Multi-metal ion mixture detection method

Abstract

The invention discloses a multi-metal ion mixture detection method which comprises the following steps: complexing a plurality of chelating agents with metal ions to form a metal ion mixture with different spectral responses; the method comprises the following steps: collecting spectral data of a metal ion mixture, and constructing a training data set comprising various concentration combinations; constructing a neural network detection model, and inputting the training data set into the neural network detection model to train a network model to obtain a trained network model; and inputting a to-be-detected detection sample into the trained network model, and outputting the concentration of each ion. The multi-metal ion mixture can be quantitatively detected in real time at the same time, the sensitivity is high, and the anti-interference performance is high.

Description

Technical Field

[0001] The present invention relates to the field of environmental monitoring, and particularly to a method for detecting a mixture of multiple metal ions. Background Art

[0002] At present, although traditional techniques such as atomic absorption spectrometry (AAS), inductively coupled plasma mass spectrometry (ICP-MS), and ion chromatography (IC) have high accuracy, their dependence on complex instruments and time-consuming sample preparation limits their applicability to real-time on-site analysis.

[0003] Currently, although X-ray fluorescence (XRF) provides a fast and non-destructive method for simultaneously detecting multiple heavy metals, its trace detection sensitivity is limited. Although portable sensors are suitable for on-site detection of multiple metal ions, in particular, fluorescence and colorimetric arrays show great potential in single analyte detection, but there are still major challenges in simultaneously quantifying mixtures of coexisting metal ions, mainly due to overlapping spectral signals, non-linear responses, and interference from shared optical properties (such as overlapping emission bands) in complex matrices.

[0004] Therefore, to solve the above problems, a method for detecting a mixture of multiple metal ions is needed, which can simultaneously quantify and detect the mixture of multiple metal ions in real time, with high sensitivity and strong anti-interference ability. Summary of the Invention

[0005] In view of this, the object of the present invention is to overcome the defects in the prior art and provide a method for detecting a mixture of multiple metal ions, which can simultaneously quantify and detect the mixture of multiple metal ions in real time, with high sensitivity and strong anti-interference ability.

[0006] The method for detecting a mixture of multiple metal ions of the present invention includes:

[0007] Complexing multiple chelating agents with metal ions to form a metal ion mixture with different spectral responses;

[0008] Collecting spectral data of the metal ion mixture and constructing a training data set including multiple concentration combinations;

[0009] Constructing a neural network detection model, inputting the training data set into the neural network detection model for network model training to obtain a trained network model;

[0010] Inputting the detection sample to be tested into the trained network model and outputting the concentrations of each ion.

[0011] Further, forming a metal ion mixture with different spectral responses specifically includes:

[0012] Mix metal ions with at least one chelating agent among a variety of them in a specific molar ratio, let it stand and then mix it with silver nanoparticles and an NaCl solution to form a complex; wherein, the chelating agent includes PAN, TPY, and Phen.

[0013] Furthermore, collect spectral data of the metal ion mixture, specifically including:

[0014] Use a laser to excite the Raman scattering of the metal ion mixture, and use an objective lens to focus the light onto the metal ion mixture for surface-enhanced Raman spectroscopy acquisition, and obtain several spectra for each sample.

[0015] Furthermore, the neural network detection model includes an initial convolutional layer, a pooling layer, a residual block group, and a fully connected layer;

[0016] The initial convolutional layer is used to perform spectral data preprocessing and extract primary spatial spectral features;

[0017] The pooling layer is used to reduce the feature dimension through downsampling and enhance the robustness of local features;

[0018] The residual block group includes several consecutive residual blocks. Adjacent residual blocks share convolutional kernel parameters, and the risk of overfitting is reduced through parameter reuse. Among them, each residual block includes a convolutional layer, batch normalization, ReLU activation, and skip connection;

[0019] Set a self-attention module at the output end of the residual block group. The self-attention module is used to calculate the long-range correlation of the feature map and output the reconstructed feature; and determine the high-dimensional feature X through the following formula out :

[0020] X out = X in + λ·Z; where X in is the feature input to the self-attention module, λ is the scaling factor, and Z is the reconstructed feature;

[0021] The fully connected layer is used to map the high-dimensional feature to the output space of the metal ion concentration and / or category.

[0022] Furthermore, from the initial convolutional layer to the pooling layer and then to the residual block group, a feature pyramid from bottom to top is formed, and spectral feature representations are established through layer-by-layer abstraction.

[0023] Furthermore, calculate the attention weight matrix in the self-attention module, and optimize the attention weight matrix through backpropagation, so that the neural network detection model automatically strengthens the feature response to the spectral overlap region and interference peaks during the training process.

