An Early Corrosion Detection Method Based on Eddy Current Thermography and Optical Flow-PCA

Abstract

The present invention discloses an early corrosion detection method based on eddy current thermal imaging and optical flow PCA, which relates to the field of corrosion detection technology. The method comprises the following steps: analyzing the temperature response curves of the corroded and non-corroded areas of an early-stage corrosion sample and selecting the time frame interval with the maximum relative tangent slope in the temperature response curve; obtaining a sequence of optical flow amplitude images of consecutive frames, and calculating the optical flow amplitude difference based on the sequence of optical flow amplitude images to assess the corrosion degree of the early-stage corrosion sample; reconstructing the sequence of optical flow amplitude images of consecutive frames, extracting features from the reconstructed matrix using principal component analysis, and performing weighted reconstruction of the principal component features to obtain the heat flow characteristics of the corroded area. The present invention can accurately reflect the subtle differences in heat flow direction and intensity between corroded and non-corroded areas, greatly improving the sensitivity of early-stage corrosion signals.

Description

Technical Field

[0001] The present invention relates to the technical field of corrosion detection, and in particular to an early corrosion detection method based on eddy current thermal imaging and optical flow-PCA. Background Art

[0002] Pulsed eddy current thermography actively excites and heats the surface of conductive materials, capturing infrared image sequences to obtain material corrosion information. However, in early corrosion detection, the thermal conductivity characteristics of the corrosion area change little, resulting in unclear local temperature differences. The contrast between the corroded and non-corroded areas in the infrared image is low, making it difficult to effectively identify early corrosion areas. In addition, infrared thermal image sequences are susceptible to interference from factors such as environmental noise, heating unevenness, and background heat reflection, which further reduces the accuracy of feature extraction in early corrosion areas. On the other hand, existing data processing methods are insufficiently sensitive to capturing small dynamic heat flow changes in time series, and are unable to effectively characterize the heat flow distribution characteristics of the corrosion area, further limiting the accurate identification and visualization of early corrosion areas.

[0003] The existing fast Fourier transform (FFT) method has obvious limitations in early corrosion detection, which are mainly reflected in the following three aspects:

[0004] First, the FFT method ignores the dynamic heat flow characteristics of the temperature field and cannot accurately capture early corrosion signals. Theoretically, FFT extracts amplitude-frequency and phase-frequency characteristics by performing static frequency-domain decomposition on infrared image sequences. However, it cannot characterize the changes in heat flow direction and intensity between corroded and non-corroded areas over time. In practice, early corrosion signals manifest as weak temperature distribution and insignificant dynamic response. Especially during the heating and cooling phases, heat diffusion in the corroded area exhibits time delays and localized conduction blockages. This dynamic heat flow characteristic is particularly critical in the time domain, while FFT only extracts global frequency information and ignores the temporal correlation of temperature response. Furthermore, due to the uneven heat flow distribution caused by variations in thermal conductivity and resistivity in the corroded area, the amplitude and phase characteristics extracted by FFT cannot effectively reflect these subtle changes. This results in low contrast between the corroded area and the background, easily masking the corrosion signal, and ultimately affecting detection sensitivity and visualization.

[0005] Secondly, the FFT method is extremely sensitive to noise, which seriously affects the accuracy and stability of detection. In infrared thermal image sequences, environmental noise, background heat reflection, and heating unevenness are the main sources of interference. Especially in the early stages of corrosion, the temperature change amplitude of the corroded area is small, the thermal response is weak, and the real signal is easily superimposed on the noise. In the frequency domain, FFT lacks an effective separation mechanism for these noise components, and the noise is easily amplified, thereby obscuring the true characteristics of the corroded area. Specifically, the signal-to-noise ratio (SNR) is reduced, the feature contrast between the corroded and non-corroded areas is insufficient, false signals or feature loss appear in the reconstructed image, and the boundaries and morphology are difficult to clearly present. Therefore, noise interference makes it difficult for the FFT method to stably extract key information of the corrosion area, limiting its actual detection accuracy.

[0006] Finally, the FFT method lacks an objective basis for frequency selection, further reducing its feasibility in practical operations. Although FFT can extract the amplitude and phase characteristics of temperature field signals, in actual applications, infrared image sequences usually contain multiple frequency components, and different frequencies have different response effects on the corrosion area. Therefore, to obtain the best detection effect, it is necessary to select the optimal frequency for image reconstruction. However, the current FFT lacks a unified frequency selection standard and usually relies on subjective experience or repeated experiments, making it difficult to ensure the objectivity and consistency of the reconstruction results. In addition, the differences in corrosion degree, thickness, and material properties of different samples make the selection of the optimal frequency component more complex and unstable, further increasing the difficulty of operation and application cost. This results in the unstable effect of the FFT method in practical applications, and the reconstructed corrosion area image cannot fully and accurately reflect the true characteristics, reducing the reliability of early corrosion detection.

