Complex substance artificial intelligence detection system based on thermal radiation chip and construction method

Abstract

The invention provides a complex substance detection system based on a thermal radiation super-structure surface chip and a construction method, and relates to the technical field of infrared detection. The method comprises the following steps: preparing a thermal radiation chip, constructing an artificial intelligence model, heating the thermal radiation chip, and obtaining thermal imaging data of the thermal radiation chip after penetrating through different standard samples by using an infrared camera; and finally, inputting the thermal imaging data of different standard samples into the artificial intelligence model to adjust the parameters of the artificial intelligence model so as to train the artificial intelligence model. According to the technical scheme, the trained artificial intelligence model is used for processing the infrared thermal imaging data, the components of the to-be-detected sample can be detected, the concentration of the components can be predicted, the recognition precision and sensitivity of the system on a complex mixture are improved, and the detection efficiency and accuracy are improved.

Description

Technical Field

[0001] The present invention relates to the field of infrared detection technology, and particularly to an artificial intelligence detection system for complex substances based on a thermal radiation chip and a construction method thereof. Background Art

[0002] In recent years, infrared thermal imaging and spectroscopy detection technologies based on micro-nano structured metasurfaces and artificial intelligence have received extensive attention. The introduction of metasurfaces and the application of infrared thermal imaging technology provide a new direction for improving the performance of infrared detection systems.

[0003] Infrared absorption spectroscopy detection is a technology that analyzes the composition and physical properties of materials by measuring the absorption intensity of materials for infrared light of different wavelengths. Infrared spectra can provide rich information on chemical composition, molecular structure, and material state, and are widely used in chemical analysis, environmental monitoring, and medical diagnosis. Due to the specific infrared absorption characteristics of different materials, infrared spectral data can effectively identify and analyze the composition of materials and their physical changes.

[0004] A metasurface is an artificial material composed of sub-wavelength scale micro-nano structures and has unique electromagnetic wave regulation capabilities. By designing the shape, size, and arrangement of its structural units (such as nanocolumns, nanoantennas, etc.), the metasurface can achieve precise regulation of electromagnetic waves in the infrared band. With the development of nano-manufacturing technology, the application prospect of metasurface technology in infrared spectroscopy and thermal radiation regulation is becoming increasingly broad. Using metasurfaces, efficient infrared absorption chips can be designed, providing new solutions for infrared thermal imaging, temperature detection, etc.

[0005] Infrared thermal imaging technology detects the infrared light radiated by an object and converts it into an image to display the temperature distribution of the object. This technology is widely used in fields such as medical treatment, industrial inspection, and environmental monitoring. By combining micro-nano structured metasurfaces, modern infrared thermal imaging systems can achieve better absorption and radiation performance in different infrared bands, thereby improving the sensitivity and accuracy of the system. Current traditional and mainstream infrared substance detection technologies mainly focus on using spectral analysis for substance identification and quantitative analysis. Among them, Fourier Transform Infrared Spectroscopy (FTIR) technology is still one of the most widely used methods. An FTIR system emits an infrared beam through high-precision optical devices such as a Michelson interferometer, a tunable laser, and a blackbody radiation source, and irradiates the sample to be measured. Substances will absorb at specific wavelengths of infrared light to form characteristic absorption spectra. The characteristic peaks of the spectra are closely related to the chemical composition and structure of the substances, and can provide accurate substance identification and quantitative analysis. FTIR technology is usually used in multiple fields such as gas analysis, food safety, environmental monitoring, and the pharmaceutical industry. In a traditional FTIR system, the key to obtaining high-quality spectral data is to use a stable and sufficiently intense infrared light source, especially in the mid-infrared (MIR) range that is more sensitive in the infrared band. FTIR systems usually use a blackbody radiation source to provide infrared light with a wide spectrum, and adjust the phase of the light source through a Michelson interferometer to achieve interference-resolved spectroscopy. By the absorption characteristics of the infrared detection spectrum, the chemical components in the sample can be judged, and it is widely used in the qualitative and quantitative analysis of substances. Although FTIR technology is widely used in substance analysis, there are some deficiencies. First, the spectral resolution is limited, which may lead to peak overlap in complex samples and affect the analysis accuracy. Second, the sample preparation requirements are relatively high. Solid or viscous samples need special treatment, which may lead to measurement errors. In addition, moisture interferes with the infrared absorption spectrum. Especially in the mid-infrared band, the absorption peak of water may overlap with the characteristic peak of the target substance. In terms of equipment, the cost of an FTIR system is relatively high, and it needs to be maintained regularly, increasing the economic burden and technical threshold. Finally, the absorption spectra of some samples are not obvious or overlap, resulting in difficulties in quantitative analysis. Although these technologies can provide relatively high spectral resolution, their sensitivity and resolution are still insufficient when dealing with complex mixtures or trace components. Therefore, there is an urgent need to design a method that can accurately detect the substance composition of products without spectral measurement. In addition, although some systems use metasurface filters, they still require an additional infrared light source, thereby increasing the cost and complexity of the system. Summary of the Invention

[0006] An object of the present invention is to provide a construction method for an artificial intelligence detection system for complex substances based on a thermal radiation chip, so as to solve the technical problem of spectral overlap of mixed substance components in the prior art.

[0007] Another object of the present invention is to provide an artificial intelligence detection system for complex substances based on a thermal radiation chip.

[0008] In particular, the present invention provides a method for constructing an artificial intelligence detection system for complex substances based on a thermal radiation chip, comprising the following steps:

[0009] Prepare a thermal radiation chip and construct an artificial intelligence model based on the thermal radiation chip;

[0010] Place different standard samples between the thermal radiation chip and the infrared camera in sequence, and heat the thermal radiation chip so that the infrared camera obtains reaction signal data after the thermal radiation chip undergoes absorption reactions with different standard samples respectively;

[0011] Input the reaction signal data of different standard samples into the artificial intelligence model to adjust the parameters of the artificial intelligence model and train the artificial intelligence model.

