A spectral multi-metal ion concentration detection method based on temperature modulation

Abstract

The present invention discloses a method for detecting the concentration of multiple metal ions using a spectrum based on temperature modulation, which relates to the field of spectral quantitative analysis technology. The method comprises the following steps: step S1, designing a temperature modulation strategy to obtain a temperature-modulated spectral signal; step S2, constructing a two-dimensional spectral matrix based on temperature modulation; step S3, constructing a high-order feature extraction network based on a neural network, and establishing an ion concentration regression network; step S4, training the feature extraction network and the ion concentration regression network; step S5, obtaining a temperature-modulated spectral matrix of a sample to be tested; and step S6, inputting the temperature-modulated spectrum of the sample to be tested into the trained model to obtain the concentration of the metal ion components. The detection method of the present invention analyzes the absorption characteristics and temperature-sensitive characteristics of each metal ion spectrum under temperature changes by analyzing the differences in the shape and intensity changes of the spectrum under temperature changes, constructing high-order features for feature separation and extraction, and improving the accuracy and anti-interference performance of the information separation and analysis of each component of the overlapping spectrum.

Description

Technical Field

[0001] The present invention relates to the technical field of spectral quantitative analysis, and in particular to a spectral multi-metal ion concentration detection method based on temperature modulation. Background Art

[0002] Heavy metal ions are one of the common pollutants in water. In particular, industrial wastewater, metallurgical liquid, electroplating wastewater and other solutions contain a variety of metal ions. The ratio of different metal ion concentrations may vary greatly. For example, the concentration ratio of zinc ions to trace metal ions can be as high as 10. 3 This complexity requires that the detection method must be able to accurately and simultaneously measure the concentrations of multiple metal ions to ensure environmental safety and human health.

[0003] Commonly used physical and chemical methods, such as electrochemical methods, have limited accuracy when analyzing complex systems, and the electrode processing process is relatively cumbersome. Although the colorimetric method has an intuitive measurement process, its lower limit of measurement is relatively high, and its sensitivity to heavy metal ions depends largely on the thickness of the membrane. The chemical precipitation method is suitable for measuring water bodies with high heavy metal ion content, but it is more sensitive to the pH of the measurement environment. These methods often require a pretreatment step to separate different metal ions using physical and chemical means, and then measure their concentrations separately. Therefore, most instruments using the corresponding methods can only measure the concentration of a single ion. When it is necessary to measure the concentration of multiple heavy metal ions, these methods are often difficult to meet the needs of multi-metal ion concentration detection due to the complex and time-consuming pretreatment process.

[0004] In recent years, spectral analysis, with its advantages of sensitivity and rapidity, has become an important method for qualitative, quantitative, and structural analysis of substances. It has broad application prospects in industrial online component detection. Its simplicity, rapid detection speed, and low construction and operating costs make it a powerful tool for online detection of multiple metal ion concentrations. However, in water, signals in the water body have high overlap and mutual interference, and are affected by factors such as the on-site environment, stray light, and light source fluctuations, making it difficult to distinguish the spectral signals of individual metal ions. Summary of the Invention

[0005] In response to the problems that multiple metal ions in water have similar properties, high spectral signal overlap, and difficulty in distinguishing the signals of each ion, the present invention provides a spectral multi-metal ion concentration detection method based on temperature modulation. By examining the differences in the spectral shape and intensity changes of each metal ion under temperature changes, comprehensively analyzing the absorption characteristics and temperature sensitivity of the metal ion spectra, and constructing corresponding high-order features, it is possible to separate and extract the characteristics of each component, thereby improving the accuracy and anti-interference ability of the separation and analysis of the information of each component in the overlapping spectra.

