination result and equivalent particle size are obtained by matching the standard polymer Raman spectroscopy library. The identification results of single-particle events are combined with the unmixing identification results of multi-particle overlapping events, and cumulative counting is performed within the monitoring window according to material type and particle size range. By combining the sampling flow rate, the particle concentration of various microplastics in a unit volume of seawater is calculated, generating monitoring data on the classification and concentration of microplastic materials in marine water.

[0014] This invention provides a marine water microplastic particle monitoring system, comprising: The signal synchronization acquisition module is used to acquire the photoacoustic signal, stimulated Raman loss signal and forward scattered light pulse signal generated by the particle group in the ocean water sampling flow after being irradiated by pulsed laser, and to perform synchronization alignment of the three signals according to a unified clock. The event window marking module is used to perform peak detection on the forward-scattered light pulse signal and mark the event window as a single-particle event window or a multi-particle overlapping event window based on the comparison between the time interval of adjacent pulses and the duration threshold. The photoacoustic signal separation module is used to perform sparse deconvolution on the photoacoustic signals within the multi-particle overlapping event window, using the scattering event timestamp as a priori constraint, to obtain the separated photoacoustic amplitude estimate and separation delay of each particle. The physical parameter extraction module is used to extract the equivalent particle size of each particle based on the estimated value of the separated photoacoustic amplitude, and to calculate the volume percentage weight vector of each particle. The cross-modal constrained unmixing module is used to take the number of particles derived from photoacoustics as the component hard constraint of non-negative matrix decomposition, and take the volume proportion weight vector after Raman scattering cross section conversion as the initial value of abundance coefficient and relaxation range constraint. It performs constrained non-negative matrix decomposition on the overlapping stimulated Raman spectrum to obtain the abundance coefficient and candidate material identifier of each component.

[0015] The present invention has the following beneficial effects.

[0016] By using forward scattering pulse timestamp constraints and sparse deconvolution to separate aliased photoacoustic signals, physical parameters such as the number and volume weight of each particle are obtained, providing prior information from independent modes for subsequent stimulated Raman spectroscopy demixing, thus avoiding the uncertainty caused by the need to estimate the number of components from the stimulated Raman spectrum itself.

[0017] By using the number of particles as a hard constraint on the number of components in nonnegative matrix factorization, stimulated Raman spectroscopy unmixing is transformed from an unconstrained blind separation with unknown components into a constrained optimization problem with known components, thus eliminating the problem of false material determination when the number of components is estimated incorrectly.

[0018] The volume proportion weight vector is converted and used as the initial value of the abundance coefficient and the relaxation range constraint. The search range of endmember combination is limited to a finite subset compatible with the physical parameters, so that sparse regularization can obtain a unique solution in the narrowed solution space.

[0019] Photoacoustic mode provides particle separation capabilities at the physical parameter level, compensating for the uncertainty in component fractionation in stimulated Raman spectroscopy unmixing; the chemical specificity of stimulated Raman mode compensates for the inability of photoacoustic mode to accurately distinguish polymer types. The two modes complement each other at the information level, enabling the monitoring of high-concentration marine microplastics to simultaneously obtain particle counting and material identification results.

[0020] By introducing a secondary matching step between residual spectra and extended spectral libraries, it is possible to cover microplastic particles whose Raman spectral characteristic peaks have shifted or broadened due to weathering degradation and whose corresponding entries are missing in the standard polymer Raman spectral library, thereby improving the coverage of material identification. Attached Figure Description

[0021] Figure 1 This is a flowchart of the method for monitoring microplastic particles in marine water provided in an embodiment of the present invention; Figure 2 This is a schematic diagram showing the arrival time and peak intensity of the scattered pulses in the three signal channels provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the precise time delay and photoacoustic amplitude coefficient of each particle after sparse deconvolution separation provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the equivalent particle size and volume percentage weight of each particle provided in the embodiments of the present invention; Figure 5 This is a schematic diagram comparing the initial and final values ​​of the abundance coefficients in the constrained nonnegative matrix decomposition provided in an embodiment of the present invention. Figure 6 This is a schematic diagram comparing the Raman scattering cross sections of four candidate materials provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of the cumulative particle count of microplastic materials according to an embodiment of the present invention; Figure 8 This is a schematic diagram of the particle concentration distribution of microplastic materials according to an embodiment of the present invention; Figure 9 This is a schematic diagram comparing the reconstruction accuracy index provided in this embodiment of the invention with a preset threshold. Detailed Implementation

