A mixed suspended particle density detection method of multi-spectral characteristic peak joint inversion

Abstract

The application discloses a mixed suspended particle density detection method of multi-spectrum characteristic peak joint inversion, constructs a four-wavelength detection system by using 430nm, 480nm, 680nm absorption characteristic peaks and a 730nm near-infrared reference waveband; scattering correction is completed by using a near-infrared waveband, and a multi-dimensional physical ratio characteristic is constructed; the particle growth / aggregation state is identified based on the characteristic ratio; a spectral physical constraint is embedded into MEP symbolic regression to generate an explicit density calculation formula; finally, the output result is checked by a spectral physical rule; the application does not need to transform the spectral hardware, the model has a clear physical expression, the determination coefficient R 2 of three-fold cross-validation of single-component and multi-component density inversion can reach 0.99 at most, the frequent calibration requirement can be reduced, and the application is suitable for online nondestructive density detection of microalgae and other microbial suspended particles with pigment absorption characteristics.

Description

Technical Field

[0001] This invention belongs to the field of quantitative spectral detection technology, specifically relating to a method for detecting the density of mixed suspended particles by joint inversion of multispectral characteristic peaks. Background Technology

[0002] Suspended particle density is a core quantitative indicator for bio-fermentation, aquaculture, industrial fluid processing, and water environment monitoring. Currently, the mainstream optical quantitative methods are single-wavelength optical density method, multi-wavelength PLS machine learning modeling method, and hyperspectral imaging fitting detection method. The above-mentioned existing technologies have core technical defects: in mixed suspended particle systems, particle Mie scattering and multiple scattering effects will cause transmission light intensity attenuation and distortion, and the absorption spectra of multiple components will overlap. Existing data-driven fitting models have no clear physical expression and lack interpretability. When the particle growth cycle, degree of aggregation, and component ratio change, the model's generalization ability will drop significantly, and full sample recalibration must be carried out. It is impossible to achieve high-precision non-destructive density detection of multiple components simultaneously, stably, and without frequent calibration.

[0003] Existing single-wavelength optical density methods rely entirely on the linear assumption of Beer-Lambert's law. In high-concentration, high-scattering systems, multiple scattering deviates significantly from the linear relationship, and the loss of scattered light intensity is misjudged as absorption attenuation. The absolute error can reach as high as 0.15 normalized density, making them only suitable for ideal samples with low concentration, single component, and no particle aggregation. They are completely ineffective in mixed multi-component scenarios. While machine learning models such as multi-wavelength partial least squares (PLS), random forests, and neural networks can fit nonlinear spectral-density mapping relationships, their internal computational logic is a black box structure. There are no derivable and verifiable physical calculation formulas, and they cannot relate to the fundamental spectral laws of scattering and absorption. When particles enter different growth stages, particle size distribution changes, or the proportion of mixed components is adjusted, the prediction error of the original model increases significantly. Dozens of new calibration samples must be collected to retrain the model, resulting in an extremely large calibration workload in batch continuous online monitoring scenarios.

[0004] For microalgae and microorganism mixed suspension systems, the pigment absorption bands of different biological particles highly overlap, and cross scattering between particles further distorts the characteristic spectra of each component. Existing technologies cannot separate the independent spectral contributions of each component and can only output the total density of the mixed system, but cannot simultaneously quantify the individual particle concentrations of multiple components. At the same time, existing technologies do not distinguish the differences in spectral characteristics of particles during the lag phase, logarithmic phase, stationary phase, and decay phase. The unified model cannot be adapted to the detection of the entire growth cycle, which further aggravates model drift and requires periodic recalibration.

[0005] In addition, existing spectral detection algorithms all require dedicated modified spectral acquisition hardware, or read spectral data through driver interception and hardware modification. Commercial standardized spectrometers cannot be directly compatible with the algorithm process, resulting in high equipment modification costs and deployment difficulties. Furthermore, the existing detection process relies on manual step-by-step operations, with filtering, peak extraction, correction, inversion, and verification performed in segments. Manual operation introduces random errors, leading to poor consistency in batch sample detection.

[0006] In summary, existing spectral suspended particle density detection technologies suffer from core technical bottlenecks such as the inability to quantitatively compensate for scattering interference, the inability to decouple the spectral contributions of multiple components, the lack of physical interpretability of the model, and the need for frequent recalibration after changes in particle state. The industry needs to design an automated density detection method that relies on the physical laws of spectroscopy, can quantitatively correct scattering, distinguish particle growth states, generate explicit physical inversion formulas, and is adaptable to single / multi-component mixed particles. Summary of the Invention

[0007] This invention addresses the core technical problems of existing spectral detection methods, such as the inability to quantitatively compensate for multiple scattering interference, the difficulty in separating overlapping spectra of multi-component mixed particles, the lack of physical interpretability of data-driven models, and the need for frequent recalibration due to model drift after changes in particle growth state. It provides a method for detecting the density of mixed suspended particles by joint inversion of multispectral characteristic peaks, enabling non-invasive, automated, high-precision, and calibration-free simultaneous quantitative detection of the density of single-component, two-component, and multi-component mixed suspended particles.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] A method for detecting the density of mixed suspended particles by joint inversion of multispectral characteristic peaks includes the following steps:

[0010] Step 1, Spectral Preprocessing and Characteristic Peak Extraction: The host computer reads the raw spectral data of suspended particles in CSV format exported from the spectrometer. Noise reduction is achieved using a Savitzky-Golay filter with a window of 5 data points and a polynomial order of 2. Four characteristic wavelengths—430nm, 480nm, 680nm, and 730nm—are fixed, and local transmission peaks are retrieved within a wavelength range of ±20nm for each characteristic wavelength. , , , The original transmitted light intensities extracted from the locations are respectively , , , ;

[0011] Step 2, Quantitative compensation for near-infrared reference peak scattering: Using 730nm as the near-infrared reference band where pigment absorption response is relatively weak, the correction coefficients β1, β2, β3, and β4 are dynamically calculated according to a preset formula; through... , , Calculate the corrected transmitted light intensity at 430nm, 480nm, and 680nm; define the near-infrared corrected transmission reference signal. ;

[0012] Step 3, Calculation of multidimensional physical ratio characteristics: based on , , , Calculate the four types of ratio characteristics: / , / , / , / Set a minimum threshold for the denominator; when the denominator is below the threshold, mark the sample as abnormal and replace it with a stable value.

[0013] Step 4, Threshold determination of particle growth state: based on features Based on the preset algal species multi-level threshold, the particles are divided into the lag phase, logarithmic growth phase, stationary phase, and death phase, and the output state discrimination label is used as an auxiliary input for density inversion.

[0014] Step 5, density inversion based on the physical constraint MEP model: Density normalization mapping is completed using blank samples and baseline density standard samples. Three physical rules—non-negative density, attenuation of transmitted light intensity with increasing density, and overall attenuation of near-infrared transmitted reference signal with increasing density—are used as constraint evolution multivariate expressions to program the inversion equation. The relative density is calculated using the corresponding piecewise explicit equations based on the sample algal species and the state labels output from Step 4. The true absolute density of the particles is restored using a bilinear interpolation mapping formula. For multi-component mixed samples, a multi-output joint MEP equation is constructed to simultaneously calculate the density of each component.

[0015] Step 6, Physical Consistency Verification and Automatic Output: Perform four-layer verification on the true density output in Step 5: density value non-negation verification; value falling within a preset detection range verification; if density increases... , , Verification of a monotonically decreasing trend; if density increases... Overall downward trend verification; samples that fail verification are marked as abnormal and prompted for retesting; if verification passes, density results are stored; the host computer monitors the spectrometer output directory, and automatically loops through steps 1 to 6 after detecting a new spectral CSV file, batch exporting test reports in PDF, Excel, and CSV formats.

[0016] In a preferred embodiment of the present invention, in step 2, the formula for calculating the correction coefficient is:

[0017]

[0018]

[0019]

[0020]

[0021] in, The value range is 0.02 ± 0.001. The value range is 0.15 ± 0.003. The value range is 0.01 ± 0.001.

[0022] In a preferred embodiment of the present invention, in step 4, the threshold discrimination rule for *Streptococcus* is as follows: ≥1.1 Determination of delay period, 1.1> A value ≥0.9 indicates the logarithmic growth phase; 0.9 > A value ≥0.7 indicates a stable period; 0.7 > Determining the death phase; the threshold discrimination rule for *Microcystis aeruginosa* is as follows: ≥0.7 indicates a delay period; 0.7> A value ≥0.6 indicates the logarithmic growth phase; 0.6 > A value ≥0.55 indicates a stable period; 0.55 > Determine the death period.

[0023] In a preferred embodiment of the present invention, in step 5, the relative density of *Streptococcus* is... The calculation formula is:

[0024]

[0025]

[0026] In a preferred embodiment of the present invention, in step 5, the relative density of *Microcystis aeruginosa* is... The calculation formula is

[0027]

[0028] in , for:

[0029]

[0030]

[0031] In a preferred embodiment of the present invention, in step 5, the formula for the true density of the bilinear interpolation mapping is:

[0032]

[0033] Among them, the age interpolation weight ω(t) and the relative density piecewise calibration function fA,d(ρ) are defined as follows:

[0034]

[0035]

[0036] Among them, D real (M,A,t,ρ) represents the true density obtained from the mapping; M is the pre-established calibration matrix; A represents the type of particle to be tested; t is the actual culture age of the sample; ρ is the relative density factor output by the MEP model; d i and d i+1 The adjacent calibrated age nodes surrounding t; q j and y j These are the j-th input node and density node of the piecewise calibration curve, respectively. When t or ρ exceeds the calibration range, boundary extrapolation should be performed and the out-of-range state should be marked.

