Metallic oxy-corrosion product film protective detection method and system based on multispectral and color features

Abstract

This invention relates to the field of metal oxide corrosion detection and protection technology, and is a method and system for detecting the protective properties of metal oxide corrosion product films based on multispectral and color characteristics. The method includes acquiring the multispectral reflectance and CIE color parameters of the characteristic bands of the metal sample to be tested; using the multispectral reflectance and CIE color parameters of the metal sample as input to a protective assessment model; and outputting the logarithmic predicted value of the low-frequency impedance modulus and the predicted protection level of the metal sample. This invention establishes a mapping relationship between optical characteristics and protection levels through standardized acquisition, effective area screening, and logarithmic low-frequency impedance modulus, exhibiting good objectivity and repeatability, making it suitable for field engineering applications.

Description

Technical Field

[0001] This invention relates to the field of metal oxygen corrosion detection and protection technology, and is a method and system for detecting the protective properties of metal oxygen corrosion product films based on multispectral and color characteristics. Background Technology

[0002] During service, metallic materials form oxygen corrosion product films of varying types and densities. The continuity, density, and stability of these films directly affect the transport of corrosive media to the metal substrate. Therefore, rapidly determining whether the oxygen corrosion product film is protective is a critical issue in equipment operation and maintenance, life prediction, and corrosion control.

[0003] Chinese patent document CN114708623B discloses a method for evaluating the corrosion level of metals based on image processing. This method uses CNN combined with a sliding window to locate the corrosion area, employs the K-means algorithm to quantify the corrosion color, and establishes a standard color spectrum. Finally, it completes the corrosion level evaluation based on the color of the corrosion products and the corrosion area. Although this method involves technologies related to image recognition, color clustering, color spectrum construction, and model classification, its core only focuses on classifying and evaluating the corrosion appearance level and the proportion of corrosion area. It cannot quantitatively characterize the barrier ability and protective properties of the corrosion product film itself, nor does it establish a calibration relationship between color features and the gold standard of film protection. Therefore, it cannot provide a core basis for corrosion prevention and control and operation and maintenance decisions regarding film protection performance.

[0004] Chinese patent application CN111028229A discloses a method for detecting corrosion of metals or coatings based on image processing technology. It achieves automatic assessment of corrosion level by preprocessing, segmenting, and extracting binary features from color corrosion images. Although this method can achieve automatic judgment of on-site corrosion images and reduce the subjective bias of human visual inspection to a certain extent, it only performs binary segmentation of corrosion areas and appearance level assessment based on ordinary RGB images. It does not use the fusion features of multispectral reflectance information and color parameters, nor does it complete feature calibration through electrochemical impedance index. Therefore, it cannot accurately reflect the intrinsic protective performance of corrosion product film.

[0005] Chinese patent application CN113298766A discloses a quantitative evaluation method for metal corrosion damage based on image recognition. This method acquires corrosion images of steel structures, extracts damage type, damage area, and damage degree features, and combines a BP neural network and a weight calculation model to obtain the comprehensive damage level of the steel structure. Although this method adopts a comprehensive evaluation approach combining image recognition and neural networks, its evaluation object is the overall appearance damage degree of the steel structure, rather than the barrier ability and protective performance of the corrosion product film. It cannot accurately determine the protective properties of the corrosion product film itself, which is fundamentally different from the technical objective of this application.

[0006] Chinese patent application CN112067565A discloses a metal corrosion state assessment device and method based on wavelength variation. It determines the corrosion state of the metal by collecting the wavelength information and band gap width of corrosion products and matching them with a pre-set corrosion state database. Although this scheme involves wavelength information acquisition and database matching technology for corrosion products, it only completes the corrosion state determination through simple database comparison. It does not establish a model training relationship between multi-band reflectivity, CIE L*a*b* color parameters and calibration sample sets with electrochemical impedance labels, and therefore cannot achieve accurate quantitative prediction of the protective properties of corrosion product films. The reliability and generalization of the detection results are insufficient.

[0007] Furthermore, existing published research literature on hyperspectral imaging mainly focuses on using VNIR / HSI spectral information to distinguish the severity of metal corrosion, identify the types of corrosive minerals, or locate coating defects. Although it involves related technologies such as spectral acquisition, feature extraction, and classification, its core research objectives are the presence or absence of corrosion, the severity of corrosion, or the type of corrosion product minerals. It has never taken the protective properties of corrosion product films as the core evaluation object, nor has it established a calibration relationship between spectral characteristics and the gold standard for film protection. Therefore, it cannot meet the actual needs of rapid evaluation of the protective performance of oxygen corrosion product films in engineering sites. Summary of the Invention

[0008] This invention provides a method and system for detecting the protective properties of metal oxygen corrosion product films based on multispectral and color characteristics, which can effectively solve the problem that existing methods are difficult to directly evaluate the protective properties of oxygen corrosion product films.