[0022] Microplastic particle monitoring in marine waters typically employs a combined photoacoustic and stimulated Raman scattering (SRS) detection method, performing online flow cytometry analysis on microplastic particles in sampled water. In areas with normal concentrations, particles pass through the detection area one by one, with photoacoustic signals and stimulated Raman scattering independently extracting features and performing material matching. This process assumes that only a single particle is within the laser irradiation area at any given time. However, in high-concentration areas such as nearshore sewage outfalls and estuaries, particle concentrations can reach tens of thousands per liter, with multiple particles frequently passing through the detection area simultaneously. In these cases, photoacoustic signals undergo temporal aliasing, and stimulated Raman scattering produces a superposition of multiple polymer spectra. While photoacoustic signal demixing can separate the physical parameters of each particle, photoacoustic features lack the ability to distinguish the chemical composition of polymers, making material identification impossible. Raman spectroscopy demixing faces unknown component numbers and a vast endmember combination space when dealing with overlapping spectra; sparse regularization alone cannot guarantee the uniqueness of the solution, leading to false material determinations when component estimation is incorrect. The two modalities, when processed independently, each have their own irreparable information gaps, leading to a decrease in the accuracy of material identification and distortion of concentration statistics in high-concentration marine monitoring.

[0023] The detection method provided in this embodiment requires the following hardware environment support: a pulsed laser, a photoacoustic transducer, a stimulated Raman loss detector, and a forward scattering light detector are configured in the streaming detection channel, and the four signal acquisition channels share a unified clock source.

[0024] A method for monitoring microplastic particles in marine water according to an embodiment of the present invention, such as... Figure 1 As shown, it includes the following steps: Step 1: Obtain the three-mode synchronization signal data stream.

[0025] The system acquires a continuous photoacoustic signal sequence, a continuous stimulated Raman loss signal sequence, and a continuous forward-scattered light pulse signal sequence generated when a particle group in an ocean water sampling stream is irradiated by a pulsed laser. The three signals are synchronized and aligned using a unified clock to generate a time-synchronized three-mode continuous signal data stream.

[0026] Step 2: Generate an event window marker sequence based on forward scattering pulse analysis.

[0027] Peak detection and pulse width analysis are performed on the forward-scattered light pulse signal to obtain the arrival time and peak intensity of each scattered pulse. Based on the comparison between the time interval of adjacent pulses and the typical duration of the photoacoustic signal, windows with time intervals less than the duration threshold are marked as multi-particle overlapping event windows, and the remaining windows are marked as single-particle event windows, generating an event window marking sequence and a set of scattering event timestamps within each window.

[0028] It should be noted that the aforementioned duration threshold refers to the typical time length during which a single-particle photoacoustic pulse decays from its generation to the background noise level. When the arrival time difference between two adjacent scattered pulses is less than the duration threshold, the photoacoustic response of the preceding particle has not yet completely decayed before the photoacoustic response of the following particle is superimposed, resulting in temporal aliasing of the photoacoustic signal within this window. The duration threshold is obtained based on the statistical characteristics of the photoacoustic pulses collected within a single-particle event window, taking the average of multiple single-particle photoacoustic pulse durations as the duration threshold.

[0029] Step 3: Perform sparse deconvolution separation on the photoacoustic signals within the multi-particle overlapping event window.