[0007] According to the above analysis, the existing FFT method has defects such as ignoring dynamic heat flow characteristics, being sensitive to noise, and lacking objective basis for frequency selection. These technical defects cause the FFT method to exhibit the shortcomings of low sensitivity, low signal-to-noise ratio, and high complexity in early corrosion detection, limiting its reliability and feasibility in practical applications, and unable to effectively meet the needs of weak signal extraction and feature visualization in early corrosion detection.

[0008] Currently, no effective solutions have been proposed for the problems in related technologies. Summary of the Invention

[0009] In view of this, the present invention provides an early corrosion detection method based on eddy current thermal imaging and optical flow-PCA to solve the above-mentioned problems.

[0010] In order to solve the above problems, the specific technical solutions adopted by the present invention are as follows:

[0011] An early corrosion detection method based on eddy current thermography and optical flow-PCA, the method comprising the following steps:

[0012] S1. Based on a sequence of infrared thermal images of an early-stage corrosion sample acquired in advance, analyze the temperature response curves of the corrosion area and the non-corrosion area of the early-stage corrosion sample, and select the time frame interval with the largest relative tangent slope in the temperature response curve;

[0013] S2. Based on the Lucas-Kanade optical flow method, optical flow is calculated for the selected time frame interval, and an optical flow amplitude matrix is constructed based on the calculation results. Based on the optical flow amplitude matrix, an optical flow amplitude image sequence of consecutive frames is obtained, and the optical flow amplitude difference is calculated based on the optical flow amplitude image sequence to evaluate the corrosion degree of the early corrosion sample.

[0014] S3. Reconstruct the optical flow amplitude image sequence of continuous frames to obtain a reconstruction matrix, and use the principal component analysis method to extract features of the reconstruction matrix to obtain principal component features. By performing weighted reconstruction on the principal component features, the heat flow features of the corrosion area are obtained.

[0015] Preferably, the step of analyzing the temperature response curves of the corroded area and the non-corroded area of the early corrosion sample based on the infrared thermal image sequence acquired in advance, and selecting the time frame interval with the largest relative tangent slope in the temperature response curve comprises the following steps:

[0016] S11. Acquire an infrared thermal image sequence of an early-stage corrosion sample pre-detected using eddy current pulse thermography technology, and record the temperature changes of the corroded area and the non-corroded area of the early-stage corrosion sample during the heating and cooling processes;

[0017] S12. Based on the temperature changes of the corrosion area and the non-corrosion area of the early corrosion sample during the heating and cooling processes, the temperature change curves of the corrosion area and the non-corrosion area are analyzed, and the time frame interval with the largest relative tangent slope is selected.

[0018] Preferably, the method of performing optical flow calculation on a selected time frame interval based on the Lucas-Kanade optical flow method and constructing an optical flow amplitude matrix according to the calculation results; obtaining an optical flow amplitude image sequence of consecutive frames according to the optical flow amplitude matrix, and calculating the optical flow amplitude difference based on the optical flow amplitude image sequence to evaluate the corrosion degree of the early corrosion sample includes the following steps:

[0019] S21. Calculate pixel displacement of adjacent frames within the selected time frame interval using the Lucas-Kanade optical flow method, and obtain an optical flow component for each pixel based on the calculation result.

[0020] S22. Calculate the optical flow amplitude according to the optical flow component of each pixel using the optical flow amplitude calculation formula;

[0021] S23, normalizing the optical flow amplitude using a normalization method to obtain a normalized optical flow amplitude, and constructing an optical flow amplitude matrix based on the normalized optical flow amplitude;

[0022] S24, converting the normalized optical flow amplitude into a color image using a Jet color mapping scheme to obtain a sequence of optical flow amplitude images of consecutive frames;

[0023] S25. Based on the normalized optical flow amplitude, the optical flow amplitudes of the corroded area and the non-corroded area are selected, and the average optical flow amplitudes of the corroded area and the non-corroded area are calculated respectively. The corrosion degree of the early corrosion sample is evaluated according to the difference between the average optical flow amplitudes of the corroded area and the non-corroded area.