[0012] Optionally, after the step of preparing a thermal radiation chip and constructing an artificial intelligence model based on the thermal radiation chip, the following steps are further included:

[0013] Heat the thermal radiation chip and obtain the background signal of the thermal radiation chip without passing through the standard sample;

[0014] Use the background signal to correct the reaction signals of different standard samples to obtain corrected reaction signal data;

[0015] Use the corrected reaction signal data to train the artificial intelligence model to obtain a trained model.

[0016] Optionally, the artificial intelligence model is constructed through the following steps:

[0017] Construct a first Conv1D layer, and the first Conv1D layer is configured to extract local features of the corrected reaction signal data;

[0018] Construct a MaxPooling1D layer, and the MaxPooling1D layer is configured to reduce the dimension of the local features;

[0019] Construct a second Conv1D layer, and the second Conv1D layer is configured to enhance the extraction ability of the local features;

[0020] Use a Flatten layer to flatten the two-dimensional feature map of the local features into a one-dimensional vector;

[0021] Construct a Dense layer, which is configured to integrate the feature information of the local features;

[0022] Construct an output layer, which is configured to output the component detection result of the standard sample and the prediction result of the concentration of the component.

[0023] Optionally, prepare a thermal radiation chip through the following steps:

[0024] Construct a model of a metasurface array;

[0025] Use simulation software to simulate the model of the metasurface array to obtain the target thickness parameters of the materials of each layer of the model of the metasurface array;

[0026] Prepare the thermal radiation chip according to the target thickness parameters of the materials of each layer.

[0027] Optionally, the step of preparing the thermal radiation chip according to the target thickness parameters of the materials of each layer specifically includes the following steps:

[0028] Provide a substrate;

[0029] Deposit a metal layer on the substrate, and the metal layer serves as a reflective substrate;

[0030] Prepare a transition layer and a silicon layer on the metal layer in sequence by magnetron sputtering or chemical vapor deposition method, and the transition layer is located between the metal layer and the silicon layer;

[0031] Use photolithography or etching process to perform micro-nano processing on the silicon layer, so as to prepare and obtain the thermal radiation chip with a metasurface array.

[0032] Optionally, the step of using photolithography or etching process to perform micro-nano processing on the silicon layer, so as to prepare and obtain the thermal radiation chip with a metasurface array specifically includes the following steps:

[0033] Use the simulation software to simulate the period, diameter and height of the silicon cylinders of the metasurface array of the silicon layer to screen out the target period, target diameter and target height of the silicon cylinders;

[0034] Use photolithography or etching process to perform micro-nano processing on the silicon layer according to the target period, target diameter and target height of the silicon cylinders, so as to prepare and obtain the thermal radiation chip with a metasurface array.

[0035] Optionally, in the step of inputting the reaction signal data of different standard samples into the artificial intelligence model to adjust the parameters of the artificial intelligence model for training the artificial intelligence model, the corrected reaction signal data is divided into a training set and a test set. The training set is used to train the artificial intelligence model, and the test set is used to test the artificial intelligence model.

[0036] In particular, the present invention also provides an artificial intelligence detection system for complex substances based on a thermal radiation chip, which is applied to the above construction method, and includes:

[0037] A heating device;

[0038] A thermal radiation chip, arranged on one side of the heating device, and the thermal radiation chip is configured to generate an infrared thermal radiation signal with preset radiation characteristics under the heating of the heating device;

[0039] An infrared camera, arranged on the side of the thermal radiation chip away from the heating device. A standard sample is placed between the thermal radiation chip and the infrared camera. The infrared camera is configured to acquire the signal after the infrared thermal radiation signal reacts with the standard sample and process the signal to generate a data packet;

[0040] An artificial intelligence model based on the thermal radiation chip, arranged on the side of the infrared camera away from the thermal radiation chip. The artificial intelligence model is configured to receive the data packet and detect the components of the standard sample according to the data packet and predict the concentration of the components.

[0041] The present invention first prepares a thermal radiation chip and constructs an artificial intelligence model, then successively places different standard samples between the thermal radiation chip and the infrared camera, and heats the thermal radiation chip so that the infrared camera acquires the thermal imaging data after the thermal radiation chip passes through different standard samples. Finally, the thermal imaging data of different standard samples is input into the artificial intelligence model to adjust the parameters of the artificial intelligence model for training the artificial intelligence model. The above technical solution uses the thermal imaging data of different standard samples to train the artificial intelligence model, and uses the trained artificial intelligence model to process the infrared thermal imaging data, which can detect the components of the sample to be tested and predict the concentration of its components, improving the recognition accuracy and sensitivity of the system for complex mixtures, and improving the detection efficiency and accuracy.

[0042] From the following detailed description of specific embodiments of the present invention in conjunction with the accompanying drawings, those skilled in the art will become more clear about the above and other objects, advantages and features of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Some specific embodiments of the present invention will be described in detail hereinafter with reference to the accompanying drawings in an illustrative rather than restrictive manner. The same reference numerals in the drawings denote the same or similar components or parts. Those skilled in the art should understand that these drawings are not necessarily drawn to scale. In the drawings:

[0044] Figure 1 is a schematic flow chart of a method for constructing an artificial intelligence detection system for complex substances based on a thermal radiation chip according to an embodiment of the present invention;

[0045] Figure 2 is a schematic flow chart of a method for constructing an artificial intelligence detection system for complex substances based on a thermal radiation chip according to another embodiment of the present invention;

[0046] Figure 3 is a schematic flow chart of a method for constructing an artificial intelligence detection system for complex substances based on a thermal radiation chip according to still another embodiment of the present invention;

[0047] Figure 4 is a schematic structural diagram of a thermal radiation chip according to an embodiment of the present invention;

[0048] Figure 5 is a schematic diagram of an infrared absorption spectrum of a thermal radiation chip according to an embodiment of the present invention;