[0009] One of the technical solutions of this invention is achieved through the following measures: a method for detecting the protective properties of metal oxide corrosion product films based on multispectral and color characteristics, comprising: Obtain the multispectral reflectance and CIE color parameters of the characteristic bands of the metal sample to be tested; Using the multispectral reflectance and CIE color parameters of the characteristic bands of the metal sample to be tested as inputs, the protective assessment model is input, and the predicted values ​​of the logarithmic low-frequency impedance modulus and the protective level of the metal sample to be tested are output. The protective assessment model includes a regression model that correlates color features with the logarithmic low-frequency impedance modulus and a protective level classification model. The regression model that correlates color features with the logarithmic low-frequency impedance modulus is a regression model constructed from the CIE color parameters, multispectral reflectance of the characteristic bands, and logarithmic low-frequency impedance modulus of the metal calibration sample. The protective level classification model is obtained by unsupervised learning using several training samples. The training samples include the logarithmic low-frequency impedance modulus and the protective level of the metal calibration sample, and the metal type and service environment of the metal sample to be tested correspond to those of the metal calibration sample.

[0010] The following are further optimizations and / or improvements to the above-mentioned technical solution: Furthermore, the multispectral reflectance of the characteristic bands of the above-mentioned metal calibration samples was obtained using the following method: Using a metal calibration sample as the processing sample, after surface pretreatment, a multispectral reflectance image is acquired. The acquisition band of the multispectral reflectance image is from 400nm to 1000nm, and the acquisition band includes multiple discrete bands including the visible band and the near-infrared band. The multispectral reflectance image is divided into multiple sub-regions. Invalid regions are removed from these sub-regions to obtain several valid regions. The multispectral reflectance of multiple discrete bands in these valid regions is obtained. The mean spectral reflectance of each discrete band is obtained based on the multispectral reflectance of the multiple discrete bands in these valid regions. Principal component analysis (PCA) is used to reduce the dimensionality of the mean spectral reflectance of each discrete band, and principal components whose cumulative contribution rate reaches a threshold are extracted. The principal components are feature bands selected from the discrete bands, and the multispectral reflectance of the feature bands is the mean spectral reflectance of the corresponding discrete band.

[0011] Furthermore, the aforementioned discrete bands are combinations of the 405nm, 430nm, 450nm, 490nm, 515nm, 590nm, 630nm, 660nm, 690nm and 850nm bands.

[0012] Furthermore, the regression model relating the aforementioned color features to the logarithmic low-frequency impedance modulus is constructed as follows: The CIE color parameters are CIE L*a*b* color parameters. The CIE L*a*b* color parameters include the brightness parameter L*, the red-green parameter a*, and the yellow-blue parameter b*. The color information corresponding to the effective area is obtained, and the corresponding color information is converted to the CIEL*a*b* color space to obtain the CIE L*a*b* color parameters. Through multiple linear regression analysis, a regression model was constructed to correlate color features with the logarithmic value of low-frequency impedance modulus. The expression is as follows: lg|Z|=C0+C1×L*+C2×a*+C3×b*+Σ(C i ×B i ) Where |Z| is the low-frequency impedance magnitude; lg|Z| is the logarithmic value of the low-frequency impedance magnitude; C0 is a constant term; C1, C2, and C3 are the regression coefficients of L*, a*, and b*, respectively; C i Characteristic band B i The regression coefficient, B i The multispectral reflectance is the characteristic band.

[0013] Furthermore, the aforementioned low-frequency impedance modulus logarithm value is the impedance modulus logarithm value at frequencies from 0.005 Hz to 0.1 Hz, preferably the impedance modulus logarithm value at 0.01 Hz.

[0014] Furthermore, the above-mentioned protection level classification model is constructed using the following method: Obtain a calibration sample set, which includes the logarithmic values ​​of the low-frequency impedance modulus of several metal calibration samples and their corresponding protection levels; The calibration sample set is divided into a training sample set and a test sample set; The unsupervised learning model is trained using the training sample set to obtain the classification model for predicting the initial protection level. The initial protection level classification model is tested using a test sample set. If the model accuracy is met, the initial protection level classification model is output as the optimal protection level classification model. Otherwise, the parameters of the unsupervised learning model are adjusted, and the unsupervised learning model is retrained using a training sample set.

[0015] Furthermore, when acquiring the multispectral reflectance images, the calibrated lighting conditions are as follows: illuminance of 500 lux to 1000 lux; light incident angle preferably of 30° to 60°; and shooting distance of 20cm to 50cm.