[0006] In order to solve the above technical problems, the technical solution proposed by the present invention is:

[0007] A method for detecting the concentration of multiple metal ions by spectroscopy based on temperature modulation comprises the following steps:

[0008] Step S1, designing a temperature modulation strategy to obtain a temperature-modulated spectral signal;

[0009] Step S2, constructing a two-dimensional spectrum matrix based on temperature modulation;

[0010] Step S3, constructing a high-order feature extraction network based on a neural network, extracting spectral features in the two-dimensional matrix, and establishing an ion concentration regression network;

[0011] Step S4, training the feature extraction network and the ion concentration regression network;

[0012] Step S5, obtaining a temperature modulation spectrum matrix of the sample to be tested;

[0013] Step S6, inputting the temperature modulation spectrum of the sample to be tested into the network model trained in step S4 to obtain the concentration of each metal ion component;

[0014] In step S1, the temperature modulation spectrum signal acquisition steps are as follows:

[0015] S11: Determine the fluctuation range of the concentration of each metal ion and analyze the law of the change of the spectrum of each metal ion with temperature;

[0016] S12: Acquire spectral data collected from the sample at different temperature points to form an original data set, thereby forming a spectral curve of the sample.

[0017] As a further improvement of the above technical solution:

[0018] Preferably, in step S2, the process of constructing a two-dimensional spectrum matrix is:

[0019] S21: Obtain a set of standard sample spectral signals for modeling And the corresponding component content , the spectral data of each sample is a one-dimensional array, which contains the response values ​​of L spectral wavelengths at k temperature points, and the entire standard sample data set is a two-dimensional array;

[0020] S22: Arrange the spectral data of the one-dimensional array in sequence according to the temperature modulation order to construct a two-dimensional spectral matrix of training samples based on temperature modulation ,The one-dimensional spectrum data of each sample is reconstructed into a two-dimensional matrix, and each element in the two-dimensional matrix represents the spectral response value at a specific spectral wavelength and a specific temperature point.

[0021] Preferably, the step S3 specifically includes the following steps:

[0022] S31: Constructing a spectral high-order feature extraction network based on neural network N F , used to capture complex patterns and features in spectral data; among them, the high-order feature extraction network N F The input is the two-dimensional spectral matrix of the training sample in step S2 ;

[0023] S32: Using high-order feature extraction network N F Extract the two-dimensional spectral matrix of the training samples High-order features of , and transferred to the ion concentration regression model;

[0024] S33: Ion concentration regression model N R Receive from the high-order feature extraction network N F The output of , through the calculation of the regression network, the predicted value of the metal ion concentration of each sample is obtained .

[0025] Preferably, in step S3, the feature extraction network N F Use convolutional neural network.

[0026] Preferably, the feature extraction network N F Based on two-dimensional convolutional layers; including convolutional layers, BatchNorm normalization layers, and LeakyRelu nonlinear activation layers.

[0027] Preferably, the ion concentration regression model N R It consists of a fully connected layer and a regression layer. The fully connected layer maps the features extracted by the convolutional layer to a high-dimensional space, and the regression layer predicts the concentration value of the ion based on the features.

[0028] Preferably, in step S4, a back propagation algorithm is used to train the network weights.

[0029] Preferably, the step S4 specifically includes the following steps:

[0030] S41: Define the loss function, use the mean square error as the main loss term, and add the L2 regularization term to the loss function;

[0031] S42: Training the model and updating the ion concentration regression model N using the error back propagation algorithm R and feature extraction network N F The weight in

[0032] S43: Repeat S42 until the training is completed and the model is established.

[0033] Preferably, the step S5 specifically includes the following steps:

[0034] S51: Obtain the spectral signal of the sample to be tested ;

[0035] S52: Arrange the spectral curves of the samples in sequence according to the modulation order, and construct a two-dimensional spectral matrix of the test samples based on temperature modulation .

[0036] Preferably, the step S6 specifically includes the following steps:

[0037] S61: Modulate the temperature of the sample to be tested to test the sample's two-dimensional spectrum matrix Input high-order feature extraction network N F , get the high-order feature map of the sample to be tested ;

[0038] S62: High-order feature maps Input the quantitative analysis model to obtain the concentration results of each metal ion component in the sample.

