Coating quality detection method and device based on spectrum method

Abstract

The invention discloses a coating quality detection method and device based on a spectrum method, and relates to the technical field of quality detection. The method comprises the following steps: carrying out accelerated aging on a coating sample product, and carrying out spectrum acquisition to obtain a first spectrum data set; performing baseline correction on the first spectral data set by adopting a neural network to obtain a second spectral data set; performing de-noising processing on the second spectral data set by adopting modal decomposition and a wavelet threshold algorithm to obtain a third spectral data set; and performing quality evaluation according to the third spectral data set. By introducing neural network automatic baseline correction and combining modal decomposition and wavelet threshold denoising technologies, efficient and accurate processing of coating spectral data is realized. According to the method, the time and errors of traditional manual correction are remarkably reduced, and the detection automation level and efficiency are improved. Meanwhile, signal details are effectively reserved through multiple denoising means, the extraction accuracy of aging features is enhanced, and the reliability of quality evaluation is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of quality detection, and in particular to a coating quality detection method and device based on spectroscopy. Background Art

[0002] In the coatings industry, testing and evaluating the coating's weathering resistance is a critical step in quality control to ensure long-term product stability and performance reliability in practical applications. Weathering resistance refers to a coating's ability to maintain its physical, chemical, and mechanical properties when exposed to natural or simulated environmental conditions (such as UV light, high temperature and humidity, and wind and rain corrosion). It directly impacts the coating's protective effect on the substrate and its service life. To verify the coating's stability and durability in outdoor environments, manufacturers typically conduct accelerated aging tests before shipping to simulate the material's response to environmental stresses during long-term service.

[0003] At present, paint quality detection based on spectroscopy has gradually become a research hotspot. Using spectroscopy technology to perform non-destructive testing on paint samples can capture the chemical composition and structural changes caused by aging at the microscopic level of the material, and has the advantages of fast detection speed and high sensitivity.

[0004] However, in the process of spectral data processing, baseline correction and signal noise reduction are key steps to ensure data accuracy. Traditional baseline correction methods usually require manual selection of base points or baseline intervals, which consumes a lot of time and leads to low detection efficiency. Especially when a large number of samples or spectral data are collected periodically, semi-automatic correction greatly limits the detection efficiency and automation level. In addition, the existing denoising processing technology is single, such as wavelet denoising, Savitzky-Golay filtering denoising, etc., which directly processes the full length of the signal, making it difficult to take into account both signal detail retention and noise suppression, further affecting the accuracy of quality assessment. Summary of the Invention

[0005] The purpose of the present invention is to solve the problem of low detection efficiency mentioned in the above background technology, and to propose a coating quality detection method and device based on spectroscopy.

[0006] The first aspect of the present invention provides a method for detecting coating quality based on spectroscopy, the method comprising: Performing accelerated aging on a sample product of the coating, and collecting spectra according to a preset collection period to obtain a first spectral data set; performing baseline correction on each spectral signal in the first spectral dataset using a neural network to obtain a second spectral dataset; performing denoising on each spectral signal in the second spectral dataset using modal decomposition and wavelet threshold algorithm to obtain a third spectral dataset; generating aging characterization data based on the third spectral data set; A quality assessment is performed based on the aging characterization data to obtain a quality test result.

[0007] Optionally, performing baseline correction on each spectral signal in the first spectral dataset includes: Preprocessing the first spectral signal to obtain a fourth spectral signal; the first spectral signal is any spectral signal in the first spectral data set; Using the fourth spectral signal as input to a pre-trained baseline recognition model to obtain a target baseline; the baseline recognition model is a fully connected neural network with an output dimension identical to an input dimension; The target baseline is subtracted from the fourth spectral signal to obtain a corrected spectral signal, which is included in the second spectral data set.

[0008] Optionally, performing denoising processing on each spectral signal in the second spectral dataset includes: Decomposing the second spectral signal using a variational modal decomposition algorithm to obtain a modal component set and a center frequency of each modal component; the second spectral signal is any spectral signal in the second spectral data set; Calculate the sample entropy of each modal component; Determine the noise modal component based on the sample entropy and center frequency of each modal component; Eliminating the noise modal component from the modal component set to obtain a valid component set; Perform wavelet threshold denoising on each effective component to obtain the denoised modal component; Reconstructing the signal according to the denoised modal components to obtain a fifth spectral signal; The fifth spectral signal is smoothed and filtered to obtain a third spectral signal, which is included in a third spectral data set.