[0016] The second technical solution of the present invention is achieved through the following measures: a system for detecting the protective properties of metal oxide corrosion product films based on multispectral and color characteristics, comprising: Data acquisition module: Acquires the multispectral reflectance and CIE color parameters of the characteristic bands of the metal sample under test; The determination module takes the multispectral reflectance and CIE color parameters of the characteristic band of the metal sample to be tested as inputs, inputs the protection determination model, and outputs the predicted value of the logarithmic value of the low-frequency impedance modulus and the predicted value of the protection level of the metal sample to be tested. The protection determination model includes a regression model that correlates the color features with the logarithmic value of the low-frequency impedance modulus and a protection level classification model. The regression model that correlates the color features with the logarithmic value of the low-frequency impedance modulus is a regression model constructed by the CIE color parameters, multispectral reflectance of the characteristic band, and logarithmic value of the low-frequency impedance modulus of the metal calibration sample. The protection level classification model is obtained by unsupervised learning using several training samples. The training samples include the logarithmic value of the low-frequency impedance modulus and the protection level of the metal calibration sample, and the metal type and service environment of the metal sample to be tested correspond to those of the metal calibration sample.

[0017] The beneficial effects of this invention are: Compared with existing image recognition methods that only target the presence or degree of corrosion, this invention directly targets the oxygen corrosion product film as the technical object, predicts the protection level, and outputs results that are closer to the needs of oxygen corrosion prevention and control and operation and maintenance decisions.

[0018] Compared with single-point electrochemical detection, this invention obtains the surface area information of metal samples by scanning with a multispectral imaging device, which can quickly evaluate the surface state over a larger range and reduce the dependence on extrapolation of single-point results.

[0019] Compared with ordinary RGB image methods, this invention uses multiple discrete band reflectance and CIE L*a*b* color parameters simultaneously, which helps to reduce the impact of illumination changes and visual ambiguity.

[0020] This invention establishes a mapping relationship between optical characteristics and protection levels through standardized acquisition, effective area screening, and logarithmic low-frequency impedance modulus values. It has good objectivity and repeatability, making it suitable for engineering field applications. Attached Figure Description

[0021] Appendix Figure 1 The flowchart shows a method for detecting the protective properties of metal oxide corrosion product films based on multispectral and color characteristics. Appendix Figure 2 A schematic diagram of the standardized acquisition geometry for multispectral images; Appendix Figure 3 This is a schematic diagram of effective region extraction.

[0022] The codes in the attached figures are as follows: 1 represents the multispectral imaging device, 2 represents the lens optical axis, 3 represents the standard D65 light source, 4 represents the shooting distance, 5 represents the standard reference white board, 6 represents the metal calibration sample or the metal sample to be tested, and α represents the incident angle. Detailed Implementation

[0023] The present invention is not limited to the following embodiments, and the specific implementation can be determined according to the technical solution of the present invention and the actual situation.

[0024] like Figure 2 As shown, the multispectral imaging equipment is a conventional device for acquiring multispectral reflectance, and its supporting standard reference whiteboard and other accessories are also standard.

[0025] like Figure 1 As shown, this invention provides a method for detecting the protective properties of metal oxide corrosion product films based on multispectral and color characteristics, comprising: Obtain the multispectral reflectance and CIE color parameters of the characteristic bands of the metal sample to be tested; Using the multispectral reflectance and CIE color parameters of the characteristic bands of the metal sample to be tested as inputs, the protective assessment model is input, and the predicted values ​​of the logarithmic low-frequency impedance modulus and the protective level of the metal sample to be tested are output. The protective assessment model includes a regression model that correlates color features with the logarithmic low-frequency impedance modulus and a protective level classification model. The regression model that correlates color features with the logarithmic low-frequency impedance modulus is a regression model constructed from the CIE color parameters, multispectral reflectance of the characteristic bands, and logarithmic low-frequency impedance modulus of the metal calibration sample. The protective level classification model is obtained by unsupervised learning using several training samples. The training samples include the logarithmic low-frequency impedance modulus and the protective level of the metal calibration sample, and the metal type and service environment of the metal sample to be tested correspond to those of the metal calibration sample.

[0026] Low-frequency impedance modulus is measured by electrochemical impedance spectroscopy. Calibration metal samples can be derived from natural exposure tests, accelerated corrosion tests in the laboratory, and validated in-service component samples. Protection levels are classified from 4 to 8, with 8 being preferred, for rapid on-site comparison and result visualization in engineering projects.

[0027] The multispectral reflectance of the characteristic band is obtained by the following method: The multispectral reflectance of the characteristic bands of the metal calibration sample is obtained using the following method: Surface pretreatment: The surface of the sample is cleaned without damaging the oxygen corrosion product film, removing dust, oil and obvious loose attachments, and making the surface ready for repeated sampling.