[0039] The temperature modulation-based spectral multi-metal ion concentration detection method provided by the present invention has the following advantages over the prior art:

[0040] (1) The present invention's temperature-modulated spectral multi-metal ion concentration detection method utilizes differences in temperature sensitivity among metal ion spectra. By using temperature modulation, the temperature-sensitive characteristics of ion spectra are acquired, and the temperature-sensitive properties of each metal ion are analyzed to form high-order features, enabling the separation and extraction of the characteristics of each component. This method can effectively separate and extract the characteristics of each component in a complex sample, and accurately identify and determine the concentration of each metal ion even when the spectral signals are highly overlapped.

[0041] (2) The temperature modulation-based spectral multi-metal ion concentration detection method of the present invention can realize the separation and extraction of the characteristics of each component, improve the accuracy and anti-interference of the separation and analysis of the information of each component of the overlapping spectrum, and solve the problem that the properties of multi-metal ions are similar, the spectral signal overlap is high, and the signals of each ion are difficult to distinguish. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 This is a flow chart of the temperature modulation-based spectral multi-metal ion concentration detection method of the present invention.

[0043] Figure 2 In the embodiment of the present invention, the content of 16 mg / L Zn + 、1.6 mg / L Cu + and 1.6mg / L Ni + Schematic diagram of the solution spectrum curve changing with temperature.

[0044] Figure 3 In the embodiment of the present invention, the content of Zn is 8 mg / L + 、8 mg / L Cu + and 1.6mg / L Ni + Schematic diagram of the solution spectrum curve changing with temperature.

[0045] Figure 4 In the embodiment of the present invention, the content of Zn is 8 mg / L + 、1.6mg / L Cu + and 4 mg / L Ni + Schematic diagram of the solution spectrum curve changing with temperature.

[0046] Figure 5 Schematic diagram of the intensity and shape of the zinc ion spectrum curve changing with temperature in an embodiment of the present invention.

[0047] Figure 6 Schematic diagram showing how the intensity and shape of the copper ion spectrum curve change with temperature in an embodiment of the present invention.

[0048] Figure 7 Schematic diagram showing how the intensity and shape of the nickel ion spectrum curve change with temperature in an embodiment of the present invention.

[0049] Figure 8 Schematic diagram of the feature extraction network and ion concentration regression model in an embodiment of the present invention. DETAILED DESCRIPTION

[0050] The specific embodiments of the present invention are described in detail below. It should be understood that the described embodiments are only a part of the embodiments of the present invention, not all of them. Based on these examples, those skilled in the art can understand and implement all other possible implementations of the present technical solution without performing creative work, and these implementations also fall within the scope of protection of the present technical solution. The protection of the present technical solution is not limited to the specific embodiments described herein, but also includes all equivalent implementations, which are all organic components of the present invention.

[0051] like Figure 1 As shown, the temperature modulation-based spectral multi-metal ion concentration detection method of the present invention comprises the following steps:

[0052] Step S1: designing a temperature modulation strategy to obtain a temperature-modulated spectral signal.

[0053] The implementation steps are as follows:

[0054] S11: Determine the fluctuation range of each metal ion concentration based on field data and historical data, and analyze how the spectra of each metal ion change with temperature. Understanding the concentration changes of metal ions in the sample provides a basis for subsequent temperature point design.

[0055] S12: Acquire spectral data collected from the sample at different temperature points to form a raw data set. Spectra of the sample are collected at the selected spectral measurement temperature point to obtain spectral data under different temperature conditions. The spectral data collected at different temperature points are integrated to form a spectral curve for the sample, forming a raw spectral data set based on temperature modulation.

[0056] Step S2: Arrange the spectral curves of the samples in sequence according to the modulation order to construct a two-dimensional spectral matrix based on temperature modulation.

[0057] The two-dimensional spectrum matrix of the training samples is constructed The process is as follows:

[0058] S21: Obtain a set of standard sample spectral signals for modeling and its corresponding component content , The standard sample data set consists of M samples. The spectral data of each sample contains the spectral response of the sample at L spectral wavelengths at k temperature points. The spectral data of each sample is considered as a one-dimensional array, containing the response values ​​of L spectral wavelengths at k temperature points. The entire standard sample data set is represented as a two-dimensional array of size M × (L × k).