[0037] In a preferred embodiment of the present invention, in step 1, the host computer reads data in a non-intrusive manner, without modifying the spectrometer hardware, driver, or communication program, but only reading the standard CSV file exported locally by the spectrometer, without intercepting the instrument's real-time communication data stream.

[0038] In a preferred embodiment of the present invention, step 6 includes three types of abnormal sample processing methods after verification: marking abnormalities, removing the current detection data, outputting the corrected estimated density, and adding a retest reminder in the report notes.

[0039] In a preferred embodiment of the present invention, the detection targets adapted to the method include microalgae and microbial suspended particles with pigment absorption characteristics; for different test species, the threshold in step 4 and the MEP inversion equation in step 5 are adjusted according to their pigment absorption band, culture conditions and calibration data.

[0040] A multispectral characteristic peak joint inversion mixed suspended particle density detection system for implementing the above detection method includes a light source unit, a sample carrier unit, and a spectral acquisition unit arranged sequentially along the transmission light propagation direction, as well as a data processing unit connected to the spectral acquisition unit. The sample carrier unit includes a three-dimensional adjustable base and a suspended particle sample to be tested. The light source unit includes a constant current driven LED backlight and an LED power supply. The spectral acquisition unit includes a transmission-type quartz fiber optic probe and a miniature fiber optic spectrometer. The data processing unit includes a data processing device equipped with the algorithm program described above, possessing functions such as file directory monitoring, automatic spectral analysis, feature calculation, state discrimination, MEP density inversion, physical verification, and automatic report export.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] 1. This invention uses the 730 nm near-infrared band as a reference band with relatively weak pigment-related absorption response and relatively obvious scattering response. It compensates for the scattering influence of absorption-related characteristic peaks such as 430 nm, 480 nm, and 680 nm, which can reduce the systematic deviation caused by suspended particle scattering on transmitted light intensity measurement. This makes the corrected spectral characteristics and particle density maintain a more stable correspondence within the preset detection range, thereby improving the accuracy and repeatability of density detection.

[0043] 2. This invention utilizes spectral information from multiple characteristic wavelengths for joint inversion. Different types of suspended particles have different absorption and scattering response characteristics at different wavelengths. By combining multiple wavelength features, the ability to distinguish the spectral contributions of different components in the mixed system can be enhanced, thereby improving the applicability of density inversion for single-component, two-component, and multi-component suspended particles.

[0044] 3. This invention incorporates the Beer-Lambert law, the relationship between scattering and absorption response, and the variation law between density, transmitted light intensity, and scattering response into a multivariate expression programming model, enabling the model to generate a density inversion formula with a clear mathematical form. Compared to methods that rely solely on black-box data fitting, the inversion process of this invention is clearer, and the model results are easier to interpret, verify, and deploy in engineering.

[0045] 4. This invention identifies the growth or aggregation state of suspended particles by analyzing changes in spectral feature ratios, and uses the state identification results as auxiliary information for density inversion. This reduces the impact of particle state changes on the inversion results, reduces the need for repeated calibration, and is suitable for continuous or batch detection scenarios.

[0046] 5. This invention can complete subsequent calculations by reading the standard data file output by the spectrometer through the host computer software, without modifying the spectrometer hardware structure or changing the original spectral acquisition process, making it easy to deploy and use on existing spectral detection equipment.

[0047] 6. This invention establishes an automated processing flow from data reading, filtering and denoising, feature peak extraction, scattering effect compensation, feature ratio calculation, state recognition, density inversion to result output, which can reduce the errors caused by manual step-by-step operation and improve the efficiency and consistency of batch sample processing.

[0048] 7. This invention is applicable to microalgae and other microbial suspended particles with pigment absorption characteristics. For different test species, the state discrimination threshold in step 4 and the MEP density inversion equation in step 5 can be adjusted according to their pigment absorption bands, culture conditions, and calibration data, thereby adapting to the spectral response characteristics of different biological suspended particles. It can be used in laboratory analysis, aquatic biological monitoring, and culture process monitoring.

[0049] 8. The present invention sets up a result consistency verification mechanism before outputting density results, which can judge the non-negativity of density, numerical range, transmission light intensity change trend and scattering response change trend; for results that obviously do not conform to the preset spectral change law, abnormal marking, rejection, prompting retesting or correction can be performed, thereby improving the reliability of output results.

[0050] 9. This invention simultaneously constructs multiple features such as the corrected transmission intensity ratio and the absorption-scattering combined ratio, combining the information required for particle component identification, state judgment and density estimation. This reduces the risk of single features being affected by noise, baseline drift or changes in sample state, and improves the stability of the inversion model in complex suspension systems. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments or prior art, the drawings used in the description of the embodiments or prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 The flowchart illustrates a method for detecting the density of mixed suspended particles by joint inversion of multispectral characteristic peaks, as provided in this embodiment of the invention.

[0053] Figure 2 This is a schematic diagram of a multispectral characteristic peak joint inversion system for detecting the density of mixed suspended particles, provided in an embodiment of the present invention.

[0054] Figure 3 The normalized transmission light intensity spectrum curves of *Streptococcus* at different concentration gradients are provided in the embodiments of the present invention.

[0055] Figure 4The normalized transmission light intensity spectrum curves of *Microcystis aeruginosa* at different concentration gradients are provided in the embodiments of the present invention.

[0056] Figure 5 The diagram shows the original light intensity variations at four characteristic wavelengths under different normalized density gradients of *Streptococcus*, as provided in this embodiment of the invention.

[0057] Figure 6 This is a comparison chart of the measured and predicted normalized density of *Streptococcus* provided in an embodiment of the present invention. Each predicted density in the chart is a cross-validation out-of-bounds prediction result obtained when the sample did not participate in the training of the corresponding model.

[0058] Figure 7 The absolute error distribution diagram of the three-round repeated experiment of *Streptococcus* provided in the embodiment of the present invention is shown. The absolute error in the diagram is calculated from the absolute value of the difference between the predicted density outside the cross-validation and the corresponding measured density.

[0059] Figure 8 The diagram shows the original light intensity changes at four characteristic wavelengths under different normalized density gradients of *Microcystis aeruginosa*, as provided in an embodiment of the present invention.

[0060] Figure 9 This is a comparison chart of the measured and predicted normalized densities of *Microcystis aeruginosa* provided in an embodiment of the present invention. Each predicted density in the chart is a cross-validation out-of-bounds prediction result obtained when the sample did not participate in the training of the corresponding model.

[0061] Figure 10 The absolute error distribution diagram of the three-round repeated experiment of *Microcystis aeruginosa* provided in the embodiment of the present invention is shown. The absolute error in the diagram is calculated from the absolute value of the difference between the predicted density outside the cross-validation and the corresponding measured density.

[0062] Figure 11 This is a comparison chart of the measured and predicted normalized relative densities of a mixed algae sample provided in an embodiment of the present invention. Figure 11 (a) is a comparison chart of the measured and predicted normalized relative density of *Microcystis aeruginosa*. Figure 11 (b) is a comparison chart of the measured and predicted normalized relative density of Streptococcus.

[0063] Figure 12 This is an absolute error distribution diagram of three repeated experiments on a mixed algae sample provided in an embodiment of the present invention, wherein... Figure 12 (a) is a graph showing the absolute error distribution of the three repeated experiments on *Microcystis aeruginosa*. Figure 12 (b) is a graph showing the absolute error distribution of the three repeated experiments on Streptococcus pyogenes.

[0064] Figure 13 This is a comparison chart of the measured and predicted normalized relative densities of a mixed sample of three algae provided in an embodiment of the present invention. Figure 13(a) is a comparison chart of the measured and predicted normalized relative density of *Microcystis aeruginosa*. Figure 13 (b) is a comparison chart of the measured and predicted normalized relative density values ​​of golden algae. Figure 13 (c) is a comparison chart of the measured and predicted normalized relative density of Streptococcus.

[0065] Figure 14 This is a diagram showing the absolute error distribution of a three-round repeated experiment on a mixed algae sample provided in an embodiment of the present invention. Figure 14 (a) is a graph showing the absolute error distribution of the three repeated experiments on *Microcystis aeruginosa*. Figure 14 (b) is a graph showing the distribution of absolute error in the three-round repeated experiments on golden algae. Figure 14 (c) is a graph showing the distribution of absolute error in the three-round repeated experiments of Streptococcus pyogenes. Detailed Implementation

[0066] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application. In the accompanying drawings of this embodiment, directions or orientations are marked with letters such as a, b, and c. These letter markings are only used to distinguish different components and describe their relative positional relationships, and do not represent the installation orientation of the device or functional components in actual use.

[0067] like Figure 1 As shown, this invention provides a method for detecting the density of mixed suspended particles by joint inversion of multispectral characteristic peaks. This method addresses technical problems in the spectral detection of mixed suspended particles, such as systematic bias caused by multiple scattering, indistinguishable overlapping of multi-component spectra, lack of physical interpretability of the model, and the need for frequent calibration after particle state changes. The method includes the following steps:

[0068] Step 1, Spectral Preprocessing and Characteristic Peak Extraction: The host computer reads the raw spectral data of suspended particles in CSV format exported from the spectrometer. Noise reduction is achieved using a Savitzky-Golay filter with a window of 5 data points and a polynomial order of 2. Four characteristic wavelengths—430nm, 480nm, 680nm, and 730nm—are fixed, and local transmission peaks are retrieved within a wavelength range of ±20nm for each characteristic wavelength. , , , The original transmitted light intensities extracted from the locations are respectively , , , ; , , , Together, they form a four-wavelength raw light intensity dataset, which serves as the basic input for subsequent scattering effect compensation, spectral feature construction, and density inversion.