[0030] For the photoacoustic signal segment within the multi-particle overlapping event window, the scattering event timestamp is used as a priori constraint for the start time of the photoacoustic signal of each particle. The aliased photoacoustic signal is modeled as the superposition of multiple time-shifted photoacoustic pulse responses. The amplitude coefficient and precise time delay of each component are solved to generate the estimated value of the separated photoacoustic amplitude and the separation time delay of each overlapping particle.

[0031] It should be noted that the signal model of the above sparse deconvolution is expressed as:

[0032] in, These are aliased photoacoustic signals acquired within a multi-particle overlap event window. For time variables, The number of particles indicated by the set of timestamps of scattering events within this window. For granular index and The range of values ​​is to , For the first The photoacoustic amplitude coefficient of each particle The convolution kernel is the average photoacoustic pulse template. For the first Precise time delay of each particle, This is the noise term. Convolution kernel. The convolution kernel is obtained by statistical estimation from multiple photoacoustic pulse signals collected within a single-particle event window, and the average of the multiple single-particle photoacoustic pulse waveforms is used as the convolution kernel. The timestamps of the scattering events are for each delay. Provide initial estimates, limit the solution range to a narrow interval near the timestamp, and solve for each under sparsity regularization constraints. and The optimal value.

[0033] Step 4: Extract particle physical parameters based on the separated photoacoustic signals.

[0034] Short-time Fourier transforms are performed on each separated photoacoustic pulse response to extract frequency domain features. Based on the forward mapping relationship between photoacoustic frequency and particle size, the equivalent particle size of each overlapping particle is estimated. Using the estimated separated photoacoustic amplitude and equivalent particle size, the volume percentage weight of each particle is calculated to generate the number of particles derived from the photoacoustic model. Equivalent particle size sequence and volume percentage weight vector ,in This represents the total number of particles within the window. Indicates the first The proportion of the scattering volume of a single particle in the total scattering volume of all particles within the window, and the equivalent particle size, are obtained from the peak frequency of the photoacoustic spectrum through a forward mapping relationship.

[0035] It should be noted that the above volume percentage weighting The calculation method is as follows: Let the first... The equivalent particle size of each particle is Then its equivalent sphere volume is proportional to , No. The volume percentage weight of each particle is:

[0036] in, For the first The equivalent particle size of each particle. This represents the total number of particles within the window. For summation index and The range of values ​​is to , The ratio is the sum of the equivalent sphere volumes of all particles within the window. It is composed of the ratio of volumes with the same dimension, and the calculated result is a dimensionless ratio value.

[0037] Step 5: Perform constrained nonnegative matrix decomposition on the overlapping stimulated Raman spectra based on cross-modal constraints.

[0038] Number of particles As a hard constraint on the components of nonnegative matrix factorization, the volume proportion weight vector is used. After scaling up the Raman scattering cross section to the volume, the values ​​were used as the initial values ​​for the abundance coefficients and as constraints for the relaxation range. All reference spectra from the standard polymer Raman spectral library were used as endmember matrices. Overlapping stimulated Raman spectra collected within the same time window Perform constrained nonnegative matrix factorization to generate the abundance coefficients of each component under cross-modal constraints. And candidate material identifiers.

[0039] Step 5 may specifically include: Step 501, based on the volume proportion weight vector For each candidate polymer material, the Raman scattering cross section is used to convert the volume percentage weight into the Raman signal intensity percentage, generating an initial vector of abundance coefficients. .

[0040] It should be noted that the above conversion method is as follows: Let the first... The Raman scattering cross section of each particle corresponding to the candidate polymer material is Then the first The initial values ​​for the abundance coefficients of each component are:

[0041] in, For the first The volume percentage weight of each particle For the first Each particle corresponds to a Raman scattering cross section of the candidate polymer material. This represents the total number of particles within the window. For summation index and The range of values ​​is to , As a normalized benchmark, This is a dimensionless proportionality value, reflecting the expected contribution of the particle to the intensity of the mixed Raman signal.