[0059] S22: Arrange the spectral data of the one-dimensional array in sequence according to the temperature modulation order to construct a two-dimensional spectral matrix of training samples based on temperature modulation , the one-dimensional spectral data of each sample is reconstructed into a two-dimensional matrix with a size of L×k. Each element in this two-dimensional matrix represents the spectral response value at a specific spectral wavelength and a specific temperature point. It contains the spectral response information of the sample at different temperatures, and through its structured form, provides a basis for subsequent data analysis and model building.

[0060] Step S3: construct a feature extraction network based on deep learning, extract spectral features in the two-dimensional matrix, and establish an ion concentration regression network.

[0061] The specific steps include:

[0062] S31: Constructing a spectral high-order feature extraction network based on a two-dimensional convolutional neural network (CNN) F , used to process two-dimensional spectral matrices and capture complex patterns and features in spectral data. Among them, the high-order feature extraction network N FThe input is the two-dimensional spectral matrix of the training sample in step S2 .

[0063] Feature extraction network N F The design adopts the convolutional neural network (CNN) architecture in deep learning. F It consists of two two-dimensional convolutional layers, Conv1 and Conv2. Conv1 includes 8 convolution kernels, each of size 3×3, and Conv2 includes 16 convolution kernels, each of size 3×3. The two-dimensional convolutional layer operation process includes convolution operation, BatchNorm normalization operation, and LeakyRelu nonlinear activation operation. Convolution operation is the core of CNN. It performs weighted summation of input data through sliding convolution kernels to extract local features. BatchNorm normalization operation is used to adjust and normalize the output of the layer to accelerate the training process and improve the stability of the model. LeakyRelu activation operation introduces nonlinearity, enabling the model to learn more complex feature representations.

[0064] S32: Using high-order feature extraction network N F Extract the two-dimensional spectral matrix of the training samples The network gradually extracts the deep features of the spectral data through multiple convolutional layers and activation functions to form a high-order feature map , and transferred to the ion concentration regression model. The high-order feature map can more accurately reflect the spectral characteristics of the sample and provide rich information for subsequent regression analysis.

[0065] S33: Regression Network N R Receive from the high-order feature extraction network N F The output of , and further processing and analysis are carried out to obtain the predicted value of metal ion concentration of each sample through regression network calculation .

[0066] Ion concentration regression model N R It consists of a fully connected layer and a regression layer. The fully connected layer maps the features extracted by the convolutional layer to a high-dimensional space, while the regression layer is responsible for predicting the concentration of ions based on these features.

[0067] Step S4: training the feature extraction network and the ion concentration regression network.

[0068] Network Figure 8As shown in Figure 1, the network weights are trained using backpropagation and the Adamw optimizer. Adamw is an improved Adam optimizer that uses a more rigorous weight decay. Unlike traditional weight decay, its weight decay is only used in the final update formula and is not included in the momentum calculation.

[0069] The specific steps include:

[0070] S41: Define the loss function Loss, use the mean square error as the main loss term, and add the L2 regularization term to the loss function to prevent the model from overfitting. The expression of the loss function Loss is:

[0071] ;

[0072] in, Indicates the i The predicted concentration value of the sample, Indicates the i The true concentration value of the sample, w represents N R and N F All weight matrices in , M represents the number of training samples, and λ is the regularization parameter.

[0073] S42: Update N using the error back propagation algorithm R and N F The weight in .

[0074] Assume that the model parameter to be optimized is θ, the learning rate is η, the iteration period epoch is t, and the gradient at the tth epoch is , first calculate the exponentially weighted average of the squared gradients :

[0075] ;

[0076] Then calculate the exponentially weighted average of the gradients :

[0077] ;

[0078] in and They are the decay rates of the two moving averages, 0.9 and 0.999 respectively.

[0079] Correct the deviation:

[0080] ;

[0081] ;

[0082] Calculate parameters and update parameters:

[0083] ;

[0084] in, is the learning rate, It is a small constant used to avoid the denominator being 0, usually with a value of 1e-8. is the regularization coefficient used to prevent overfitting.