[0028] Using a metal calibration sample as the processing sample, after surface pretreatment, a multispectral reflectance image is acquired. The acquisition band of the multispectral reflectance image is from 400nm to 1000nm, and the acquisition band includes multiple discrete bands including the visible band and the near-infrared band. After performing whiteboard correction, dark field correction, and noise suppression on the acquired multispectral reflectance image, the multispectral reflectance image is divided into multiple sub-regions. From these sub-regions, invalid regions (such as regions with strong light spots, shadows, exposed substrates, or obvious contaminant coverage) are removed, resulting in several effective regions. The multispectral reflectance of multiple discrete bands in these effective regions is obtained. Based on the multispectral reflectance of multiple discrete bands in these effective regions, the mean spectral reflectance of each discrete band is obtained. Principal component analysis (PCA) is used to reduce the dimensionality of the mean spectral reflectance of each discrete band, and principal components whose cumulative contribution rate reaches a threshold are extracted. These principal components are feature bands selected from the discrete bands, and the multispectral reflectance of these feature bands is the mean spectral reflectance of the corresponding discrete band.

[0029] Since the metal type and service environment of the metal sample to be tested correspond to those of the metal calibration sample, the characteristic bands of the metal sample to be tested are the same as those of the metal calibration sample. Therefore, for the metal sample to be tested, it is only necessary to collect the multispectral reflectance of the characteristic band of the effective area, and there is no need to collect the multispectral reflectance of 10 bands.

[0030] The regression model relating the color features to the logarithmic value of the low-frequency impedance modulus is constructed as follows: The CIE color parameters are CIE L*a*b* color parameters. The CIE L*a*b* color parameters include the brightness parameter L*, the red-green parameter a*, and the yellow-blue parameter b*. The color information corresponding to the effective area is obtained, and the corresponding color information is converted to the CIEL*a*b* color space to obtain the CIE L*a*b* color parameters. Through multiple linear regression analysis, a regression model was constructed to correlate color features with the logarithmic value of low-frequency impedance modulus. The expression is as follows: lg|Z| 0.01 Hz=C0+C1×L*+C2×a*+C3×b*+Σ(C i ×B i ) Where |Z| is the low-frequency impedance magnitude; lg|Z| is the logarithmic value of the low-frequency impedance magnitude; C0 is a constant term; C1, C2, and C3 are the regression coefficients of L*, a*, and b*, respectively; C i Characteristic band B i The regression coefficient, B i Let be the multispectral reflectance of the characteristic band, and i be the number of characteristic bands. The goodness of fit Ri of the regression model is ensured through fitting the model. 2 With a value ≥0.95, the sum of squared residuals is minimized, enabling accurate quantitative prediction of the logarithmic value of the low-frequency impedance modulus of the metal sample under test.

[0031] When constructing a regression model relating color features to the logarithmic value of the low-frequency impedance modulus, |Z| represents the low-frequency impedance modulus of the metal-calibrated sample measured by electrochemical impedance spectroscopy; lg|Z| represents the logarithmic value of the low-frequency impedance modulus measured by electrochemical impedance spectroscopy; B i This refers to the multispectral reflectance of the characteristic bands of the metal calibration sample; the CIE L*a*b* color parameters are the CIE L*a*b* color parameters of the metal calibration sample. This is determined by the lg|Z| and B values ​​of the metal calibration sample. i Using the CIE L*a*b* color parameters, determine the constant term and regression coefficients in the regression model.

[0032] When predicting the logarithmic value of the low-frequency impedance modulus of a metal sample using a regression model that correlates color features with the logarithmic value of the low-frequency impedance modulus, B i It is the multispectral reflectance of the characteristic band of the metal sample to be tested; the CIE L*a*b* color parameter is the CIE L*a*b* color parameter of the metal sample to be tested.

[0033] The multiple discrete bands are band combinations of 405nm, 430nm, 450nm, 490nm, 515nm, 590nm, 630nm, 660nm, 690nm and 850nm bands.