[0069] Step 2, Quantitative compensation for near-infrared reference peak scattering: Using 730nm as the near-infrared reference band where pigment absorption response is relatively weak, the correction coefficients β1, β2, β3, and β4 are dynamically calculated according to a preset formula; through... , , Calculate the corrected transmitted light intensity at 430nm, 480nm, and 680nm; define the near-infrared corrected transmission reference signal. ;

[0070] Step 3, Calculation of multidimensional physical ratio characteristics: based on , , , Calculate the four types of ratio characteristics: / , / , / , / Set a minimum threshold for the denominator; when the denominator is below the threshold, mark the sample as abnormal and replace it with a stable value.

[0071] Step 4, Threshold determination of particle growth state: based on features Based on the preset algal species multi-level threshold, the particles are divided into the lag phase, logarithmic growth phase, stationary phase, and death phase, and the output state discrimination label is used as an auxiliary input for density inversion.

[0072] Step 5, density inversion based on the physical constraint MEP model: Density normalization mapping is completed using blank samples and baseline density standard samples. Three physical rules—non-negative density, attenuation of transmitted light intensity with increasing density, and overall attenuation of near-infrared transmitted reference signal with increasing density—are used as constraint evolution multivariate expressions to program the inversion equation. The relative density is calculated using the corresponding piecewise explicit equations based on the sample algal species and the state labels output from Step 4. The true absolute density of the particles is restored using a bilinear interpolation mapping formula. For multi-component mixed samples, a multi-output joint MEP equation is constructed to simultaneously calculate the density of each component.

[0073] Step 6, Physical Consistency Verification and Automatic Output: Perform four-layer verification on the true density output in Step 5: density value non-negation verification; value falling within a preset detection range verification; if density increases... , , Verification of a monotonically decreasing trend; if density increases... Overall downward trend verification; samples that fail verification are marked as abnormal and prompted for retesting; if verification passes, density results are stored; the host computer monitors the spectrometer output directory, and automatically loops through steps 1 to 6 after detecting a new spectral CSV file, batch exporting test reports in PDF, Excel, and CSV formats.

[0074] The specific steps are detailed below:

[0075] In step 1, the host computer reads data in a non-intrusive manner, without modifying the spectrometer hardware, drivers, or communication programs. It only reads the standard CSV file exported locally by the spectrometer and does not intercept the instrument's real-time communication data stream.

[0076] This invention, through multiple sets of comparative experiments, determined that the SG filter window size is 5 data points and the polynomial order is 2. If the filter window is too large or the polynomial order is too low, it can easily cause loss of spectral peak details; if the filter window is too small or the polynomial order is too high, the suppression effect on random noise is insufficient. After smoothing, preferably, the system uses... , , , The target characteristic wavelength, where 430 nm 480 nm It is 680 nm. The wavelength is 730 nm, and representative values ​​of local transmitted light intensity are extracted within ±20 nm of each target wavelength as the original transmitted light intensity signal for the corresponding characteristic wavelength. By searching for local peaks in the neighborhood of the target wavelength, instead of directly reading the light intensity value of a single wavelength point, the influence of slight spectral peak shifts and interference from adjacent bands on the results can be reduced, improving the stability and consistency of feature extraction.

[0077] In step 2, the compensation method for the four-peak scattering effect based on the near-infrared reference band is as follows. Suspended particulate samples, especially microalgae, microorganisms, and suspensions containing solids, usually exhibit both absorption and scattering effects. Mie scattering and multiple scattering generated by particles weaken the intensity of transmitted light received by the detector, making it easy for traditional spectroscopic methods to misinterpret the intensity loss caused by scattering as a change in absorption, thus causing density inversion bias.

[0078] In step 2, the 730 nm near-infrared band is used as a reference band with weak transmission response and significant scattering response. Combined with a preset scattering correction coefficient, the scattering influence of absorption-related characteristic peaks at 430 nm, 480 nm, and 680 nm is compensated to reduce the systematic bias caused by suspended particle scattering in the transmitted light intensity measurement. Selecting... , , , A scattering correction system is constructed using four characteristic wavelengths, and the scattering correction coefficients are determined based on the relative light intensity relationships between the four bands. Let... , , , They are respectively , , , If the original transmitted light intensity is at a certain point, then the correction coefficients β1 to β4 can be determined by the following formula:

[0079]

[0080]

[0081]

[0082]

[0083] in, Located in the blue-violet light transmission sensitive range, it mainly reflects the transmission attenuation characteristics caused by chlorophyll a and auxiliary pigments; Corresponding to the blue region auxiliary pigment-related wavelength band; This is the main response band of chlorophyll a and an important absorption correlation band for density inversion; Located in the near-infrared region, pigment-related transmission response is weak, and it is mainly used to characterize reference scattering levels. The scattering correction coefficients obtained under different conditions are allowed to fluctuate due to factors such as experimental environment, light source status, culture medium batch, sample container, and optical path conditions. The value range is 0.02 ± 0.001. The value range is 0.15 ± 0.003. The value range is 0.01 ± 0.001.

[0084] After the above processing, the influence of scattering on the absorption-related transmission signal can be reduced within the preset detection range, so that the corrected spectral features and the density of suspended particles maintain a more stable correspondence, providing more reliable input data for subsequent density inversion.

[0085] In step 3, the multidimensional spectral feature extraction method is as follows. Based on the corrected transmitted light intensity and near-infrared corrected transmitted light reference signal obtained in step 2, this invention constructs a multidimensional spectral feature set composed of the corrected transmitted light intensity ratio, the absorption-near-infrared reference joint ratio, and the original band ratio. This feature set retains, on the one hand, the spectral physical meaning related to particle absorption, scattering, and state changes, and on the other hand, retains some of the original transmitted light intensity ratio features to improve the fitting ability of the inversion model to complex nonlinear relationships.

[0086] Step 3: Based on the scattering-corrected transmitted light intensity signal and the near-infrared corrected transmitted light reference signal, calculate multi-dimensional spectral features such as the corrected transmitted light intensity ratio feature and the absorption / near-infrared reference ratio feature to form a feature set for characterizing the composition, concentration and state changes of suspended particles.

[0087] Multidimensional spectral features include the following four types of ratio features:

[0088] First, and The corrected transmitted light intensity ratio between them: / This ratio reflects the relative transmission change between the 680 nm and 480 nm wavelength bands, and is used to characterize changes in pigment composition and particle growth state.

[0089] Second, the ratio between the absorbed correlated transmitted light intensity and the near-infrared corrected transmitted reference signal: / This ratio reflects Corrected transmitted light intensity and The relative relationship between near-infrared corrected transmission reference signals can be used to characterize particle density variations and scattering effects.

[0090] third, and The corrected transmitted light intensity ratio between them: / This ratio is used to characterize the relative spectral response difference between the blue-violet band and the red band, and can help reflect particle composition, pigment state or light adaptation state.

[0091] fourth, and The original transmitted light intensity ratio between them: / This ratio preserves the original band relationships before scattering compensation, and is used to capture nonlinear information in spectral data, providing supplementary input for multivariate expression programming models.

[0092] By constructing the aforementioned features, this invention combines the corrected spectral features with the original light intensity ratio features, preserving the physical meaning of the spectral detection process while improving the model's adaptability to multi-component, strongly scattering, or nonlinear sample systems. To avoid numerical anomalies caused by excessively small denominators, the system can set a minimum threshold. When the denominator is less than the preset threshold, the sample is marked as an anomaly or a preset numerical stabilization method is applied. Furthermore, a numerical protection threshold ε>0 is set: the independent variables of the logarithm and square root should be positive, and the absolute value of the divisor should not be less than ε; when the absolute value of V1 is less than ε, or any operation does not meet the domain condition, the sample is marked as an anomaly and a retest is prompted, without directly outputting the inversion density.

[0093] In step 4, the method for identifying the growth or aggregation state of suspended particles is as follows: For biological suspended particles such as microalgae or microorganisms, the number of particles, cell morphology, pigment composition, and degree of aggregation will change at different growth stages. These changes will be reflected in the corrected spectral features and their ratio changes. Therefore, this invention identifies the growth or aggregation state of the sample based on the feature ratios obtained in step 3, and uses the state identification results as auxiliary input for subsequent density inversion. Taking microalgae samples as an example, the microalgae growth cycle can be divided into a lag phase, a logarithmic growth phase, a stationary phase, and a death phase. During the lag phase, the ratios of various spectral features change relatively little and are generally stable; during the logarithmic growth phase, cells proliferate rapidly, and the ratios of pigment composition and transmitted light intensity change significantly; during the stationary phase, the growth of cell number slows down, and scattering-related features and absorption-related features tend to stabilize; during the death phase, the number of cells decreases or cell morphology changes, and the corresponding feature ratios decline or shift. Based on the above patterns, this invention establishes corresponding threshold discrimination rules for different algal species to avoid identification errors caused by using a uniform threshold.

[0094] Step 4: Based on the variation law of spectral feature ratio, identify the growth state or aggregation state of suspended particles, and use the state identification results as auxiliary information for subsequent density inversion, so as to reduce the impact of particle state changes on the inversion results and reduce the need for repeated calibration.

[0095] In step 4, the threshold discrimination rule for *Streptococcus* is as follows: ≥1.1 Determination of delay period, 1.1> A value ≥0.9 indicates the logarithmic growth phase; 0.9 > A value ≥0.7 indicates a stable period; 0.7 > Determining the death phase; the threshold discrimination rule for *Microcystis aeruginosa* is as follows: ≥0.7 indicates a delay period; 0.7> A value ≥0.6 indicates the logarithmic growth phase; 0.6 > A value ≥0.55 indicates a stable period; 0.55 > Determine the death period.