[0085] S43: Repeat S42 until the training is completed and the final model is established.

[0086] Step S5: obtaining a temperature modulation spectrum of the sample to be tested.

[0087] The specific steps include:

[0088] S51: Obtain the spectral signal of the sample to be tested .

[0089] S52: Arrange the spectral curves of the samples in sequence according to the modulation order, and construct a two-dimensional spectral matrix of the test samples based on temperature modulation .

[0090] Step S6: Input the temperature modulation spectrum of the sample to be tested into the trained high-order feature extraction network N F and regression network N R The concentration of each metal ion component is obtained.

[0091] The specific steps include:

[0092] S61: Modulate the temperature of the sample to be tested to test the sample's two-dimensional spectrum matrix Input high-order feature extraction network N F , get the high-order feature map of the sample to be tested .

[0093] S62: High-order feature maps Input regression model N R , and obtain the concentration results of each metal ion component in the sample.

[0094] Experimental verification:

[0095] Taking zinc, copper and nickel as examples, the difference of spectral temperature sensitivity characteristics is detected. Figure 2 、 3 As shown in Figure 4, the spectral curves of solutions containing different concentrations of zinc, copper, and nickel ions change with temperature. Taking solutions containing different concentrations of zinc, copper, and nickel metal ions as the object, the spectrum of the solution is temperature modulated, and a two-dimensional spectrum matrix based on temperature modulation is constructed. Then, a high-order feature extraction network N based on a neural network is used to extract the spectrum. FExtract high-order feature maps and then input them into the regression network N R The prediction results of the samples are obtained. The loss function is calculated based on the deviation between the predicted value and the true value, and the error back propagation algorithm is used to update the network weights, ultimately obtaining a multi-metal ion concentration detection model.

[0096] The specific process is as follows:

[0097] Step S1: designing a temperature modulation strategy to obtain a temperature-modulated spectral signal.

[0098] First, if Figure 5 、 Figure 6 and Figure 7 As shown, the intensity and shape of the spectral curves for zinc, copper, and nickel ions vary significantly with temperature, providing information needed to separate and resolve overlapping signals. A series of solutions containing these metal ions were prepared based on the concentration ranges of zinc, copper, and nickel ions commonly found in hydrometallurgical zinc smelting solutions. The ion concentrations of these solutions were controlled within a range of 0 mg / L to 10.0 mg / L to simulate the various concentration levels encountered in actual industrial processes.

[0099] Next, at different test temperatures, a UV-visible spectrophotometer was used to measure the UV-visible absorption spectrum intensity of the solution. When designing the temperature points for spectral modulation and spectral acquisition, the principle of maximizing the spectral difference between ions was followed. This means that the selected temperature point can maximize the difference between the spectra of different ions, thereby improving the discrimination and accuracy of spectral analysis. Specifically, within the temperature range of 40°C to 90°C, the spectral signal of the ion solution was measured every 10°C to obtain a series of spectral data collected at different temperature points. The collected spectral data was integrated to form a spectral curve of the sample, forming an original spectral data set based on temperature modulation.

[0100] Step S2: Arrange the spectral curves of the samples in sequence according to the modulation order to construct a two-dimensional spectral matrix based on temperature modulation.

[0101] The spectral matrix consists of 100 samples, including the spectral response data of these samples at 400 wavelengths (400~800nm) at 6 temperature points (40℃-90℃). Using the SPXY segmentation method, 80 samples are selected from these 100 samples to form the training set signal. and , the remaining 20 samples constitute the test set signal and , The size is 80×2400, The size is 20×2400.

[0102] In the process of constructing a two-dimensional spectral matrix, the collected sample spectral response data are first arranged according to the temperature point sequence, i.e., 40°C, 50°C, 60°C, 70°C, 80°C, and 90°C, to maintain the consistency and comparability of the data and obtain a two-dimensional spectral matrix.

[0103] Next, the spectral signals at the same temperature in the dataset are normalized to eliminate the dimensional differences by scaling the data to a common range. The two-dimensional spectral matrix of the training set is The size is 80×400×6.