[0034] The characteristic bands are selected from these 10 bands. For example, the multispectral reflectance of the characteristic bands is obtained: if there are 5 effective regions (see...). Figure 3 Each effective region has 10 discrete bands (405nm, 430nm, 450nm, 490nm, 515nm, 590nm, 630nm, 660nm, 690nm, and 850nm bands) with multispectral reflectance. The mean multispectral reflectance of each band across these 5 effective regions is calculated (for example, for the 405nm band, if its multispectral reflectances in the 5 effective regions are A, B, C, D, and E respectively, then the mean multispectral reflectance of the 405nm band is the mean of A, B, C, D, and E). Then, principal component analysis (PCA) is used to analyze the multispectral reflectance of each band... The dimensionality of the mean spectral reflectance of discrete bands is reduced, and principal components whose cumulative contribution rate reaches the cumulative contribution rate threshold are extracted. The principal components are the feature bands selected from the discrete bands, and the multispectral reflectance of the feature bands is the mean spectral reflectance of the corresponding discrete bands (for example, through PCA principal component analysis, if the cumulative contribution rate of the mean spectral reflectance of the 405nm, 430nm, and 450nm bands reaches the cumulative contribution rate threshold, then the 405nm, 430nm, and 450nm bands are the required feature bands, and the corresponding mean spectral reflectance is the multispectral reflectance of the feature bands).

[0035] The low-frequency impedance magnitude logarithm value is the impedance magnitude logarithm value at frequencies from 0.005 Hz to 0.1 Hz, preferably the impedance magnitude logarithm value at 0.01 Hz.

[0036] Due to the impedance modulus of the oxygen corrosion system at a frequency of 0.01 Hz (|Z| 0.01 (Hz) spans multiple orders of magnitude; to achieve high-precision linear fitting, its logarithmic value lg|Z| 0.01 Hz is the gold standard for determining the level of protection (the impedance modulus at this frequency can accurately characterize the barrier ability of metal oxygen corrosion products and is positively correlated with the level of protection).

[0037] The protection level classification model is constructed using the following method: Obtain a calibration sample set, which includes the logarithmic values ​​of the low-frequency impedance modulus of several metal calibration samples and their corresponding protection levels; The calibration sample set is divided into a training sample set and a test sample set; The unsupervised learning model is trained using the training sample set to obtain the classification model for predicting the initial protection level. The initial protection level classification model is tested using a test sample set. If the model accuracy is met, the initial protection level classification model is output as the optimal protection level classification model. Otherwise, the parameters of the unsupervised learning model are adjusted, and the unsupervised learning model is retrained using a training sample set.

[0038] The unsupervised learning model employs one or more of the following: support vector machine, random forest, gradient boosting tree, and one-dimensional convolutional neural network.

[0039] like Figure 2 As shown, when acquiring multispectral reflectance images, a fixed incident angle, shooting distance, and whiteboard / dark field correction are used to reduce the impact of illumination drift and specular reflection on the data.

[0040] (1) Calibrated lighting conditions: It is preferred to use an equivalent D65 white light source, and the illuminance is preferably 500 lux to 1000 lux; (2) Shooting geometry: The incident angle of light is preferably 30° to 60°, more preferably about 45°; the optical axis of the lens is preferably approximately perpendicular to the surface to be measured; (3) Shooting distance: Set according to the lens field of view and resolution, preferably 20cm to 50cm.

[0041] For new samples subsequently verified by electrochemical impedance spectroscopy, they can be incorporated into the calibration sample set for offline updating of model parameters (such as updating the regression coefficients of the regression model) or optimizing the protection level threshold.

[0042] The present invention will be further described below with reference to embodiments: Example 1: Protective testing of oxygen corrosion product film on X70 carbon steel for long-distance pipelines This embodiment focuses on a section of X70 carbon steel natural gas pipeline that has been in service for 3 years. The service environment of this pipeline is a western soil environment, and a dark brown oxygen corrosion product film has formed on its surface. The method of this invention is used to conduct protective testing of the oxygen corrosion product film. The specific steps are as follows: 1. Database matching: The object of this test is X70 carbon steel, and the service environment is soil environment. The calibration sample set of carbon steel-soil environment (calibration sample size 186 groups) in the database of this invention is matched, and the regression model and protection level classification model corresponding to the color feature and low frequency impedance modulus value are called.

[0043] Regression model showing the logarithmic correlation between color characteristics and low-frequency impedance magnitude: lg|Z| 0.01 Hz=C0+C1×L*+C2×a*+C3×b*+Σ(C i ×B i (1) Where |Z| is the low-frequency impedance magnitude; lg|Z| 0.01 Hz represents the logarithmic value of the low-frequency impedance modulus at a sampling frequency of 0.01Hz; C0 is a constant term, C1, C2, and C3 are the regression coefficients of L*, a*, and b*, respectively. i Characteristic band B i The regression coefficient, B i denoted as , where i is the multispectral reflectance of the characteristic band and i is the number of characteristic bands.

[0044] In this regression model, the multispectral reflectance of the characteristic bands is obtained using the following method: 1) Surface pretreatment: Cut a pipe section sample (50mm×50mm×10mm) and use the pipe section sample as the metal calibration sample. Clean it with anhydrous ethanol for 30s with an ultrasonic power of 120W and dry it with cold air until the surface relative humidity is 55%. Use an HB hardness brush to remove loose rust from the surface and retain the complete oxygen corrosion product film. After surface pretreatment, the surface is free of oil, dust and exposed substrate, which meets the testing requirements.