[0096] The aforementioned thresholds can be adjusted based on different sample types, culture conditions, detection optical paths, and calibration data. The state recognition results are used to assist in subsequent model selection, feature weight adjustment, or inversion formula invocation, thereby reducing the impact of sample state changes on density inversion results and minimizing the need for repeated calibration.

[0097] In step 5, the density inversion and result verification method based on multivariate expression programming is as follows.

[0098] This invention employs a multivariate expression programming method to construct a density inversion model. Multivariate expression programming, a symbolic regression method, can automatically generate explicit mathematical expressions based on input features. Compared to black-box models such as neural networks and random forests, the inversion formulas generated by multivariate expression programming are clearly defined, facilitating deployment in host computer software or embedded devices, and also making it easier to check the model output by incorporating spectral variation patterns.

[0099] During the model establishment and evolution phase, considering that the absolute value of microalgal cell density in actual production is usually around 10... 9 If the absolute density value, on the order of cells per liter, is directly introduced as the dependent variable into the nonlinear model evolution process, the large numerical range may reduce computational stability and increase the risk of numerical overflow, accuracy loss, or model convergence difficulties during power, exponential, or composite function operations. Therefore, this invention introduces a baseline proportional mapping mechanism during model construction: using standard samples with known densities as calibration data, the preset baseline density samples in the calibration system are mapped to a relative density factor of 1, the blank control samples are mapped to a relative density factor of 0, and the densities of other samples are mapped to the corresponding normalized numerical ranges according to their proportional relationship with the baseline density. Through this processing, the original high-order-of-magnitude density data can be converted into relative density data that is convenient for model calculation and evolution, thereby improving numerical computational stability, reducing model complexity, and facilitating the conversion between subsequent density inversion results and actual cell densities.

[0100] Corrected transmitted light intensity , , Near-infrared corrected transmission reference signal eigenratio , , , Using the state recognition result as input and the actual sample density as output, the density inversion formulas corresponding to different sample types are obtained through multivariate expression programming evolution. During model construction, spectral variation rules can be introduced as constraints; for example, density variation should match the attenuation trend of absorption-related transmitted light intensity, density variation should be coordinated with the attenuation trend of near-infrared transmitted reference signal, and the output density should be non-negative. In actual continuous detection, when the actual sample density exceeds the calibration upper limit, the calculated relative density result is allowed to exceed the relative density limit of 1.

[0101] Step 5: Call the density inversion formula constructed based on spectral physics rules and multivariate expression programming (MEP) method to perform density inversion on single-component, two-component, and multi-component mixed suspended particles to obtain the density calculation results of each component or the overall suspended particles.

[0102] For a mixed sample containing C known target components, the corrected light intensity, physical ratio characteristics, and state labels are combined to form a unified input vector z:

[0103]

[0104] A multi-output MEP consists of C explicit expressions that share the same input vector and output the relative density of each component, respectively. Its output vector is defined as:

[0105]

[0106] The model evolution uses the weighted sum of prediction errors and physical consistency penalties for each component as the overall fitness:

[0107]

[0108] Where C represents the number of target components, MSEc is the mean square error of the c-th component, Pc represents the nonnegativity, transmission attenuation trend, and domain anomaly penalty of the c-th component, wc is the component weight, and λ is the physical penalty weight. For a two-algae mixture sample, C=2 is used, outputting the relative densities of *Microcystis aeruginosa* and *Streptococcus*; for a three-algae mixture sample, C=3 is used, outputting the relative densities of *Microcystis aeruginosa*, *Golden Algae*, and *Streptococcus*. Each output component is restored to the true density of its corresponding component through the same calibration and mapping process.

[0109] In step 5, the relative density of Streptococcus is... The calculation formula is:

[0110]

[0111]

[0112] In step 5, the relative density of *Microcystis globulus* The calculation formula is

[0113]

[0114] in , for:

[0115]

[0116]

[0117] In step 5, the formula for the true density mapped by bilinear interpolation is:

[0118]

[0119] Among them, the age interpolation weight ω(t) and the relative density piecewise calibration function fA,d(ρ) are defined as follows:

[0120]

[0121]

[0122] Among them, D real (M,A,t,ρ) represents the true density obtained from the mapping; M is the pre-established calibration matrix; A represents the type of particle to be tested; t is the actual culture age of the sample; ρ is the relative density factor output by the MEP model; d i and d i+1 The adjacent calibrated age nodes surrounding t; q j and y j These are the j-th input node and density node of the piecewise calibration curve, respectively. When t or ρ exceeds the calibration range, boundary extrapolation should be performed and the out-of-range state should be marked.

[0123] To improve the reliability of the output results, this invention includes a consistency check after density inversion. The checks include nonnegativity verification, numerical range verification, transmission intensity variation trend verification, and scattering response variation trend verification. Specifically, within a preset applicable range, if particle density increases, the absorbed transmission intensity generally shows a decreasing trend; when the particle dispersion is relatively stable, an increase in the number of particles usually causes an overall attenuation of the near-infrared corrected transmission reference signal. For samples that meet the above rules, the system retains the model output results; for samples that clearly do not meet the above rules, the system marks them as abnormal, removes them, prompts for retesting, or processes them using preset correction rules. This verification process does not change the calculation results of normal samples and is mainly used to intercept obviously abnormal outputs.

[0124] Step 6 involves performing a physical consistency check on the inversion results. Within a preset applicable range, it is determined whether the density results satisfy basic laws such as changes in transmitted light intensity, changes in scattering response, and non-negativity. Abnormal results are marked, removed, or corrected, and the final density detection result is output. Step 6 includes three methods for handling abnormal samples after verification: marking abnormalities, removing the current detection data, and outputting a corrected estimated density, with a retest reminder noted in the report. The method is suitable for detecting microalgae and microbial suspended particles with pigment absorption characteristics. For different test species, the threshold in Step 4 and the MEP inversion equation in Step 5 are adjusted according to their pigment absorption bands, culture conditions, and calibration data.

[0125] This invention enables non-intrusive reading and automated processing of spectrometer output data through host computer software. Non-intrusive reading means that it does not modify the spectrometer's original hardware, driver program, or intercept instrument communication data; instead, it completes subsequent calculations by reading the standard CSV file exported by the spectrometer. This method does not affect the spectrometer's original acquisition process or certification status, making it easy to deploy on existing commercial spectrometers.

[0126] To meet the needs of continuous and batch detection, the host computer software can set the spectrometer data output directory as the monitoring directory and use an operating system-level file system monitoring mechanism to detect newly added CSV files. When a new file is detected and the file writing is confirmed to be complete, the software automatically reads the file and sequentially executes steps 1 to 5: spectral preprocessing, characteristic peak extraction, scattering effect compensation, characteristic ratio calculation, state identification, density inversion, and result verification.

[0127] The software interface displays information such as the original spectral curve, target peak, scattering compensation results, characteristic ratio, growth or aggregation state, density inversion results, and verification status. All intermediate results are correlated with the original spectral file and automatically saved to a log file or results folder. The system can export test reports in CSV, Excel, or PDF formats according to user settings to support batch sample processing, historical data tracing, and subsequent statistical analysis.

[0128] Through the above six steps, this invention achieves an automated processing flow from spectral data reading, target peak extraction, scattering effect compensation, multi-dimensional feature construction, state recognition, density inversion to result output without changing the spectrometer hardware structure. It can be used for non-invasive density detection of single-component, two-component, and multi-component mixed suspended particles.

[0129] Example results:

[0130] Cross-validation method: To evaluate the density inversion capability of this invention under small sample conditions, three-fold cross-validation was used for single-component, two-algae mixture, and three-algae mixture experiments. For each type of experiment, 11 concentrations or mixing ratios were set, and each sample underwent three rounds of independent spectral acquisition, resulting in a total of N=33 observation data. The 33 observation data were divided into K=3 non-overlapping subsets S1, S2, and S3 according to the acquisition rounds, and each subset contained 11 observation data obtained from the corresponding round of acquisition.

[0131] The three-fold cross-validation process is as follows: Fold 1 uses the 22 observations obtained by merging S2 and S3 as the training set and the 11 observations from S1 as the validation set; Fold 2 uses S1 and S3 as the training set and S2 as the validation set; Fold 3 uses S1 and S2 as the training set and S3 as the validation set. Within each fold, the spectral feature processing rules remain consistent. The evolution of the MEP explicit expression, parameter determination, and model selection only use the training set of that fold. The validation set does not participate in the model fitting for that fold, and the density inversion is performed on the 11 observations in the validation set by the model of that fold to obtain the out-of-fold predicted values.

[0132] After completing the three-fold validation, the out-of-fold predicted values ​​obtained from the three validation sets are merged according to the correspondence between the original samples and the collection rounds to form a cross-validation prediction result covering all 33 observation data. The integrated measured density and out-of-fold predicted density are then subjected to unified statistical processing. Figure 6 , Figure 7 , Figures 9 to 14 The predicted density and coefficient of determination R are shown. 2 Both the mean absolute error and the average absolute error are calculated based on the 33 outlier prediction results, without using the direct fitting results of the model to the same observation data in the training set. For multi-component samples, evaluation indicators are calculated separately for each target component.

[0133] Coefficient of determination R 2 The mean absolute error (MAE) is calculated using the following formula:

[0134]

[0135]

[0136] In this embodiment, N=33, K=3; Di is the measured density of the component corresponding to the i-th observation data, D̂i is the out-of-fold prediction density obtained when the observation data does not participate in the corresponding model training, and D̄ is the average value of the measured densities of the 33 observation data.

[0137] To verify the applicability of the method of this invention to different algal species and samples of different concentrations, this embodiment selected *Microcystis aeruginosa* and *Streptococcus aureus* samples, and prepared samples of different concentrations according to the density gradients shown in Table 1. Subsequently, spectral acquisition, characteristic peak extraction, scattering effect compensation, characteristic ratio calculation, and density inversion were performed on each concentration sample. The density unit in the table is cells / liter, and each density gradient is used to correspond to subsequent... Figures 5 to 10 The results of the normalized density experiment.

