Adhesive component quantitative detection method and system based on spectral analysis

Abstract

The invention provides an adhesive component quantitative detection method and system based on spectral analysis, and relates to the technical field of analysis and detection.According to the method, near infrared spectral data of an adhesive sample are obtained, and environmental background spectral data are synchronously collected for background correction, so that interference is eliminated. And performing signal enhancement processing on the corrected data, extracting spectral features under different scales, and separating overlapped absorption peaks. In the decomposition process, for the possible layering or interfacial effect of the adhesive, an interfacial feature enhancement treatment is introduced to focus on the absorption features of the internal layered structure. A compensation factor of flow-induced spectral distortion is introduced based on the transmission speed of the production line, and characteristics related to the internal structure are corrected. And combining the corrected characteristics into a set, inputting the set into a quantitative prediction model established through supervised learning in advance, and outputting the concentration value of the adhesive component, so that high-precision quantitative detection of the adhesive component can be realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of analytical detection, and in particular to a method and system for quantitative detection of adhesive components based on spectral analysis. BACKGROUND

[0002] In the production and quality control process of adhesives, it is crucial to quickly and accurately determine the concentration of key components (such as curing agents, plasticizers or solvents). In particular, in a continuous production line, real-time online quantitative detection of product components is a core technical requirement to ensure product quality consistency, improve production efficiency and reduce raw material loss. The traditional offline sampling-laboratory analysis method has a serious time lag and cannot meet the requirements of modern intelligent manufacturing for real-time process control.

[0003] To meet the above requirements, near-infrared spectral analysis technology has been applied to the detection of adhesive components. An existing specific solution is to directly obtain the near-infrared spectrum of flowing adhesive through an online probe, and then compare the collected spectral data with a quantitative calibration model established by chemometrics methods such as partial least squares to directly predict the concentration of a specific component. This solution attempts to transition from offline analysis to online detection and has the potential for rapid and non-destructive detection. However, this existing solution has obvious defects in actual industrial applications.

[0004] Firstly, adhesives are complex mixtures with severely overlapping spectral absorption peaks and low signal-to-noise ratio, making it difficult to extract sufficient effective feature information through direct modeling, resulting in insufficient model robustness. Secondly, the continuous flow of adhesives in the production line can cause spectral baseline drift and distortion, and complex internal interface effects (such as delamination and precipitation) further interfere with the characteristic signals directly related to the bulk components. The existing solution lacks a mechanism to specifically correct these dynamic physical disturbances, making the detection results prone to significant deviations under changing production conditions, and the accuracy and reliability are difficult to guarantee. SUMMARY

[0005] The present application aims to provide a method and system for quantitative detection of adhesive components based on spectral analysis to solve the problem of low detection accuracy in the prior art.

[0006] To solve the above technical problems, in a first aspect, the present application provides a method for quantitative detection of adhesive components based on spectral analysis, comprising:

[0007] Obtaining near-infrared spectral data of an adhesive sample to be detected, synchronously collecting background spectral data of an environment in which the adhesive sample is located, and performing background correction on the near-infrared spectral data according to the background spectral data;

[0008] Performing signal enhancement processing on the background-corrected near-infrared spectrum data, and performing multi-scale decomposition on the near-infrared spectrum data after signal enhancement processing to extract spectral features at different scales and separate overlapping absorption peaks;

[0009] In the multi-scale decomposition process, in view of the possible stratification or interface effect of the adhesive, an interface feature enhancement processing is introduced to focus on the absorption features characterizing the internal layered structure of the adhesive;

[0010] According to the transmission speed of the adhesive production line, a flow-induced spectrum distortion compensation factor is introduced to correct the spectral features related to the internal structure extracted after the interface feature enhancement processing;

[0011] The corrected spectral features are combined into a spectral feature set, and the spectral feature set is input into a quantitative prediction model previously established by supervised learning, and the concentration value of the adhesive component is output by the quantitative prediction model to realize quantitative detection.

[0012] Optionally, in the multi-scale decomposition process, in view of the possible stratification or interface effect of the adhesive, an interface feature enhancement processing is introduced to focus on the absorption features characterizing the internal layered structure of the adhesive, comprising:

[0013] Among the multiple groups of spectral shape features obtained by multi-scale decomposition, the wavelength region where the adhesive has significant differences between adjacent scales is identified;

[0014] The difference between the spectral shape features of adjacent scales in the wavelength region is calculated;

[0015] According to the difference, an appropriate prominence coefficient is assigned to the interface feature of the wavelength region;

[0016] Using the prominence coefficient, the feature value of the corresponding wavelength region is adjusted, so that the interface feature characterizing the layered structure is relatively enhanced, focusing on the absorption features characterizing the internal layered structure of the adhesive.

[0017] Optionally, performing multi-scale decomposition on the near-infrared spectrum data after signal enhancement processing to extract spectral features at different scales and separate overlapping absorption peaks, comprising:

[0018] Using each width of a preset group of primitive waveforms with different widths, respectively, to perform matching operation with the near-infrared spectrum data after signal enhancement processing, to obtain a sequence of the degree of fit between the spectral data and the primitive waveforms at the corresponding width;

[0019] For each width of the primitive waveform, according to the sequence of the degree of fit, the spectral shape feature characterized by the primitive waveform in the spectral data is extracted;

[0020] Multiple sets of spectral shape features extracted from primitive waveforms of different widths are combined to form spectral features at multiple scales.

[0021] Based on the spectral characteristics at different scales, overlapping absorption peak regions are identified by comparing the feature differences at different scales, and a peak separation algorithm is applied to deconvolve the overlapping peaks to achieve the separation of absorption peaks.

[0022] Optionally, a compensation factor for flow-induced spectral distortion is introduced based on the conveying speed of the adhesive production line to correct the spectral features extracted after interface feature enhancement processing that are related to the internal structure, including:

[0023] Establish the correspondence between the transmission speed of the adhesive production line and the wavelength shift corresponding to the spectral characteristics at different scales;

[0024] Based on the transmission speed at the current detection time, the compensation factor for flow-induced spectral distortion is calculated using the corresponding relationship;

[0025] The compensation factor is applied to the spectral features related to the internal structure extracted after interface feature enhancement processing, and the wavelength positions corresponding to the spectral features are repositioned.

[0026] Based on the repositioned wavelength, the values ​​of the spectral features are corrected and adjusted accordingly.

