CN-121721003-B - High-flux biosensing system for early warning of algal bloom and pathogen

CN121721003BCN 121721003 BCN121721003 BCN 121721003BCN-121721003-B

Abstract

The invention belongs to the technical field of data analysis, and particularly relates to a high-flux biosensing system for early warning of algal bloom and pathogens, which comprises a plurality of sampling units, a microfluidic chip with a multichannel structure and a high-flux surface-enhanced Raman detection unit integrated with the surface of a metal nano structure, wherein the system acquires an original surface-enhanced Raman spectrum of a water body sample through a Raman spectrum acquisition mechanism. The system reduces the data volume and maintains the integrity of spectrum information by utilizing a compressed spectrum sampling and sparse reconstruction method, constructs a pathogen Raman fingerprint dictionary by a dictionary learning mode, and realizes the identification of different algae metabolites, algae toxins and pathogen related molecular components and the concentration interval judgment thereof by combining a sparse representation classification mechanism. Based on pathogen type and concentration interval, the system further generates algal bloom risk early warning and pathogen risk early warning, and early detection of abnormal change of the water body is achieved. The invention has remarkable advantages in the aspects of stability, sensitivity, flux and real-time performance.

Inventors

LU ZHIYUAN
HUANG HAIYAN
ZHANG GUANGFU
CHEN LU
FU YICHENG
ZHANG JIAN
WEN JIE

Assignees

云南水利水电职业学院
中国水利水电科学研究院

Dates

Publication Date: 20260508
Application Date: 20260227

Claims (7)