[0138]

[0139] Figure 3 The normalized transmission light intensity spectra of *Streptococcus* at different concentration gradients are shown. Figure 3It is evident that the spectral responses in the culture medium and at different concentrations of *Streptococcus* samples are mainly concentrated in the 400–750 nm range, exhibiting distinct characteristic responses in the blue-violet, green-to-yellow, and near-red light regions. With changes in sample concentration, the overall intensity and peak shape of the spectral curves change accordingly, but the characteristic peak distribution remains relatively consistent across all concentrations, indicating that *Streptococcus* samples possess stable spectral response patterns across multiple characteristic wavelength bands. Specifically, the wavelengths around 430 nm and 480 nm reflect absorption changes in the blue-violet and blue light regions, while the wavelength around 680 nm corresponds to the chlorophyll-related red light response region. The pigment-related absorption influence around 730 nm is relatively weak and can be used as a reference band reflecting scattering changes. These results demonstrate that selecting 430 nm, 480 nm, 680 nm, and 730 nm as multispectral characteristic wavelengths can effectively characterize the spectral changes of *Streptococcus* samples at different concentrations.

[0140] Figure 4 The normalized transmission light intensity spectra of *Microcystis globulus* at different concentration gradients are shown. Figure 4 As can be seen, *Microcystis globulus* samples at different concentrations also exhibited obvious multi-peak spectral response characteristics, with the main response region located in the 400–750 nm range. The spectral curves of samples at different concentrations showed differences in peak intensity, peak width, and local response amplitude, indicating that changes in *Microcystis globulus* concentration caused regular changes in transmission spectral characteristics. Compared with *Streptococcus*, *Microcystis globulus* showed differences in peak shape and relative intensity distribution at certain wavelengths, indicating that different algal species have distinguishable spectral response characteristics. These results further demonstrate that constructing joint inversion features using multiple characteristic wavelengths can enhance the distinguishability between different algal species and samples at different concentrations, providing a foundation for subsequent scattering effect compensation, characteristic ratio calculation, growth state identification, and density inversion.

[0141] Figure 5 The results show the original light intensity variations of *Streptococcus* samples at four characteristic wavelengths (430 nm, 480 nm, 680 nm, and 730 nm) under different normalized density gradients. Figure 5As the normalized density of *Streptococcus* gradually increases from 0 to 1.0, the original light intensities at 430 nm, 480 nm, and 680 nm generally show a decreasing trend. This indicates that the increased concentration of *Streptococcus* enhances its absorption and scattering of visible light, leading to a decrease in transmitted light intensity. The intensity change is most pronounced at 480 nm, while the 430 nm and 680 nm bands also exhibit a decreasing trend with increasing density. The 730 nm band shows relatively low overall intensity and a smaller variation, indicating that it is less affected by pigment absorption and can be used as a scattering reference band for subsequent scattering effect compensation. These results demonstrate that the four characteristic wavelengths of 430 nm, 480 nm, 680 nm, and 730 nm can reflect the differences in spectral response caused by changes in *Streptococcus* density, providing experimental evidence for the joint inversion of multiple spectral characteristic peaks.

[0142] Figure 6 This figure compares the measured and model-predicted normalized density of *Streptococcus*. The horizontal axis represents the measured normalized density of *Streptococcus*, and the vertical axis represents the model-predicted normalized density. Three different symbols represent data from three repeated experiments, and the dashed line represents the ideal prediction line. Figure 6 As can be seen, the data points from the three rounds of experiments are generally distributed near the ideal prediction line, indicating a good consistency between the model's predictions and the measured density. The coefficient of determination R in the figure... 2 The value of 0.99 indicates that the Streptococcus faecium density detection model established based on multispectral characteristic peaks, scattering effect compensation, and MEP inversion formula has good fitting effect and prediction ability. The predicted densities in the figure are cross-validation out-of-place prediction results obtained when the samples did not participate in the corresponding model training.

[0143] Figure 7 The figure shows the absolute error distribution results for three rounds of experiments on *Streptococcus*. The figure also presents the box plots and corresponding scatter plots of the absolute errors for the first, second, and third rounds of experiments. Figure 7 As can be seen, the absolute errors of the three rounds of experiments are generally at a low level, with most sample errors concentrated in a small range, and the average absolute error is 0.0068. Although some samples have relatively large error points, the overall error distribution does not show a significant systematic shift, indicating that the method of the present invention has good repeatability and stability in the density inversion of *Streptococcus*. This result further verifies the feasibility of the feature extraction, scattering effect compensation, state recognition, and MEP density inversion method described in this invention for the detection of *Streptococcus* samples. The absolute error in the figure is calculated from the absolute value of the difference between the cross-validation out-of-fold predicted density and the corresponding measured density.

[0144] Figure 8The original light intensity variations of *Microcystis aeruginosa* samples at four characteristic wavelengths (430 nm, 480 nm, 680 nm, and 730 nm) under different normalized density gradients are shown. Figure 8 As the normalized density of *Microcystis globulus* gradually increases, the original light intensities at 430 nm, 480 nm, and 680 nm all show varying degrees of decrease, indicating that changes in *Microcystis globulus* concentration cause regular changes in the transmission response of visible light characteristic bands. The intensity attenuation at 480 nm is particularly significant, indicating that this band has a high response to changes in *Microcystis globulus* concentration. The 430 nm and 680 nm bands also show a decreasing trend with increasing density and can be used to supplement the characterization of pigment absorption and particle concentration changes. The 730 nm band shows a smaller overall change and is relatively less affected by pigment absorption, and can be used as a scattering reference band for subsequent correction and feature construction. These results demonstrate that the four selected characteristic wavelengths can reflect the differences in the spectral response of *Microcystis globulus* samples under different concentration conditions.

[0145] Figure 9 This figure shows a comparison between the measured and model-predicted normalized density of *Microcystis aeruginosa*. Different symbols represent data from three repeated experiments, and the dashed line represents the ideal prediction line. Figure 9 As can be seen, the data points from all three rounds of experiments are distributed near the ideal prediction line, indicating a good consistency between the model's predicted density and the measured density. (Coefficient of Determination R0) 2 The value of 0.99 indicates that the density detection model for *Microcystis aeruginosa*, established based on multispectral characteristic peaks, scattering effect compensation, and MEP inversion formula, can accurately reflect sample density changes. The predicted densities in the figure are cross-validation out-of-place predictions obtained when the sample did not participate in the corresponding model training.

[0146] Figure 10 This presents the absolute error distribution results for three rounds of repeated experiments on *Microcystis aeruginosa*. Figure 10 As can be seen, the overall error of the three rounds of testing is at a low level, with most sample points concentrated within a small error range, and the average absolute error is 0.0031. Although there is some dispersion between different rounds, the overall error distribution does not show a significant shift, indicating that this method has good repeatability and stability when used for density inversion of *Microcystis aeruginosa*. Figure 8 and Figure 9 It can be seen that *Microcystis globulus* possesses a stable spectral response at the selected characteristic wavelengths that can be used for density inversion, and the established inversion model can effectively predict the normalized density of *Microcystis globulus*. The absolute error in the figure is calculated from the absolute value of the difference between the predicted density outside the cross-validation and the corresponding measured density.

[0147] Mixed samples were prepared using pre-defined unequal-proportion templates. Different gradient numbers corresponded to different formulation combinations. Sample analysis was based on the actual liquid volume given in the table, and gradient values ​​were not directly equated to concentrations or proportions. The concentration of the mother liquor for each algal species was calibrated before the experiment; when the final volume was fixed at 50, the true concentration of each component could be calculated from the mother liquor concentration and the actual liquid volume taken.

[0148] Table 2. Preparation scheme for mixed samples of two algae

[0149]

[0150] Note: The total volume of each mixed sample of the two algae is 50 mL; the actual liquid volume of the two algae is expressed in the order of Micrococcus: Streptococcus.

[0151] To verify the joint inversion capability of this invention for the density of each component under the condition of overlapping spectra of two suspended particles, this embodiment selected *Microcystis aeruginosa* and *Streptococcus aureus* to prepare a mixed sample of the two algae, as shown in Table 2. The total volume of each group of samples was fixed at 50 mL, and the proportion of the total volume of the two algal solutions to the total sample volume was set sequentially to 0%, 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90%, and 100%, with any portion less than 50 mL supplemented by culture medium. As the proportion of the total volume of the mixed algae increased, the volume of the culture medium was gradually reduced from 50 mL to 0 mL.

[0152] While varying the total volume ratio of the mixed algae, the composition ratios of *Microcystis aeruginosa* and *Streptococcus* were set in a non-uniform, non-monotonic manner. For example, when the total volume ratio of the mixed algae was 10%, the actual liquid volumes of *Microcystis aeruginosa* and *Streptococcus* were 1 mL and 4 mL, respectively; when the total volume ratio was 50%, they were 20 mL and 5 mL, respectively; when the total volume ratio was 60%, they were 6 mL and 24 mL, respectively; when the total volume ratio was 90%, they were 4.5 mL and 40.5 mL, respectively; and when the total volume ratio was 100%, they were 25 mL and 25 mL, respectively. Through this preparation method, the total algal concentration of the mixed system and the composition ratio of the two algae varied independently, avoiding the simultaneous increase of the densities of the two components at a fixed ratio. This experimental design increases the difficulty of mixed spectral inversion, verifying whether the present invention can extract the density information of *Microcystis aeruginosa* and *Streptococcus* separately from the composite spectrum formed by the overlap of the absorption spectra of the two algae and the cross-scattering of particles, rather than simply estimating the total density of the mixed algae based on the overall transmitted light intensity.