[0027] Optionally, the spectral feature set is input into a pre-established quantitative prediction model through supervised learning, and the quantitative prediction model outputs the concentration value of the adhesive component to achieve quantitative detection, including:

[0028] A quantitative prediction model is pre-established, which includes a processing network consisting of multiple layers of computing nodes;

[0029] Each feature in the set of spectral features, which is composed of the corrected spectral features, is input to the corresponding computing node in the starting layer of the processing network.

[0030] For each computing node in the processing network, the output value from the computing node in the previous layer is received, and weighted summation is performed according to the predetermined connection relationship and internal parameters. Then the calculation result is passed to the computing node in the next layer.

[0031] After being passed and calculated layer by layer through all layers of the processing network, the final layer's calculation node outputs one or more numerical values. These output values ​​represent the concentration values ​​of one or more analytes in the adhesive, thus achieving quantitative detection of adhesive components.

[0032] Optionally, background correction is performed on the near-infrared spectral data based on the background spectral data, including:

[0033] Obtain background spectral data under the same environmental conditions when the adhesive sample is absent;

[0034] Align the background spectral data with the near-infrared spectral data obtained from the adhesive sample at each wavelength point;

[0035] Subtract the values ​​of the near-infrared spectral data at the corresponding wavelength points from the values ​​of the background spectral data at each wavelength point to obtain a set of differential spectral data.

[0036] A baseline adjustment operation is performed on the differential spectral data so that the initial values ​​of the differential spectral data in the specified wavelength range are brought close to zero, thereby achieving background correction of the near-infrared spectral data.

[0037] Optionally, signal enhancement processing is performed on the background-corrected near-infrared spectral data, including:

[0038] For the background-corrected near-infrared spectral data, calculate the degree of variation at different wavelengths;

[0039] Based on the degree of change, wavelength regions with drastic changes are enhanced, while wavelength regions with gradual changes are smoothed.

[0040] The entire near-infrared spectral data is divided into multiple continuous wavelength intervals, and the above-mentioned enhancement and smoothing processes are performed independently on each wavelength interval.

[0041] The data segments, which have been processed independently in each wavelength range, are connected and combined to form near-infrared spectral data that has undergone signal enhancement processing.

[0042] Secondly, this application provides a quantitative detection system for adhesive components based on spectral analysis, comprising:

[0043] The acquisition module is used to acquire near-infrared spectral data of the adhesive sample to be tested, simultaneously acquire background spectral data of the environment in which the adhesive sample is located, and perform background correction on the near-infrared spectral data based on the background spectral data.

[0044] The decomposition module is used to perform signal enhancement processing on the background-corrected near-infrared spectral data, and at the same time, to perform multi-scale decomposition on the signal-enhanced near-infrared spectral data to extract spectral features at different scales and separate overlapping absorption peaks.

[0045] The processing module is used to introduce interface feature enhancement processing to address the possible layering or interface effects of the adhesive during the multi-scale decomposition process, so as to focus on the absorption characteristics that characterize the layered structure of the adhesive.

[0046] The correction module is used to introduce a compensation factor for flow-induced spectral distortion based on the transmission speed of the adhesive production line, and to correct the spectral features extracted after interface feature enhancement processing that are related to the internal structure.

[0047] The input module is used to combine the corrected spectral features into a spectral feature set, and input the spectral feature set into a quantitative prediction model established in advance through supervised learning. The quantitative prediction model outputs the concentration value of the adhesive component to achieve quantitative detection.

[0048] Thirdly, this application provides an electronic device, comprising:

[0049] Memory, used to store computer programs;

[0050] A processor is configured to execute the computer program to implement the steps of a method for quantitative detection of adhesive components based on spectral analysis as described in the first aspect above.

[0051] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the steps of the quantitative detection method for adhesive components based on spectral analysis as described in the first aspect above.

[0052] This application provides a method for quantitative detection of adhesive components based on spectral analysis. By simultaneously acquiring background spectra and performing background correction, it can effectively eliminate environmental noise and baseline interference, improving the purity of spectral data. By performing signal enhancement processing on the corrected data, it can improve the signal-to-noise ratio and highlight effective spectral information. By simultaneously performing multi-scale decomposition, it can extract spectral features at different scales and effectively separate overlapping absorption peaks, thereby resolving masked details. By introducing interface feature enhancement processing, it can focus on the absorption features characterizing the internal layered structure of the adhesive, strengthening key signals generated by layering or interface effects. By introducing a compensation factor for flow-induced spectral distortion based on transmission speed, it can offset the spectral distortion caused by the flow state of the production line, improving the accuracy of online detection. Finally, by inputting the corrected feature set into a pre-established quantitative prediction model, it can achieve high-precision and rapid quantitative analysis of adhesive component concentration.

[0053] Furthermore, in the spectral shape features obtained from multi-scale decomposition, wavelength regions with significant differences between adjacent scales are identified, their differences are calculated, and appropriate prominence coefficients are assigned to the interface features accordingly. This adjusts the feature values, thereby relatively enhancing the interface features characterizing layered structures. This approach can accurately capture and amplify subtle spectral changes caused by the internal layered structure or interface effects of adhesives, effectively separating and highlighting interface features from complex spectral backgrounds. This provides the model with more direct and significant key discriminative features, fundamentally improving the model's prediction accuracy and reliability for sample components exhibiting layering or interface effects. Attached Figure Description

[0054] To more clearly illustrate the technical solutions of the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1 A schematic flowchart of a method for quantitative detection of adhesive components based on spectral analysis provided in this application embodiment;

[0056] Figure 2 A schematic diagram illustrating a specific implementation of a method for quantitative detection of adhesive components based on spectral analysis, provided in this application embodiment;

[0057] Figure 3 This is a schematic diagram of a quantitative detection system for adhesive components based on spectral analysis, provided in an embodiment of this application. Detailed Implementation

[0058] In the online quantitative detection of adhesives using near-infrared spectroscopy, existing direct modeling methods face a core contradiction: the single model established for component analysis is fundamentally mismatched with the complex physicochemical state of adhesives in actual production. Specifically, the spectral absorption peaks of adhesive mixtures severely overlap, masking the characteristic signals of key components; simultaneously, the continuous flow of the production line causes significant drift and distortion of the spectral baseline, and interface effects such as stratification and phase separation within the product further introduce interference signals unrelated to the components. Existing solutions lack mechanisms for targeted analysis and removal of these dynamic physical interferences, resulting in insufficient robustness of the established quantitative models in stable and changing industrial environments, making it difficult to guarantee detection accuracy.