[0045] 2) Multispectral Image Acquisition: A handheld multispectral imaging device (spectral resolution 3nm, spatial resolution 80μm) was used to acquire multispectral reflectance images of each metal calibration sample in 10 bands at a frequency of 0.01Hz. The acquisition environment was a standard D65 white light source with an illuminance of 800 lux, a light incident angle of 45°, a shooting distance of 30cm, an ambient temperature of 25℃, and a relative humidity of 50%. The surface of the metal sample to be tested was divided into 9 sub-regions using the nine-square grid method. Two sub-regions with slight shadows were removed, and 7 effective regions were selected. The mean multispectral reflectance of all metal calibration samples in 10 bands and the CIEL*a*b* color parameter values ​​were calculated as follows: L*=32.6, a*=12.8, b*=8.2.

[0046] 3) Feature band extraction: Principal component analysis (PCA) was used to reduce the dimensionality of the 10 band data and extract three principal components with a cumulative contribution rate of 96.8% (cumulative contribution rate threshold). Specifically, the 450nm, 590nm, and 660nm bands were selected as the three feature bands. The average multispectral reflectance of the 450nm, 590nm, and 660nm bands is the multispectral reflectance of these three feature bands.

[0047] Logarithmic values ​​of the low-frequency impedance modulus (obtained by electrochemical impedance spectroscopy), multispectral reflectance of three characteristic bands (450nm, 590nm, and 660nm), L*, a*, and b* were analyzed using multiple linear analysis to determine the constant term C0 in formula (1), the regression coefficients C1, C2, and C3 of L*, a*, and b*, and the characteristic band B*. i The regression coefficient C i Thus, a regression model is obtained that correlates the color features with the logarithmic value of the low-frequency impedance modulus: lg|Z| 0.01 Hz=4.28+0.032×L*+0.056×a*-0.028×b*+0.12×B 450 +0.15×B 590 +0.11×B 660 (2) Among them, lg|Z| 0.01 Hz represents the logarithmic value of the low-frequency impedance modulus at a sampling frequency of 0.01Hz; L* represents the brightness parameter; a* represents the red-green parameter; b* represents the yellow-blue parameter; B 450 B 590 B 660 The multispectral reflectance is the characteristic band.

[0048] Protection level classification model: Level I: lg|Z|≥7.0; Level II: 7.0>lg|Z|≥6.0; Level III: 6.0>lg|Z|≥5.7; Level IV: 5.7>lg|Z|≥5.0; Level V: 5.0>lg|Z|≥4.0; Level VI: 4.0>lg|Z|≥3.7; Level VII: 3.7>lg|Z|≥3.0; Level VIII: lg|Z|<3.0; Among them, Level I and Level II are high protection; Level III and Level IV are good protection; Level V is medium protection; Level VI is weak protection; and Level VII and Level VIII are no protection.

[0049] 2. Determination of Protection Level: The metal sample to be tested is processed according to steps 1) and 2) above to obtain the CIE L*a*b* color parameters of the metal sample to be tested: L*=32.6, a*=12.8, b*=8.2; the multispectral reflectance of the three characteristic bands (450nm, 590nm, 660nm bands) is substituted into formula (2) to calculate lg|Z| 0.01 The Hz value is 6.28, meaning the logarithmic predicted value of the impedance modulus is 6.28, corresponding to a predicted impedance modulus of 1.91 × 10⁻⁶. 6 Ω·cm 2 .

[0050] According to the protection level classification model, the oxygen corrosion product film is classified as strong protection level II. Subsequent electrochemical impedance spectroscopy measurements showed lg|Z|0.01Hz to be 6.32, indicating that the predicted result can effectively reflect the protective and barrier properties of the oxygen corrosion product film.

[0051] Example 2: Protective Testing of Oxygen Rust Layer on Q345qDNH Weathering Steel for Cross-Sea Bridges The test subject was a Q345qDNH weathering steel component (metal sample) that had been in service for approximately 5 years, and the environment was a marine atmospheric environment. Surface pretreatment: After lightly wiping away sea salt particles and dust from the surface of the metal sample, a portable multispectral device was used to collect data on-site, and a whiteboard was simultaneously calibrated.