[0153] Figure 11The figure shows the comparison between the measured and predicted normalized relative densities of *Microcystis aeruginosa* and *Streptococcus aureus* in the mixed algae samples prepared according to the formulation shown in Table 2. Different symbols in the figure represent data from three independent replicate experiments, and the dashed line represents the ideal prediction line where the predicted density perfectly matches the measured density. Figure 11 As can be seen, under the condition that the total volume ratio of mixed algae changes from 0% to 100%, and the composition ratio of *Microcystis aeruginosa* and *Streptococcus aureus* does not change monotonically, the predicted density of *Microcystis aeruginosa* is generally distributed near the ideal prediction line. The data from the three rounds of repeated experiments show good consistency, and the coefficient of determination R between the predicted density and the measured density is [value missing]. 2 The value reached 0.99. The prediction densities in the figure are the cross-validation out-of-place predictions obtained when the samples did not participate in the corresponding model training.

[0154] Especially under different mixing conditions—high proportion of *Microcystis* and low proportion of *Streptococcus*, and low proportion of *Microcystis* and high proportion of *Streptococcus*—the predicted results for *Microcystis* still change accordingly with its actual density. This indicates that the model output is not solely determined by the total density of the mixed algae, but can extract characteristic information related to the density of *Microcystis* from the composite spectrum of the two algae. Figure 11 b. As can be seen, the predicted density of *Streptococcus* also maintains good consistency with the measured density, with a determination coefficient R0. 2 The value reached 0.98. Although the ratio of Microcystis globulus to Streptococcus globulus varied considerably among different samples, the data from the three rounds of experiments were still mainly distributed near the ideal prediction line, indicating that the present invention can reduce the cross-interference caused by the absorption and scattering responses of Microcystis globulus to the density inversion of Streptococcus globulus. Figure 11 The results show that by using multiple wavelength features of 430 nm, 480 nm, 680 nm and 730 nm in combination with near-infrared reference band scattering influence compensation, multidimensional physical ratio features and MEP explicit inversion equation, the present invention can obtain the relative density of Microcystis and Streptococcus under the condition of asynchronous change of the density of the two algae, instead of only outputting the total density of the two algae.

[0155] Figure 12 The figure shows the absolute error distribution of the density inversion results for *Microcystis aeruginosa* and *Streptococcus aureus* after three independent and repeated experiments according to the two-algae mixed sample preparation scheme shown in Table 2. In the figure, A, B, and C represent the first, second, and third rounds of testing, respectively, and the boxes and scatter points represent the absolute error and its dispersion for each mixing ratio. The absolute error in the figure is calculated from the absolute value of the difference between the predicted density from cross-validation and the corresponding measured density.

[0156] Depend on Figure 12As can be seen, the absolute errors of the three rounds of testing for *Microcystis aeruginosa* were generally at a low level, with an average absolute error of 0.021. Although some mixed-ratio samples showed relatively high error points, most of the errors in each round of testing were concentrated within a small range, and no systematic shift with a continuous increase occurred between the three rounds of testing. This indicates that even with significant changes in the content of *Streptococcus* and the total volume ratio of mixed algae, the inversion of *Microcystis aeruginosa* density using this invention still exhibits good repeatability and stability.

[0157] Depend on Figure 12 As shown in b, the average absolute error of the three rounds of testing for *Streptococcus* was 0.013, with the absolute errors of most samples concentrated in a low range. The error distributions and median levels of different experimental rounds were quite similar, indicating that changes in the proportion of *Microcystis* did not cause significant systematic interference to the density inversion of *Streptococcus*. Individual samples with larger errors can be flagged as abnormal or prompted for retesting through the physical consistency verification mechanism set up in this invention.

[0158] Combination Figure 11 and Figure 12 As can be seen, when the overall density and component ratios of the two algae changed, the prediction determination coefficients of this invention for *Microcystis aeruginosa* and *Streptococcus aureus* reached 0.99 and 0.98, respectively, with mean absolute errors of 0.021 and 0.013, respectively. These results demonstrate that this invention can perform component-level density inversion on mixed samples of two algae with spectral absorption overlap and cross-scattering interference, and maintains good accuracy and stability in multiple repeated experiments.

[0159] Compared with the method of establishing a fitting model for the total density of mixed algae based on a fixed component ratio, this embodiment demonstrates that the prediction results obtained by the present invention are not apparent correlations caused by changes in the total concentration of mixed algae by allowing the total volume ratio of mixed algae to vary independently from the composition ratio of the two algae.

[0160] This invention utilizes a near-infrared reference band to compensate for the scattering effects of visible light absorption-related characteristics, and further constructs corrected light intensity ratios and absorption-scattering joint ratios. The densities of two algae are then inverted using physically constrained MEP explicit equations. Even under conditions of superimposed spectral responses of the two algae and asynchronous changes in the density of each component, R0 is still obtained. 2 The technical effects of 0.99 and 0.98 respectively, and mean absolute errors of 0.021 and 0.013 respectively, indicate that the above technical features have a synergistic effect, improving the ability to distinguish the density of each component in the mixed suspended particulate system, the inversion accuracy, and the stability of repeated detection.

[0161] Table 3. Preparation schemes for mixed samples of three algae

[0162]

[0163] Note: The total volume of each group of mixed samples of the three algae is 50 mL; the actual liquid volume of the three algae is expressed in the order of Microcystis: Chrysophyta: Streptococcus.

[0164] To further verify the ability of this invention to simultaneously retrieve the density of each component in a three-component mixed suspension particle, this embodiment selected *Microcystis aeruginosa*, *Golden Algae*, and *Streptococcus* to prepare a mixed sample of the three algae, as shown in Table 3. The total volume of each sample was fixed at 50 mL, and the proportion of the total volume of the three algae solutions to the total sample volume was set to 0%, 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90%, and 100% respectively, with the remaining volume made up by culture medium. As the proportion of the total volume of the mixed algae increased, the volume of the culture medium gradually decreased from 50 mL to 0 mL.

[0165] While varying the total volume ratio of the mixed algae, the composition ratios of *Microcystis aeruginosa*, *Chlorella vulgaris*, and *Streptococcus* were set in a non-fixed ratio. For example, when the total volume ratio of the mixed algae was 10%, the actual liquid volumes of the three algal solutions were 1 mL, 1 mL, and 3 mL, respectively; when the total volume ratio was 40%, they were 10 mL, 6 mL, and 4 mL, respectively; when the total volume ratio was 70%, they were 17.5 mL, 12.5 mL, and 5 mL, respectively; and when the total volume ratio was 100%, they were 20 mL, 17.5 mL, and 12.5 mL, respectively. This caused the total algal density of the mixed system and the component ratios of the three algae to change simultaneously, and prevented a fixed linear relationship between the densities of each algal species. This formulation scheme can be used to rule out the possibility that the model only fits the overall density of the mixed algae, and to test whether the present invention can extract the density information corresponding to different algal species separately under the conditions of overlapping absorption spectra of the three algae and mutual interference of particle scattering.

[0166] Figure 13 The figure shows the comparison between the measured and predicted normalized densities of *Microcystis aeruginosa*, *Chlorella vulgaris*, and *Streptococcus aureus* in the mixed algae sample prepared according to the formulation shown in Table 3. Different symbols in the figure represent data from three independent and repeated experiments, and the dashed line represents the ideal prediction line where the predicted density perfectly matches the measured density. Figure 13 As can be seen, under the conditions of coexistence of *Streptococcus* and *Golden Algae*, and with the continuously changing mixing ratio of the three algae, the predicted density of *Microcystis* generally falls near the ideal prediction line. The data from the three rounds of experiments show good consistency, and the coefficient of determination R between the predicted density and the measured density is [value missing]. 2 The result reached 0.98. This indicates that the present invention can extract feature information related to the density of *Microcystis aeruginosa* from the composite spectrum of the three algae, and reduce the absorption overlap and scattering interference produced by the other two algae. The predicted densities in the figure are cross-validation out-of-place prediction results obtained when the samples did not participate in the corresponding model training.

[0167] Depend on Figure 13As can be seen from b, the predicted density of golden algae maintains good consistency with the measured density, and the coefficient of determination R0 2 The value reached 0.98. Even under conditions where the proportions of *Microcystis* and *Streptococcus* changed asynchronously, the predicted value of golden algae still changed accordingly with their actual density, indicating that the model output does not depend solely on the overall density of the three algae, but can distinguish the spectral contribution corresponding to *Streptococcus*. Figure 13 c. As can be seen, the predicted density of *Streptococcus* is generally concentrated near the ideal prediction line, with a determination coefficient R0. 2 The value reached 0.98. Even when the content and relative proportion of Microcystis and Chrysophytes changed, the predicted density of Streptococcus remained in good agreement with the measured density, indicating that the present invention can retain and identify multi-wavelength spectral information related to the density of Streptococcus even when there is interference from two other algae.

[0168] Figure 14 This table shows the absolute error distribution of density inversion results for *Microcystis aeruginosa*, *Chlorella vulgaris*, and *Streptococcus aureus* after three independent and repeated experiments according to the formulation schemes shown in Table 3. Different test chambers correspond to the three rounds of testing results for each algal species. The test chamber and scatter plot represent the absolute error and dispersion of samples with different mixing ratios. Figure 14 As can be seen, the absolute errors of the three rounds of testing for *Microcystis aeruginosa* were generally at a low level, with an average absolute error of 0.010776. The errors for most samples were concentrated within a small range, and no sustained increase or overall shift was observed between the three rounds of testing. This indicates that the density inversion of *Microcystis aeruginosa* has good accuracy and repeatability under varying *Streptococcus* and *Golden Algae* content. The absolute error in the figure was calculated from the absolute value of the difference between the predicted density and the corresponding measured density after cross-validation.

