An underwater microplastic sampling concentration calculation system

Abstract

The invention provides an underwater microplastic sampling concentration calculation system, which belongs to the technical field of underwater microplastic sampling. The system is composed of a water pump, a black tube and a control chip. Six filter screens are arranged in sequence in the black tube, and the mesh aperture is gradually reduced. Two chambers are formed between the fourth filter screen and the fifth filter screen and the sixth filter screen. The first chamber is provided with a carbon dioxide generator, and carbon dioxide enters the first chamber. The second chamber is provided with multiple groups of laser transceivers, and the laser sensor sends the received laser signal to the control chip. An underwater microplastic concentration calculation module is arranged in the control chip, and the concentration of microplastics in water is calculated according to the laser sensor signal and the output parameter of the water pump. Each laser transmitter emits lasers of different colors. The device can monitor the content of microplastics in water in real time. The invention solves the technical problem that the prior art cannot realize the rapid online detection of the concentration of microplastics in water, and manual sampling, pretreatment and analysis are required, which is inefficient.

Description

Technical Field

[0001] The present invention belongs to the technical field of underwater microplastic sampling, and in particular, relates to an underwater microplastic sampling concentration calculation system. Background Art

[0002] In recent years, with the continuous increase of human activities, the problem of microplastic pollution in oceans and freshwater has become increasingly serious. Plastic is a very common and widely used material. Its excellent properties make it indispensable in many fields such as packaging, daily necessities, and industrial products. However, plastic is difficult to degrade naturally. After long-term use and improper disposal, it will remain in the environment, especially those microplastic particles with a diameter of less than 5 mm. These microplastics will enter lakes, rivers and oceans with the flow of rivers and sewage, causing water pollution.

[0003] Microplastic particles are small in size and can be ingested by marine organisms and enter the food chain, posing a serious threat to the ecosystem. In addition, microplastics may also adsorb and enrich some toxic and harmful substances such as persistent organic pollutants and heavy metal ions, further exacerbating the harm to organisms. Therefore, timely and effective detection and monitoring of the concentration distribution of microplastics in water bodies is crucial for preventing and controlling environmental pollution and protecting marine ecosystems.

[0004] At present, the main methods for detecting microplastics in water bodies include:

[0005] 1. Sampling and filtration method: Sampling is done in water bodies, filtered using filter membranes of different pore sizes, and then the solids remaining on the filter membranes are observed under a microscope and manually identified. This method is cumbersome to operate, has low analysis efficiency, and cannot give quantitative results for the number and volume of microplastic particles.

[0006] 2. Spectroscopic analysis: Use infrared spectroscopy, Raman spectroscopy and other technologies to analyze and identify the composition of microplastics in water samples. This method requires manual sorting of the particles to be tested, and cannot achieve rapid online detection. In addition, for microplastics made of composite materials, the spectral characteristics are easily obscured.

[0007] 3. Labeling and tracing method: Mark tracers such as fluorescent dyes or isotopes on microplastics, and then track the distribution of tracers in water bodies. This method requires manual labeling of all microplastic particles, which is complicated and costly.

[0008] The above existing technologies all have the technical problem of being unable to achieve rapid online detection of microplastic concentrations in water bodies, requiring manual sampling, pretreatment and analysis, and being inefficient. Summary of the invention

[0009] In view of this, the present invention provides an underwater microplastic sampling concentration calculation system, which can solve the technical problems that the prior art cannot realize rapid online detection of microplastic concentration in water bodies, requires manual sampling, preprocessing and analysis, and is inefficient.

[0010] The present invention is achieved in that:

[0011] The first aspect of the present invention provides an underwater microplastic sampling concentration calculation system, which includes a water pump, a black tube and a control chip, the output end of the water pump is connected to one end of the black tube, and the first filter screen, the second filter screen, the third filter screen, the fourth filter screen, the fifth filter screen and the sixth filter screen are sequentially arranged in the black tube, and the sixth filter screen is arranged at the other end of the black tube. The mesh apertures of the first filter screen, the second filter screen, the third filter screen and the fourth filter screen are gradually reduced, and the mesh apertures of the fourth filter screen, the fifth filter screen and the sixth filter screen are the same. A first chamber and a second chamber are formed between the fourth filter screen, the fifth filter screen and the sixth filter screen, and a carbon dioxide generator is arranged on the outer wall of the first chamber. The carbon dioxide generated by the carbon dioxide generator is introduced into the first chamber through the through hole on the outer wall of the first chamber. A plurality of laser transceiver groups are arranged on the opposite side walls of the second chamber, including laser transmitters and laser sensors arranged relatively to each other. The laser beam emitted by the laser transmitter is received by the laser sensor. The laser sensor is electrically connected to the control chip and sends the laser signal generated after receiving the laser beam to the control chip. An underwater microplastic concentration calculation module is arranged in the control chip, which is used to analyze the concentration of microplastics in the water according to the laser signals collected by multiple laser sensors and the output parameters of the water pump. The laser emitted by each laser transmitter has a different color.