[0024] Furthermore, calculate the attention weight matrix Attention(Q,K) according to the following formula:

[0025]

[0026] Among them, τ is a set coefficient, and the adaptive adjustment of the feature focusing granularity is realized through the gradient update of the set coefficient τ.

[0027] The beneficial effects of the present invention are as follows: A detection method for a multi-metal ion mixture disclosed by the present invention forms complexes with different spectral responses through the complexation of multiple chelating agents with metal ions, expanding the feature dimension; uses surface-enhanced Raman spectroscopy (SERS) technology to collect high-dimensional spectral data of the metal ion mixture, and constructs a training data set including multiple concentration combinations; adopts a residual network (ResNet) architecture, introduces a self-attention mechanism to dynamically adjust the attention to key spectral features, solves the problems of spectral overlap and non-linear interference, and realizes the simultaneous quantitative and accurate detection of multiple metal ions. Brief Description of the Drawings

[0028] The present invention will be further described below in conjunction with the drawings and embodiments:

[0029] Figure 1 It is a schematic flow chart of the metal ion mixture detection method of the present invention;

[0030] Figure 2 It is a schematic diagram of the concentration composition of 84 kinds of mixed metal ion samples of the present invention;

[0031] Figure 3 It is a simplified schematic diagram of the network model framework for the quantitative analysis of metal ion mixtures of the present invention;

[0032] Figure 4 It is a violin plot showing the prediction error distribution of each metal ion of the present invention. Detailed Embodiments

[0033] The following further describes the present invention in conjunction with the drawings of the specification, as shown in the figure:

[0034] This embodiment discloses a detection method for a multi-metal ion mixture, including the following steps:

[0035] S1. Complex multiple chelating agents with metal ions to form a metal ion mixture with different spectral responses;

[0036] S2. Collect spectral data of the metal ion mixture and construct a training data set including multiple concentration combinations;

[0037] S3. Construct a neural network detection model, input the training data set into the neural network detection model for training of the network model, and obtain a trained network model;

[0038] S4. Input the test sample to be detected into the trained network model, and output the concentrations of various ions.

[0039] The present invention provides a method capable of simultaneously quantitatively detecting a mixture of multiple metal ions, solving the problems of signal overlap, insufficient sensitivity, and inability to perform real-time on-site analysis in the prior art, and is particularly suitable for the rapid monitoring of heavy metal ions in complex water environments.

[0040] In this embodiment, in step S1, multiple chelating agents are complexed with metal ions to form a metal ion mixture with different spectral responses, specifically including:

[0041] Mix the metal ions with at least one of the multiple chelating agents in a specific molar ratio, let it stand and then mix with silver nanoparticles and an NaCl solution to form a complex; the chelating agents include three kinds: PAN, TPY, and Phen. Among them, the molar ratio is 1:2 or 1:3.

[0042] After complexing with metal ions, different chelating agents will exhibit unique electronic structures and molecular configurations, thus causing significantly different spectral responses, such as absorption, fluorescence, or Raman signals. This makes the mixture show diverse characteristics in the spectrum, which helps to enhance the discrimination of metal ions and effectively alleviate the problem of signal overlap in multi-component detection. Moreover, the combined action of multiple chelating agents can cover a wider range of metal ion types and concentration ranges, thereby improving the detection sensitivity and selectivity. The differences in the affinity of different metal ions for various chelating agents will be reflected in the response intensity and spectral shape, and then be captured by the neural network model for efficient identification and quantification.

[0043] In addition, through the combined action with silver nanoparticles (SERS enhancement substrate) and NaCl, the chelated composite system can further enhance the Raman signal, which is beneficial to achieving rapid, sensitive, and on-site operable detection in a complex background. Therefore, a multi-dimensional spectral feature is formed by using multiple chelating agents, expanding the feature dimension and constructing a robust and accurate metal ion detection model.

[0044] In this embodiment, in step S2, collect the spectral data of the metal ion mixture, specifically including:

[0045] Use a 532 nm laser to excite the Raman scattering of the metal ion mixture, and use a 50× objective lens (NA = 0.60) to focus the light on the metal ion mixture for surface-enhanced Raman spectroscopy collection. 50 spectra are obtained for each sample.

[0046] With the above settings, the 532 nm laser wavelength is in the visible light range, which can effectively excite the Raman scattering signals of most metal chelates. Especially when these complexes form a SERS active substrate with silver nanoparticles, the signal intensity of characteristic peaks can be significantly enhanced, and the signal-to-noise ratio can be improved. This laser wavelength also has the advantages of high excitation efficiency and low fluorescence background, which helps to obtain clear spectral data. Using a 50× objective lens (numerical aperture NA = 0.60) can achieve high spatial resolution and laser focusing accuracy, accurately concentrate the laser energy in a small area of the sample, and maximize the enhancement effect in the local hot spot area, thereby further improving the detection sensitivity.