[0042] It should be further explained that the Raman scattering cross-sections of each candidate polymer material in step 501... Since the specific material was not yet determined during the initial conversion, the cross-sectional values ​​of each reference material in the standard polymer Raman spectroscopy library were taken, and the results were calculated based on the endmember matrix. The corresponding material values ​​in each column are assigned sequentially, and the Raman scattering cross section is updated synchronously with the endmember selection during the iteration process in step 502. The value of is taken until the iteration converges.

[0043] Step 502, using the initial value vector of abundance coefficients Starting from the group number, with the group number fixed as follows: Under the constraint of overlapping stimulated Raman spectra With endmember matrix Alternating least squares iterative optimization is performed. In each iteration, the abundance coefficients are updated by fixing the endmember selection, while applying nonnegativity constraints and sparsity regularization terms, and each abundance coefficient is... The value of is restricted to the interval Inside, among which This is the initial value for this component. These are the preset relaxation parameters.

[0044] Step 503: When the iteration converges, output the final abundance coefficients. Based on the index of the endmembers corresponding to the non-zero components in the abundance coefficient in the standard polymer Raman spectroscopy library, the candidate material identifiers for each component are obtained.

[0045] In this embodiment of the application, by increasing the number of particles As a hard constraint on the component fractions, stimulated Raman spectroscopy unmixing no longer requires searching through all possible component fractions, eliminating spurious material determinations caused by component fraction estimation errors. By converting volume fraction weights into initial values ​​of abundance coefficients and relaxation range constraints, the endmember combination space is reduced from the full combination of the standard polymer Raman spectral library to a finite subset compatible with physical parameters, thus reducing the dimensionality of the unmixing space.

[0046] Step 6: Perform reconstruction verification on the candidate material identifiers.

[0047] Using the reference spectrum and abundance coefficients of the candidate materials Linear reconstruction of mixed spectra, generating reconstructed spectra ,in This represents the reference spectral submatrix corresponding to the candidate material. The reconstructed spectrum is then calculated. Overlapping stimulated Raman spectrum with the original Reconstruction accuracy indicators between :

[0048] in, The original superimposed stimulated Raman spectrum, To reconstruct the spectrum, For L2 norm operations, numerator With denominator They have the same dimensions. This is a dimensionless relative error. A smaller value indicates a higher degree of agreement between the reconstructed spectrum and the original overlapping stimulated Raman spectrum. When Below the preset threshold At that time, the candidate material identifier is confirmed as the material identification result of each particle. Exceeding the preset threshold At that time, it was determined that the current candidate material combination failed to fully explain the original overlapping stimulated Raman spectrum.

[0049] It should be noted that the above-mentioned preset threshold The maximum acceptable relative error between the reconstructed spectrum and the original overlapping stimulated Raman spectrum can be determined based on the statistical distribution of reconstruction errors of known material particles within a single-particle event window. The mean of the reconstruction accuracy index under a single-particle event plus a certain number of standard deviations is taken as the preset threshold. .

[0050] In this embodiment of the application, in order to process the reconstruction accuracy index Exceeding the preset threshold In this case, the following steps are also included: calculating the residual spectrum. The residual spectra are then subjected to a secondary matching with an extended spectral library containing the spectra of weathered and degraded polymers to identify material components in the residual spectra that may not be covered by the standard polymer Raman spectral library. After adding the newly identified materials to the candidate component list, the constrained nonnegative matrix factorization iteration in step 5 is re-executed to update the abundance coefficients and recalculate the reconstruction accuracy index, generating the final material identification results and reconstruction accuracy evaluation values ​​for each overlapping particle. This process can cover microplastic particles in marine environments whose polymer surface chemical structures have changed due to factors such as ultraviolet radiation and saltwater erosion. These particles exhibit Raman spectral characteristic peaks that are shifted or broadened compared to the original polymers, and for which corresponding entries are lacking in the standard polymer Raman spectral library.

[0051] Step 7: Combine the identification results of single-particle events and multi-particle events and calculate the concentration.