[0012] A second aspect of the present invention provides an underwater microplastic sampling concentration calculation system, wherein the underwater microplastic concentration calculation module is used to perform the following steps:

[0013] S10, acquiring laser signals detected by multiple laser sensors;

[0014] S20, classifying the received laser signals according to the different laser colors used by each laser transmitter, including direct signals, refracted signals and scattered signals;

[0015] S30, filtering and denoising the three types of laser signals to obtain a net direct laser signal, a net refracted laser signal, and a net scattered laser signal;

[0016] S40, inputting the net direct laser signal into the pre-trained model to estimate the transmittance in the second chamber;

[0017] S50, inputting the net direct laser signal, the net refracted laser signal and the net scattered laser signal into a pre-trained laser-polymer calculation model to estimate the number and volume of the microplastic-bubble aggregates in the second chamber;

[0018] S60. Obtain the flow output parameters of the water pump, and calculate the microplastic concentration value per unit volume of the water sample according to the second chamber permeability, water pump flow, flow rate, and the number and volume of microplastic-bubble aggregates.

[0019] Furthermore, the method for filtering and denoising the three types of laser signals is median filtering or Gaussian filtering.

[0020] The step S10 comprises the following steps:

[0021] Step 101: Establishing a communication connection with each laser sensor through the control chip, reading and acquiring laser signal data detected by each laser sensor;

[0022] Step 102: The control chip uses an analog-to-digital converter to convert the analog signal output by each laser sensor into a digital signal, and samples and records the digital signal;

[0023] Step 103: After acquiring the signal data of all laser sensors, the control chip temporarily stores the signal data in the memory.

[0024] The step S20 comprises the following steps:

[0025] Step 201, the control chip reads the laser signal data temporarily stored in the memory, and classifies the signal data according to the laser color used by each laser emitter-sensor combination;

[0026] Step 202: Based on the principle of dispersion, using the signal intensity difference of laser signals of different wavelengths received by the sensor, all laser signal data are divided into three categories: direct signal, refracted signal and scattered signal;

[0027] Step 203: The control chip sets three initial signal arrays for temporarily storing direct signals, refracted signals and scattered signals;

[0028] Step 204, traverse the data of each laser sensor, and store the data into the corresponding direct signal array, refracted signal array and scattered signal array.

[0029] Furthermore, the method for obtaining the net direct laser signal, the net refracted laser signal and the net scattered laser signal is to first convert each signal from a time domain signal into a frequency domain signal using Fourier transform, perform filtering and denoising on the frequency domain signal, and then perform inverse Fourier transform to restore the frequency domain signal to the time domain signal.

[0030] The step S30 comprises the following steps:

[0031] Step 301: The control chip uses digital signal processing technology to filter and denoise the direct signal array, the refracted signal array, and the scattered signal array obtained in the previous step;

[0032] Step 302: For each signal array, the control chip uses a Fourier transform method to convert the time domain signal into a frequency domain signal;

[0033] Step 303: Use a bandpass filter in the frequency domain to retain the useful signal frequency components and remove the useless noise components;

[0034] Step 304: perform an inverse transform on the filtered frequency domain signal to obtain a net direct signal, a net refracted signal, and a net scattered signal after filtering.

[0035] Furthermore, the pre-training model adopts a decision tree regression model, and the training steps of the pre-training model are specifically:

[0036] Establishing a first training data set, including multiple groups of first historical experimental data, each group of the first historical experimental data includes a net direct laser signal obtained in S10-S30, and a transmittance in the second chamber corresponding to the net direct laser signal measured in actual experiments;

[0037] The training model specifically uses the first training data set to train a decision tree regression model to obtain the pre-trained model.

[0038] The step S40 comprises the following steps:

[0039] Step 401, normalizing the net direct laser signal intensity data to obtain a relative intensity value between 0 and 1;

[0040] Step 402: input the normalized direct signal intensity value and the corresponding laser emission intensity value as input into a pre-trained regression model;

[0041] Step 403: the output result of the regression model is the estimated transmittance of the second chamber;

[0042] Step 404: The regression model is trained using a machine learning algorithm such as support vector machine regression, decision tree regression or neural network, preferably a decision tree regression model.

[0043] Furthermore, the laser-polymer computing model adopts a convolutional neural network model, and the training steps of the laser-polymer computing model are specifically:

[0044] Establishing a second training data set, including multiple groups of second historical experimental data, each group of second historical experimental data includes a net direct laser signal, a net refracted laser signal, and a net scattered laser signal obtained from S10 to S30, and the number and volume of microplastic-bubble aggregates in the second chamber corresponding to the three net signals obtained by the experimental method;

[0045] The training model specifically uses the second training data set to train a convolutional neural network to obtain the laser-polymer computing model.

[0046] The step S50 comprises the following steps:

[0047] Step 501: input the net direct laser signal, the net refraction signal and the net scattered signal as input features into a pre-trained laser-polymer computational model;

[0048] Step 502: the laser-polymer computing model is a deep learning model based on a convolutional neural network;

[0049] Step 503, the convolutional neural network includes a convolution layer and a pooling layer for automatic feature extraction, and a fully connected layer for compressing features;

[0050] Step 504: The last layer of the convolutional neural network is a regression output layer, which outputs the number of aggregates and the proportion of aggregates of different volumes.

[0051] Furthermore, the method for calculating the microplastic concentration value per unit volume of the water sample is: using multiple sets of historical data to obtain the flow output parameters of the water pump, and training a neural network regression model based on the second chamber permeability, water pump flow, flow rate and the number and volume of microplastic-bubble aggregates, as well as the corresponding microplastic concentration value per unit volume of the water sample in the second chamber measured experimentally, and inputting the current water pump flow output parameters into the neural network regression model based on the second chamber permeability, water pump flow, flow rate and the number and volume of microplastic-bubble aggregates to obtain the microplastic concentration value per unit volume of the water sample in the second chamber, that is, the microplastic concentration value per unit volume of the water sample actually collected by the underwater microplastic sampling concentration calculation system.