[0059] To fundamentally address the aforementioned problems, this application proposes a quantitative detection method for adhesive components based on multi-level feature analysis and adaptive correction. The core of this method lies in constructing a step-by-step spectral processing flow: First, background correction and signal enhancement preprocessing improve the quality of the original signal. Then, innovative multi-scale decomposition technology is employed to remove overlapping spectral peaks, and a specialized interface feature enhancement algorithm is introduced during this process to focus and amplify the feature signals generated by the internal layered structure. Simultaneously, a compensation factor is dynamically introduced based on the real-time production line transmission speed to accurately correct spectral distortions caused by flow. Finally, the spectral feature set, significantly improved in terms of anti-interference capability after the above steps, is input into a pre-trained quantitative prediction model. This method, by decoupling and specifically processing chemical component signals, physical interface effects, and dynamic flow interference at each level, achieves high-fidelity extraction of target component information from complex backgrounds. This effectively overcomes the shortcomings of existing technologies, such as poor adaptability to complex working conditions and low model accuracy, providing key technical support for achieving high-precision and high-reliability online quantitative detection.

[0060] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0061] The core of this application is to provide a method for quantitative detection of adhesive components based on spectral analysis, and a flowchart of one specific implementation is shown below. Figure 1 As shown, the method includes:

[0062] S101. Obtain the near-infrared spectral data of the adhesive sample to be tested, simultaneously collect the background spectral data of the environment in which the adhesive sample is located, and perform background correction on the near-infrared spectral data based on the background spectral data.

[0063] Optionally, step S101 may specifically include the following steps:

[0064] S1011. Obtain background spectral data under the same environmental conditions when the adhesive sample is not present;

[0065] S1012. Align the background spectral data with the near-infrared spectral data obtained from the adhesive sample at each wavelength point;

[0066] S1013. Subtract the values ​​of the near-infrared spectral data at the corresponding wavelength points from the values ​​of the background spectral data at each wavelength point to obtain a set of differential spectral data.

[0067] S1014. Perform a baseline adjustment operation on the differential spectral data so that the initial value of the differential spectral data in the specified wavelength range returns to near zero, thereby achieving background correction of the near-infrared spectral data.

[0068] In the above scheme, near-infrared spectral data refers to a set of numerical sequences acquired by a near-infrared spectrometer that reflect the degree of near-infrared light absorption by the adhesive sample. The horizontal axis represents wavelength or wavenumber, and the vertical axis represents absorbance. Background spectral data refers to spectral data acquired under identical environmental conditions, but only when there is no adhesive sample to be tested. This data includes environmental noise and inherent signals of the instrument itself. Differential spectral data refers to the spectral data obtained by subtracting the sample's near-infrared spectral data from the background spectral data at corresponding wavelength points. Its purpose is to initially eliminate common background interference. Baseline adjustment is a mathematical processing of the spectral data, designed to eliminate the tilt or shift of the overall spectral curve caused by light source fluctuations, particle scattering, etc., so that it has a consistent reference level within the specific wavelength range of interest.

[0069] In this embodiment, firstly, in step S1011, before or during detection, it is ensured that there is no adhesive sample in the sample detection area. A set of background spectral data is collected by a near-infrared spectrometer, which serves as a template for pure noise and interference in the current environment. For example, on Company A's production line, before each batch change, the operator starts the equipment's built-in background acquisition program, and the spectrometer scans the empty transmission pipeline to obtain the spectral background of the current environment. Secondly, in step S1012, the background spectral data obtained in step S1011 is matched one-to-one with the near-infrared spectral data actually measured from the adhesive sample to ensure that subsequent calculations are performed on the same wavelength scale. For example, the spectrometer control software ensures that both the background spectrum and the sample spectrum are within the wavelength range of 1100 nm to 2500 nm, and is collected at 2 nm intervals, for a total of 701 data points, thereby ensuring strict alignment of the two spectral sequences. Next, in step S1013, at each aligned wavelength point, the sample near-infrared spectrum data is subtracted from the background spectral data to obtain a new set of differential spectral data. For example, at the wavelength point of 1200 nm, the absorbance value of the background spectrum is 0.05, while the absorbance value of the sample spectrum is 0.88. Therefore, the calculated value of the differential spectrum at this point is 0.05 minus 0.88, which equals -0.83. This result initially removes the influence of the background environment at this wavelength point. Finally, in step S1014, baseline adjustment is performed on the differential spectral data obtained in S1013. Methods such as standard normal transformation are typically used to bring the initial values ​​of the spectral curve to zero within a specified characteristic wavelength range that is insensitive to the target component, thereby correcting baseline drift caused by physical scattering. For example, the range of 1800 nm to 1900 nm, where solvent absorption is not significant, is selected as the reference range. Through mathematical fitting, the average value of the differential spectrum in this range is adjusted to zero, ensuring that the entire spectral curve falls on a unified reference line, completing the final background correction.

[0070] In practical applications, on the continuous adhesive production line of Chemical Company A, the online near-infrared spectrometer C, located at the outlet of reactor B, initiates the detection process. First, spectrometer C acquires background spectral data D of the current environment (including the detection cell window and ambient light) the instant the adhesive sample does not flow through the detection cell. Then, when adhesive sample E flows through the detection cell, spectrometer C acquires its near-infrared spectral data F. The data processing unit G aligns the background spectrum D with the sample spectrum F within the wavelength range of 1100 nm to 2500 nm. Then, at each wavelength point, the value of the sample spectrum F is subtracted from the value of the background spectrum D to obtain the preliminary differential spectrum H. Finally, baseline correction is performed on the differential spectrum H within the specified wavelength range of 1550 nm to 1650 nm, shifting the spectral values ​​in this range to near zero, ultimately outputting clean spectral data I after sufficient background correction for subsequent analysis.

[0071] This method, by simultaneously acquiring and subtracting the environmental background spectrum and performing baseline correction, can effectively eliminate interference signals introduced by factors such as the detection instrument itself, the surrounding environment, and non-specific scattering in the sample. This results in a near-infrared spectrum with a flat and pure baseline that is only related to the chemical composition of the adhesive sample itself, laying a reliable data foundation for subsequent accurate feature extraction and quantitative analysis.

[0072] S102. Perform signal enhancement processing on the near-infrared spectral data after background correction, and at the same time perform multi-scale decomposition on the near-infrared spectral data after signal enhancement processing to extract spectral features at different scales and separate overlapping absorption peaks.

[0073] Optionally, step S102 may specifically include the following steps:

[0074] S1021. Calculate the degree of variation at different wavelengths for the background-corrected near-infrared spectral data.

[0075] S1022. Based on the degree of change, the wavelength region with drastic changes is enhanced, and the wavelength region with gentle changes is smoothed.