[0052] Features were independently extracted from the photoacoustic signal and stimulated Raman spectrum within a single-particle event window. Material identification results and equivalent particle sizes were obtained by matching with a standard polymer Raman spectral library. The identification results of single-particle events were combined with the unmixing identification results of multi-particle overlapping events, and cumulative counting was performed within the monitoring window according to material category and particle size range. Combined with the sampling flow rate, the particle concentration of various microplastics per unit volume of seawater was calculated, generating marine microplastic material classification and concentration monitoring data.

[0053] This implementation addresses the problem of temporal aliasing and stimulated Raman spectral superposition caused by multiple particles simultaneously passing through the detection area in high-concentration marine environments. It utilizes forward scattering pulse timestamp constraints and sparse deconvolution to separate the aliased photoacoustic signals, obtaining physical parameters such as the number and volume percentage of each particle. The particle number... As a hard constraint on the components of nonnegative matrix factorization, the volume proportion weight vector After conversion, these values ​​serve as initial values ​​for abundance coefficients and relaxation range constraints, transforming stimulated Raman spectroscopy unmixing from an unconstrained blind separation problem with unknown components into a constrained optimization problem with defined components and limited abundance coefficients. Since the components are independently determined by the photoacoustic mode rather than estimated from the stimulated Raman spectrum itself, the problem of false material identification due to component estimation errors is eliminated. Because the search range of abundance coefficients is constrained within a reasonable range by physical parameters, the endmember combination space is significantly reduced, and sparse regularization can obtain unique solutions within the narrowed solution space. The photoacoustic mode provides particle separation capability at the physical parameter level, compensating for the uncertainty of components in stimulated Raman spectroscopy unmixing; the chemical specificity of the stimulated Raman mode compensates for the inability of the photoacoustic mode to accurately distinguish polymer types. The two modes complement each other at the information level, enabling the monitoring of high-concentration marine microplastics to simultaneously obtain accurate particle counting and material identification results.

[0054] The following is an example of an application of the present invention, such as Figures 2-9 As shown, the implementation process is as follows: In a microplastic pollution monitoring mission at a nearshore sewage outfall, the sampled water flowed through a flow cytometry channel at a rate of approximately 120 ml per minute. The monitoring period was spring of 20XX, and the particle concentration in this area was approximately 32,000 particles per liter, a typical high-concentration scenario. The following section illustrates the data flow process at each step, focusing on a multi-particle overlap event window captured during a specific continuous sampling period.

[0055] The three signal acquisition channels operate synchronously under a unified clock, with the photoacoustic transducer, stimulated Raman loss detector, and forward scattering detector each outputting a continuous signal sequence. During this time period, the forward scattering channel captures several scattering pulses, the photoacoustic channel synchronously records aliased waveforms, and the stimulated Raman loss channel records superimposed spectra. After synchronization and alignment, the three signals form a time-corresponding three-mode continuous signal data stream, providing a unified time reference for subsequent event window division.

[0056] Table 1. Tri-mode synchronous signal acquisition parameters

[0057] Peak detection was performed on the forward-scattered light pulse signal, and multiple scattered pulses were identified within the sampling period. A typical time window was selected below, in which three scattered pulses were detected, denoted as pulse A, pulse B, and pulse C. The arrival time difference between pulse A and pulse B is 1.8 microseconds, and the arrival time difference between pulse B and pulse C is 2.1 microseconds. The typical duration threshold of a single-particle photoacoustic pulse was obtained statistically from single-particle event windows, with an average value of 4.5 microseconds. Since the time interval between adjacent pulses is less than 4.5 microseconds, this window is marked as a multi-particle overlapping event window, and the set of timestamps for the three scattering events is denoted as... The corresponding times are 12.3 microseconds, 14.1 microseconds and 16.2 microseconds, respectively.

[0058] Table 2 Forward Scatter Pulse Detection and Event Window Marking

[0059] Using three scattering event timestamps as prior constraints, sparse deconvolution separation is performed on the aliased photoacoustic signals within the window. The signal model is as follows:

[0060] in This is the template for the average photoacoustic pulse statistically estimated from a single particle event window. The search range is limited to a narrow interval of ±0.5 microseconds around each scattering timestamp. After solving under sparsity regularization constraints, the precise time delay and photoacoustic amplitude coefficient of each particle are obtained.