[0052] Following the method described in Example 1, the characteristic wavelengths of the calibrated metal sample of the test object are known to be 430nm, 515nm, and 850nm. The regression coefficients and constants in the regression model relating the color characteristics to the logarithmic correlation of the low-frequency impedance modulus are known. By collecting the multispectral reflectance of the metal sample in the 430nm, 515nm, and 850nm wavelengths and the corresponding color parameters: L*=28.4, a*=8.7, b*=5.3, and inputting them into the regression model relating the color characteristics to the logarithmic correlation of the low-frequency impedance modulus, lg|Z| is calculated. 0.01Hz = 6.74. Based on the protection level classification model, the tested object is determined to belong to Strong Protection Level I. Sampling verification shows the measured lg|Z| of the electrochemical impedance spectroscopy. 0.01 The Hz value is 6.81, indicating that the method described in this invention can be used for on-site determination of stable oxygen rust layer on weathering steel.

[0053] Example 3: Protective Test of Oxygen Corrosion Product Film on Hot-Dip Galvanized Pipelines in Chemical Plants The object of the inspection was a hot-dip galvanized pipe that had been in service for approximately one year, and the environment was a chemical medium. Multispectral images were acquired after surface pretreatment.

[0054] Following the method described in Example 1, the characteristic wavelengths of the calibrated metal sample for testing are known to be 490nm, 630nm, and 690nm. The regression coefficients and constants in the regression model relating the color characteristics to the logarithmic low-frequency impedance modulus are clearly defined. By collecting the multispectral reflectance of the metal sample in the 490nm, 630nm, and 690nm wavelengths and the corresponding color parameters: L*=82.6, a*=-1.2, b*=3.8, and inputting them into the regression model relating the color characteristics to the logarithmic low-frequency impedance modulus, lg|Z|0.01Hz=3.42 is calculated. Based on the protection level classification model, the oxygen corrosion product film is determined to be unprotected, Class I. The measured electrochemical impedance spectroscopy lg|Z|0.01Hz is 3.37, indicating that the barrier ability of the oxygen corrosion product film in this zinc plating layer is weak, requiring timely maintenance measures.

[0055] As can be seen from the above embodiments, the method of the present invention can evaluate the protective level of oxygen corrosion product films.

[0056] The above technical features constitute various embodiments of the present invention, which have strong adaptability and implementation effect. Unnecessary technical features can be added or removed according to actual needs to meet the needs of different situations.

Claims

1. A method for detecting the protective properties of metal oxide corrosion product films based on multispectral and color characteristics, characterized in that, include: Obtain the multispectral reflectance and CIE color parameters of the characteristic bands of the metal sample to be tested; Using the multispectral reflectance and CIE color parameters of the characteristic bands of the metal sample to be tested as inputs, the protective assessment model is input, and the predicted values ​​of the logarithmic low-frequency impedance modulus and the protective level of the metal sample to be tested are output. The protective assessment model includes a regression model that correlates color features with the logarithmic low-frequency impedance modulus and a protective level classification model. The regression model that correlates color features with the logarithmic low-frequency impedance modulus is a regression model constructed from the CIE color parameters, multispectral reflectance of the characteristic bands, and logarithmic low-frequency impedance modulus of the metal calibration sample. The protective level classification model is obtained by unsupervised learning using several training samples. The training samples include the logarithmic low-frequency impedance modulus and the protective level of the metal calibration sample, and the metal type and service environment of the metal sample to be tested correspond to those of the metal calibration sample.

2. The method for detecting the protective properties of metal oxide corrosion product films based on multispectral and color characteristics according to claim 1, characterized in that, The multispectral reflectance of the characteristic bands of the metal calibration sample is obtained using the following method: Using a metal calibration sample as the processing sample, after surface pretreatment, a multispectral reflectance image is acquired. The acquisition band of the multispectral reflectance image is from 400nm to 1000nm, and the acquisition band includes multiple discrete bands including the visible band and the near-infrared band. The multispectral reflectance image is divided into multiple sub-regions. Invalid regions are removed from these sub-regions to obtain several valid regions. The multispectral reflectance of multiple discrete bands in these valid regions is obtained. The mean spectral reflectance of each discrete band is obtained based on the multispectral reflectance of the multiple discrete bands in these valid regions. Principal component analysis (PCA) is used to reduce the dimensionality of the mean spectral reflectance of each discrete band, and principal components whose cumulative contribution rate reaches a threshold are extracted. The principal components are feature bands selected from the discrete bands, and the multispectral reflectance of the feature bands is the mean spectral reflectance of the corresponding discrete band.

3. The method for detecting the protective properties of metal oxide corrosion product films based on multispectral and color characteristics according to claim 2, characterized in that, Multiple discrete bands are combinations of bands at 405nm, 430nm, 450nm, 490nm, 515nm, 590nm, 630nm, 660nm, 690nm and 850nm.