[0104] Step S3: construct a high-order feature extraction network N based on neural network F , extract the spectral features in the two-dimensional matrix and establish the ion concentration regression model N R .

[0105] Step S4: training the feature extraction network N F and ion concentration regression network N R .

[0106] When the error increases for 10 consecutive cycles or the preset number of training epochs is completed, the final quantitative analysis model is established.

[0107] Step S5, obtaining the temperature modulation test sample two-dimensional spectrum matrix of the sample to be tested .

[0108] Use the spectral signal of the sample to be tested Test the performance of the regression model. Process according to steps S1 and S2 to obtain , which is 20×400×6 in size and contains the spectral response data of 20 samples at 400 wavelengths at 6 temperature points.

[0109] Step S6: Modulate the temperature of the sample to be tested to obtain a two-dimensional spectrum matrix of the sample to be tested. The input is fed into the trained high-order feature extraction network and regression network to obtain the concentration of each metal ion component.

[0110] The test sample two-dimensional spectrum matrix Input the trained feature extraction network and regression network to get the output To evaluate the performance of the model, we calculated and Root mean square error RMSE and coefficient of determination R 2 , these two indicators are commonly used indicators to measure the accuracy of regression model prediction. RMSE measures the difference between the predicted value and the actual value, while R 2It measures how well the model explains the variability of the data. 2 The calculation formula is as follows:

[0111] ;

[0112] ;

[0113] Where N is the number of test set samples, N is 20, and Respectively represent i The actual and predicted values ​​of the samples.

[0114] The measurement results of the solution were compared and analyzed based on the original spectrum at room temperature, the spectrum at each temperature point combined with one-dimensional CNN, and the temperature-modulated two-dimensional spectrum combined with two-dimensional CNN. To ensure the reliability of the results, the model was repeated 100 times using each method and the results were statistically analyzed. The determination coefficient R 2 The root mean square error (RMSE) is shown in Table 1.

[0115] Table 1:

[0116] method RMSE <![CDATA[R 2 ]]> Normal temperature-CNN 0.0439±0.0058 0.9329±0.0162 40℃-CNN 0.0484±0.0053 0.9225±0.0184 50℃-CNN 0.0527±0.0063 0.9128±0.0201 60℃-CNN 0.0517±0.0065 0.9369±0.0128 70℃-CNN 0.0491±0.0069 0.9326±0.0228 80℃-CNN 0.0489±0.0061 0.9291±0.0161 90℃-CNN 0.0512±0.0049 0.9339±0.0148 Temperature Modulated Two-Dimensional Spectroscopy - 2DCNN 0.0382±0.0021 0.9541±0.0042

[0117] As shown in Table 1, the RMSE of the temperature modulation-based spectral multi-metal ion concentration detection method of the present invention is 0.0382±0.0021, R 2 The model of this method not only has higher explanatory power but also shows good stability compared with other methods.

[0118] In summary, the present invention's temperature-modulated spectral multi-metal ion concentration detection method and device effectively separates and extracts the characteristics of each component, resolving the problem of similar properties of multiple metal ions, high spectral signal overlap, and difficulty distinguishing individual ion signals. By combining temperature modulation with deep learning technology, this method provides a new and efficient solution for multi-component analysis in complex samples.

[0119] The above examples are merely preferred embodiments of the present invention and are not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, they are not intended to limit the present invention. Therefore, any simple modifications, equivalent variations, and modifications to the above examples that do not depart from the technical solution of the present invention and are based on the technical essence of the present invention shall fall within the scope of protection of the technical solution of the present invention.