[0061] Table 3. Results of sparse deconvolution separation

[0062] A short-time Fourier transform is performed on each separated photoacoustic pulse response to extract the peak frequency in the frequency domain. The equivalent particle size is estimated using the forward mapping relationship between photoacoustic frequency and particle size. The peak frequency of particle A corresponds to the equivalent particle size. It is 58 micrometers, and particle B corresponds to The particle size is 43 micrometers, and particle C corresponds to... The value is 52 micrometers. The volume percentage weight is calculated using the following formula:

[0063] The sum of the volumes of the three equivalent spheres is based on the following standard: The calculation result is , , .

[0064] Table 4 Results of particle physical parameter extraction

[0065] Based on the number of particles As a hard constraint for the components of the nonnegative matrix factorization, four reference materials—polyethylene (PE), polypropylene (PP), polystyrene (PS), and polyethylene terephthalate (PET)—were selected from the standard polymer Raman spectroscopy library to form the endmember matrix. Their Raman scattering cross sections are respectively , , , (Unit: relative cross section, au).

[0066] In step 501, the initial abundance coefficient is calculated based on the scattering cross-section of each initial candidate material particle:

[0067] In the initial iteration phase, the candidate materials for each particle were tentatively set as PE, PP, and PS, respectively. After conversion, the following results were obtained. Relaxation parameters The value is set to 0.08, and the search range for each abundance coefficient is limited to ±0.08 of the corresponding initial value.

[0068] In the alternating least squares iteration in step 502, the abundance coefficients are updated with fixed endmember selection, nonnegativity constraints and sparsity regularization terms are applied, and the iteration continues until convergence. Step 503 outputs the final abundance coefficients. The non-zero components correspond to the end members PE, PP, and PS in sequence, and the candidate material identifiers are confirmed to be three polymers.

[0069] Table 5. Abundance Coefficients Results of Constrained Nonnegative Matrix Decomposition

[0070] Using the reference spectral sub-matrices corresponding to candidate materials PE, PP, and PS With the final abundance coefficient Linear reconstruction of the mixed spectrum, calculation of reconstruction accuracy metrics:

[0071] The window's reconstruction accuracy index The calculated result is 0.047. (Preset threshold) The reconstruction error was determined by the statistical distribution of the single-particle event window, taking the mean plus twice the standard deviation, resulting in a value of 0.12. Because... The candidate material identifiers were verified through reconstruction, confirming that particle A is PE, particle B is PP, and particle C is PS.

[0072] Table 6 Reconstruction Verification Results

[0073] Throughout the monitoring period, material identification results from all single-particle event windows and multi-particle overlapping event windows were merged and accumulated. For single-particle event windows, material and equivalent particle size were obtained through direct matching using a standard spectral library, and these results were combined with the unmixing results from the multi-particle overlapping events and counted according to material category and particle size range. Combined with a sampling flow rate of 120 ml / min, the accumulated particle count was converted into the particle concentration of various microplastics per unit volume of seawater, generating material classification concentration monitoring data.

[0074] Table 7. Statistical Results of Microplastic Material Classification and Concentration

[0075] The entire data flow process reflects the logical transmission relationship between each step. The forward scattering pulse timestamp is passed from step 2 to step 3, providing a priori constraints for sparse deconvolution and ensuring that the time delay search range is narrowed to the vicinity of the scattering event. The separated photoacoustic amplitude coefficients and time delay output from step 3 are entered into step 4 and transformed into equivalent particle size and volume percentage weight vectors. . The physical parameter information is carried into step 5, and after Raman scattering cross-section conversion, it provides initial values ​​of abundance coefficients and relaxation range constraints for stimulated Raman spectroscopy unmixing, while the number of particles is also determined. The component hard constraints lock the solution space dimension. The candidate material identifiers and abundance coefficients output in step 5 are reconstructed and verified in step 6, and finally merged with the single-particle event results in step 7 to form material classification concentration monitoring data. Photoacoustic mode completes particle separation at the physical parameter level, while stimulated Raman mode completes material identification at the chemical specificity level. The information from both complements each other in the constraint transfer in step 5, enabling particle counting and material identification to be completed simultaneously in high-concentration marine multi-particle overlapping scenarios.