[0047] Collecting 50 spectral maps for each sample is conducive to multi-angle, multi-region and multiple repeated observations of the spectral response of the sample, reducing data deviation, and enhancing the robustness and generalization ability of model training. This high-repeatability data acquisition method can effectively reflect the overall characteristics of the sample, and then improve the prediction stability and accuracy of the subsequent neural network model.

[0048] Using three chelating agents to generate a spectral database of Cu 2+ , Fe 3+ , Ni 2+ , Pb 2+ mixed solution, as Figure 2 shown. Each ion will cause different response patterns, and the data of multiple chelating agents × multiple wavelengths constitute a high-dimensional data array. Each photo has information in multiple dimensions, including color, intensity, and position.

[0049] In this embodiment, in step S3, the neural network detection model includes an initial convolutional layer, a pooling layer, a residual block group, and two fully connected layers;

[0050] The initial convolutional layer is used to perform spectral data preprocessing and extract primary spatial spectral features;

[0051] The pooling layer is used to reduce the feature dimension through downsampling and enhance the robustness of local features;

[0052] The residual block group includes six consecutive residual blocks. Adjacent residual blocks share convolutional kernel parameters to reduce the risk of overfitting through parameter reuse. Among them, each residual block includes a convolutional layer, batch normalization, ReLU activation, and skip connection;

[0053] A self-attention module is set at the output end of the residual block group. The self-attention module is used to calculate the long-range correlation of the feature map and output the reconstructed features; and determine the high-dimensional feature X out :

[0054] Xout = X in + λ·Z; where X in is the feature input into the self-attention module; λ is the scaling factor, and its value can be set according to the actual working conditions; Z is the reconstructed feature;

[0055] The fully connected layer is used to map the high-dimensional features to the output space of the metal ion concentration and / or category.

[0056] Through the above settings, the initial convolutional layer can perform local feature extraction and preprocessing on the spectral signal, capture local patterns in the wavelength dimension, such as the positions, intensities, and change trends of characteristic peaks, which is beneficial for distinguishing different metal ions. The pooling layer effectively compresses the data dimension through downsampling operations, reduces the computational amount while retaining key information, thereby improving the robustness of the model to local perturbations and noises.

[0057] The introduction of the residual block group significantly increases the network depth while avoiding the problems of gradient vanishing and degradation in deep networks. The residual connection allows the model to learn the residual mapping between the input and output, which can accelerate the convergence speed and improve the generalization ability. The structure of six consecutive residual blocks deepens the expression ability of the network, and the parameter sharing strategy between adjacent blocks effectively reduces the model complexity and the risk of overfitting.

[0058] Furthermore, a self-attention module is introduced after the residual block group. By calculating the long-range dependence relationships between different bands, the self-attention mechanism can model global information interactions, make up for the deficiency that the convolutional layer can only perceive locally, and enable the model to have a stronger expression ability for the complex response patterns that may exist between different metal ions. Its output is introduced into the weighted fusion of the original information and the global reconstructed feature in the form of residuals, which not only retains the stability of the original features but also injects global context information, further improving the detection accuracy and stability.

[0059] The fully connected layer maps the extracted deep high-dimensional features to the output space of the metal ion concentration and / or category, realizing the final quantitative prediction or classification judgment. The overall network structure takes into account both local detail capture and global information modeling, is suitable for processing multi-component, mixed, and overlapping spectral data, and has obvious advantages in the high-precision identification and quantitative detection of metal ions in complex water samples.

[0060] In this embodiment, from the initial convolutional layer to the pooling layer and then to the residual block group, a feature pyramid is formed from bottom to top, and spectral feature representations are established through layer-by-layer abstraction.

[0061] The "bottom-up feature pyramid" structure constructed from the initial convolutional layer to the pooling layer and then to the residual block group realizes the hierarchical abstraction and multi-scale expression of spectral features. The initial convolutional layer is responsible for extracting local features at the bottom layer, such as the positions and intensities of characteristic peaks; the pooling layer retains key patterns while compressing data, improving the model's tolerance to noise; the residual block group further extracts complex spectral structures at the middle and high levels, achieving deeper feature representations. This pyramid-like structure facilitates the model to understand the fine-grained differences and global trends of spectral data at different scales, thereby improving the accuracy and robustness of metal ion identification and quantitative prediction.

[0062] In this embodiment, the attention weight matrix is calculated in the self-attention module, and the attention weight matrix is optimized through backpropagation, enabling the neural network detection model to automatically strengthen the feature response to the spectral overlap region and interference peaks during the training process.