[0052] The step S60 comprises the following steps:

[0053] Step 601, the control chip obtains the real-time flow output parameters of the water pump, including flow rate, flow rate, etc.;

[0054] Step 602: Calculate the residence time of the water sample in the second chamber according to the known size and flow rate of the second chamber;

[0055] Step 603, using the estimated values ​​of the number and volume distribution of polymers, the estimated value of the permeability, and the above-mentioned flow parameters obtained in step S50 as input features;

[0056] Step 604: input the above input features into a pre-trained neural network regression model, and the model outputs the microplastic concentration value per unit volume of the water sample.

[0057] Furthermore, it also includes S70, digitally filtering the calculated microplastic concentration value to eliminate the influence of abnormal values, obtain the final microplastic concentration result and output it.

[0058] The step S70 comprises the following steps:

[0059] Step 701: The control chip maintains a ring buffer of concentration values ​​for temporarily storing the concentration values ​​calculated in the latest N time steps;

[0060] Step 702: For each newly calculated concentration value, first store it in a ring buffer, and then execute a digital filtering algorithm on all concentration values ​​in the buffer;

[0061] Step 703: The digital filtering algorithm adopts a nonlinear filtering method such as a median filter or a Gaussian filter;

[0062] Step 704: The control chip outputs the microplastic concentration value after filtering as the final result.

[0063] Among them, the pore sizes of the six layers of filter are 5mm, 1mm, 500μm, 300μm, 300μm and 300μm from left to right.

[0064] Wherein, the number of laser transceiver groups is 8.

[0065] Compared with the prior art, the beneficial effects of the underwater microplastic sampling concentration calculation system provided by the present invention are:

[0066] 1. Realize rapid online detection of microplastic concentration in water bodies. The system directly samples and analyzes microplastics in underwater environments without the need for manual sampling and pretreatment, greatly improving detection efficiency and meeting the needs of real-time monitoring.

[0067] 2. The detection results are highly quantitative. The system combines laser signal analysis with machine learning algorithms to accurately estimate the number, volume distribution, and concentration of microplastics in water samples, providing a quantitative basis for pollution assessment and control.

[0068] 3. Simple operation and high degree of automation. The system can work by simply placing it in the waters to be tested, without complicated manual operation, which is conducive to large-scale deployment and promotion and application.

[0069] In summary, the solution of the present invention solves the technical problem that the prior art cannot realize rapid online detection of microplastic concentration in water bodies, requires manual sampling, pretreatment and analysis, and is inefficient. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0071] Figure 1 A schematic diagram of the partial composition structure of an underwater microplastic sampling concentration calculation system provided for the invention;

[0072] Figure 2 Flowchart of steps executed for underwater microplastic concentration calculation module;

[0073] In the accompanying drawings, the components represented by the reference numerals are listed as follows:

[0074] 01. Water pump; 02. Black pipe; 21. First filter; 22. Second filter; 23. Third filter; 24. Fourth filter; 25. Fifth filter; 26. Sixth filter; 31. Carbon dioxide generator; 41. Laser transmitter; 42. Laser sensor. DETAILED DESCRIPTION

[0075] In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0076] like Figure 1As shown, it is a schematic diagram of the partial composition structure of an underwater microplastic sampling concentration calculation system provided by the present invention. The system includes a water pump 01, a black tube 02 and a control chip. The output end of the water pump is connected to one end of the black tube. The first filter 21, the second filter 22, the third filter 23, the fourth filter 24, the fifth filter 25 and the sixth filter 26 are arranged in sequence in the black tube. The sixth filter is arranged at the other end of the black tube. The mesh apertures of the first filter, the second filter, the third filter and the fourth filter are gradually reduced. The mesh apertures of the fourth filter, the fifth filter and the sixth filter are the same. A first chamber and a second chamber are formed between the fourth filter, the fifth filter and the sixth filter, and a first chamber is arranged on the outer wall of the first chamber. There is a carbon dioxide generator 31, and the carbon dioxide generated by the carbon dioxide generator is introduced into the first chamber through the through hole on the outer wall of the first chamber. A plurality of laser transceiver groups are arranged on the opposite side wall of the second chamber, including a laser transmitter 41 and a laser sensor 42 arranged relatively. The laser beam emitted by the laser transmitter is received by the laser sensor, and the laser sensor is electrically connected to the control chip and sends the laser signal generated after receiving the laser beam to the control chip; an underwater microplastic concentration calculation module is arranged in the control chip, which is used to analyze and obtain the microplastic concentration in the water according to the laser signals collected by the multiple laser sensors and the output parameters of the water pump; the laser emitted by each laser transmitter has a different color. The black tube represents an opaque tube.