[0009] Optionally, the method further includes using an improved particle swarm optimization algorithm to optimize the number of decomposition layers and penalty factors of the variational mode decomposition algorithm. Specifically, an improved particle velocity update formula is used for particle iteration, and the improved particle velocity update formula is: ; Among them, w is the inertia weight, w min and w max is the minimum and maximum value of the inertia weight; t is the current iteration number, T max is the maximum number of iterations; and are two types of learning factors for particle i, and 、 and are the minimum and maximum values ​​of the two types of learning factors respectively; is the normalized value of the fitness of particle i; is the position of particle i at iteration t; is the historical optimal position of particle i; is the global optimal position; r1 and r2 are random numbers; is the updated particle velocity.

[0010] Optionally, determining the noise modal component according to the sample entropy and center frequency of each modal component includes: According to the sample entropy of each modal component, the interquartile range method is used to determine the upper limit of abnormality, and the modal components with sample entropy greater than the upper limit of abnormality are included in the candidate set; According to the center frequency of each modal component, the K-means clustering algorithm is used to cluster and obtain the low-frequency set, the medium-frequency set and the high-frequency set; The intersection of the candidate set and the high-frequency set is taken as the noise modal component set.

[0011] A second aspect of the present invention provides a coating quality detection device based on spectroscopy, the device comprising: A spectrum acquisition module, configured to perform accelerated aging on a sample product of the coating and acquire spectra according to a preset acquisition cycle to obtain a first spectrum data set; a baseline correction module, configured to perform baseline correction on each spectral signal in the first spectral data set using a neural network to obtain a second spectral data set; a signal denoising module, configured to perform denoising processing on each spectral signal in the second spectral dataset by using modal decomposition and wavelet threshold algorithm to obtain a third spectral dataset; a feature expression module, configured to generate aging characterization data based on the third spectral dataset; The indicator evaluation module is used to perform quality evaluation based on the aging characterization data to obtain quality detection results.

[0012] Optionally, the baseline correction module includes: a preprocessing module, configured to preprocess the first spectral signal to obtain a fourth spectral signal; the first spectral signal is any spectral signal in the first spectral data set; a baseline generation module, configured to use the fourth spectral signal as input to a pre-trained baseline recognition model to obtain a target baseline; the baseline recognition model is a fully connected neural network whose output dimension is the same as the input dimension; The baseline elimination module is used to subtract the target baseline from the fourth spectral signal to obtain a corrected spectral signal and incorporate it into the second spectral data set.

[0013] Optionally, the signal denoising module includes: a signal decomposition module, configured to decompose the second spectral signal using a variational modal decomposition algorithm to obtain a modal component set and a center frequency of each modal component; the second spectral signal is any spectral signal in the second spectral data set; Sample entropy calculation module, used to calculate the sample entropy of each modal component; A noise determination module is used to determine the noise modal component according to the sample entropy and center frequency of each modal component; An effective set determination module, configured to remove noise modal components from the modal component set to obtain an effective component set; Component denoising module, used to perform wavelet threshold denoising on each effective component to obtain denoised modal components; a signal reconstruction module, configured to reconstruct the signal according to the denoised modal components to obtain a fifth spectral signal; The signal smoothing module is used to perform smoothing filtering on the fifth spectral signal to obtain a third spectral signal and incorporate it into a third spectral data set.

[0014] Optionally, the device further includes a parameter optimization module for optimizing the number of decomposition layers and penalty factors of the variational mode decomposition algorithm using an improved particle swarm optimization algorithm; specifically, particle iteration is performed using an improved particle velocity update formula, and the improved particle velocity update formula is: ; Among them, w is the inertia weight, w min and w max is the minimum and maximum value of the inertia weight; t is the current iteration number, T max is the maximum number of iterations; and are two types of learning factors for particle i, and 、 and are the minimum and maximum values ​​of the two types of learning factors respectively; is the normalized value of the fitness of particle i; is the position of particle i at iteration t; is the historical optimal position of particle i; is the global optimal position; r1 and r2 are random numbers; is the updated particle velocity.