[0049] Figure 6 is a schematic structural diagram of an artificial intelligence detection system for complex substances based on a thermal radiation chip according to an embodiment of the present invention;

[0050] Figure 7 is a schematic diagram of infrared transmission spectra of methanol, ethanol, and acetone measured by FTIR according to an embodiment of the present invention;

[0051] Figure 8 is a schematic diagram of obtaining characteristic temperature data of methanol, ethanol, and acetone by using an artificial intelligence detection system for complex substances based on a thermal radiation chip according to an embodiment of the present invention;

[0052] Figure 9 is a schematic diagram of a confusion matrix for classifying methanol, ethanol, and acetone by using an artificial intelligence detection system for complex substances based on a thermal radiation chip according to an embodiment of the present invention;

[0053] Figure 10 is a schematic diagram of a confusion matrix for classifying 50 samples with different mixing ratios by using an artificial intelligence detection system for complex substances based on a thermal radiation chip according to an embodiment of the present invention;

[0054] Figure 11It is a schematic diagram of the result of predicting the methanol concentration by an artificial intelligence detection system for complex substances based on a thermal radiation chip according to an embodiment of the present invention;

[0055] Figure 12 It is a schematic diagram of the result of predicting the ethanol concentration by using an artificial intelligence detection system for complex substances based on a thermal radiation chip according to an embodiment of the present invention;

[0056] Figure 13 It is a schematic diagram of the result of predicting the acetone concentration by using an artificial intelligence detection system for complex substances based on a thermal radiation chip according to an embodiment of the present invention.

[0057] Reference numerals:

[0058] 100 - Detection system, 200 - Standard sample, 10 - Heating device, 20 - Thermal radiation chip, 30 - Infrared camera, 40 - Artificial intelligence model. Detailed implementation manners

[0059] The embodiments of the present invention will be described in detail below. The examples of the embodiments are shown in the drawings, where the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions from beginning to end. The embodiments described below by referring to the drawings are exemplary and are intended to explain the present invention, and should not be construed as a limitation to the present invention.

[0060] Figure 1 It is a schematic flow chart of a construction method of an artificial intelligence detection system for complex substances based on a thermal radiation chip according to an embodiment of the present invention. As Figure 1 shown, in a specific embodiment, the construction method of the artificial intelligence detection system for complex substances based on a thermal radiation chip includes the following steps:

[0061] Step S100, prepare a thermal radiation chip and construct an artificial intelligence model based on the thermal radiation chip;

[0062] Step S200, sequentially place different standard samples between the thermal radiation chip and the infrared camera, and heat the thermal radiation chip to enable the infrared camera to obtain thermal imaging data of the thermal radiation chip passing through different standard samples;

[0063] Step S300, input the reaction signal data of different standard samples into the artificial intelligence model to adjust the parameters of the artificial intelligence model and train the artificial intelligence model.

[0064] In this embodiment, the thermal imaging data of different standard samples are used to train an artificial intelligence model, and the trained artificial intelligence model is used to process the infrared thermal imaging data, which can improve the recognition accuracy and sensitivity of complex mixtures, and can detect the components of the target sample and predict the concentration of the components.

[0065] On the one hand, in this embodiment, a thermal radiation signal is generated by heating a thermal radiation chip, and substance detection can be performed by combining infrared thermal imaging and the trained artificial intelligence model. Heating the thermal radiation chip provides a low-cost and stable infrared light source, replacing expensive lasers and blackbody radiation sources, significantly reducing the system cost, and realizing miniaturization and integration, solving the problem that traditional infrared spectroscopy systems are bulky and difficult to be portable. On the other hand, using the trained artificial intelligence model to process the infrared thermal imaging data improves the recognition accuracy and sensitivity of the system to complex mixtures and enhances the accuracy of detection.

[0066] In the stage of training the artificial intelligence model, first, the metasurface chip is heated by a heating stage to generate infrared thermal radiation with specific spectral characteristics. When the infrared thermal radiation passes through the standard sample, the radiation signals of different wavelengths react with the standard sample by absorption. The infrared camera integrates these signals to capture the overall radiation change after passing through the standard sample, and the recorded data contains comprehensive information on the absorption characteristics of the standard sample.

[0067] This embodiment develops a set of one-dimensional convolutional neural network model (1D-CNN) based on the measurement data of a series of standard samples and the radiation characteristics of the thermal radiation chip, realizing two functions of substance classification and concentration prediction of the sample. After data acquisition and preprocessing are completed, the thermal imaging data is input into the convolutional neural network for analysis and processing.

[0068] Specifically, the convolutional neural network model specifically includes: a one-dimensional convolutional layer that uses 64 convolutional kernels (kernel size is 3), with ReLU as the activation function, to extract local features of the sample; subsequently, a pooling layer (pool size is 2) is used to reduce the feature dimension; the second convolutional layer contains 128 convolutional kernels (kernel size is 3) to further enhance the feature extraction ability. The flattened features are input into a fully connected layer with 128 neurons, and then a Dropout layer (dropout rate is 0.4) is used to prevent overfitting. Finally, the output layer with 64 neurons and a linear activation function completes the substance classification and concentration prediction. In the substance classification task, the convolutional neural network model extracts key features from the spectral data through multiple convolutional and feature extraction operations, and finally accurately outputs the category information of the sample. The classification results are visualized through a confusion matrix to display the classification accuracy and analyze possible misclassifications, thereby ensuring the reliability and accuracy of the classification results. In the concentration prediction task, the convolutional neural network model establishes a mapping relationship between spectral features and the concentration of sample components through deep learning of the spectral data, and outputs the specific concentration values of each component (such as A%, B%, C%). The concentration prediction results are evaluated by the mean square error (MSE) and the coefficient of determination (R 2 ), to verify the prediction accuracy of the model within different concentration ranges.

[0069] Through the training and optimization with standard sample data, the artificial intelligence model, namely the convolutional neural network model, can efficiently complete the two core functions of substance classification and concentration prediction of samples, significantly improving the intelligent level and practical application ability of the detection system.