[0077] like Figure 2 As shown, the underwater microplastic concentration calculation module is used to perform the following steps:

[0078] S10, acquiring laser signals detected by multiple laser sensors;

[0079] S20, classifying the received laser signals according to the different laser colors used by each laser transmitter, including direct signals, refracted signals and scattered signals;

[0080] S30, filtering and denoising the three types of laser signals to obtain a net direct laser signal, a net refracted laser signal, and a net scattered laser signal;

[0081] S40, inputting the net direct laser signal intensity into the pre-trained model to estimate the transmittance in the second chamber;

[0082] S50, inputting the net direct laser signal, the net refracted laser signal and the net scattered laser signal into a pre-trained laser-polymer calculation model to estimate the number and volume of the microplastic-bubble aggregates in the second chamber;

[0083] S60, obtaining the flow output parameter of the water pump, and calculating the microplastic concentration value per unit volume of the water sample according to the second chamber permeability, the water pump flow rate, the flow rate, and the number and volume of microplastic-bubble aggregates;

[0084] S70: Digitally filter the calculated microplastic concentration value to eliminate the influence of outliers, obtain the final microplastic concentration result and output it.

[0085] Since microplastic particles have a relatively strong adsorption capacity for carbon dioxide in water, the formed microplastic-bubble aggregates will block the laser. Compared with water without microplastic particles, the blocking rate, refraction and scattering of the laser are different. Therefore, the water pump is started to extract water samples from the water body and enter the black tube; the water samples pass through the first filter, the second filter, the third filter and the fourth filter in turn to filter out particles with a diameter larger than the aperture; the carbon dioxide generator injects carbon dioxide gas into the first chamber, so that the microplastic particles and carbon dioxide form microplastic-bubble aggregates; the microplastic-bubble aggregates enter the second chamber , multiple groups of laser transceivers emit laser beams, and the polymers flowing with the water flow will block, refract, and scatter the laser beams to reach the laser sensor. The laser sensor receives the residual laser beam that passes through the polymer and sends the collected laser signal to the control chip; the underwater microplastic concentration calculation module in the control chip is based on each laser signal. Since the laser color emitted by each laser transmitter is different, the laser signal collected by each laser sensor can be distinguished, including direct laser, refracted laser, and scattered laser. The model is then used to calculate the number and volume of bubble aggregates.

[0086] The specific implementation methods of the above steps are described in detail below:

[0087] The specific implementation of step S10 is: establish a communication connection with each laser sensor through the control chip, read and obtain the laser signal data detected by each sensor. Since the laser color emitted by each laser emitter is different, different laser sensors receive laser signals of different wavelengths. The control chip uses an analog-to-digital converter to convert the analog signal output by each laser sensor into a digital signal, and samples and records the digital signal. After obtaining the signal data of all laser sensors, the control chip temporarily stores these signal data in the memory to prepare for the next step.

[0088] The specific implementation of step S20 is: the control chip reads the laser signal data temporarily stored in the memory, and classifies the signal data according to the laser color used by each laser emitter-sensor combination. Since different colors of lasers have different intensities and distribution characteristics of direct signals, refraction signals and scattered signals after interacting with microplastic-bubble aggregates, all laser signal data can be divided into three categories of direct signals, refraction signals and scattered signals based on the dispersion principle and using the signal intensity differences of laser signals of different wavelengths received by the sensor. Specifically, the control chip can set three initial signal arrays for temporarily storing these three types of signals, and then traverse the data of each laser sensor, and store the data in the corresponding array according to the wavelength range. The critical value of the wavelength can be tested in advance by testing microplastic particles of different materials, and the wavelength range of the direct, refracted and scattered light beams formed by different color lasers after interacting with them is obtained, thereby determining the critical value of the wavelength.

[0089] The specific implementation of step S30 is: for the three types of laser signal arrays obtained in the previous step, use digital signal processing technology to filter and denoise them respectively. Specifically, the control chip can use the Fourier transform method for each signal array to convert the time domain signal into a frequency domain signal, use a bandpass filter in the frequency domain to retain the useful signal frequency component, remove the useless noise component, and then perform inverse transformation to obtain the net signal after filtering. The cutoff frequency of the bandpass filter can be determined by performing spectrum analysis on the test data. For direct signals, since the optical path is relatively simple, a wider bandpass filter can be used to retain more effective signal components; for refracted signals and scattered signals, which contain more noise, a narrower bandpass filter can be used to filter out more noise components. In addition, the source of noise may also include other factors, such as light source oscillation, sensor noise, etc. At this time, it is also necessary to design a filter specifically to eliminate the influence of these noises on the final result.

[0090] The specific implementation of step S40 is: based on the processed net direct laser signal, the control chip estimates the transmittance in the second chamber by inputting a pre-trained model. Specifically, the control chip will first normalize the net direct signal intensity data to obtain a relative intensity value between 0 and 1. Then, the normalized direct signal intensity value and the corresponding laser emission intensity value are used as inputs and input into a pre-trained regression model. The model can use traditional machine learning algorithms such as support vector machine (SVM) regression and decision tree regression, or use deep learning models such as neural networks. The output result of the model is the estimated transmittance of the second chamber. The training process of the regression model is: first, a large number of second chamber samples with different known transmittances are collected, and different transmittances are simulated by artificially setting different concentrations of microplastic-bubble aggregates, and the corresponding direct signal intensity and laser emission intensity are recorded, and these data are used as training sets to train the model. In the process of model training, it is also necessary to pay attention to normalizing the input data, selecting appropriate model hyperparameters, and preventing overfitting. After training, the model can more accurately map the ratio of the direct signal intensity to the emitted laser intensity to the transmittance of the second chamber.