[0015] Optionally, the noise determination module includes: A high entropy determination module is used to determine the upper bound of abnormality based on the sample entropy of each modal component using the interquartile range method, and include the modal components with sample entropy greater than the upper bound of abnormality into the candidate set; The high-frequency determination module is used to cluster the center frequency of each modal component using the K-means clustering algorithm to obtain a low-frequency set, a medium-frequency set, and a high-frequency set; The noise selection module is used to take the intersection of the candidate set and the high-frequency set as the noise modal component set.

[0016] Beneficial effects of the present invention: The present invention proposes a coating quality detection method based on spectroscopy, which includes: subjecting a sample product of the coating to accelerated aging and performing spectral acquisition according to a preset acquisition period to obtain a first spectral data set; using a neural network to perform baseline correction on each spectral signal in the first spectral data set to obtain a second spectral data set; using modal decomposition and a wavelet threshold algorithm to perform denoising on each spectral signal in the second spectral data set to obtain a third spectral data set; generating aging characterization data based on the third spectral data set; and performing quality assessment based on the aging characterization data to obtain a quality detection result.

[0017] By introducing automatic baseline correction using a neural network, combined with modal decomposition and wavelet threshold denoising techniques, efficient and accurate processing of coating spectral data is achieved. This method significantly reduces the time and errors associated with traditional manual calibration, improving detection automation and efficiency. Furthermore, multiple denoising techniques effectively preserve signal details, enhance the accuracy of aging feature extraction, and improve the reliability of quality assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A flow chart of a coating quality detection method based on spectroscopy is provided for an embodiment of the present invention; Figure 2 The present invention provides a structural diagram of a coating quality detection device based on spectroscopy. DETAILED DESCRIPTION

[0019] In order to further illustrate the technical means and effects adopted by the present invention to achieve the predetermined purpose of the invention, the specific implementation methods, structures, features and effects of the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments.

[0020] The embodiment of the present invention provides a method for detecting coating quality based on spectroscopy. Figure 1 , Figure 1 A flow chart of a coating quality detection method based on spectroscopy provided in an embodiment of the present invention. The method comprises the following steps: S101 , performing accelerated aging on a sample product of a coating, and collecting spectra according to a preset collection period to obtain a first spectral data set.

[0021] S102 , performing baseline correction on each spectral signal in the first spectral data set using a neural network to obtain a second spectral data set.

[0022] S103 , performing denoising processing on each spectral signal in the second spectral dataset using modal decomposition and wavelet threshold algorithm to obtain a third spectral dataset.

[0023] S104: Generate aging characterization data based on the third spectral data set.

[0024] S105: Perform quality assessment based on the aging characterization data to obtain quality test results.

[0025] The collected spectral data is Raman spectrum.

[0026] A spectroscopy-based coating quality detection method, based on an embodiment of the present invention, achieves efficient and accurate processing of coating spectral data by introducing automatic baseline correction using a neural network, combined with modal decomposition and wavelet threshold denoising techniques. This method significantly reduces the time and errors associated with traditional manual calibration, improving detection automation and efficiency. Furthermore, multiple denoising techniques effectively preserve signal details, enhance the accuracy of aging feature extraction, and improve the reliability of quality assessment.

[0027] In one implementation, accelerated aging can be performed using a UV light and condensation cycle. The acquisition cycle can be set to 12 hours, that is, spectral data is collected every 12 hours.

[0028] In one embodiment, step S102, using a neural network to perform baseline correction on each spectral signal in the first spectral dataset includes: Step 1: preprocess the first spectral signal to obtain a fourth spectral signal.

[0029] Step 2: Use the fourth spectral signal as input to the pre-trained baseline recognition model to obtain the target baseline.

[0030] Step 3: Subtract the target baseline from the fourth spectral signal to obtain a corrected spectral signal, which is included in the second spectral data set.

[0031] The first spectral signal is any spectral signal in the first spectral data set.

[0032] In one implementation, preprocessing includes normalization. The baseline recognition model is a fully connected neural network with an output dimension that matches the input dimension and the signal length. It has three hidden layers with 1024, 2048, and 1024 neurons, respectively, and uses the ReLU function for activation. The training process uses mean squared error loss to optimize model parameters.