[0169] Depend on Figure 14 As can be seen from b, the average absolute error of the three rounds of testing on golden algae was 0.006963, the lowest among the three algae. The absolute errors of most samples were concentrated within 0.012, and the error distributions of different rounds were generally similar, indicating that the present invention has a good ability to identify the spectral contribution of golden algae and can effectively suppress the cross-interference caused by changes in the ratio of *Microcystis* and *Streptococcus*. Figure 14 As can be seen from c, the average absolute error of the three rounds of testing for *Streptococcus* was 0.008278, with the absolute error of most samples remaining within a small range. Although some samples with relatively large mixing ratios had relatively large error points, no obvious systematic shift was observed in the three rounds of testing, indicating that the density inversion of *Streptococcus* has good stability under different mixing ratios and different experimental rounds.

[0170] Combination Figure 13 and Figure 14It can be seen that, under the condition that the total volume ratio and the proportion of each component of the three algae change simultaneously, the prediction determination coefficients of *Microcystis aeruginosa*, *Chlorella vulgaris*, and *Streptococcus aureus* all reach 0.98, with mean absolute errors of 0.010776, 0.006963, and 0.008278, respectively. These results indicate that the present invention can perform component-level density joint inversion of three-component mixed suspended particles and maintains good accuracy and stability in multiple repeated experiments.

[0171] The technical effects and advantages of this method are as follows:

[0172] 1. Strong ability to compensate for scattering effects. This method uses the 730 nm near-infrared band as a reference band with relatively weak pigment-related absorption effects and can reflect scattering changes. It compensates for the scattering effects of absorption-related characteristic peaks such as 430 nm, 480 nm, and 680 nm, which can reduce the systematic deviation caused by suspended particle scattering on transmitted light intensity measurement. This ensures a more stable correspondence between the corrected spectral characteristics and particle density within the preset detection range, thereby improving the accuracy and repeatability of density detection.

[0173] 2. Applicable to joint inversion of multi-component mixed suspended particles. This method utilizes spectral information from multiple characteristic wavelengths for joint analysis. Different types of suspended particles have different absorption and scattering response characteristics at different wavelengths. By combining multiple wavelength features, the ability to distinguish the spectral contributions of different components in the mixed system can be enhanced, thereby improving the applicability of density inversion for single-component, two-component, and multi-component suspended particles.

[0174] 3. The model results have good interpretability. This method incorporates spectral physics rules such as the Beer-Lambert law, the relationship between scattering and absorption response, and the variation law between density, transmitted light intensity, and scattering response into a multivariate expression programming model, enabling the model to generate density inversion formulas with explicit mathematical forms. Compared to methods that rely solely on black-box data fitting, the inversion process of this method is clearer, and the model results are easier to interpret, verify, and deploy in engineering.

[0175] 4. It can reduce the impact of particle state changes on the inversion results. This method identifies the growth or aggregation state of suspended particles by analyzing changes in spectral characteristic ratios, and uses the state identification results as auxiliary information for density inversion. This reduces the impact of particle growth stage, aggregation degree, or optical state changes on the density inversion results, reduces the need for repeated calibration, and is suitable for continuous or batch detection scenarios.

[0176] 5. The device has a simple structure and the detection process is non-invasive. The detection device used in this method mainly includes an LED backlight panel, a fiber optic probe, a spectrometer, and data processing equipment, which has a relatively simple structure. During the detection process, there is no need to centrifuge, stain, or perform any destructive treatment on the sample, which helps to maintain the original state of the sample and facilitates its use with existing spectroscopic detection equipment.

[0177] 6. High degree of automation throughout the entire process. This method establishes an automated processing flow from spectral data reading, filtering and denoising, characteristic peak extraction, scattering effect compensation, characteristic ratio calculation, state recognition, density inversion to result output. This can reduce errors caused by manual step-by-step operation and improve the efficiency and consistency of batch sample processing.

[0178] 7. The application scope is expandable. This method is not only applicable to biological suspended particles such as microalgae and microorganisms, but can also be extended to detect organic suspensions, industrial inorganic suspended particles, and other objects based on the spectral response characteristics of the target sample. It can be used in laboratory analysis, water monitoring, culture process monitoring, and industrial process detection.

[0179] 8. High reliability of results. This method sets up a consistency verification mechanism before outputting density results, which can judge the non-negativity of density, numerical range, transmission light intensity variation trend, and scattering response variation trend; for results that obviously do not conform to the preset spectral variation law, anomaly marking, rejection, prompting retesting or correction can be performed, thereby improving the reliability of output results.

[0180] 9. More comprehensive feature construction method. This method simultaneously constructs multiple types of features, such as the corrected transmission intensity ratio and the absorption-scattering combined ratio, to combine and express the information required for particle component identification, state judgment, and density estimation. This reduces the risk of single features being affected by noise, baseline drift, or changes in sample state, and improves the stability of the inversion model in complex suspension systems.

[0181] In addition, this method also has the following engineering application characteristics:

[0182] 1. This method can complete subsequent calculations by reading the standard data file output by the spectrometer, without modifying the spectrometer hardware structure or changing the original spectral acquisition process, making it easy to deploy and use on existing spectral detection equipment.

[0183] 2. This method does not require a complicated sample pretreatment process. It can complete the detection of suspended particle density with a common spectrometer and supporting data processing software, which helps to reduce the complexity of detection operations and improve sample processing efficiency.

[0184] 3. This method can be applied to scenarios such as microalgae cultivation, aquaculture water quality monitoring, industrial suspension process control, and environmental water monitoring. It can be used to assist in the rapid detection of suspended particle concentration or density, and provide data support for process control and state assessment.

[0185] 4. This method adopts a technical route that combines multispectral characteristic peak joint inversion, scattering effect compensation, state identification and multivariate expression programming density inversion. It can invert the density of different components or the total suspended particles without separating the mixed system to be tested, thus improving the applicability of detection of complex suspended systems.

[0186] 5. This method adopts a non-invasive data reading and software automatic processing approach, which can expand the functionality of existing spectrometers, reduce the difficulty of system deployment, and support batch processing, result storage, and traceability of the detection process.

[0187] like Figure 2 As shown, this embodiment of the invention provides a multispectral characteristic peak joint inversion mixed suspended particle density detection system for implementing the above-mentioned detection method. The system includes a light source unit, a sample carrier unit, and a spectral acquisition unit arranged sequentially along the transmission light propagation direction, as well as a data processing unit connected to the spectral acquisition unit. The sample carrier unit includes a three-dimensional adjustable base 2 and a suspended particle sample 1 to be tested. The light source unit includes a constant current driven LED backlight 3 and an LED power supply 4. The spectral acquisition unit includes a transmission-type quartz fiber optic probe 5 and a miniature fiber optic spectrometer 6. The data processing unit includes a data processing device 7 configured to execute the data processing program of any one of claims 1 to 9, possessing functions such as file directory monitoring, automatic spectral analysis, feature calculation, state discrimination, MEP density inversion, physical verification, and automatic report export.

[0188] The suspended particle sample 1 to be tested is placed in a transparent sample container and placed on an adjustable base 2. The adjustable base 2 is used to adjust the position of the sample container relative to the LED backlight 3 and the fiber optic probe 5 to adapt to different sample volumes, detection heights, or optical path alignment requirements.

[0189] The LED backlight panel 3 is located on one side of the sample container to provide stable back illumination to the suspended particle sample 1 to be tested; the LED power supply 4 is connected to the LED backlight panel 3 to provide stable power to the LED backlight panel 3, so as to reduce the influence of light source fluctuations on the spectral acquisition results.

[0190] The fiber optic probe 5 is located on the other side of the sample container and is used to receive the transmission spectral signal after passing through the suspended particle sample 1 to be tested. The fiber optic probe 5 is connected to the spectrometer 6, which is used to collect transmission spectral data and transmit the spectral data to the data processing device 7, or save the spectral data as a standard data file for the data processing device 7 to read.

[0191] The data processing device 7 can be a computer, industrial control computer, or embedded processing device, used to receive or read the spectral data output by the spectrometer 6, and to execute the spectral preprocessing, target feature peak extraction, scattering influence compensation, multidimensional feature calculation, state recognition, density inversion, and result output process described in this invention.

[0192] During operation, the LED power supply 4 drives the LED backlight panel 3 to emit light, and the LED backlight panel 3 provides stable illumination light. After the illumination light passes through the suspended particle sample 1 to be tested, it is received by the fiber optic probe 5 and transmitted to the spectrometer 6. After the spectrometer 6 collects the transmission spectrum data, the data processing device 7 performs subsequent algorithm processing to obtain the density detection result of the suspended particle sample to be tested.

[0193] Preferably, the adjustable base 2 is a three-dimensional adjustable mechanical platform with a vertical adjustment stroke of 0 to 100 mm and an adjustment accuracy of not less than 0.1 mm. It can be continuously adjusted by a manual knob or a stepper motor to control the relative position between the suspended particle sample 1 to be tested and the fiber optic probe 5, so as to ensure the consistency of optical path alignment, optical path stability and repeatability of detection at different heights or in different areas.

[0194] Preferably, the sample container holding the suspended particulate sample 1 to be tested is a rectangular transparent container, which can be made of acrylic, glass or quartz material, with a detection volume of 5 to 500 mL. The light transmittance of the sample container in the detection band is not less than 90% to reduce the influence of the container wall on the spectral signal.

[0195] Preferably, the LED backlight panel 3 is a cool white light high uniformity surface light source with a color temperature of 5000-6500 K, a color rendering index (CRI) of not less than 90, and a light-emitting surface uniformity of not less than 95%, so as to provide stable, uniform, and low-flicker backlighting.