Claims

1. A method for detecting the concentration of multiple metal ions by spectroscopy based on temperature modulation, characterized in that: The following steps are involved: Step S1, designing a temperature modulation strategy to obtain a temperature modulated spectral signal; when designing the temperature points for spectral modulation and spectral acquisition, the selected temperature points can maximize the difference between the spectra of different ions; Step S2, constructing a two-dimensional spectrum matrix based on temperature modulation; Step S3, constructing a high-order feature extraction network based on a neural network, extracting spectral features in the two-dimensional matrix, and establishing an ion concentration regression network; Step S4, training the feature extraction network and the ion concentration regression network; Step S5, obtaining a temperature modulation spectrum matrix of the sample to be tested; Step S6, inputting the temperature modulation spectrum of the sample to be tested into the network model trained in step S4 to obtain the concentration of each metal ion component; In step S1, the temperature modulation spectrum signal acquisition steps are as follows: S11: Determine the fluctuation range of the concentration of each metal ion and analyze the law of the change of the spectrum of each metal ion with temperature; S12: Acquire spectral data collected from the sample at different temperature points to form an original data set, thereby forming a spectral curve of the sample; In step S2, a two-dimensional spectrum matrix of training samples is constructed. The process is: S21: Obtain a set of standard sample spectral signals for modeling And the corresponding component content , the spectral data of each sample is a one-dimensional array, which contains the response values ​​of L spectral wavelengths at k temperature points, and the entire standard sample data set is a two-dimensional array; S22: Arrange the spectral data of the one-dimensional array in sequence according to the temperature modulation order to construct a two-dimensional spectral matrix of training samples based on temperature modulation ,The one-dimensional spectral data of each sample is reconstructed into a two-dimensional matrix, and each element in the two-dimensional matrix represents the spectral response value at a specific spectral wavelength and a specific temperature point; The step S3 specifically includes the following steps: S31: Constructing a spectral high-order feature extraction network based on neural network N F , used to capture complex patterns and features in spectral data; among them, the high-order feature extraction network N F The input is the two-dimensional spectral matrix of the training sample in step S2 ; S32: Using high-order feature extraction network N F Extract the two-dimensional spectral matrix of the training samples High-order features of , and transferred to the ion concentration regression model; S33: Ion concentration regression model N R Receive from the high-order feature extraction network N F The output of , through the calculation of the regression network, the predicted value of the metal ion concentration of each training sample is obtained .

2. The method for detecting the concentration of metal ions by spectrometry based on temperature modulation according to claim 1, wherein: In step S3, the feature extraction network N F Use convolutional neural network.

3. The method for detecting the concentration of metal ions by spectrometry based on temperature modulation according to claim 2, wherein: In step S3, the feature extraction network N F Based on two-dimensional convolutional layers; including convolutional layers, BatchNorm normalization layers, and LeakyRelu nonlinear activation layers.

4. The method for detecting the concentration of metal ions by spectrometry based on temperature modulation according to claim 1, wherein: The ion concentration regression model N R It consists of a fully connected layer and a regression layer. The fully connected layer maps the features extracted by the convolutional layer to a high-dimensional space, and the regression layer predicts the concentration value of the ion based on the features.

5. The method for detecting the concentration of metal ions by spectrometry based on temperature modulation according to claim 1, wherein: In step S4, the network weights are trained using a back propagation algorithm.

6. The method for detecting the concentration of multiple metal ions by spectrometry based on temperature modulation according to claim 5, characterized in that: The step S4 specifically includes the following steps: S41: Define the loss function, use the mean square error as the main loss term, and add the L2 regularization term to the loss function; S42: Training the model and updating the ion concentration regression model N using the error back propagation algorithm R and feature extraction network N F The weight in S43: Repeat S42 until the training is completed and the model is established.

7. The method for detecting the concentration of multiple metal ions by spectrometry based on temperature modulation according to claim 6, characterized in that: The step S5 specifically includes the following steps: S51: Obtain the spectral signal of the sample to be tested ; S52: Arrange the spectral curves of the samples in sequence according to the modulation order, and construct a two-dimensional spectral matrix of the sample to be tested based on temperature modulation .

8. The method for detecting the concentration of multiple metal ions by spectrometry based on temperature modulation according to claim 6, wherein: The step S6 specifically includes the following steps: S61: Modulate the temperature of the sample to be tested into a two-dimensional spectral matrix Input high-order feature extraction network N F , get the high-order feature map of the sample to be tested ; S62: High-order feature maps Input the quantitative analysis model to obtain the concentration results of each metal ion component in the sample.

Images