[0033] Traditional baseline correction methods (such as polynomial fitting and filtering) often perform poorly when dealing with complex or nonlinear background interference. This embodiment uses a deep, fully connected neural network, which has strong nonlinear modeling capabilities. This allows for accurate fitting of complex background trends and effectively identifies and removes background baselines. Furthermore, it eliminates the need for manual selection of base points or baseline regions, significantly improving the efficiency of baseline correction.

[0034] In one embodiment, step S103, performing denoising on each spectral signal in the second spectral dataset using modal decomposition and wavelet threshold algorithm includes: In step 1, the variational mode decomposition (VMD) algorithm is used to decompose the second spectral signal to obtain a set of modal components and the center frequency of each modal component.

[0035] Step 2: Calculate the sample entropy of each modal component; Step 3: Determine the noise modal component based on the sample entropy and center frequency of each modal component; Step 4: remove the noise modal components from the modal component set to obtain the effective component set; Step 5: Perform wavelet threshold denoising on each effective component to obtain the denoised modal component; Step six, reconstructing the signal based on the denoised modal components to obtain a fifth spectral signal; Step seven: perform smoothing filtering on the fifth spectral signal to obtain a third spectral signal, which is incorporated into a third spectral data set.

[0036] The second spectral signal is any spectral signal in the second spectral data set.

[0037] This embodiment uses variational modal decomposition to decompose the complex original spectral signal into multiple modal components of different frequencies, allowing subsequent processing to operate at a finer granularity. Wavelet threshold denoising further processes the detail noise in the effective component, achieving joint denoising from coarse to fine and from structure to detail, which is more powerful than a single method. To verify the effectiveness of the denoising method proposed in this invention, the performance of the proposed denoising method was compared with traditional denoising methods (global wavelet threshold denoising and Savitzky-Golay filtering denoising), and evaluation indicators included root mean square error (RMSE) and signal-to-noise ratio (SNR). Table 1 shows the experimental results of different denoising methods on the test set.

[0038] Table 1:

[0039] In one embodiment, the improved particle swarm optimization algorithm is used to adjust the decomposition layer number K and penalty factor of the variational mode decomposition algorithm. Optimize. The particle swarm optimization process includes: Step 1: Define the optimization objective, use envelope entropy as the fitness function, and minimize the sum of the envelope entropies of each modal component. The smaller the envelope entropy, the better.

[0040] Step 2: Initialize the particle swarm parameters, set the population size, maximum number of iterations, inertia weight w, learning factors c1, c2, and parameter search range.

[0041] Step 3: Initialize the position and velocity of the particles. The position of each particle is a (K, )right.

[0042] Step 4: Calculate the fitness of each particle.

[0043] Step 5: Update the historical optimal position and global optimal position of each particle.

[0044] Step 6: Update the particle's velocity and position. The standard update formula is: .

[0045] Step 7: Determine the termination condition. If the maximum number of iterations is reached or the global optimal position converges, the optimization process ends and the number of decomposition layers and penalty factors corresponding to the global optimal position are output. Otherwise, return to step 4 and continue iterating.

[0046] In the traditional VMD algorithm, the decomposition layer K and penalty factor Manual settings are required, relying on experience and lacking universality. This embodiment introduces particle swarm optimization to search for the optimal combination, avoiding the manual trial and error process and greatly improving the stability and universality of VMD.

[0047] In one implementation, an improved particle velocity update formula is used for particle iteration. The improved particle velocity update formula is: ; Among them, w is the inertia weight, w min and w max is the minimum and maximum value of the inertia weight; t is the current iteration number, T max is the maximum number of iterations; and are two types of learning factors for particle i, and 、 and are the minimum and maximum values ​​of the two types of learning factors respectively; is the normalized value of the fitness of particle i; is the position of particle i at iteration t; is the historical optimal position of particle i; is the global optimal position; r1 and r2 are random numbers between (0, 1); is the updated particle velocity.

[0048] The inertia weight w uses a cosine decay strategy, which promotes strong exploration in the early stages and gradual convergence in the later stages. This improves global search capabilities and avoids premature convergence. The learning factor is dynamically adjusted based on the normalized fitness value. When a particle's fitness is poor, it learns more from the swarm (decreasing c1 and increasing c2), enhancing its learning ability. When a particle's fitness is good, its exploration direction is effective, and a higher learning weight is assigned to that direction (increasing c1 and decreasing c2). Dynamic adjustment of algorithm parameters improves the algorithm's optimization efficiency.