[0196] Preferably, the spectral range of the LED backlight panel 3 covers 400–800 nm, and the output power can be adjusted according to the detection requirements. The light intensity is adjustable in the range of 0–10000 lux, preferably 2000–5000 lux, so as to take into account both the spectral signal intensity and the influence of sample illumination.

[0197] Preferably, the LED power supply 4 is a constant current drive power supply with an output current fluctuation of no more than ±1% and a ripple coefficient of no more than 5%, so as to ensure the stability of the output light intensity of the LED backlight panel 3 and reduce the influence of light source drift on the spectral measurement results.

[0198] Preferably, the fiber optic probe 5 is a transmission fiber optic probe with a quartz end face, a numerical aperture (NA) of 0.22–0.37, and a fiber core diameter of 200–600 μm to ensure sufficient light throughput and signal strength.

[0199] Preferably, the fiber optic probe 5 is fixed by a bracket, which allows for three-dimensional position fine-tuning with an adjustment accuracy of no less than 0.5 mm. The probe end face of the fiber optic probe 5 is kept aligned with the light-emitting surface of the LED backlight panel 3 to ensure consistent transmission light acquisition.

[0200] Preferably, the working wavelength range of the fiber optic probe 5 is 200–1100 nm, which can cover the acquisition requirements of the characteristic wavelengths of 430 nm, 480 nm, 680 nm and 730 nm described in this invention.

[0201] Preferably, the spectrometer 6 is a miniature fiber optic spectrometer with a wavelength range of 350–1000 nm, an optical resolution of no more than 2 nm, and a signal-to-noise ratio of no less than 400:1, in order to meet the requirements for target characteristic peak localization and light intensity measurement.

[0202] Preferably, the integration time of the spectrometer 6 is adjustable, ranging from 1 ms to 10 s, and preferably set to 10 to 100 ms. It can be adjusted according to the sample concentration, light source intensity, or signal strength to avoid signal saturation or insufficient signal-to-noise ratio.

[0203] Preferably, the spectrometer 6 has a built-in USB, RS-232 or other data interface for data transmission with the data processing device 7; or, the spectrometer 6 can export the collected spectral data as a standard data file for non-intrusive reading and processing by the data processing device 7.

[0204] Preferably, the data processing device 7 is a computer, industrial control computer, or embedded processing device, equipped with the suspended particle density detection software of the present invention, which can realize spectral data reading, filtering and noise reduction, characteristic peak extraction, scattering influence compensation, characteristic ratio calculation, state recognition, density inversion, result verification, and result output.

[0205] Preferably, the data processing device 7 has automatic data storage and report generation functions, and can output the detection results in CSV, Excel or PDF format, and supports batch processing and historical data backtracking.

[0206] Preferably, the data processing device 7 has a built-in result consistency verification module, which can verify the inversion results according to preset rules such as density non-negativity, numerical range, transmission light intensity change trend and scattering response change trend, and output a test report with different states such as normal, abnormal, need to be retested or need to be corrected according to the verification results.

[0207] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for detecting the density of mixed suspended particles by joint inversion of multispectral characteristic peaks, characterized in that, Includes the following steps: Step 1, Spectral Preprocessing and Characteristic Peak Extraction: The host computer reads the raw spectral data of suspended particles in CSV format exported from the spectrometer. Noise reduction is achieved using a Savitzky-Golay filter with a window of 5 data points and a polynomial order of 2. Four characteristic wavelengths—430nm, 480nm, 680nm, and 730nm—are fixed, and local transmission peaks are retrieved within a wavelength range of ±20nm for each characteristic wavelength. , , , The original transmitted light intensities extracted from the locations are respectively , , , ; Step 2, Quantitative compensation for near-infrared reference peak scattering: Using 730nm as the near-infrared reference band where pigment absorption response is relatively weak, the correction coefficients β1, β2, β3, and β4 are dynamically calculated according to a preset formula; through... , , Calculate the corrected transmitted light intensity at 430nm, 480nm, and 680nm; define the near-infrared corrected transmission reference signal. ; Step 3, Calculation of multidimensional physical ratio characteristics: based on , , , Calculate the four types of ratio characteristics: / , / , / , / Set a minimum threshold for the denominator; when the denominator is below the threshold, mark the sample as abnormal and replace it with a stable value. Step 4, Threshold determination of particle growth state: based on features Based on the preset algal species multi-level threshold, the particles are divided into the lag phase, logarithmic growth phase, stationary phase, and death phase, and the output state discrimination label is used as an auxiliary input for density inversion. Step 5, density inversion based on the physical constraint MEP model: Density normalization mapping is completed using blank samples and baseline density standard samples. Three physical rules—non-negative density, attenuation of transmitted light intensity with increasing density, and overall attenuation of near-infrared transmitted reference signal with increasing density—are used as constraint evolution multivariate expressions to program the inversion equation. The relative density is calculated using the corresponding piecewise explicit equations based on the sample algal species and the state labels output from Step 4. The true absolute density of the particles is restored using a bilinear interpolation mapping formula. For multi-component mixed samples, a multi-output joint MEP equation is constructed to simultaneously calculate the density of each component. Step 6, Physical Consistency Verification and Automatic Output: Perform four-layer verification on the true density output in Step 5: density value non-negation verification; value falling within a preset detection range verification; if density increases... , , Verification of a monotonically decreasing trend; if density increases... Overall downward trend verification; samples that fail verification are marked as abnormal and prompted for retesting; if verification passes, density results are stored; the host computer monitors the spectrometer output directory, and automatically loops through steps 1 to 6 after detecting a new spectral CSV file, batch exporting test reports in PDF, Excel, and CSV formats.

2. The method for detecting the density of mixed suspended particles by joint inversion of multispectral characteristic peaks according to claim 1, characterized in that, In step 2, the formula for calculating the correction coefficient is: ； ； ； ； in, The value range is 0.02 ± 0.

001. The value range is 0.15 ± 0.

003. The value range is 0.01 ± 0.

001.

3. The method for detecting the density of mixed suspended particles by joint inversion of multispectral characteristic peaks according to claim 1, characterized in that, In step 4, the threshold discrimination rule for *Streptococcus* is as follows: ≥1.1 Determination of delay period, 1.1> A value ≥0.9 indicates the logarithmic growth phase; 0.9 > A value ≥0.7 indicates a stable period; 0.7 > Determining the death phase; the threshold discrimination rule for *Microcystis aeruginosa* is as follows: ≥0.7 indicates a delay period; 0.7> A value ≥0.6 indicates the logarithmic growth phase; 0.6 > A value ≥0.55 indicates a stable period; 0.55 > Determine the death period.

4. The method for detecting the density of mixed suspended particles by joint inversion of multispectral characteristic peaks according to claim 1, characterized in that, In step 5, the relative density of Streptococcus is... The calculation formula is: ； 。 5. The method for detecting the density of mixed suspended particles by joint inversion of multispectral characteristic peaks according to claim 1, characterized in that, In step 5, the relative density of *Microcystis globulus* The calculation formula is ； in, , for: ； 。 6. The method for detecting the density of mixed suspended particles by joint inversion of multispectral characteristic peaks according to claim 1, characterized in that, In step 5, the formula for the true density mapped by bilinear interpolation is: ； Among them, the age interpolation weight ω(t) and the relative density piecewise calibration function fA,d(ρ) are defined as follows: ； ； Among them, D real (M,A,t,ρ) represents the true density obtained from the mapping; M is the pre-established calibration matrix; A represents the type of particle to be tested; t is the actual culture age of the sample; ρ is the relative density factor output by the MEP model; d i and d i+1 The adjacent calibrated age nodes surrounding t; q j and y j These are the j-th input node and density node of the piecewise calibration curve, respectively; when t or ρ exceeds the calibration range, boundary extrapolation should be performed and the out-of-range state should be marked.

7. The method for detecting the density of mixed suspended particles by joint inversion of multispectral characteristic peaks according to claim 1, characterized in that, In step 1, the host computer reads data in a non-intrusive manner, without modifying the spectrometer hardware, drivers, or communication programs. It only reads the standard CSV file exported locally by the spectrometer and does not intercept the instrument's real-time communication data stream.

8. The method for detecting the density of mixed suspended particles by joint inversion of multispectral characteristic peaks according to claim 1, characterized in that, In step 6, the processing methods for abnormal samples after verification include three categories: marking abnormal samples, removing the data from this test, and outputting the corrected estimated density, with a retest reminder noted in the report.

9. The method for detecting the density of mixed suspended particles by joint inversion of multispectral characteristic peaks according to claim 1, characterized in that, The method is adapted to detect microalgae and other microbial suspended particles with pigment absorption characteristics. For different test species, the state discrimination threshold in step 4 and the MEP density inversion equation in step 5 are adjusted according to their pigment absorption band, culture conditions and calibration data.

10. A system for detecting the density of mixed suspended particles by joint inversion of multispectral characteristic peaks according to any one of claims 1 to 9, characterized in that, The system includes a light source unit, a sample carrier unit, and a spectral acquisition unit arranged sequentially along the direction of transmitted light propagation, as well as a data processing unit connected to the spectral acquisition unit. The sample carrier unit includes a three-dimensional adjustable base (2) and a suspended particle sample (1) to be tested. The light source unit includes a constant current driven LED backlight plate (3) and an LED power supply (4). The spectral acquisition unit includes a transmission-type quartz fiber optic probe (5) and a miniature fiber optic spectrometer (6). The data processing unit includes a data processing device (7) configured to execute the data processing program of the detection method described in any one of claims 1 to 9, and has functions such as file directory monitoring, automatic spectral analysis, feature calculation, state discrimination, MEP density inversion, physical verification, and automatic report export.

Images