[0076] S1023. Divide the entire near-infrared spectral data into multiple continuous wavelength intervals, and perform the above-mentioned enhancement and smoothing processing independently on each wavelength interval.

[0077] S1024. Connect the data segments that have been processed independently in each wavelength range and combine them to form near-infrared spectral data that has undergone signal enhancement processing.

[0078] S1025. Using a preset set of primitive waveforms with different widths, each of the primitive waveforms is matched with the near-infrared spectral data after signal enhancement processing to obtain a sequence of the degree of matching between the spectral data and the primitive waveform at the corresponding width.

[0079] S1026. For each width of the primitive waveform, extract the spectral shape features represented by the primitive waveform in the spectral data according to the matching degree sequence.

[0080] S1027. The multiple sets of spectral shape features extracted from primitive waveforms of different widths are combined to form spectral features at multiple scales.

[0081] S1028. Based on the spectral characteristics at different scales, overlapping absorption peak regions are identified by comparing the feature differences at different scales, and a peak separation algorithm is applied to deconvolve the overlapping peaks to achieve the separation of absorption peaks.

[0082] In the above scheme, the degree of change refers to the rate of change or derivative of the absorbance value at each wavelength point in the near-infrared spectral data, used to quantify the severity of local fluctuations in the spectral curve, and is usually calculated using the first or second derivative; the wavelength region refers to a continuous wavelength range in the spectrum; enhancement processing is a signal processing technique aimed at increasing the amplitude or contrast of the effective signal; smoothing processing is another signal processing technique aimed at suppressing random noise and making the curve smoother; data segment refers to the spectral data portion corresponding to a single wavelength interval after independent processing. The degree of fit sequence refers to a numerical sequence obtained through operations such as convolution, which characterizes the matching similarity between the primitive waveform and the spectral signal at each position as the primitive waveform slides along the wavelength axis; spectral shape features are parameters extracted from the degree of fit sequence that can represent the morphological information such as the local spectral peak shape and peak width; multi-scale spectral features refer to the set of features extracted using primitive waveforms of different widths, reflecting the details and overall shape of the spectral signal at different scales; deconvolution is a mathematical algorithm used to estimate the original shape of each individual component signal from an overlapping composite signal.

[0083] In this embodiment, firstly, in step S1021, the first derivative or moving window variance of the background-corrected spectral data is calculated to quantify the degree of local variation of the signal near each wavelength point. For example, when analyzing the spectrum of an adhesive, the calculated derivative value is large in the region near 1700 nm corresponding to the absorption of the C=O bond in the curing agent, indicating that the spectrum in this region changes drastically; while in the region near 1850 nm where there is no characteristic absorption, the derivative value is close to zero, indicating that the variation is gradual. Secondly, in step S1022, based on the variation degree map calculated in S1021, for regions with drastic variations, such as near 1700 nm, an appropriate algorithm, such as weighted enhancement, is used to make its characteristics more prominent; at the same time, for regions with gradual variations, such as near 1850 nm, smoothing processing such as moving average filtering is performed to suppress random noise. Next, in step S1023, the entire spectral range, for example, 1100-2500 nm, is divided into multiple consecutive smaller wavelength intervals, such as intervals of 100 nm each, and the enhancement and smoothing strategies of S1022 are applied independently to each interval. This segmented processing method avoids local feature distortion caused by using the same enhancement parameter globally. Then, in step S1024, all data segments within the wavelength range that have been processed independently are seamlessly reconnected according to their original wavelength order, combining them into a complete signal-enhanced spectral curve with improved global signal-to-noise ratio. Subsequently, in step S1025, multi-scale decomposition is performed. A set of preset Gaussian functions with different widths are used as primitive waveforms, and convolution operations are performed with the signal-enhanced spectral data respectively. The convolution operation formula can be expressed as Ci(λ)=∫Gi(u)·S(λ-u)du, where Gi(u) is the Gaussian primitive waveform of the i-th width, S is the signal-enhanced spectrum, and Ci(λ) is the sequence of fit under the primitive waveform of that width. For example, convolution with a narrow-width Gaussian function can obtain a detail sequence with a strong response to sharp absorption peaks; while convolution with a wide-width Gaussian function can obtain a trend sequence reflecting a broad and gentle absorption background. Next, in step S1026, for each width primitive waveform, the obtained fit sequence Ci(λ) is identified, along with its local extrema, amplitude, full width at half maximum (FWHM), and other parameters. These parameters constitute the spectral shape features extracted at that scale. Narrow-scale features capture fine peak information, while wide-scale features reflect the overall envelope. Then, in step S1027, multiple sets of spectral shape features extracted using primitive waveforms of different widths are aggregated to form a multi-scale spectral feature set that comprehensively describes the spectrum from microscopic details to macroscopic trends. Finally, in step S1028, based on the obtained multi-scale features, the differences in feature responses at different scales are compared. For example, a strong response at a certain wavelength position at a narrow scale but a weak response at a wide scale may indicate the presence of overlapping sharp peaks.Based on this, peak separation algorithms such as Gaussian fitting or nonlinear least squares fitting are applied to the identified overlapping peak regions for deconvolution, thereby estimating key information such as the peak position and peak area of ​​each individual component, and realizing the separation of absorption peaks.

[0084] In practical applications, within Company A's online adhesive detection system, the corrected spectrum J output from step S101 undergoes processing in step S102. The system first calculates the first derivative of spectrum J, identifying two drastically changing regions near 1680 nm and 1720 nm, presumably representing absorption peaks of key components. The system then enhances the entire 1680-1730 nm range while smoothing the gentler regions on either side. Next, the system divides the entire spectrum into 14 equally wide intervals and performs this adaptive processing on each, finally stitching them together to form the enhanced spectrum K. Subsequently, the system performs multi-scale convolution operations on spectrum K using a set of Gaussian functions with width parameters σ of 5 nm, 15 nm, and 30 nm, obtaining three sets of consistency sequences L1, L2, and L3. Parameters for multiple sharp peaks are extracted from L1, and the trend of the gentle background is extracted from L3. All these features are aggregated into a multi-scale feature set M. By comparison, it was found that L1 showed a double peak at 1690 nm, while L2 almost merged into a single broad peak. Based on this, the system determined that there were severely overlapping peaks at this point and successfully separated the two sub-peaks N1 and N2 by calling the quadratic Gaussian fitting algorithm, providing a purer feature input for quantitative analysis.