[0049] In one implementation, determining the noise modal component based on the sample entropy and center frequency of each modal component includes: Step 1: Based on the sample entropy of each modal component, the interquartile range method is used to determine the upper limit of the abnormality , the modal components whose sample entropy is greater than the abnormal upper bound are included in the candidate set. Specifically, ; Among them, Q1 is the 25% quantile of the sample entropy, and Q3 is the 75% quantile of the sample entropy.

[0050] Step 2: Based on the center frequency of each modal component, the K-means clustering algorithm is used to cluster the components into low-frequency, medium-frequency, and high-frequency clusters. Specifically, the number of clusters k = 3, and the minimum, maximum, and median of the center frequencies are used as the initial cluster heads.

[0051] Step 3: Take the intersection of the candidate set and the high-frequency set as the noise modal component set.

[0052] This implementation method combines the signal complexity reflected by sample entropy with the spectral characteristics of the center frequency to comprehensively determine whether it is a noise mode, avoiding misjudgment caused by single feature judgment, and effectively improving the accuracy and robustness of noise recognition.

[0053] In one implementation, the smoothing filter may adopt Gaussian filtering or Savitzky-Golay filtering.

[0054] In one embodiment, step S104, generating aging characterization data according to the third spectral dataset includes: In step 1, the main Raman peak positions are extracted using peak detection algorithms (such as first-order derivative zero crossing, Lorentz / Gauss fitting).

[0055] Step 2: Select several representative characteristic peaks (such as benzene ring breathing peak, C–H / C=O / C–C and other functional group characteristic peaks).

[0056] Step three: track the parameters of these characteristic peaks that change over time, and establish the change curves of each parameter (peak position shift, peak height change, peak width increase, area change, etc.) as aging characterization data.

[0057] In one embodiment, step S105 can be used to perform quality assessment based on the aging characterization data in various ways. For example, a classification method based on threshold rules can be used. Several threshold conditions can be set to determine whether the failure point has been reached prematurely, such as: peak position shift Δλ > X cm⁻¹; peak intensity decrease > 30%. If any of the aging characteristics of the sample exceeds the limit before a preset time point, the sample is judged to be of unqualified quality.

[0058] The embodiment of the present invention provides a coating quality detection device based on spectroscopy. Figure 2 , Figure 2 This is a structural diagram of a coating quality detection device based on spectroscopy provided in an embodiment of the present invention. The device includes: The spectrum acquisition module is used to perform accelerated aging on the sample product of the coating and to acquire spectra according to a preset acquisition cycle to obtain a first spectrum data set.

[0059] The baseline correction module is used to perform baseline correction on each spectral signal in the first spectral data set using a neural network to obtain a second spectral data set.

[0060] The signal denoising module is used to perform denoising processing on each spectral signal in the second spectral data set by using modal decomposition and wavelet threshold algorithm to obtain a third spectral data set.

[0061] The feature expression module is used to generate aging characterization data based on the third spectral data set.

[0062] The index evaluation module is used to perform quality assessment based on aging characterization data and obtain quality detection results.

[0063] The collected spectral data is Raman spectrum.

[0064] A spectroscopy-based paint quality detection device, based on an embodiment of the present invention, achieves efficient and accurate processing of paint spectral data by introducing automatic baseline correction using a neural network, combined with modal decomposition and wavelet threshold denoising techniques. This method significantly reduces the time and errors associated with traditional manual calibration, improving detection automation and efficiency. Furthermore, multiple denoising techniques effectively preserve signal details, enhance the accuracy of aging feature extraction, and improve the reliability of quality assessment.

[0065] In one embodiment, the baseline correction module includes: The preprocessing module is used to preprocess the first spectral signal to obtain a fourth spectral signal.

[0066] The baseline generation module is used to use the fourth spectral signal as the input of a pre-trained baseline recognition model to obtain a target baseline; the baseline recognition model is a fully connected neural network with an output dimension the same as an input dimension.

[0067] The baseline elimination module is used to subtract the target baseline from the fourth spectral signal to obtain a corrected spectral signal and incorporate it into the second spectral data set.

[0068] In one embodiment, the signal denoising module includes: The signal decomposition module is used to decompose the second spectral signal using a variational modal decomposition algorithm to obtain a modal component set and a center frequency of each modal component.