[0070] Figure 2 is a schematic flowchart of a method for constructing a complex substance artificial intelligence detection system based on a thermal radiation chip according to another embodiment of the present invention. As Figure 2 shown, after step S100, the following steps are further included:

[0071] Step S110, heating the thermal radiation chip and obtaining the background signal of the thermal radiation chip without passing through the standard sample;

[0072] Step S120, using the background signal to correct the reaction signals of different standard samples to obtain the corrected reaction signal data;

[0073] Step S300’, using the corrected reaction signal data to train the artificial intelligence model to obtain the trained artificial intelligence model.

[0074] After the reaction signal is corrected using the background signal in this embodiment, the accuracy of the reaction signal data can be improved, and further, the artificial intelligence model can be better trained. During the training of the artificial intelligence model, first, the background signal without the standard sample is recorded for subsequent background correction. Then, the micro-nano structure of the thermal radiation chip is activated and its radiation characteristics are recorded. Next, the transmitted light intensities of air and the standard sample are measured to obtain the system baseline data and the radiation characteristic data of the standard sample respectively. All measurements are repeated to ensure the accuracy of the data. After background correction, the transmittance of the standard sample is calculated to obtain reliable standard sample signal data for the training of the artificial intelligence model.

[0075] The construction method of this embodiment significantly reduces the influence of external interference on the detection results through precise background correction and standardized data preprocessing, as well as the wide-spectrum signal generation mechanism based on the thermal radiation chip, ensuring high accuracy and repeatability of the data. Combining the efficient detection process of the infrared camera and the artificial intelligence model, this embodiment can quickly realize the material classification and concentration prediction of the sample to be measured, and has the characteristics of high efficiency and low cost. It has broad application prospects in the rapid detection and intelligent analysis of multiple fields such as material detection, component analysis, and environmental monitoring.

[0076] Figure 3 It is a schematic flowchart of the construction method of an artificial intelligence detection system for complex substances based on a thermal radiation chip according to another embodiment of the present invention. As Figure 3 shown, an artificial intelligence model is constructed through the following steps:

[0077] Step S11, construct a first Conv1D layer, and the first Conv1D layer is set to extract local features of the corrected reaction signal data;

[0078] Step S12, construct a MaxPooling1D layer, and the MaxPooling1D layer is set to reduce the dimension of the local features;

[0079] Step S13, construct a second Conv1D layer, and the second Conv1D layer is set to enhance the extraction ability of local features;

[0080] Step S14, use a Flatten layer to flatten the two-dimensional feature map of the local features into a one-dimensional vector;

[0081] Step S15, construct a Dense layer, and the Dense layer is set to integrate the feature information of the local features;

[0082] Step S16, construct an output layer, and the output layer is set to output the component detection result of the standard sample and the prediction result of the concentration of the component.

[0083] In some embodiments, the metasurface chip is prepared through the following steps:

[0084] Step 1: Construct a model of the metasurface array;

[0085] Step 2: Use simulation software to simulate the model of the metasurface array to obtain the target thickness parameters of the materials for each layer of the model of the metasurface array;

[0086] Step 3: Fabricate a thermal radiation chip according to the target thickness parameters of the materials for each layer.

[0087] In Step 2, parameters such as the metal layer, the transition layer, and the height of the silicon cylinders can be adjusted as variables in the simulation software: for example, the thickness of the metal layer can vary within the range of 50–150 nm, the thickness of the transition layer is set between 50–150 nm, and the height of the silicon cylinders can be adjusted within the range of 300–800 nm. These adjustments are made by systematically scanning combinations of layer parameters and observing the effects on the electromagnetic resonance modes and the positions of the absorption peaks, so as to find the optimal layer thickness combination to ensure efficient and precise absorption and radiation control within the target infrared band.

[0088] In some embodiments, the step of constructing a thermal radiation chip according to the target thickness parameters of the materials for each layer specifically includes the following steps:

[0089] Step 4: Provide a substrate;

[0090] Step 5: Deposit a metal layer on the substrate, and the metal layer serves as a reflective substrate;

[0091] Step 6: Sequentially prepare a transition layer and a silicon layer on the metal layer by magnetron sputtering or chemical vapor deposition, and the transition layer is located between the metal layer and the silicon layer;

[0092] Step 7: Use photolithography or etching processes to perform micro-nano processing on the silicon layer, thereby fabricating a thermal radiation chip with a metasurface array.

[0093] After Step 7, that is, after the thermal radiation chip is processed, cleaning and drying treatments are performed to remove impurities and ensure a clean surface. It should be noted that the design method of this embodiment is not limited to the above processing flow, and various flexible process technologies such as nanoimprinting and electron beam lithography can be used for the fabrication of the thermal radiation chip to adapt to different preparation requirements and conditions. Through the above process flow, a periodic metasurface array located above the metal layer and the transition layer is obtained.

[0094] In some embodiments, Step 7 specifically includes the following steps:

[0095] Step 8: Use simulation software to simulate the period of the metasurface array of the silicon layer, the diameter and height of the silicon cylinders, so as to screen out the target period, the target diameter and the target height of the silicon cylinders;

[0096] Step 9: Use photolithography or etching process to perform micro-nano processing on the silicon layer according to the target period, the target diameter, and the target height of the silicon cylinder, so as to prepare a thermal radiation chip with a metasurface array.

[0097] Through the above steps, this embodiment constructs a thermal radiation chip to achieve efficient thermal radiation regulation in the mid-infrared band (7-12 μm). The design concept of the thermal radiation chip is to utilize the electromagnetic characteristics of the metasurface structure. Place the silicon cylinder array on top of the metal layer and the transition layer, and precisely regulate multi-pole resonance modes such as electric dipole (ED) and magnetic dipole (MD) by optimizing geometric parameters (such as cylinder diameter, height, and period), thereby enhancing the absorption and radiation performance within a specific wavelength range. In addition, by adjusting the proportional parameter (scaling parameter s) of the geometric dimensions of the metasurface chip, the position of the absorption peak can be flexibly tuned to meet the requirements of different application scenarios.