[0085] This scheme effectively improves the signal-to-noise ratio of useful component features in the spectrum through adaptive signal enhancement processing, while suppressing random fluctuations. Furthermore, through multi-scale decomposition, complex spectral information is separated and extracted according to feature scales, which can simultaneously capture subtle sharp absorption peaks and broad overall background trends. Finally, based on the difference identification of multi-scale features, overlapping absorption peaks are separated, thereby clearly resolving the feature information of masked individual components, providing rich, pure, and hierarchical feature inputs for subsequent high-precision quantitative modeling.

[0086] S103. In the multi-scale decomposition process, in view of the possible layering or interface effects of the adhesive, an interface feature enhancement process is introduced to focus on the absorption characteristics that characterize the layered structure of the adhesive.

[0087] Optionally, step S103 may specifically include the following steps:

[0088] S1031. Among the multiple sets of spectral shape features obtained through multi-scale decomposition, wavelength regions in which there are significant differences in adhesives between adjacent scales are identified.

[0089] S1032. Calculate the difference in spectral shape features between adjacent scales within the wavelength region;

[0090] S1033. Based on the difference, assign a suitable prominence coefficient to the interface features of the wavelength region;

[0091] S1034. Using the prominence coefficient, the characteristic values ​​of the corresponding wavelength region are adjusted so that the interface characteristics representing the layered structure are relatively enhanced, so as to focus on the absorption characteristics representing the content layered structure of the adhesive.

[0092] In the above scheme, adjacent scales refer to the analysis levels corresponding to primitive waveforms with adjacent width parameters used in multi-scale decomposition. For example, primitive waveforms with widths of 5 nanometers and 15 nanometers are a pair of adjacent scales. Significantly different wavelength regions refer to bands in which the spectral shape characteristic parameters of different scales, such as amplitude or full width at half maximum (FWHM), show significant inconsistencies. This inconsistency is often caused by interface effects. The difference quantity is a numerical indicator used to quantify the degree of feature difference between adjacent scales. It can be an absolute difference, a relative ratio, or other difference measure. The salience coefficient is a weighting factor determined according to the magnitude of the difference quantity. Regions with more significant differences are assigned a larger coefficient for subsequent feature enhancement.

[0093] In this embodiment, firstly, through S1031, among the multiple sets of spectral shape features obtained through multi-scale decomposition, the system compares the feature responses at adjacent scales one by one, identifying wavelength regions where the features are significant at finer scales but significantly weaken or disappear at adjacent coarser scales. For example, when analyzing a composite adhesive of Company A's Model B, the system found that near the 1675 nm wavelength corresponding to the characteristic absorption of a certain polymer interface layer component, the feature amplitude extracted using a 5 nm narrow primitive waveform is very high, showing a sharp peak; while the feature extracted using an adjacent 15 nm wide primitive waveform at the same position becomes very flat, with a significant decrease in amplitude. This phenomenon of significant attenuation of response with scale suggests that the spectral features of this wavelength region may be related to a scale-sensitive interfacial layered structure, and therefore are marked by the system as candidate interfacial feature regions. Secondly, through S1032, for each marked candidate wavelength region, the system quantitatively calculates the difference between features at adjacent scales to objectively assess its significance. For example, for the aforementioned 1675 nm region, the system calculates the relative difference ratio between the characteristic amplitude at the 5 nm scale and the characteristic amplitude at the 15 nm scale. Assuming the fine-scale amplitude is 0.75 and the coarse-scale amplitude is 0.25, the calculated difference will be much greater than the difference calculated in regions with consistent responses at other scales (such as the stable absorption peak of the main component), thus numerically confirming the specificity of this region in characterizing interface effects. Next, through S1033, the system assigns appropriate enhancement coefficients to the corresponding wavelength regions based on the magnitude of the calculated difference. For example, for a large difference calculated at 1675 nm, the system assigns a high enhancement coefficient, such as 2.0, according to a preset mapping rule; while for regions with negligible differences, a coefficient close to 1.0 is assigned, meaning that there is basically no enhancement. This step ensures that the enhancement is proportional to the specificity of the feature. Finally, through S1034, the system uses the calculated enhancement coefficients to scale and adjust the original fine-scale spectral shape feature values. For example, multiplying the original fine-scale feature amplitude of 0.75 at 1675 nm by its corresponding salience factor of 2.0 yields an enhanced new feature value of 1.50. This operation significantly increases the weight of interfacial features that effectively characterize the layered structure inside the adhesive in the overall feature set, allowing them to be more clearly identified and utilized by the model in subsequent quantitative modeling, ultimately achieving a more accurate prediction of the concentration of interface-related components.

[0094] In practical applications, when Company A conducted online testing on Model B composite adhesive, after S102 multi-scale decomposition, three sets of spectral shape features based on 5 nm, 15 nm, and 30 nm width primitive waveforms were obtained. The system first compared the features of the adjacent 5 nm and 15 nm scales. At a wavelength of 1675 nm, the system identified a significant difference: the 5 nm feature amplitude was as high as 0.75, while the 15 nm feature amplitude was only 0.28. This region was then labeled as R1. The system calculated the relative difference D at this location as |0.75 - 0.28| / 0.75 ≈ 0.627. According to the preset rule, regions with a difference D greater than 0.6 were assigned a prominence coefficient K = 2.5. Subsequently, the system multiplied the original feature amplitude of 0.75 at the 1675 nm scale by the coefficient 2.5 to obtain an enhanced feature amplitude of 1.875. Similarly, a difference was also identified at 1710 nm, and similar enhancement was performed. For other wavelength regions where differences are not significant, the original eigenvalues ​​are kept unchanged. Finally, all fine-scale features enhanced by the interface feature processing are output for subsequent analysis.

[0095] This scheme accurately locates sensitive spectral bands that may be caused by the layered structure or interface effects inside the adhesive by identifying regions of difference in characteristic response between adjacent scales. By calculating the difference and assigning prominence coefficients accordingly, it achieves quantitative evaluation of interface-related features and differentiated enhancement strategies. Finally, by adjusting the coefficients, it significantly increases the proportion of weak interface features representing the layered structure in the overall features, making them stand out from the complex spectral background, thereby effectively improving the quantitative model's ability to perceive and predict changes in the internal structure of the adhesive.

[0096] S104. Based on the transmission speed of the adhesive production line, a compensation factor for flow-induced spectral distortion is introduced to correct the spectral features extracted after interface feature enhancement processing that are related to the internal structure.