[0069] The sample entropy calculation module is used to calculate the sample entropy of each modal component.

[0070] The noise determination module is used to determine the noise modal component according to the sample entropy and center frequency of each modal component.

[0071] The effective set determination module is used to remove the noise modal components from the modal component set to obtain the effective component set.

[0072] The component denoising module is used to perform wavelet threshold denoising on each effective component to obtain the denoised modal component.

[0073] The signal reconstruction module is used to reconstruct the signal according to the denoised modal component to obtain a fifth spectrum signal.

[0074] The signal smoothing module is used to perform smoothing filtering on the fifth spectral signal to obtain a third spectral signal and incorporate it into the third spectral data set.

[0075] In one embodiment, the apparatus further includes a parameter optimization module for optimizing the number of decomposition layers and penalty factors of the variational mode decomposition algorithm using an improved particle swarm optimization algorithm.

[0076] In one embodiment, the noise determination module includes: The high entropy determination module is used to determine the upper bound of the anomaly based on the sample entropy of each modal component using the interquartile range method, and include the modal components with sample entropy greater than the upper bound of the anomaly into the candidate set.

[0077] The high frequency determination module is used to cluster the center frequency of each modal component using the K-means clustering algorithm to obtain a low frequency set, a medium frequency set and a high frequency set.

[0078] The noise selection module is used to take the intersection of the candidate set and the high-frequency set as the noise modal component set.

[0079] The above description is merely a preferred embodiment of the present invention and does not constitute any form of limitation to the present invention. Although the present invention has been disclosed as above in terms of a preferred embodiment, it is not intended to limit the present invention. Any person skilled in the art can, without departing from the scope of the technical solution of the present invention, make some changes or modifications to equivalent embodiments using the technical contents disclosed above. However, any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are within the scope of the technical solution of the present invention.

Claims

1. A coating quality detection method based on spectroscopy, characterized in that: The method comprises: Performing accelerated aging on a sample product of the coating, and collecting spectra according to a preset collection period to obtain a first spectral data set; performing baseline correction on each spectral signal in the first spectral dataset using a neural network to obtain a second spectral dataset; performing denoising on each spectral signal in the second spectral dataset using modal decomposition and wavelet threshold algorithm to obtain a third spectral dataset; generating aging characterization data based on the third spectral data set; A quality assessment is performed based on the aging characterization data to obtain a quality test result.

2. The coating quality detection method based on spectroscopy according to claim 1, characterized in that: Performing baseline correction on each spectral signal in the first spectral data set includes: Preprocessing the first spectral signal to obtain a fourth spectral signal; the first spectral signal is any spectral signal in the first spectral data set; Using the fourth spectral signal as input to a pre-trained baseline recognition model to obtain a target baseline; the baseline recognition model is a fully connected neural network with an output dimension identical to an input dimension; The target baseline is subtracted from the fourth spectral signal to obtain a corrected spectral signal, which is included in the second spectral data set.

3. The coating quality detection method based on spectroscopy according to claim 1, characterized in that: Performing denoising processing on each spectral signal in the second spectral data set includes: Decomposing the second spectral signal using a variational modal decomposition algorithm to obtain a modal component set and a center frequency of each modal component; the second spectral signal is any spectral signal in the second spectral data set; Calculate the sample entropy of each modal component; Determine the noise modal component based on the sample entropy and center frequency of each modal component; Eliminating the noise modal component from the modal component set to obtain a valid component set; Perform wavelet threshold denoising on each effective component to obtain the denoised modal component; Reconstructing the signal according to the denoised modal components to obtain a fifth spectral signal; The fifth spectral signal is smoothed and filtered to obtain a third spectral signal, which is included in a third spectral data set.

4. The coating quality detection method based on spectroscopy according to claim 3, characterized in that: The method further includes using an improved particle swarm optimization algorithm to optimize the number of decomposition layers and the penalty factor of the variational mode decomposition algorithm. Specifically, an improved particle velocity update formula is used for particle iteration. The improved particle velocity update formula is: ； Among them, w is the inertia weight, w min and w max is the minimum and maximum value of the inertia weight; t is the current iteration number, T max is the maximum number of iterations; and are two types of learning factors for particle i, and 、 and are the minimum and maximum values ​​of the two types of learning factors respectively; is the normalized value of the fitness of particle i; is the position of particle i at iteration t; is the historical optimal position of particle i; is the global optimal position; r1 and r2 are random numbers; is the updated particle velocity.