[0063] The self-attention module is continuously optimized by calculating the attention weight matrix and using the backpropagation mechanism during the training process, effectively enhancing the sensitivity and recognition ability of the neural network to the spectral overlap region and interference peaks. Compared with the traditional convolutional structure that mainly relies on local receptive fields, the self-attention mechanism can model the dependencies between global features, enabling the model to still capture key difference signals in complex spectral backgrounds. Through the continuously optimized attention weights, the network can automatically allocate more attention to information-dense or key interference regions, thereby improving the ability to distinguish overlapping peaks and the robustness to noise, and enhancing the accuracy and stability of multi-metal ion mixture detection.

[0064] In this embodiment, the adaptive adjustment of the feature focusing granularity is achieved by setting the gradient update of the coefficient τ, and the attention weight matrix Attention(Q,K) is calculated according to the following formula:

[0065]

[0066] where τ is a set coefficient, and its value can be set according to actual working conditions; the feature map generates the Query / Key / Value triple (Q,K,V), Q is the query matrix, and K is the key matrix; d k is the dimension of the key matrix; softmax() acts on each row to make the attention distribution of each Query a probability, and the sum of the probabilities is 1.

[0067] As Figure 3 shown, the neural network detection model constructed in the present invention is trained using the training data set to obtain a trained network model. In step S4, the detection sample to be measured is input into the trained network model, and the concentrations of each ion are output. For example, for blind testing of real water samples, the model outputs the prediction errors of the concentrations of each ion, and the errors follow a normal distribution as Figure 4 shown.

[0068] The present invention can simultaneously and quantitatively detect Cu 2+ , Fe 3+ , Ni 2+ , Pb 2+ in a quaternary metal ion mixture with an average relative error < 10%. Through a high-dimensional spectral array and a self-attention mechanism, the influence of spectral overlap and matrix interference is significantly reduced. Combining with the rapid detection characteristics of SERS, it is applicable to on-site environmental monitoring. The model can be adapted to more metal ion species, providing a general solution for complex water pollution scenarios.

[0069] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. A method for detecting a mixture of multiple metal ions, characterized in that: Including: Complexing multiple chelating agents with metal ions to form a metal ion mixture with different spectral responses; Collecting spectral data of the metal ion mixture and constructing a training data set including multiple concentration combinations; Constructing a neural network detection model, inputting the training data set into the neural network detection model for network model training to obtain a trained network model; Inputting the detection sample to be measured into the trained network model and outputting the concentrations of various ions.

2. The method for detecting a multi-metal ion mixture according to claim 1, wherein: Forming a metal ion mixture with different spectral responses, specifically including: Mixing metal ions with at least one chelating agent among multiple ones in a specific molar ratio, standing and then mixing with silver nanoparticles and an NaCl solution to form a complex; wherein, the chelating agent includes PAN, TPY, and Phen.

3. The method for detecting a multi-metal ion mixture according to claim 1, characterized in that: Collecting spectral data of the metal ion mixture, specifically including: Using a laser to excite the Raman scattering of the metal ion mixture, using an objective lens to focus the light on the metal ion mixture for surface-enhanced Raman spectroscopy collection, and obtaining several spectra for each sample.

4. The method for detecting a multi-metal ion mixture according to claim 1, characterized in that: The neural network detection model includes an initial convolutional layer, a pooling layer, a residual block group, and a fully connected layer; The initial convolutional layer is used to perform spectral data preprocessing and extract primary spatial spectral features; The pooling layer is used to reduce the feature dimension through downsampling and enhance the robustness of local features; The residual block group includes several consecutive residual blocks. Adjacent residual blocks share convolutional kernel parameters, and the risk of overfitting is reduced through parameter reuse. Among them, each residual block includes a convolutional layer, batch normalization, ReLU activation, and skip connection; An self-attention module is arranged at the output end of the residual block group, and the self-attention module is used to calculate the long-range correlation of the feature map and output a reconstructed feature; and a high-dimensional feature X is determined by the following formula out :[[]]END]] X out = X in + λ · Z; where, X in is the feature input to the self-attention module, λ is the scaling factor, and Z is the reconstructed feature; The fully connected layer is used to map high-dimensional features to the output space of metal ion concentrations and / or categories.

5. The method for detecting a multi-metal ion mixture according to claim 4, characterized in that: From the initial convolutional layer to the pooling layer and then to the residual block group, a feature pyramid from bottom to top is formed, and spectral feature representations are established through layer-by-layer abstraction.

6. The method for detecting a multi-metal ion mixture according to claim 4, wherein: Calculating an attention weight matrix in the self-attention module, and optimizing the attention weight matrix through backpropagation, so that the neural network detection model automatically strengthens the feature response to the spectral overlap region and interference peaks during the training process.

7. The method for detecting a multi-metal ion mixture according to claim 6, wherein: Calculating the attention weight matrix Attention(Q,K) according to the following formula: where τ is a set coefficient, and the adaptive adjustment of the feature focusing granularity is realized through the gradient update of the set coefficient τ.

Images