[0097] Optionally, step S104 may specifically include the following steps:

[0098] S1041. Establish the correspondence between the transmission speed of the adhesive production line and the wavelength shift corresponding to the spectral characteristics at different scales;

[0099] S1042. Based on the transmission speed at the current detection time, calculate the compensation factor for flow-induced spectral distortion using the corresponding relationship;

[0100] S1043. The compensation factor is applied to the spectral features related to the internal structure extracted after the interface feature enhancement process, and the wavelength positions corresponding to the spectral features are repositioned.

[0101] S1044. Based on the repositioned wavelength position, the values ​​of the spectral features are corrected and adjusted accordingly.

[0102] In the above scheme, the transmission speed refers to the linear velocity of the adhesive flowing through the detection point on the production line; flow-induced spectral distortion refers to the phenomenon that the shape of the absorption peak of the near-infrared spectrum is broadened or the peak wavelength is shifted due to physical effects such as shear force and turbulence generated when the adhesive flows in the pipeline; wavelength drift refers to the distance that the center wavelength of a specific absorption peak deviates from its standard position in its static state due to the flow effect; the compensation factor is a correction value calculated based on the current flow rate, used to offset the deviation of the spectral feature position caused by the flow.

[0103] In this embodiment, firstly, through step S1041, during the system debugging phase, a quantitative correspondence between the transmission speed of the adhesive production line and the spectral characteristic wavelength shift is established experimentally. For example, in Company A's testing system, technicians control the production line to run at multiple different known speeds, such as 0.5 m / s, 1.0 m / s, and 1.5 m / s, and measure the center wavelength position of the characteristic absorption peak of a key component of a certain standard adhesive sample at a specific scale under static conditions and at each flow rate. Through data analysis, it is found that when the flow rate is 1.0 m / s, the center wavelength of a certain characteristic absorption peak shifts from 1710 nm at rest to a longer wavelength direction by about 0.8 nm, thereby establishing a "speed-drift" correspondence dataset or fitting formula. Secondly, through step S1042, during the actual testing process, the system reads the current transmission speed of the production line in real time and uses the correspondence established in S1041 to calculate the wavelength correction amount that should be compensated at the current flow rate, i.e., the compensation factor. For example, at the current detection moment, the velocity sensor reading is 1.2 m / s. The system calculates the corresponding wavelength compensation factor as 0.5 nm shifting towards shorter wavelengths by consulting a preset relationship curve or substituting into a fitting formula. Next, in step S1043, the calculated compensation factor is applied to the spectral features related to the internal structure extracted after interface feature enhancement processing. Specifically, the wavelength positions corresponding to these features are repositioned. For example, if the original wavelength of a feature after interface enhancement is 1710.5 nm, after applying a compensation factor of -0.5 nm, the feature is repositioned to 1710.0 nm. Finally, in step S1044, the values ​​of the spectral features are adjusted accordingly based on the repositioned precise wavelength position. This involves small-scale resampling or interpolation of the spectral curve to ensure that the feature values, such as absorbance amplitude, accurately reflect the signal intensity at the corrected wavelength position. For example, after the feature is corrected from 1710.5 nm to 1710.0 nm, the system will calculate the feature amplitude at that precise position based on the spectral data points near 1710.0 nm using an interpolation algorithm, thereby completing the correction.

[0104] In practical applications, in Company A's online inspection system for adhesive model B, the correlation between production line conveyor speed and characteristic wavelength drift was pre-determined through experiments: for every 0.1 m / s increase in speed, the wavelength of a key absorption peak shifts approximately 0.1 nanometers towards longer wavelengths. During a particular inspection, the production line conveyor speed was 1.2 m / s, and the system calculated the required compensation factor to be -1.2 nanometers based on this relationship. Subsequently, the system applied this compensation factor to a feature at 1675.2 nanometers obtained in step S103, which had undergone interface enhancement, and repositioned its wavelength to 1674.0 nanometers. The system then interpolated the spectral data at 1674.0 nanometers to obtain the corrected characteristic amplitude. All important features related to the internal structure were corrected through this process, thereby eliminating the spectral distortion caused by flow.

[0105] This scheme establishes a quantitative relationship between flow rate and spectral distortion, and introduces a compensation factor in real time to reposition and numerically correct the characteristic wavelengths. This effectively offsets the spectral peak position shift and shape distortion induced by the flow of adhesive on the production line, ensuring that the extracted spectral features, especially the internal structural features after interface enhancement, have positional accuracy and numerical consistency. This greatly improves the adaptability, robustness, and reliability of the online quantitative detection model in the face of production flow rate fluctuations.

[0106] S105. The corrected spectral features are combined into a spectral feature set, and the spectral feature set is input into a quantitative prediction model established in advance through supervised learning. The quantitative prediction model outputs the concentration value of the adhesive component to achieve quantitative detection.

[0107] Optionally, step S105 may specifically include the following steps:

[0108] S1051. Establish a quantitative prediction model in advance that includes a processing network consisting of multiple layers of computing nodes;

[0109] S1052. Each feature in the set of spectral features formed by combining the corrected spectral features is input to the corresponding computing node of the starting layer of the processing network.

[0110] S1053. For each computing node in the processing network, receive the output value from the computing node of the previous layer, perform weighted accumulation calculation according to the predetermined connection relationship and internal parameters, and then pass the calculation result to the computing node of the next layer.

[0111] S1054. After passing through and calculating through all layers of the processing network, the final layer's calculation node outputs one or more numerical values. The output numerical values ​​represent the concentration values ​​of one or more components to be tested in the adhesive, thereby achieving quantitative detection of adhesive components.

[0112] In the above scheme, the spectral feature set refers to the summary of all representative spectral features extracted after a series of processes such as background correction, signal enhancement, multi-scale decomposition, interface feature enhancement, and flow distortion correction; the quantitative prediction model refers to a mathematical model trained by supervised learning methods that can establish a complex nonlinear mapping relationship between spectral features and component concentrations; the processing network composed of multi-layer computing nodes is the specific implementation of the quantitative prediction model, usually referring to an artificial neural network, where each layer contains multiple nodes, and each node performs a weighted summation of the input and applies a nonlinear transformation; the internal parameters refer to the weights and biases learned during model training, which determine how features are passed from the input layer to the output layer.