1. The high-flux biosensing system for early warning of algal bloom and pathogens is characterized by comprising a plurality of sampling units, a plurality of detection units and a detection unit, wherein the sampling units are used for collecting water samples in a water body to be monitored; the system comprises a microfluidic chip, a spectrum processing module, a dictionary learning module, a classification module, a pre-warning module and a risk pre-warning module, wherein the microfluidic chip is provided with a sample inlet communicated with a sampling unit and a plurality of micro-channels for transporting a water body sample, a plurality of high-flux surface-enhanced Raman detection units are arranged in the micro-channels along a water body sample flow path, each high-flux surface-enhanced Raman detection unit comprises a metal nanostructure surface integrally formed with the bottom of the micro-channel and a laser irradiation window correspondingly arranged with the metal nanostructure surface; a branch micro-channel structure and a confluence micro-channel structure are arranged in the micro-fluidic chip, the branch micro-channel structure is used for distributing each water sample into a plurality of parallel flow channels or a plurality of mutually independent micro-droplet flows, the high-flux surface-enhanced Raman detection units are sequentially arranged along the flow paths of the parallel flow channels or the micro-droplet flows, the metal nano-structure surface of each high-flux surface-enhanced Raman detection unit is composed of a gold nano-structure surface and a silver nano-structure surface, and the laser irradiation windows are oppositely arranged with the corresponding areas of the gold nano-structure surface and the silver nano-structure surface, so that the high-flux collection of the surface-enhanced Raman scattering signals is realized when the water sample passes through the gold nano-structure surface and the silver nano-structure surface along the micro-channels; The dictionary learning module is configured to construct a pathogen Raman fingerprint dictionary in a system calibration stage, and specifically comprises the steps of preparing standard water samples respectively containing single algae metabolites, single algae toxins and single pathogen related molecular components, introducing the standard water samples into a microfluidic chip, collecting corresponding original surface enhanced Raman spectra at each high-flux surface enhanced Raman detection unit, sequentially performing baseline subtraction, background signal deduction and integral intensity normalization processing on the original surface enhanced Raman spectra of each standard water sample along a Raman shift direction, and arranging light intensity values of a preprocessed spectrum curve on each Raman shift channel according to a fixed sequence to form a standard spectrum data sequence, selecting a plurality of standard spectrum data sequences subjected to intensity normalization from all the standard spectrum data sequences as an initial dictionary atom set, and inputting each dictionary atom in the initial dictionary atom set as a pathogen Raman fingerprint candidate template into a dictionary learning process so as to update and form the pathogen Raman fingerprint dictionary in subsequent iterations; the classification module adopts a sparse representation classification mode to judge a water body sample, specifically, according to a class mark relation established by dictionary atoms in a pathogen Raman fingerprint dictionary in a dictionary learning stage, the dictionary atoms are divided into a plurality of class subsets, each class subset corresponds to an algae metabolite type, an algae toxin type or a pathogen type and a concentration interval thereof, for each sparse coefficient vector output by the spectrum processing module, the sparse coefficient vector output by the spectrum processing module is copied to form a class coefficient vector, all sparse coefficients on dictionary atom positions except the current class subset are set to zero, the sparse coefficients on the dictionary atom positions belonging to the current class subset are kept unchanged, the class reconstruction spectrum is obtained by linear combination channel by channel based on the sparse coefficients in the spectrum curve of all dictionary atoms in the pathogen Raman fingerprint dictionary and the class coefficient vector, the class reconstruction spectrum and the absolute value of the light intensity difference value of the reconstructed surface enhanced Raman spectrum of the corresponding water body sample are summed channel by channel to obtain a class difference value, the class mark corresponding to the class subset with the smallest class difference value is used as a pathogen type judgment result of the water body sample, and the concentration of the corresponding class difference value is used as a pathogen type judgment result of the water body sample in the water body sample judgment module.
2. The system of claim 1, wherein the raman spectrum acquisition module is configured to spectrally decompose scattered light acquired at the high-throughput surface-enhanced raman detection unit to obtain raw surface-enhanced raman spectra corresponding to the plurality of water samples, and arrange light intensity values of each raw surface-enhanced raman spectrum on each raman shift channel in a fixed order to form a raw spectrum data sequence, and aggregate all the raw spectrum data sequences to form a raw spectrum data set, and output the raw spectrum data set to the spectrum processing module.
3. The system of claim 2, wherein the spectral processing module is configured to construct a compressed spectral sampling mask, in particular, a set of raman-shifted channel index sets for surface-enhanced raman detection is predetermined in the spectral acquisition system, the raman-shifted channel index sets are used as candidate band sets, a plurality of random integer sequences are generated on the candidate band sets by using a random number generator, each random integer sequence corresponds to one compressed spectral sampling mask, the compressed spectral sampling mask is represented by a sequence of marks having the same length as the candidate band sets, each mark in the sequence of marks corresponds to one raman-shifted channel one by one, the marks at positions selected by the random integer sequence are in a sampling state, the marks at the remaining positions are in a non-sampling state, and each high-throughput surface-enhanced raman detection unit corresponds to one compressed spectral sampling mask and stores each compressed spectral sampling mask in the spectral processing module.
4. The system of claim 3, wherein the spectral processing module is further configured to obtain compressed spectral data by, for each of the original spectral data sequences in the original spectral data set, invoking a compressed spectral sampling mask corresponding to a high throughput surface enhanced raman detection unit that generates a current original spectral data sequence, reading light intensity values at raman-shifted channel positions in a sampled state from the original spectral data sequence, and arranging the channel sequences in the original spectral data sequence to form a compressed spectral data sequence, aggregating all of the compressed spectral data sequences to form a compressed spectral data set, and using the compressed spectral data set as compressed perceptually reconstructed input data.
5. The system of claim 4, wherein the dictionary learning module constructs a pathogen raman fingerprint dictionary by adopting an iterative mode combining sparse coding and dictionary atom updating, the sparse coding is based on an orthogonal matching pursuit algorithm, a sparse coefficient vector is calculated on a current dictionary atom set for each standard spectrum data sequence, the absolute value sum of a residual signal on all raman shift channels is controlled not to exceed a preset residual upper limit value or the number of dictionary atoms in a sparse support set is controlled not to exceed a preset upper limit, the dictionary atom updating is based on sparse coefficient vectors corresponding to all standard spectrum data sequences, a standard spectrum sub-set is formed by selecting standard spectrum data sequences with non-zero sparse coefficients for each dictionary atom, a residual matrix is constructed by utilizing the difference value between a linearly reconstructed spectrum after removing target dictionary atoms and an original standard spectrum data sequence, a singular value decomposition is performed on the residual matrix to obtain a characteristic spectrum with maximum energy and is used as an updated dictionary atom after intensity normalization, the sparse coefficient of the related standard spectrum data sequence on the updated dictionary atom position is updated by utilizing the corresponding characteristic coefficient sequence, the sparse coding and the dictionary atom updating is repeatedly performed until the preset iteration number is up to or two adjacent iteration number reaches the preset iteration number of times, and the corresponding relation between the current dictionary atom and the corresponding to the atomic number of the corresponding dictionary atoms is lower than the preset atomic number of the preset curve, and the pathogen index is calculated, and the pathogen index molecular class is formed by the atomic change, and the pathogen molecular relation is formed by the molecular relation of the atomic index, and the pathogen index is calculated.
6. The system of claim 5, wherein the spectral processing module performs sparse coefficient solving and surface-enhanced raman spectrum generation based on a compressed spectrum data sequence in an online detection stage, calls a compressed spectrum sampling mask corresponding to the compressed spectrum data sequence for each compressed spectrum data sequence, performs identical mask sampling on dictionary atoms in a pathogen raman fingerprint dictionary to obtain a compressed dictionary set, adopts an orthogonal matching pursuit algorithm on the compressed dictionary set, takes the compressed spectrum data sequence as a target compressed spectrum iteration to solve a compressed sparse coefficient, selects the compressed dictionary atoms and estimates the compressed sparse coefficient in a least square sense by utilizing an inner product value maximum principle, controls the sum of absolute values of compressed residual signals at all sampling positions to be not more than a preset compressed residual upper limit value or the number of compressed dictionary atoms in the compressed sparse support set to be not more than a preset upper limit value, so as to obtain a compressed sparse coefficient set, fills the compressed sparse coefficient set into a sparse coefficient vector according to the position of the compressed dictionary atoms in the pathogen raman fingerprint dictionary, sets the dictionary atoms not participating in the pathogen raman dictionary set to correspond sparse coefficient to zero, performs linear combination of the compressed dictionary atoms and the compressed dictionary atoms in the pathogen raman dictionary set to obtain a complete raman spectral coefficient vector, and performs surface-enhanced raman spectral sample reconstruction by using the linear combination of the compressed dictionary atoms and the linear combination raman spectral coefficient with the surface-enhanced raman spectral coefficient as a complete raman spectral sample.
7. The system of claim 6, further comprising a display terminal connected to the early warning determination module, wherein the early warning determination module pre-stores an algal bloom risk determination table and a pathogen risk determination table, the early warning determination module is configured to search the pathogen type determination result and the concentration interval thereof output by the classification module, generate an algal bloom risk early warning for a water body sample belonging to the algal bloom related algae metabolite or algae toxin class and having a concentration interval higher than a preset algal bloom safety concentration upper limit, generate a pathogen risk early warning for a water body sample belonging to the pathogen class and having a concentration interval higher than the preset pathogen safety concentration upper limit, and summarize the algal bloom risk early warning and the pathogen risk early warning corresponding to the plurality of water body samples according to sampling positions and sampling times to form algal bloom and pathogen early warning information and send the algal bloom and pathogen early warning information to the display terminal for the display terminal to display and for a manager to make a monitoring decision according to the algal bloom and pathogen early warning information.