[0098] Specifically, starting from setting the silicon cylinder diameter D as 0.6 times the period P. For example, if the period P is selected as 10 μm, then the diameter D is 6 μm. The metasurface chip includes a 70-nm-thick metal layer, a 100-nm-thick indium tin oxide (ITO) layer, and 600-nm-high silicon cylinders from bottom to top. Based on the silica substrate, establish a complete structure model in simulation software such as COMSOL. The initial parameters are set as period P = 10 μm, diameter D = 6 μm, height 600 nm, etc., and then parameter scanning is performed. By adjusting the dimensions such as the height, diameter, and period of the silicon cylinder, simulate the electric field distribution and multi-pole resonance response in the 8-12 μm infrared band, and focus on observing the excitation of modes such as electric dipole (ED) and magnetic dipole (MD). After each adjustment, combined with the reflection spectrum data measured by a Fourier transform infrared spectrometer (FTIR), use the formula A = 1 - R to calculate the absorption rate, and compare the simulation and experimental results. Repeatedly iterate to optimize the geometric parameters, adjust the period, diameter ratio, or height until efficient and precise absorption peak coverage is achieved within the target band, ensure that the absorption characteristics meet the expected requirements, and finally determine the most optimal parameters of the silicon cylinder.

[0099] Although this embodiment uses silicon cylinders as the basic unit, the manufacturing process is not limited to the cylindrical shape. By adjusting the design and processing process parameters, metasurface structures of other shapes can be prepared, such as square columns, cone columns, ellipsoidal columns, etc. These micro-nano structures of different shapes can also excite specific electromagnetic resonance modes to achieve efficient absorption and radiation control within the target band. Selecting different geometric shapes not only expands the design flexibility but also further optimizes the spectral response characteristics to meet more diverse application requirements.

[0100] Define the amplification parameter s as the scaling factor of the silicon pillar size, based on the optimized geometric parameters. For example, when s = 1.4, the initial period P = 10 μm and diameter D = 6 μm are amplified to P' = 14 μm and D' = 8.4 μm respectively, while the height remains unchanged at 600 nm. Different s values (such as 1.2, 1.4, 1.6, etc.) are introduced into the simulation software, and the movement of the peak position and the change of the resonance mode in the absorption spectrum are gradually observed. The simulation results show that when s = 1.4, the array resonates mainly due to the electric dipole mode at a wavelength of about 8.5 μm, and the absorption peak gradually covers the entire 8–12 μm fingerprint region. According to the simulation data, further refine the step size of s value adjustment (1, 1.2, 1.4, …, 2.6) to accurately match the target absorption peak position and optimize the absorption intensity. After each adjustment, verify through simulation and combine with FTIR experimental measurement to calibrate the final design, ensuring that under the adjusted geometric dimensions, the metasurface array chip achieves the expected infrared absorption performance and dynamic tunable characteristics.

[0101] Figure 4 is a schematic structural diagram of the thermal radiation chip 20 according to an embodiment of the present invention. As Figure 4 shown, from left to right are the schematic diagrams of the thermal radiation chip 20, a single metasurface, and the structural unit, clearly showing the multi-layer structure of the thermal radiation chip 20 and the morphology of the silicon cylinder array after micro-nano processing. Figure 5 is a schematic diagram of the infrared absorption spectrum of the thermal radiation chip 20 according to an embodiment of the present invention. Figure 5 The left diagram in [Figure] shows the infrared absorption spectrum of the metasurface array, highlighting the trend of the resonance peak changing with the structure amplification parameter. The simulation shows that the metasurface array exhibits unique absorption characteristics in the mid-infrared band, and there are significant differences in the absorption intensity at different wavelengths. At different array sizes and periods, the absorption peak can be precisely tuned to cover the range of 7 - 12 μm. This cylindrical structure can be decomposed into different electromagnetic resonance modes, including electric dipole (ED), magnetic dipole (MD), electric quadrupole (EQ), and magnetic quadrupole (MQ). The right diagram shows the heat map of the absorption intensity of the metasurface array under different amplification parameters s, intuitively reflecting the distribution of the absorption performance in the entire mid-infrared band. By simultaneously scaling up the diameter and period of the silicon cylinder in equal proportion to adjust the overall size, it is called the amplification parameter s. The simulation results show that as the amplification parameter s increases, the resonance peak dominated by the electric dipole moves correspondingly towards the long-wave direction.

[0102] Adopt a multi-level decomposition method to analyze the contribution of each mode to the total absorption. The simulation results show that the electric dipole scattering mode is the main absorption contributor, indicating that the electric field inside the silicon cylinder at this wavelength induces a significant non-uniform charge distribution, resulting in strong electric dipole resonance and forming an obvious absorption peak.

[0103] In step S300’, the corrected reaction signal data is divided into a training set and a test set. The training set is used to train the artificial intelligence model 40, and the test set is used to test the artificial intelligence model 40.

[0104] In this embodiment, by using the metasurface heating to generate a thermal radiation signal and combining the convolutional neural network model and the infrared camera 30, the detection of complex chemical mixtures is realized. As a tunable thermal radiation source, the metasurface can provide a stable and broadband infrared light signal in the mid-infrared band, covering the characteristic absorption of substances such as volatile organic compounds. The radiation signal after passing through the sample is collected by the infrared camera 30, and the radiation data is processed and analyzed in real time by means of the convolutional neural network model. This embodiment proposes an integrated, miniaturized and low-cost solution, which can be applied to fields such as environmental monitoring and industrial control to replace the traditional Fourier transform infrared spectroscopy (FTIR) system.