[0076] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. A method for monitoring microplastic particles in marine water, characterized in that, include: Acquire photoacoustic signals, stimulated Raman loss signals, and forward scattered light pulse signals generated by pulsed laser irradiation of particle groups in ocean water sampling streams, and perform synchronization alignment of the three signals according to a unified clock. Peak detection is performed on the forward-scattered light pulse signal, and the event window is marked as a single-particle event window or a multi-particle overlapping event window based on the comparison between the time interval of adjacent pulses and the duration threshold. For the photoacoustic signals within the multi-particle overlapping event window, sparse deconvolution is performed using the scattering event timestamp as a priori constraint to obtain the estimated separation photoacoustic amplitude and separation delay of each particle. The equivalent particle size of each particle is extracted based on the separated photoacoustic amplitude estimate, and the volume proportion weight vector of each particle is calculated. The number of particles derived from photoacoustic analysis is used as a hard constraint on the components of non-negative matrix decomposition. The volume proportion weight vector is converted by Raman scattering cross section and used as the initial value and relaxation range constraint of the abundance coefficient. Constrained non-negative matrix decomposition is performed on the overlapping stimulated Raman spectrum to obtain the abundance coefficient and candidate material identifier of each component. in: The step of performing sparse deconvolution using scattering event timestamps as prior constraints includes: The aliased photoacoustic signal within the multi-particle overlapping event window is modeled as the linear superposition of multiple time-shifted photoacoustic impulse responses and the sum of noise, wherein each time-shifted photoacoustic impulse response is represented by a convolution kernel and corresponding amplitude coefficients and time delay parameters. The convolution kernel is obtained by averaging the waveforms of multiple photoacoustic pulse signals collected from a single particle event window; Using the timestamp of the scattering event as the initial estimate of each delay, the solution range of each delay is limited to a narrow interval near the corresponding timestamp, and the amplitude coefficient and precise delay of each particle are solved under the sparsity regularization constraint. The constrained nonnegative matrix decomposition includes: Starting from the initial value of the abundance coefficient, and under the constraint that the number of components is fixed as the number of photoacoustic derived particles, the overlapping stimulated Raman spectra are optimized by alternating least squares iterative optimization using all reference spectra in the standard polymer Raman spectroscopy library as endmember matrices. In each iteration, fixed endmembers are selected to update the abundance coefficients, while non-negativity constraints and sparse regularization terms are applied, and the value of each abundance coefficient is restricted to an interval centered on the corresponding initial value and with a preset relaxation parameter as the radius. When the iteration converges, the candidate material identifiers of each component are obtained based on the index of the endmembers corresponding to the non-zero components in the abundance coefficients in the standard polymer Raman spectroscopy library.

2. The method for monitoring microplastic particles in marine waters according to claim 1, characterized in that, The duration threshold is obtained based on the statistical characteristics of photoacoustic pulses collected in a single-particle event window. The average duration of multiple single-particle photoacoustic pulses is taken as the duration threshold. When the difference in arrival time between two adjacent scattering pulses is less than the duration threshold, the photoacoustic response of the previous particle has not yet decayed and the photoacoustic response of the next particle has already overlapped. This window is marked as a multi-particle overlapping event window.

3. The method for monitoring microplastic particles in marine waters according to claim 1, characterized in that, The calculation of the volume percentage weight vector for each particle includes: A short-time Fourier transform is performed on each separated photoacoustic pulse response, and the equivalent particle size of each particle is obtained through a forward mapping relationship based on the peak frequency of the photoacoustic spectrum. The volume percentage weight of each particle is determined by the ratio of the cube of the equivalent particle size to the sum of the cubes of the equivalent particle sizes of all particles in the window.