[0113] In this embodiment, firstly, through step S1051, a quantitative prediction model with a defined structure is pre-established during the model training phase. The core of the quantitative prediction model is a processing network composed of multiple layers of computing nodes, such as an artificial neural network containing an input layer, several hidden layers, and an output layer. The internal parameters of the quantitative prediction model, namely the connection weights and node biases, are optimized and determined through supervised learning training using the spectral features of a large number of adhesive samples with known accurate component concentrations and their corresponding concentration values. Secondly, through step S1052, during actual detection, each feature value in the spectral feature set obtained after processing the current adhesive sample through all the aforementioned steps S101 to S104 is input one-to-one into the input layer nodes of the pre-trained quantitative prediction model processing network in a predetermined order. For example, assuming that a spectral feature set of 50 features is ultimately formed, they are respectively sent to the 50 computing nodes of the input layer. Then, through step S1053, within the quantitative prediction model, data begins to propagate forward. Each computation node receives output values ​​from all nodes in the previous layer. Each node multiplies these input values ​​by their corresponding connection weights, sums them, and adds their own bias parameters to form a weighted sum. This sum is then processed using a non-linear activation function such as ReLU, and the result is output and passed to all nodes in the next layer. This process starts from the input layer and proceeds layer by layer to the hidden and output layers. Finally, through step S1054, after layer-by-layer computation and transmission through all layers of the processing network, the computation nodes in the output layer produce the final output values. Each output node corresponds to the predicted concentration value of one analyte. For example, if the model needs to predict the concentrations of two components, curing agent and plasticizer, in an adhesive, the output layer has two nodes whose output values ​​represent the predicted concentrations of these two components, respectively, thus achieving quantitative detection of adhesive components.

[0114] In practical applications, in Company A's online detection system for type B adhesive, a neural network model was trained using hundreds of historically accumulated sample data with standard concentration values. This model consisted of an input layer with 50 nodes corresponding to 50 spectral features, two hidden layers with 30 and 15 nodes respectively, and two output layers with nodes corresponding to the curing agent and plasticizer concentrations. During a particular online detection, the system processed the flowing adhesive sample using steps S101 to S104, ultimately extracting and correcting a set of 50 feature values. This feature set was then sequentially input into the trained neural network model. After weighted summation and nonlinear transformation within the model, one node in the output layer output a value of 12.5, and the other node output a value of 8.2. The system then determined that the predicted concentration of the curing agent in the current sample was 12.5%, and the predicted concentration of the plasticizer was 8.2%, thus completing the quantitative detection.

[0115] This method combines carefully extracted and corrected spectral features with high signal-to-noise ratio and clear physical meaning into a set, and processes them using a quantitative prediction model that has been trained with a large amount of data and can capture complex nonlinear relationships. Ultimately, it achieves the accurate mapping of rich spectral information to the specific concentration values ​​of each component in the adhesive, thus completing the entire process from spectral measurement to quantitative component analysis and ensuring high accuracy and reliability of the detection results.

[0116] The quantitative detection method for adhesive components based on spectral analysis provided in this application effectively eliminates interference from environmental and instrument noise through synchronous background acquisition and correction in step S101, providing a clean spectral data foundation for analysis. Through signal enhancement and multi-scale decomposition in step S102, the spectral signal-to-noise ratio is significantly improved and overlapping absorption peaks are successfully resolved, thereby extracting detailed and overall spectral features at different scales. Through interface feature enhancement processing introduced in step S103, the characteristic signals caused by the internal layered structure or interface effects of the adhesive are accurately focused and amplified, enhancing the ability to characterize complex physical structures. Through dynamic distortion compensation based on transmission speed in step S104, the spectral feature drift caused by the flow conditions of the production line is effectively corrected, improving the robustness of online detection. Finally, through step S105, the optimized feature set obtained in the preceding steps is input into a pre-trained quantitative model, achieving rapid and high-precision concentration prediction of adhesive components. This series of interconnected steps work together to systematically solve the problem of insufficient accuracy in quantitative spectral detection under complex working conditions.

[0117] Figure 2 This is a flowchart illustrating a specific implementation of steps S1031-S1034 in the embodiments of this application, mainly demonstrating, for example:

[0118] First, identify wavelength regions with significant response differences between adjacent scales. Second, calculate the quantization difference between features at different scales within this region. Then, assign an appropriate prominence coefficient to this region based on the magnitude of the difference. Finally, use this coefficient to adjust the feature values, thereby achieving a relative enhancement of the interface-related spectral features.

[0119] Figure 3 This is a schematic diagram of a specific embodiment of a quantitative detection system for adhesive components based on spectral analysis provided in this application. (Refer to...) Figure 3 The system may include:

[0120] The acquisition module 31 is used to acquire near-infrared spectral data of the adhesive sample to be tested, simultaneously acquire background spectral data of the environment in which the adhesive sample is located, and perform background correction on the near-infrared spectral data based on the background spectral data.

[0121] The decomposition module 32 is used to perform signal enhancement processing on the background-corrected near-infrared spectral data, and at the same time perform multi-scale decomposition on the signal-enhanced near-infrared spectral data to extract spectral features at different scales and separate overlapping absorption peaks.

[0122] The processing module 33 is used to introduce interface feature enhancement processing to address the possible layering or interface effects of the adhesive during the multi-scale decomposition process, so as to focus on the absorption characteristics that characterize the layered structure of the adhesive.

[0123] The correction module 34 is used to introduce a compensation factor for flow-induced spectral distortion based on the transmission speed of the adhesive production line, and to correct the spectral features extracted after interface feature enhancement processing that are related to the internal structure.

[0124] The input module 35 is used to combine the corrected spectral features into a spectral feature set, and input the spectral feature set into a quantitative prediction model established in advance through supervised learning. The quantitative prediction model outputs the concentration value of the adhesive component to achieve quantitative detection.

[0125] This application provides a spectral analysis-based adhesive component quantitative detection system to implement the aforementioned spectral analysis-based adhesive component quantitative detection method. Therefore, the specific implementation of the spectral analysis-based adhesive component quantitative detection system can be found in the embodiment section of the spectral analysis-based adhesive component quantitative detection method above. The specific implementation can be referred to the description of the corresponding embodiments, and will not be repeated here.

[0126] This application also provides an electronic device, comprising: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of the above-described method for quantitative detection of adhesive components based on spectral analysis.

[0127] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of any of the above-described methods for quantitative detection of adhesive components based on spectral analysis.

[0128] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory, random access memory, portable hard drives, magnetic disks, or optical disks.

[0129] Embodiments of the present invention also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the embodiments of the quantitative detection method for adhesive components based on spectral analysis described above.