[0105] Compared with the prior art, the main improvement of this embodiment is that a metasurface thermal radiation light source is adopted to replace the traditional expensive and complex infrared light source (such as a blackbody radiation source or a tunable laser), thereby significantly reducing the system cost. Secondly, the infrared camera 30 is used to replace the Michelson interferometer in the FTIR, simplifying the structure of the system and enhancing the miniaturization and integration capabilities. At the same time, the convolutional neural network model is combined to efficiently analyze and process the radiation data, enhancing the system's recognition ability and sensitivity to complex mixtures. This improvement greatly reduces the overall cost and maintenance requirements of the system, making it have a broader practical application prospect.

[0106] This embodiment also provides an artificial intelligence detection system 100 for complex substances based on the thermal radiation chip 20, which is applied to the construction method of any of the above embodiments. The complex substance detection system 100 includes a heating device 10, a thermal radiation chip 20, an infrared camera 30 and an artificial intelligence model 40. The thermal radiation chip 20 is arranged on one side of the heating device 10, and the thermal radiation chip 20 is configured to generate an infrared thermal radiation signal with preset radiation characteristics under the heating of the heating device 10. The infrared camera 30 is arranged on the side of the metasurface chip away from the heating device 10. The standard sample 200 is placed between the thermal radiation chip 20 and the infrared camera 30. The infrared camera 30 is configured to obtain the signal after the infrared thermal radiation signal reacts with the standard sample 200 and process the signal to generate a data packet. The artificial intelligence model 40 is arranged on the side of the infrared camera 30 away from the thermal radiation chip 20, and the artificial intelligence model 40 is configured to receive the data packet and detect the components of the standard sample 200 and predict the concentration of the components according to the data packet.

[0107] During the detection process, the sample to be tested is placed in the optical path of the thermal radiation chip 20, and the thermal radiation chip 20 is heated by the heating device 10 to generate an infrared thermal radiation signal with a wide spectrum. When this signal passes through the sample to be tested, radiation of different wavelengths undergoes specific absorption reactions with the sample, and the infrared camera 30 captures and records the radiation changes after passing through the sample, generating multi-dimensional data containing sample characteristics. The obtained data to be tested, after background correction and normalization preprocessing, is input into a trained one-dimensional convolutional neural network (1D-CNN) model for realizing two core functions: sample substance classification and component concentration prediction.

[0108] During the substance classification process, the 1D-CNN model extracts key features in the spectral data through deep convolution, accurately identifies and judges the category information of the sample. The classification results are visualized through a confusion matrix, intuitively reflecting the classification accuracy of each category, and analyzing possible misclassification situations to ensure the reliability and accuracy of the results. During the concentration prediction process, the 1D-CNN model, through deep learning of the spectral data, establishes a non-linear mapping relationship between spectral features and component concentrations, and finally outputs the specific concentration values of each component (such as A%, B%, C%). The prediction results are quantitatively evaluated through indicators such as mean square error (MSE) and coefficient of determination (R 2 ) and further verified for the prediction accuracy and robustness of the model within different concentration ranges by combining error distribution analysis.

[0109] This embodiment, through the combination of the metasurface and the infrared camera 30, abandons complex and bulky components such as the Michelson interferometer in the FTIR system, achieving high integration and miniaturization. The infrared camera 30 is small in size and can obtain thermal radiation signals in real time. Combining with artificial intelligence (AI) technology for signal processing and analysis, it replaces the large structure of traditional spectrometers, making the entire system suitable for scenarios with high requirements for volume and integration, such as portable devices and industrial sensing nodes.

[0110] This embodiment introduces artificial intelligence technology, which can efficiently process the thermal radiation signals obtained by the infrared camera 30 and automatically analyze complex radiation data. The convolutional neural network can quickly classify and identify the components of the mixture and accurately predict the component concentrations. This greatly improves the system's ability to process complex samples, can handle mixtures with multiple components and high overlapping features, and solves the identification difficulties in the face of complex spectra in traditional technologies. The thermal radiation generated by the metasurface covers a wide frequency band in the mid-infrared spectrum and can effectively detect the spectral information in the molecular "fingerprint region". Combining with the signal processing ability of AI technology, the system can achieve high-precision detection of multiple substances without reducing sensitivity, especially for the identification of trace components. This makes the system perform particularly well in the analysis of mixtures in complex environments and is applicable to multiple fields such as environmental monitoring, material identification, and industrial quality control.

[0111] This embodiment proposes a method for classifying substances based on a thermal radiation chip 20 and a convolutional neural network (1D-CNN) model, which can efficiently classify 50 different substances.

[0112] Figure 7 It is a schematic diagram of the infrared transmission spectra of methanol, ethanol, and acetone measured by FTIR according to an embodiment of the present invention. Figure 8 It is a schematic diagram of obtaining the characteristic temperature data of methanol, ethanol, and acetone by using the complex substance artificial intelligence detection system 100 based on the thermal radiation chip 20 according to an embodiment of the present invention. Figure 9 It is a schematic diagram of the confusion matrix for classifying methanol, ethanol, and acetone by using the complex substance artificial intelligence detection system 100 based on the thermal radiation chip 20 according to an embodiment of the present invention. Figure 10 It is a schematic diagram of the confusion matrix for classifying 50 different mixed ratio samples by using the complex substance artificial intelligence detection system 100 based on the thermal radiation chip 20 according to an embodiment of the present invention. Figure 11 It is a schematic diagram of the result of predicting the methanol concentration by using the complex substance artificial intelligence detection system 100 based on the thermal radiation chip 20 according to an embodiment of the present invention. Figure 12 It is a schematic diagram of the result of predicting the ethanol concentration by using the complex substance artificial intelligence detection system 100 based on the thermal radiation chip 20 according to an embodiment of the present invention. Figure 13 It is a schematic diagram of the result of predicting the acetone concentration by using the complex substance artificial intelligence detection system 100 based on the thermal radiation chip 20 according to an embodiment of the present invention. See Figure 7 , Figure 7 shows the infrared transmission spectra of three pure substances (methanol, ethanol, and acetone). The data are from Fourier transform infrared spectroscopy (FTIR) measurements. The horizontal axis represents the wavelength (μm), and the vertical axis represents the transmittance. Each substance exhibits different absorption peaks within a specific wavelength range, which is due to different vibration modes within the molecule, resulting in the absorption of light at specific wavelengths. These characteristic peaks of the transmission spectrum are important bases for substance identification and can be used to distinguish different chemical components. The traditional substance detection is that the spectral measurement results of FTIR can be used as reference standards. Our goal is to identify by obtaining the characteristic temperature data of the radiation through the application of the complex substance artificial intelligence detection system 100 based on the thermal radiation chip 20, without relying on traditional spectrometers. This is also one of the core innovations of this research.