Description

High-flux biosensing system for early warning of algal bloom and pathogen Technical Field The invention belongs to the technical field of data analysis, and particularly relates to a high-flux biosensing system for early warning of algal bloom and pathogens. Background Algal bloom and water borne pathogen contamination are two central problems of increasing concern in the field of water environmental monitoring in recent years. With the increase of global climate change and nutrient input, the increase of algae biomass in lakes, reservoirs and watershed channels presents a more frequent and stronger trend, especially the outbreak of blue algae, green algae and flagella algae formation is more remarkable. Accumulation of algae metabolites and algae toxins not only can change the color of the water body and reduce the transparency, but also can cause the formation of odor substances, and can directly threaten the safety of drinking water. Meanwhile, various bacteria, viruses and parasite related components also show the characteristics of easier transmission and harder early warning in urban water supply systems, rural water intake and aquaculture water bodies. The algae metabolites, algae toxins and pathogen pollution in the water body often show the characteristics of space dispersion, time fluctuation, concentration mutation and the like, so that the traditional manual sampling mode and laboratory analysis mode are difficult to realize early monitoring and real-time response. Traditional water quality monitoring means mainly depend on chemical analysis methods and biological analysis methods. For example, detection of algal metabolites and algal toxins typically requires high performance liquid chromatography, enzyme-linked immunoassay, or mass spectrometry, which have high quantitative accuracy, but generally require long pretreatment, large analytical instruments, and specialized operators. Since the time from sampling to result is often several hours or even tens of hours, it is difficult to meet the requirement of a rapid reaction mechanism when algal bloom or pathogen diffusion occurs in a water body. Although the culture method has good specificity for the water-borne pathogens, the detection period is usually 24 hours or more, and it is difficult for some viruses or non-culturable bacteria to obtain accurate results in this way. In order to improve the analysis speed, methods such as fluorescent probes, rapid immunochromatography reagents and the like are introduced into part of laboratories, but target molecular types of the methods are limited, and the methods are easily influenced by complex matrixes of water bodies, so that the proportion of false positives or false negatives is improved. Disclosure of Invention In view of the above, the main object of the present invention is to provide a high-throughput biosensing system for early warning of algal bloom and pathogens, which is to organically combine a multipoint sampling structure, a microfluidic chip with multichannel parallel detection capability, a surface enhanced raman detection mode of a high-stability metal nanostructure surface, a data acquisition mechanism based on compressed spectrum sampling, a spectral fingerprint construction method based on dictionary learning, a sparse representation classification and a risk determination mechanism, so as to construct a set of algal bloom and pathogen early warning system capable of realizing continuous, high-throughput and accurate recognition in a water body. The system can capture and accurately analyze the spectral signal characteristics at the stage that the concentration of the water body components is still at a low level and the change is not obvious, automatically distinguish different algae metabolites, algae toxins and pathogen related molecular components, and provide corresponding concentration intervals and risk grades. Through sparse reconstruction and dictionary learning mechanisms, the system can still maintain high recognition accuracy and stability in the face of complex water body background, multi-component overlapping and time space fluctuation, and meanwhile, the data volume and processing burden are remarkably reduced, so that high-throughput online monitoring is possible. The invention can improve the real-time performance, sensitivity and coverage of water environment monitoring, provides reliable technical support for quick response to algal bloom events and water source pathogen diffusion, and has important application value and popularization significance. The technical scheme adopted by the invention is as follows: The system comprises a plurality of sampling units, a micro-fluidic chip, a dictionary learning module, a processing module, a risk pre-warning module and a risk pre-warning module, wherein the sampling units are used for collecting scattered light at the high-flux surface-enhanced Raman detection units and performing spectrum decomposition to generate a