[0091] The specific implementation method of step S50 is: the net direct laser signal, the net refraction signal and the net scattered signal are used as input features and input into a pre-trained laser-polymer calculation model to estimate the number and volume distribution of microplastic-bubble aggregates in the second chamber. The model can be a classification and counting model based on a convolutional neural network (CNN). Specifically, the control chip will first pre-process the above three types of net signals, including normalization, feature splicing, etc. The processed features are then input into the CNN model, and the last layer of the network is a regression output layer for outputting the number of aggregates and the proportion of aggregates of different volumes.

[0092] The network structure of the CNN model can be designed as follows: the first part is the convolution layer and pooling layer for automatic feature extraction, which extracts features and reduces the dimension of the three types of signals; the second part is the fully connected layer, which compresses the extracted features and inputs them into the final regression output layer. In the model training stage, it is necessary to first collect a large number of water samples with known microplastic particle concentrations through experiments or simulations, and obtain three types of net signal data through the above process, and manually annotate them, including the number of aggregates and the proportion of aggregates of different volumes. Use these labeled data as training sets to train the CNN model. During the training process, the back propagation algorithm is used to update the network parameters until the performance of the model on the validation set meets the requirements. After the training is completed, the model can output the corresponding number and volume distribution of aggregates according to the three types of net signals.

[0093] The specific implementation of step S60 is: according to the second chamber permeability obtained in the previous step, the flow output parameters of the water pump, and the estimated number and volume distribution of microplastic-bubble aggregates, the control chip uses a certain calculation model to calculate the concentration of microplastics in the water sample. First, the control chip obtains the real-time flow output parameters of the water pump, including flow rate, flow rate, etc. Then, according to the known size and flow rate of the second chamber, the residence time of the water sample in the second chamber is calculated. Combined with the estimated value of the number and volume distribution of aggregates obtained in step S50, and the estimated value of the permeability, the control chip uses a pre-trained machine learning or deep learning model, takes the above information as input features, and outputs the microplastic concentration value per unit volume of the water sample. The model can be a regression model based on an artificial neural network, such as a feedforward neural network or a recurrent neural network.

[0094] In the model training stage, it is necessary to first collect a large number of water samples with known microplastic concentrations through experiments or simulations, and process these water samples according to the above steps to obtain characteristic data such as transmittance, number of aggregates and volume distribution, as well as the actual microplastic concentration value measured manually, as labels, and use these data as training sets to train the neural network model. During the training process, a standard back propagation algorithm can be used, and the difference between the concentration prediction value output by the network and the actual label value can be used as the loss function. The network parameters are continuously updated through the gradient descent method, so that the performance of the model on the validation set is continuously improved until the expected accuracy requirements are met. After sufficient training, the model can accurately output the concentration value of microplastics in water samples based on characteristics such as transmittance and aggregate distribution.

[0095] The specific network structure of the neural network model can be designed as follows: the input layer receives the above-mentioned feature data, followed by several layers of fully connected hidden layers for feature extraction and nonlinear mapping, and the last layer is a regression output layer that outputs the predicted microplastic concentration value. In the hidden layer, different activation functions, such as ReLU, Sigmoid, etc., can be used to enhance the nonlinear fitting ability of the model. In addition, in order to prevent overfitting, dropout layers, L1 / L2 regularization and other technologies can be introduced into the network. During the network training process, in addition to monitoring the convergence of the loss function, it is also necessary to pay attention to tuning the hyperparameters (such as learning rate, number of hidden layer nodes, etc.) to obtain the best model performance.

[0096] The specific implementation of step S70 is: digitally filter the microplastic concentration value calculated in the previous step to eliminate abnormal values ​​caused by measurement errors or other interference factors, so as to obtain a smoother and more stable final concentration result output. Specifically, the control chip can first maintain a ring buffer of concentration values ​​for temporarily storing the concentration values ​​calculated in the last N time steps. For each newly calculated concentration value, first store it in the ring buffer, and then execute the digital filtering algorithm on all concentration values ​​in the buffer.

[0097] The digital filtering algorithm can use nonlinear filtering methods such as median filtering or Gaussian filtering. The principle of median filtering is to sort the concentration values ​​in the buffer by size and take the middle value as the output, which can effectively remove the influence of outliers. Gaussian filtering is to perform weighted averaging on the data in the buffer, and the weight is assigned according to the value of the Gaussian kernel function, which can smooth the noise while retaining the shape characteristics of the data.

[0098] In the specific implementation, a reasonable filter window size N can be set to balance the smoothing effect and response time of the filter. For example, N can be set to a value between 10 and 30. In addition, the filter output result can be further processed as needed, such as limiting, smoothing, etc., to further improve the stability and reliability of the output result. Finally, the control chip outputs the microplastic concentration value after filtering as the final result for subsequent analysis and display.

[0099] In order to better implement the present invention, the specific implementation of the underwater microplastic concentration calculation module is described in more detail below in combination with a specific formula:

[0100] The specific implementation of step S10 is as follows:

[0101] Suppose there are m laser emitter-sensor pairs, and the laser wavelength of the i-th emitter-sensor pair is λ i , the transmission power is The control chip establishes a connection with each laser sensor through a serial communication interface. For the i-th laser sensor, the control chip collects the analog signal data y at n time points. i (t j ), where j = 1, 2, ..., n, the analog signal is digitized by an analog-to-digital converter as follows:

[0102] y i [j] = Q(y i (t j ))

[0103] Where Q(·) represents the quantization function, which can be based on uniform quantization or non-uniform quantization. i [j] is the value of the ith sensor at time tj The received digitized laser signal intensity. The digital signal data of all sensors are temporarily stored in the control chip memory, waiting for further processing.