[0113] See Figure 8 , Figure 8It shows the infrared characteristic temperature data obtained by the detection equipment developed by us. Each bar chart corresponds to methanol, ethanol, and acetone respectively. Compared with the transmission spectrum measured by FTIR, the data here is based on the response of substances to infrared thermal radiation. Specifically, our detection system 100 uses a heated infrared metasurface as the radiation source. Different substances have different absorption characteristics of infrared radiation, resulting in different temperature imaging on the infrared camera 30. By analyzing these temperature characteristic data, we can use machine learning methods to train a classification model, thus realizing the automatic identification of substances. This process gets rid of the high cost of traditional FTIR spectroscopy and provides the possibility for low-cost and portable infrared detection solutions.

[0114] See Figure 9 , Figure 9 It shows the classification results based on the machine learning model, targeting single-component samples (i.e., pure samples of methanol, ethanol, and acetone). The horizontal axis of the confusion matrix represents the predicted labels of the model, and the vertical axis represents the true labels. The values in the matrix represent the number of samples. For example, the "200" on the diagonal means that 200 samples are correctly classified. Since all non-zero values are strictly distributed on the diagonal, it shows that the classification accuracy of the model for the three substances reaches 100%, that is, all test samples are correctly classified. This indicates that the convolutional neural network model has successfully captured the substance characteristics in the infrared thermal radiation data and achieved accurate substance identification.

[0115] In further experiments, we studied 50 samples with different mixing ratios, obtained data using the same infrared thermal imaging method, and then classified them through the trained neural network model. Figure 10 It shows the classification results of the mixed samples. Compared with the classification of pure substances, the complexity here increases because the characteristics of the mixed samples may be closer and more difficult to distinguish. However, from the results of the confusion matrix, it can be seen that the predicted values are still concentrated on the diagonal, indicating that the classification model of the neural network still maintains a high accuracy when dealing with complex mixtures. This result proves the feasibility of this detection method in the mixture system and lays a foundation for subsequent quantitative concentration prediction.

[0116] See Figure 11 , Figure 11Shows the predicted results of methanol concentration in 50 different mixing ratio samples. The horizontal axis represents the true concentration (vol%), and the vertical axis represents the model-predicted concentration (vol%). The distribution of data points shows the prediction performance of the model. The dashed line represents the ideal linear fitting relationship, that is, the predicted value should highly coincide with the true value. The error bars represent the uncertainty of the model prediction or the fluctuations among samples. It can be seen from the figure that the data points are basically distributed near the linear fitting curve, indicating that the model has a high prediction accuracy for methanol concentration and small errors. This shows that through infrared thermal imaging combined with neural network analysis, we can accurately estimate the methanol concentration in the mixture.

[0117] Figure 12 Shows the predicted results of ethanol concentration. The horizontal axis represents the true concentration (vol%), and the vertical axis represents the predicted concentration (vol%). Similarly, the data points are distributed around the linear fitting curve, and the error bars are small, indicating that the model is also very accurate in predicting the ethanol concentration. This means that although the infrared thermal imaging signal may be affected by the mixture, the neural network can still effectively learn the characteristics of ethanol and separate them from the mixture signal to achieve accurate quantitative analysis.

[0118] Figure 13 Shows the concentration prediction of acetone under different mixing ratios. The horizontal axis is the true concentration (vol%), and the vertical axis is the predicted concentration (vol%). The data points still closely distribute around the linear fitting curve, and the range of the error bars indicates that the model can provide stable prediction results in most cases. Generally speaking, the predicted results of methanol, ethanol, and acetone concentrations all show good linear correlations, indicating that our detection system 100 can not only classify substances but also achieve quantitative analysis of the concentrations of each component in the mixture.

[0119] This experiment demonstrates a new method for detecting substances of a complex substance artificial intelligence detection system 100 based on a thermal radiation chip 20. From the characteristic analysis of pure substances, to the classification and identification of mixtures, and then to concentration prediction, the experimental results show that this method can accurately identify and quantitatively analyze each component in the mixture. Figures 7 - 10It shows the FTIR spectrum, thermal radiation characteristic data, and the classification result confusion matrix, indicating that the machine learning model has successfully achieved substance classification based on thermal imaging data and still maintains high accuracy even in a mixture system. The second row shows the concentration prediction of each component in the mixture. The predicted data highly coincides with the true concentration with small errors, indicating that this method can replace traditional spectral measurements for non-contact concentration detection. This study proves that by combining the thermal radiation chip 20, thermal imaging, and machine learning, accurate infrared detection can be achieved, providing a new solution for low-cost and portable substance detection, with broad application potential, such as in the fields of biological detection, environmental monitoring, and food safety analysis. This embodiment proposes a concentration prediction method based on a thermal radiation metasurface chip and a 1D convolutional neural network (1D-CNN) model, aiming to provide accurate concentration prediction through the mapping relationship between spectral features and sample concentration. In the experiment, first, a heating stage is used to heat the thermal radiation metasurface chip to generate a broadband infrared thermal radiation signal. The sample to be measured is placed on the chip array, and the infrared signal interacts with the sample to generate corresponding spectral data. The infrared camera 30 is used to capture the radiation changes of the sample in real time to generate multi-dimensional data. The background signal is collected for correction, and the sample signal is standardized to ensure data consistency. The standardized signal data is input into the 1D-CNN model, and the model extracts the radiation temperature features through multiple convolutional and pooling layers to learn the relationship between the data features and the sample concentration. Finally, the artificial intelligence model 40 can accurately predict the concentration of each component in the sample, and the experimental results are as Figures 7 - 13 shown. The mean and error bars of the concentration predictions of the three components in the mixture are plotted in the figure.