4. The method for monitoring microplastic particles in marine waters according to claim 1, characterized in that, The volume percentage weight vector, after being converted using Raman scattering cross sections, is used as the initial value for the abundance coefficients, including: The volume percentage weight of each particle is multiplied by the Raman scattering cross section of the corresponding candidate polymer material, and then the product result is normalized to obtain the initial value of the abundance coefficient of each component. The Raman scattering cross section is assigned according to the cross section values ​​of each reference material in the standard polymer Raman spectroscopy library, and is updated synchronously with the endmember selection during the iterative process of constrained nonnegative matrix decomposition.

5. The method for monitoring microplastic particles in marine waters according to claim 1, characterized in that, After obtaining the candidate material identifier, the method further includes: The mixed spectrum is linearly reconstructed using the reference spectrum and abundance coefficients corresponding to the candidate material to generate the reconstructed spectrum; The ratio of the L2 norm of the difference between the reconstructed spectrum and the original overlapped stimulated Raman spectrum to the L2 norm of the original overlapped stimulated Raman spectrum is calculated and used as an indicator of reconstruction accuracy. When the reconstruction accuracy index is lower than a preset threshold, the candidate material identifier is confirmed as the material identification result of each particle.

6. The method for monitoring microplastic particles in marine waters according to claim 5, characterized in that, When the reconstruction accuracy index exceeds a preset threshold, the following further applies: Calculate the residual spectrum between the original overlapping stimulated Raman spectrum and the reconstructed spectrum; The residual spectra are matched twice with an extended spectral library containing the spectra of weathered and degraded polymers to identify material components not covered by the standard polymer Raman spectral library. After adding the newly identified materials to the candidate component list, the constrained nonnegative matrix decomposition is re-executed, the abundance coefficients are updated, and the reconstruction accuracy index is recalculated to generate the final material identification results for each overlapping particle.

7. The method for monitoring microplastic particles in marine waters according to claim 5, characterized in that, Also includes: The photoacoustic signal and stimulated Raman spectrum within a single particle event window are extracted independently, and the material determination result and equivalent particle size are obtained by matching the standard polymer Raman spectroscopy library. The identification results of single-particle events are combined with the unmixing identification results of multi-particle overlapping events, and cumulative counting is performed within the monitoring window according to material type and particle size range. By combining the sampling flow rate, the particle concentration of various microplastics in a unit volume of seawater is calculated, generating monitoring data on the classification and concentration of microplastic materials in marine water.

8. A marine microplastic particle monitoring system, used to execute the marine microplastic particle monitoring method according to any one of claims 1 to 7, characterized in that, include: The signal synchronization acquisition module is used to acquire the photoacoustic signal, stimulated Raman loss signal and forward scattered light pulse signal generated by the particle group in the ocean water sampling flow after being irradiated by pulsed laser, and to perform synchronization alignment of the three signals according to a unified clock. The event window marking module is used to perform peak detection on the forward-scattered light pulse signal and mark the event window as a single-particle event window or a multi-particle overlapping event window based on the comparison between the time interval of adjacent pulses and the duration threshold. The photoacoustic signal separation module is used to perform sparse deconvolution on the photoacoustic signals within the multi-particle overlapping event window, using the scattering event timestamp as a priori constraint, to obtain the separated photoacoustic amplitude estimate and separation delay of each particle. The physical parameter extraction module is used to extract the equivalent particle size of each particle based on the estimated value of the separated photoacoustic amplitude, and to calculate the volume percentage weight vector of each particle. The cross-modal constrained unmixing module is used to take the number of particles derived from photoacoustics as the component hard constraint of non-negative matrix decomposition, and take the volume proportion weight vector after Raman scattering cross section conversion as the initial value of abundance coefficient and relaxation range constraint. It performs constrained non-negative matrix decomposition on the overlapping stimulated Raman spectrum to obtain the abundance coefficient and candidate material identifier of each component.

Images