[0104] The specific implementation of step S20 is as follows:

[0105] For a wavelength of λ i The laser signal, according to the law of light dispersion, when interacting with the microplastic-bubble aggregate, the wavelength of the direct beam remains unchanged at λ i ; The wavelengths of the refracted beam and the scattered beam will shift to a certain extent, which are recorded as and And there are:

[0106]

[0107] Among them, n p and n m are the refractive indices of microplastics and medium (water), respectively, r and f s is a known optical function.

[0108] According to the above principle, the control chip can set three initial arrays D, R and S to store direct signals, refracted signals and scattered signals respectively. Calculate its power spectrum P i [k]:

[0109]

[0110] Find the wavelength index k0 of the main energy concentration, that is Will According to the corresponding wavelength Stored in the D, R or S array respectively.

[0111] The specific implementation of step S30 is as follows:

[0112] For each signal sequence in the direct signal array D, the refracted signal array R and the scattered signal array S The control chip can use the following filtering method to remove noise:

[0113] 1) Calculate the discrete Fourier transform (DFT) coefficients of the sequence:

[0114]

[0115] 2) Filter the transform coefficient X[k] through a bandpass filter:

[0116]

[0117] Among them, f l and f h is the lower and upper frequency of the bandpass filter. For direct signals, f l and f h Can take a larger value; for refracted and scattered signals, f l and f h A smaller value can be used to filter out more noise components.

[0118] 3) The transformed coefficients after filtering Do the inverse Fourier transform:

[0119]

[0120] but is the net signal sequence after filtering.

[0121] The specific implementation of step S40 is as follows:

[0122] For each net direct signal sequence The control chip calculates its relative power:

[0123]

[0124] Normalize it to relative intensity in the interval [0,1]:

[0125]

[0126] in, is the emission power of the corresponding laser transmitter.

[0127] Will As input, substitute the pre-trained regression model The transmittance τ of the second chamber can be estimated:

[0128]

[0129] The regression model Support vector regression (SVR), decision tree regression or artificial neural network can be used

[0130] Suppose there are N sample data with known transmittance. Then the loss function of the model can be defined as:

[0131]

[0132] in, is a suitable loss function, such as squared loss or absolute loss, etc. Minimize the loss function through an optimization algorithm (such as gradient descent) The optimal model parameters can be obtained.

[0133] The specific implementation of step S50 is as follows:

[0134] Assume that the oth net direct signal sequence is The oth net refraction signal sequence is The i-th net scattered signal sequence is Concatenate these three sequences into a feature vector:

[0135]

[0136] All m eigenvectors As input, feed a pre-trained convolutional neural network (CNN) model

[0137]

[0138] in, is the output of the model, indicating the number of aggregates corresponding to the i-th laser beam and the proportion of aggregates of different sizes Right now

[0139] The CNN model The network structure can be designed as follows: the convolution layer automatically extracts features from the input feature sequence, and the pooling layer performs downsampling; the fully connected layer compresses the extracted features and inputs them into the final regression output layer. Suppose there are M labeled training samples where y (j) is the real number and size distribution label of the jth sample. Then the loss function of the CNN model can be defined as:

[0140]

[0141] in, is a suitable loss function, such as mean square error loss, etc. Use the back propagation algorithm and optimizer (such as Adam) to minimize the loss function The optimal model parameters can be obtained.

[0142] The specific implementation of step S60 is as follows:

[0143] Assume that the real-time flow output parameter of the water pump is Q, and the volume of the second chamber is V, then the residence time of the water sample in the second chamber is T = V / Q. Assume that the estimated transmittance obtained in step S40 is The estimated value of the number of aggregates corresponding to the i-th laser beam obtained in step S50 is The aggregate size distribution was estimated to be

[0144] The control chip inputs these estimated values ​​and the pump parameter Q into another pre-trained neural network regression model middle:

[0145]

[0146] in, This is the estimated value of microplastic concentration per unit volume of water sample output by the model.

[0147] The neural network model A feedforward neural network or a recurrent neural network can be used. The input layer receives the above features, the hidden layer performs nonlinear mapping on the features, and the last layer is a regression output layer that outputs the microplastic concentration value. Suppose there are L labeled training samples where x (l) is the input feature of the lth sample, including transmittance, polymer information and flow parameters; ρ (l) is the true microplastic concentration value of the sample. Then the loss function of the model can be defined as:

[0148]

[0149] in, is a suitable loss function, such as squared loss or absolute loss, etc. Use the back propagation algorithm and optimizer (such as Adam) to minimize the loss function The optimal model parameters can be obtained.

[0150] The specific implementation of step S70 is as follows:

[0151] Assume that the buffer length is w, that is, the concentration values ​​calculated in the last w time steps are temporarily stored in the buffer. The estimated concentration value corresponding to the t-th time step is The data in the buffer is

[0152] 1) Median filtering:

[0153] Sort the data in the buffer to get an ordered sequence in The output after median filtering is:

[0154]

[0155] 2) Gaussian filtering:

[0156] Define a one-dimensional Gaussian kernel function:

[0157]

[0158] Where σ is the standard deviation parameter of the Gaussian kernel, which controls the smoothness of the kernel function.