5. The coating quality detection method based on spectroscopy according to claim 3, characterized in that: Determining the noise modal component according to the sample entropy and center frequency of each modal component includes: According to the sample entropy of each modal component, the interquartile range method is used to determine the upper limit of abnormality, and the modal components with sample entropy greater than the upper limit of abnormality are included in the candidate set; According to the center frequency of each modal component, the K-means clustering algorithm is used to cluster and obtain the low-frequency set, the medium-frequency set and the high-frequency set; The intersection of the candidate set and the high-frequency set is taken as the noise modal component set.

6. A coating quality detection device based on spectroscopy, characterized in that: The device comprises: A spectrum acquisition module, configured to perform accelerated aging on a sample product of the coating and acquire spectra according to a preset acquisition cycle to obtain a first spectrum data set; a baseline correction module, configured to perform baseline correction on each spectral signal in the first spectral data set using a neural network to obtain a second spectral data set; a signal denoising module, configured to perform denoising processing on each spectral signal in the second spectral dataset by using modal decomposition and wavelet threshold algorithm to obtain a third spectral dataset; a feature expression module, configured to generate aging characterization data based on the third spectral dataset; The indicator evaluation module is used to perform quality evaluation based on the aging characterization data to obtain quality detection results.

7. A coating quality detection device based on spectroscopy according to claim 6, characterized in that: The baseline correction module includes: a preprocessing module, configured to preprocess the first spectral signal to obtain a fourth spectral signal; the first spectral signal is any spectral signal in the first spectral data set; a baseline generation module, configured to use the fourth spectral signal as input to a pre-trained baseline recognition model to obtain a target baseline; the baseline recognition model is a fully connected neural network whose output dimension is the same as the input dimension; The baseline elimination module is used to subtract the target baseline from the fourth spectral signal to obtain a corrected spectral signal and incorporate it into the second spectral data set.

8. The coating quality detection device based on spectroscopy according to claim 6, characterized in that: The signal denoising module includes: a signal decomposition module, configured to decompose the second spectral signal using a variational modal decomposition algorithm to obtain a modal component set and a center frequency of each modal component; the second spectral signal is any spectral signal in the second spectral data set; Sample entropy calculation module, used to calculate the sample entropy of each modal component; A noise determination module is used to determine the noise modal component according to the sample entropy and center frequency of each modal component; An effective set determination module, configured to remove noise modal components from the modal component set to obtain an effective component set; Component denoising module, used to perform wavelet threshold denoising on each effective component to obtain denoised modal components; a signal reconstruction module, configured to reconstruct the signal according to the denoised modal components to obtain a fifth spectral signal; The signal smoothing module is used to perform smoothing filtering on the fifth spectral signal to obtain a third spectral signal and incorporate it into a third spectral data set.

9. The coating quality detection device based on spectroscopy according to claim 8, characterized in that: The device also includes a parameter optimization module for optimizing the number of decomposition layers and penalty factors of the variational mode decomposition algorithm using an improved particle swarm optimization algorithm. Specifically, an improved particle velocity update formula is used for particle iteration, and the improved particle velocity update formula is: ； Among them, w is the inertia weight, w min and w max is the minimum and maximum value of the inertia weight; t is the current iteration number, T max is the maximum number of iterations; and are two types of learning factors for particle i, and 、 and are the minimum and maximum values ​​of the two types of learning factors respectively; is the normalized value of the fitness of particle i; is the position of particle i at iteration t; is the historical optimal position of particle i; is the global optimal position; r1 and r2 are random numbers; is the updated particle velocity.

10. The coating quality detection device based on spectroscopy according to claim 8, characterized in that: The noise determination module includes: A high entropy determination module is used to determine the upper bound of abnormality based on the sample entropy of each modal component using the interquartile range method, and include the modal components with sample entropy greater than the upper bound of abnormality into the candidate set; The high-frequency determination module is used to cluster the center frequency of each modal component using the K-means clustering algorithm to obtain a low-frequency set, a medium-frequency set, and a high-frequency set; The noise selection module is used to take the intersection of the candidate set and the high-frequency set as the noise modal component set.

Images