[0130] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0131] The foregoing has provided a detailed description of a method and system for quantitative detection of adhesive components based on spectral analysis, as provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A method for quantitative detection of adhesive components based on spectral analysis, characterized in that, The method comprises the following steps: acquiring near-infrared spectrum data of an adhesive sample to be detected, synchronously acquiring background spectrum data of an environment in which the adhesive sample is located, and performing background correction on the near-infrared spectrum data according to the background spectrum data; performing signal enhancement processing on the near-infrared spectrum data after the background correction, and performing multi-scale decomposition on the near-infrared spectrum data after the signal enhancement processing to extract spectral features at different scales and separate overlapping absorption peaks; in the multi-scale decomposition process, interface feature enhancement processing is introduced for the possible layered or interface effect of the adhesive to focus on absorption features representing the internal layered structure of the adhesive; a compensation factor for flow-induced spectral distortion is introduced according to the transmission speed of the adhesive production line to correct the spectral features extracted after the interface feature enhancement processing and related to the internal structure; the corrected spectral features are combined into a spectral feature set, and the spectral feature set is input into a quantitative prediction model previously established through supervised learning, and the concentration value of the adhesive composition is output by the quantitative prediction model to realize quantitative detection.

2. The method of claim 1, wherein, In the multi-scale decomposition process, interface feature enhancement processing is introduced for the possible layered or interface effect of the adhesive to focus on absorption features representing the internal layered structure of the adhesive, comprising: in the multiple groups of spectral shape features obtained through multi-scale decomposition, a wavelength region in which the adhesive has significant differences between adjacent scales is identified; the difference between the spectral shape features of adjacent scales in the wavelength region is calculated; an appropriate prominence coefficient is assigned to the interface feature of the wavelength region according to the difference; the feature value of the corresponding wavelength region is adjusted using the prominence coefficient, so that the interface feature representing the layered structure is relatively enhanced, thereby focusing on absorption features representing the internal layered structure of the adhesive.

3. The method of claim 1, wherein, The method comprises the following steps: performing multi-scale decomposition on the near-infrared spectrum data after the signal enhancement processing to extract spectral features at different scales and separate overlapping absorption peaks, comprising: using each width of a preset group of primitive waveforms with different widths to respectively match the near-infrared spectrum data after the signal enhancement processing to obtain a sequence of the matching degree of the spectral data and the primitive waveforms at the corresponding width; for each width of the primitive waveforms, the spectral shape features represented by the primitive waveforms in the spectral data are extracted according to the sequence of the matching degree; multiple groups of spectral shape features extracted using primitive waveforms with different widths are collected to form spectral features at multiple scales; 4. The method of claim 1, wherein, based on the spectral features at different scales, overlapping absorption peak regions are identified by comparing the feature differences at different scales, and peak separation algorithms are applied to deconvolve overlapping peaks to separate the absorption peaks. A compensation factor for flow-induced spectral distortion is introduced according to the transmission speed of the adhesive production line to correct the spectral features extracted after the interface feature enhancement processing and related to the internal structure, comprising: establishing a corresponding relationship between the transmission speed of the adhesive production line and the wavelength drift corresponding to the spectral features at different scales; According to the transmission speed at the current detection time, a compensation factor of flow-induced spectral distortion is calculated by using the corresponding relationship; The compensation factor is applied to the extracted spectral features related to the internal structure after the interface feature enhancement processing, and the wavelength position corresponding to the spectral features is repositioned; According to the repositioned wavelength position, the numerical value of the spectral feature is adjusted.

5. The method of claim 1, wherein, The spectral feature set is input into a quantitative prediction model established by supervised learning in advance, and the concentration value of the adhesive component is output by the quantitative prediction model to realize quantitative detection, including: A quantitative prediction model including a processing network composed of multiple layers of calculation nodes is established in advance; Each feature in the spectral feature set composed of the corrected spectral features is input into the corresponding calculation node of the starting layer of the processing network; For each layer of calculation nodes in the processing network, the output value from the previous layer of calculation nodes is received, and weighted and accumulated calculation is performed according to the predetermined connection relationship and internal parameters, and then the calculation result is transmitted to the next layer of calculation nodes; After the layer-by-layer transmission and calculation of all layers of the processing network, one or more numerical values are output by the calculation nodes of the final layer, and the output of the numerical values represents the concentration value of one or more components to be measured in the adhesive, realizing quantitative detection of the adhesive components.

6. The method of claim 1, wherein, According to the background spectral data, the near-infrared spectral data is background corrected, including: Obtain the background spectral data under the same environmental conditions without the adhesive sample; Align the background spectral data with the near-infrared spectral data obtained from the adhesive sample at each wavelength point; Subtract the value of the corresponding wavelength point of the near-infrared spectral data from the value of each wavelength point of the background spectral data to obtain a set of differential spectral data; Perform baseline adjustment on the differential spectral data so that the starting value of the differential spectral data in the specified wavelength interval is close to zero, and the background correction of the near-infrared spectral data is realized.

7. The method of claim 1, wherein, Perform signal enhancement processing on the background-corrected near-infrared spectral data, including: Calculate the change degree at different wavelength points for the background-corrected near-infrared spectral data; According to the change degree, the wavelength region with sharp change is strengthened, and the wavelength region with gentle change is smoothed; Divide the entire near-infrared spectral data into multiple continuous wavelength intervals, and independently perform the above strengthening and smoothing processing on each wavelength interval; Connect the data segments processed independently in each wavelength interval to form the near-infrared spectral data processed by signal enhancement.

8. A system for quantitative detection of adhesive components based on spectral analysis, characterized by Including: The acquisition module is used for acquiring the near-infrared spectral data of the adhesive sample to be detected, synchronously acquiring the background spectral data of the environment where the adhesive sample is located, and background correcting the near-infrared spectral data according to the background spectral data; The decomposition module is used for performing signal enhancement processing on the background-corrected near-infrared spectral data, and simultaneously performing multi-scale decomposition on the near-infrared spectral data processed by signal enhancement to extract spectral features at different scales and separate overlapping absorption peaks; a processing module, configured to introduce an interface feature enhancement processing in a multi-scale decomposition process to focus on absorption features representing the layered structure of the adhesive; a correction module, configured to introduce a compensation factor for flow-induced spectral distortion according to the transmission speed of the adhesive production line to correct the spectral features related to the internal structure extracted after the interface feature enhancement processing; an input module, configured to combine the corrected spectral features into a spectral feature set and input the spectral feature set into a quantitative prediction model established by supervised learning to output a concentration value of the adhesive component and realize quantitative detection.

9. An electronic device, comprising: comprise: a memory, configured to store a computer program; a processor, configured to execute the computer program to implement the steps of the adhesive component quantitative detection method based on spectral analysis according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer program is stored in the computer readable storage medium and can be executed by the processor to implement the adhesive component quantitative detection method based on spectral analysis according to any one of claims 1 to 7.

Images