[0159] For data in the buffer Normalize it so that it is in the interval [0,1]:

[0160]

[0161] The normalized data is used as the independent variable of the Gaussian kernel function, and its weighted sum is calculated as the filtering output:

[0162]

[0163] After Gaussian filtering, This is the estimated value of the concentration after smoothing. The standard deviation parameter σ of the filter determines the degree of smoothing and can be adjusted according to actual needs.

[0164] Finally, the control chip will smooth the result Output as the final microplastic concentration.

[0165] The training process of the laser-polymer computing model is as follows:

[0166] 1. Laser-polymer computational model Training:

[0167] - Collect labeled training samples where z (j) is the net direct / refracted / scattered signal sequence of the jth sample, concatenated into a feature vector; (j) are the corresponding aggregate number and size distribution labels.

[0168] -initialization Model parameters, including convolution kernel weights, biases, etc.

[0169] -Define the loss function in The mean square error loss can be used.

[0170] -Use optimization algorithms (such as Adam) to optimize the loss function Minimize and update the network parameters through back propagation.

[0171] -Evaluate model performance on the validation set and adjust hyperparameters (such as learning rate, etc.) as needed until the model performance meets the requirements.

[0172] 2. Microplastic Concentration Regression Model Training:

[0173] - Collect labeled training samples where x (l) Contains transmittance estimates Aggregate Information and And the flow parameter Q; ρ (l) is the corresponding actual microplastic concentration value.

[0174] -initialization Parameters of the model.

[0175] -Define the loss function in Square loss, absolute loss, etc. can be used.

[0176] -Use optimization algorithms (such as Adam) to optimize the loss function Minimize and update the network parameters through back propagation.

[0177] -Evaluate model performance on the validation set and adjust hyperparameters as needed until the model performance meets the requirements.

[0178] During the above training process, the following points should be noted:

[0179] -The quality of training data has a great impact on model performance, and it is necessary to ensure the coverage and annotation accuracy of training samples.

[0180] -Proper preprocessing of input data (such as normalization, etc.) can help improve the model convergence speed.

[0181] - Properly set the network structure (number of layers, number of nodes, etc.) and hyperparameters (such as learning rate, regularization strength, etc.) to prevent overfitting or underfitting.

[0182] -Techniques such as data augmentation, early stopping, and model integration can be used to further improve the generalization ability of the model.

[0183] Specifically, the principle of the present invention is to use the interaction between laser beams and microplastic particles, combined with advanced signal processing and artificial intelligence algorithms, to accurately assess the concentration of microplastics in water bodies.

[0184] Specifically, the system works as follows:

[0185] 1. Optical principle: When the laser beam hits the microplastic particles, optical phenomena such as reflection, refraction and scattering will occur. Since the refractive index and reflectivity of microplastics (if light is transmitted) and water are different, the microplastic particles will block, refract or reflect the laser.

[0186] 2. Bubble Aggregation Principle: The surface of microplastic particles is hydrophobic to a certain extent, and they are easy to adsorb gas molecules in the air, especially carbon dioxide. When the water sample is in contact with carbon dioxide gas, the microplastic particles will form aggregates with the bubbles. The volume of these aggregates is significantly larger than that of a single microplastic particle, and the shielding, refraction and scattering effects on the laser beam will also be more obvious. Using this principle, the effect of microplastic particles on the laser signal can be enhanced, and the detection sensitivity can be improved.

[0187] 3. Principle of optical signal processing: After the laser beam interacts with the microplastic-bubble aggregate, it will generate direct light, refracted light and scattered light. After these light beams are received by the laser sensor, they form different optical signals. By performing digital signal processing such as Fourier transform and filtering on the received optical signal, the direct signal, refracted signal and scattered signal can be separated, and they can be denoised to obtain a net signal.

[0188] 4. Transmittance estimation principle: The ratio of the intensity of the net direct signal to the laser emission power reflects the degree of attenuation of the light beam when it passes through the second chamber, that is, the transmittance. Using the pre-trained regression model, the transmittance can be estimated based on the net direct signal intensity and the emission power. The size of the transmittance is closely related to the number and volume distribution of polymers in the water sample.

[0189] 5. Principle of aggregate analysis: Net direct signal, net refraction signal and net scattering signal carry information about microplastic-bubble aggregates. By inputting these signals and using artificial intelligence models such as deep convolutional neural networks for feature extraction and regression analysis, the number of aggregates and the proportion of aggregates of different sizes can be estimated.

[0190] 6. Principle of concentration calculation: Finally, the estimated permeability, number of aggregates, size distribution and other information are combined with the flow parameters of the water pump and input into another pre-trained neural network regression model to calculate the microplastic concentration value per unit volume of the water sample. The model is trained with a large amount of experimental data and can learn the mapping relationship between various factors and microplastic concentration.

[0191] 7. Filtering and smoothing principle: Due to some random errors and interference noise in the measurement process, the concentration value directly output may fluctuate greatly. By performing digital filtering (such as median filtering, Gaussian filtering, etc.) on the concentration values ​​calculated within continuous time steps, the influence of outliers can be effectively eliminated, making the final output concentration result smoother and more stable.