4. The method for detecting the protective properties of metal oxide corrosion product films based on multispectral and color characteristics according to claim 2 or 3, characterized in that, The regression model relating color characteristics to the logarithmic value of low-frequency impedance modulus was constructed as follows: The CIE color parameters are CIE L*a*b* color parameters. The CIE L*a*b* color parameters include the brightness parameter L*, the red-green parameter a*, and the yellow-blue parameter b*. The color information corresponding to the effective area is obtained, and the corresponding color information is converted to the CIE L*a*b* color space to obtain the CIE L*a*b* color parameters. Through multiple linear regression analysis, a regression model was constructed to correlate color features with the logarithmic value of low-frequency impedance modulus. The expression is as follows: lg|Z|=C0+C1×L*+C2×a*+C3×b*+Σ(C i ×B i ) Where |Z| is the low-frequency impedance magnitude; lg|Z| is the logarithmic value of the low-frequency impedance magnitude; C0 is a constant term; C1, C2, and C3 are the regression coefficients of L*, a*, and b*, respectively; C i Characteristic band B i The regression coefficient, B i denoted as , where i is the multispectral reflectance of the characteristic band and i is the number of characteristic bands.

5. The method for detecting the protective properties of metal oxide corrosion product films based on multispectral and color characteristics according to claim 4, characterized in that, The logarithmic value of the low-frequency impedance modulus is the logarithmic value of the impedance modulus at frequencies from 0.005 Hz to 0.1 Hz.

6. The method for detecting the protective properties of metal oxide corrosion product films based on multispectral and color characteristics according to claim 2, 3, or 5, characterized in that, The protection level classification model is constructed using the following method: Obtain a calibration sample set, which includes the logarithmic values ​​of the low-frequency impedance modulus of several metal calibration samples and their corresponding protection levels; The calibration sample set is divided into a training sample set and a test sample set; The unsupervised learning model is trained using the training sample set to obtain the classification model for predicting the initial protection level. The initial protection level classification model is tested using a test sample set. If the model accuracy is met, the initial protection level classification model is output as the optimal protection level classification model. Otherwise, the parameters of the unsupervised learning model are adjusted, and the unsupervised learning model is retrained using a training sample set.

7. The method for detecting the protective properties of metal oxide corrosion product films based on multispectral and color characteristics according to claim 4, characterized in that, The protection level classification model is constructed using the following method: Obtain a calibration sample set, which includes the logarithmic values ​​of the low-frequency impedance modulus of several metal calibration samples and their corresponding protection levels; The calibration sample set is divided into a training sample set and a test sample set; The unsupervised learning model is trained using the training sample set to obtain the classification model for predicting the initial protection level. The initial protection level classification model is tested using a test sample set. If the model accuracy is met, the initial protection level classification model is output as the optimal protection level classification model. Otherwise, the parameters of the unsupervised learning model are adjusted, and the unsupervised learning model is retrained using a training sample set.

8. The method for detecting the protective properties of metal oxide corrosion product films based on multispectral and color characteristics according to claim 2, 3, 5, or 7, characterized in that, When acquiring multispectral reflectance images, the calibrated lighting conditions are: illuminance of 500 lux to 1000 lux; light incident angle preferably of 30° to 60°; and shooting distance of 20cm to 50cm.

9. The method for detecting the protective properties of metal oxide corrosion product films based on multispectral and color characteristics according to claim 6, characterized in that, When acquiring multispectral reflectance images, the calibrated lighting conditions are: illuminance of 500 lux to 1000 lux; light incident angle preferably of 30° to 60°; and shooting distance of 20cm to 50cm.

10. A system for detecting the protective properties of metal oxide corrosion product films based on multispectral and color characteristics as described in any one of claims 1 to 9, characterized in that, include: Data acquisition module: Acquires the multispectral reflectance and CIE color parameters of the characteristic bands of the metal sample under test; The determination module takes the multispectral reflectance and CIE color parameters of the characteristic band of the metal sample to be tested as inputs, inputs the protection determination model, and outputs the predicted value of the logarithmic value of the low-frequency impedance modulus and the predicted value of the protection level of the metal sample to be tested. The protection determination model includes a regression model that correlates the color features with the logarithmic value of the low-frequency impedance modulus and a protection level classification model. The regression model that correlates the color features with the logarithmic value of the low-frequency impedance modulus is a regression model constructed by the CIE color parameters, multispectral reflectance of the characteristic band, and logarithmic value of the low-frequency impedance modulus of the metal calibration sample. The protection level classification model is obtained by unsupervised learning using several training samples. The training samples include the logarithmic value of the low-frequency impedance modulus and the protection level of the metal calibration sample, and the metal type and service environment of the metal sample to be tested correspond to those of the metal calibration sample.

Images