[0120] To verify the advantages of the method of the present invention, a comparative experiment was designed to compare the performance of a traditional FTIR infrared spectrometer and a system based on a thermal radiation metasurface chip in the classification and concentration prediction of 50 substances. In the substance classification task, the traditional method has a high recognition accuracy for single substances because it can be compared with the standard spectrum. However, due to spectral overlap in the spectral measurement of the three mixtures, it is difficult to distinguish the classification and mixed concentration prediction of several components with similar concentrations. The experimental results show that the method of the present invention has obvious advantages in both the accuracy of substance classification and concentration prediction, verifying the high efficiency and precision of the combination of the metasurface chip and the 1D-CNN model.

[0121] Up to this point, those skilled in the art should recognize that although multiple exemplary embodiments of the present invention have been shown and described in detail herein, many other variations or modifications that conform to the principles of the present invention can still be directly determined or derived from the content disclosed in the present invention without departing from the spirit and scope of the present invention. Therefore, the scope of the present invention should be understood and determined to cover all these other variations or modifications.

Claims

1. A method for constructing a complex material artificial intelligence detection system based on a thermal radiation chip, characterized in that: The steps include: Preparing a thermal radiation chip and constructing an artificial intelligence model based on the thermal radiation chip; Sequentially placing different standard samples between the thermal radiation chip and the infrared camera, and heating the thermal radiation chip so that the infrared camera acquires reaction signal data after the thermal radiation chip undergoes absorption reactions with different standard samples; The reaction signal data of different standard samples are input into the artificial intelligence model to adjust the parameters of the artificial intelligence model so as to train the artificial intelligence model.

2. The construction method according to claim 1, characterized in that: The steps of preparing a thermal radiation chip and constructing an artificial intelligence model based on the thermal radiation chip further include the following steps: heating the thermal radiation chip and obtaining a background signal of the thermal radiation chip that has not passed through the standard sample; Correcting the reaction signals of different standard samples using the background signal to obtain corrected reaction signal data; The artificial intelligence model is trained using the corrected response signal data to obtain a trained model.

3. The construction method according to claim 2, characterized in that: The artificial intelligence model is constructed by the following steps: constructing a first Conv1D layer, wherein the first Conv1D layer is configured to extract local features of the corrected response signal data; Constructing a MaxPooling1D layer, wherein the MaxPooling1D layer is configured to reduce the dimension of the local features; Constructing a second Conv1D layer, wherein the second Conv1D layer is configured to enhance the extraction capability of the local features; Flatten the two-dimensional feature map of the local feature into a one-dimensional vector using a Flatten layer; Constructing a Dense layer, wherein the Dense layer is configured to integrate feature information of the local features; An output layer is constructed, wherein the output layer is configured to output the component detection results of the standard sample and the predicted results of the concentrations of the components.

4. The construction method according to claim 1, characterized in that: The thermal radiation chip is prepared by the following steps: Construct models of metasurface arrays; Simulating the model of the metasurface array using simulation software to obtain target thickness parameters of each layer of material of the model of the metasurface array; The heat radiation chip is manufactured according to the target thickness parameters of each layer of material.

5. The construction method according to claim 4, characterized in that: The step of preparing the thermal radiation chip according to the target thickness parameters of each layer of material specifically includes the following steps: providing a substrate; Depositing a metal layer on the substrate, the metal layer serving as a reflective base; sequentially preparing a transition layer and a silicon layer on the metal layer by magnetron sputtering or chemical vapor deposition, wherein the transition layer is located between the metal layer and the silicon layer; The silicon layer is micro-nano processed by photolithography or etching technology, so as to prepare the thermal radiation chip with the metasurface array.

6. The construction method according to claim 5, characterized in that: The step of performing micro-nano processing on the silicon layer by using a photolithography or etching process to prepare the thermal radiation chip having a metasurface array specifically comprises the following steps: Using the simulation software to simulate the period of the metasurface array of the silicon layer, the diameter and height of the silicon cylinders, so as to screen out a target period, a target diameter and a target height of the silicon cylinders; The silicon layer is micro-nano processed by photolithography or etching process according to the target period, the target diameter of the silicon cylinder and the target height, so as to prepare the thermal radiation chip with the metasurface array.

7. The construction method according to claim 2, characterized in that: The reaction signal data of different standard samples are input into the artificial intelligence model to adjust the parameters of the artificial intelligence model. In the step of training the artificial intelligence model, the corrected reaction signal data are divided into a training set and a test set, the training set is used to train the artificial intelligence model, and the test set is used to test the artificial intelligence model.

8. An artificial intelligence detection system for complex substances based on a thermal radiation chip, applied to the construction method described in any one of claims 1 to 7, characterized in that: include: Heating device; A heat radiation chip is arranged on one side of the heating device, and the heat radiation chip is arranged to generate an infrared heat radiation signal with preset radiation characteristics under the heating of the heating device; An infrared camera is arranged on a side of the thermal radiation chip away from the heating device, and a standard sample is placed between the thermal radiation chip and the infrared camera. The infrared camera is arranged to obtain a signal after the infrared thermal radiation signal and the standard sample undergo an absorption reaction, and to process the signal to generate a data packet; The artificial intelligence model based on the thermal radiation chip is arranged on a side of the infrared camera away from the thermal radiation chip. The artificial intelligence model is configured to receive the data packet, detect the components of the standard sample according to the data packet, and predict the concentration of the components.

Images