[0192] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. An underwater microplastic sampling concentration calculation system, characterized in that: The invention comprises a water pump, a black tube and a control chip, wherein the output end of the water pump is connected to one end of the black tube, and the first filter screen, the second filter screen, the third filter screen, the fourth filter screen, the fifth filter screen and the sixth filter screen are sequentially arranged in the black tube, and the sixth filter screen is arranged at the other end of the black tube, and the mesh apertures of the first filter screen, the second filter screen, the third filter screen and the fourth filter screen are gradually reduced, and the mesh apertures of the fourth filter screen, the fifth filter screen and the sixth filter screen are the same, and a first chamber and a second chamber are formed between the fourth filter screen, the fifth filter screen and the sixth filter screen, and a carbon dioxide generator is arranged on the outer wall of the first chamber, and the carbon dioxide generated by the carbon dioxide generator passes through the The through hole on the outer wall of the first chamber leads into the first chamber, and a plurality of laser transceiver groups are arranged on the opposite side wall of the second chamber, including laser emitters and laser sensors arranged relatively, the laser beam emitted by the laser emitter is received by the laser sensor, and the laser sensor is electrically connected to the control chip and sends the laser signal generated after receiving the laser beam to the control chip; an underwater microplastic concentration calculation module is arranged in the control chip, which is used to analyze and obtain the microplastic concentration in the water according to the laser signals collected by the plurality of laser sensors and the output parameters of the water pump; the laser color emitted by the laser emitter of each laser transceiver group is different; Wherein, the underwater microplastic concentration calculation module is used to perform the following steps: S10, acquiring laser signals detected by multiple laser sensors; S20, classifying the received laser signals according to the different laser colors used by each laser transmitter, including direct signals, refracted signals and scattered signals; S30, filtering and denoising the three types of laser signals to obtain a net direct laser signal, a net refracted laser signal, and a net scattered laser signal; S40, inputting the net direct laser signal into the pre-trained model to estimate the transmittance in the second chamber; S50, inputting the net direct laser signal, the net refracted laser signal and the net scattered laser signal into a pre-trained laser-polymer calculation model to estimate the number and volume of the microplastic-bubble aggregates in the second chamber; S60, obtaining the flow output parameter of the water pump, and calculating the microplastic concentration value per unit volume of the water sample according to the second chamber permeability, the water pump flow rate, the flow rate, and the number and volume of microplastic-bubble aggregates; Among them, the transmittance estimation principle is: the ratio of the intensity of the net direct signal to the laser emission power reflects the degree of attenuation of the light beam when passing through the second chamber, that is, the transmittance.

2. An underwater microplastic sampling concentration calculation system according to claim 1, characterized in that: The method for filtering and denoising the three types of laser signals is median filtering or Gaussian filtering.

3. An underwater microplastic sampling concentration calculation system according to claim 2, characterized in that: The method for obtaining the net direct laser signal, the net refracted laser signal and the net scattered laser signal is to first convert each signal from a time domain signal to a frequency domain signal by Fourier transform, perform filtering and denoising on the frequency domain signal, and then perform inverse Fourier transform to restore the frequency domain signal to the time domain signal.

4. An underwater microplastic sampling concentration calculation system according to claim 2, characterized in that: The pre-training model adopts a decision tree regression model, and the training steps of the pre-training model are specifically: Establishing a first training data set, including multiple groups of first historical experimental data, each group of the first historical experimental data includes a net direct laser signal obtained in S10-S30, and a transmittance in the second chamber corresponding to the net direct laser signal measured in actual experiments; The training model specifically uses the first training data set to train a decision tree regression model to obtain the pre-trained model.

5. An underwater microplastic sampling concentration calculation system according to claim 2, characterized in that: The laser-polymer computing model adopts a convolutional neural network model, and the training steps of the laser-polymer computing model are specifically: Establishing a second training data set, including multiple groups of second historical experimental data, each group of second historical experimental data includes a net direct laser signal, a net refracted laser signal, and a net scattered laser signal obtained from S10 to S30, and the number and volume of microplastic-bubble aggregates in the second chamber corresponding to the three net signals obtained by the experimental method; The training model specifically uses the second training data set to train a convolutional neural network to obtain the laser-polymer computing model.

6. An underwater microplastic sampling concentration calculation system according to claim 2, characterized in that: The method for calculating the microplastic concentration value per unit volume of the water sample is as follows: using multiple sets of historical data to obtain the flow output parameters of the water pump, training a neural network regression model based on the second chamber permeability, water pump flow, flow rate and the number and volume of microplastic-bubble aggregates, as well as the corresponding microplastic concentration value per unit volume of the water sample in the second chamber measured experimentally, and inputting the current water pump flow output parameters into the neural network regression model based on the second chamber permeability, water pump flow, flow rate and the number and volume of microplastic-bubble aggregates to obtain the microplastic concentration value per unit volume of the water sample in the second chamber, that is, the microplastic concentration value per unit volume of the water sample actually collected by the underwater microplastic sampling concentration calculation system.

7. An underwater microplastic sampling concentration calculation system according to claim 2, characterized in that: The method also includes S70, which performs digital filtering on the calculated microplastic concentration value to eliminate the influence of abnormal values, obtain the final microplastic concentration result and output it.

8. An underwater microplastic sampling concentration calculation system according to claim 1, characterized in that: The pore sizes of the six-layer filter are 5mm, 1mm, 500μm, 300μm, 300μm and 300μm from left to right.

9. An underwater microplastic sampling concentration calculation system according to claim 1, characterized in that: The number of the laser transceiver groups is 8.

Images