[0024] Preferably, performing pixel displacement calculation on adjacent frames within the selected time frame interval by the Lucas-Kanade optical flow method and obtaining the optical flow component of each pixel according to the calculation result comprises the following steps:

[0025] S211, according to the pixel grayscale value at each position within the selected time frame interval, by setting the grayscale value change of the pixel within the preset interval to be negligible, a grayscale value conservation formula is obtained;

[0026] S212. Taylor expansion of the grayscale conservation formula is performed to obtain a linear approximation formula including pixel displacement; and the optical flow constraint equation is obtained in combination with the grayscale conservation formula.

[0027] S213. Construct an overdetermined system of equations based on the optical flow constraint equation, and use the least squares method to solve the overdetermined system of equations to obtain the optical flow component of each pixel.

[0028] Preferably, the optical flow amplitude calculation formula is:

[0029]

[0030] Where F represents the optical flow amplitude, u represents the optical flow component of the pixel in the x-axis direction, and v represents the optical flow component of the pixel in the y-axis direction.

[0031] Preferably, the calculation formula for normalizing the optical flow amplitude using the normalization method is:

[0032]

[0033] Where, F norm Represents the normalized optical flow amplitude, F min Indicates the minimum value of the optical flow amplitude, F max Indicates the maximum value of the optical flow amplitude.

[0034] Preferably, the process of reconstructing a sequence of optical flow amplitude images of consecutive frames to obtain a reconstructed matrix, extracting features from the reconstructed matrix using a principal component analysis method to obtain principal component features, and performing weighted reconstruction on the principal component features to obtain heat flow features of the corrosion area comprises the following steps:

[0035] S31, reconstructing the optical flow amplitude image sequence of consecutive frames into a two-dimensional matrix, and determining the mean matrix of the row vectors of the two-dimensional matrix;

[0036] S32. Constructing a covariance matrix of the two-dimensional matrix according to the mean matrix of the row vectors of the two-dimensional matrix and the two-dimensional matrix;

[0037] S33, decomposing the covariance matrix by singular value decomposition method to obtain an eigenvector matrix and an eigenvalue matrix;

[0038] S34. Sort the eigenvectors according to the eigenvalue matrix, select several principal components, and use the several principal components to perform image weighted reconstruction to obtain the heat flow characteristics of the corrosion area.

[0039] Preferably, the calculation formula for constructing the covariance matrix of the two-dimensional matrix based on the mean matrix of the row vectors of the two-dimensional matrix and the two-dimensional matrix is:

[0040] C=(RR mean )(RR mean ) T

[0041] Where C represents the covariance matrix, R mean is the mean matrix of the row vectors of the two-dimensional matrix R, where R represents a two-dimensional matrix and T represents a transposed matrix.

[0042] Preferably, the calculation formula for decomposing the covariance matrix by singular value decomposition is:

[0043] C=QΛQ T

[0044] Where C represents the covariance matrix, Q is the matrix consisting of the eigenvectors of the covariance matrix, Λ is the corresponding diagonal eigenvalue matrix, and T represents the transposed matrix.

[0045] Preferably, the heat flow characteristics of the corrosion area obtained by weighted image reconstruction using several principal components are calculated as follows:

[0046]

[0047] Where, represents the reconstructed data matrix, represents the reconstruction vector, μ represents the mean vector of the data, y irepresents the coefficient of the i-th principal component, w i represents the eigenvector corresponding to the i-th principal component, k represents the number of principal components, N x Indicates the width of the image, N y Indicates the height of the image.

[0048] Compared with the existing technology, the present invention provides an early corrosion detection method based on eddy current thermal imaging and optical flow-PCA, which has the following beneficial effects:

[0049] (1) The present invention uses the LK optical flow method to capture the dynamic motion information of pixel points in the infrared image sequence, and converts the spatiotemporal changes of the temperature field into optical flow amplitude and direction characteristics, thereby making up for the deficiency of the FFT method in ignoring the dynamic evolution characteristics of temperature. This process can accurately reflect the subtle differences in heat flow direction and intensity between the corroded area and the non-corroded area, greatly improving the sensitivity of the early corrosion signal.

[0050] (2) The present invention calculates the temperature difference between the corroded area and the non-corroded area and selects the key frame with the most significant change in optical flow value for analysis. This adaptive frame selection mechanism effectively reduces the interference of background noise and heating unevenness and improves the stability and accuracy of feature extraction.

[0051] (3) The present invention combines PCA dimensionality reduction and reconstruction, extracts the main components of the optical flow field, and performs weighted reconstruction on the corrosion image, highlighting the heat flow characteristics of the corrosion area while suppressing the redundant information of the background area, significantly enhancing the contrast and boundary clarity between the corrosion area and the background area. This method avoids the subjectivity of FFT frequency selection and ensures the objectivity and consistency of the reconstruction results.

[0052] (4) The present invention combines the optical flow method with PCA, breaking through the limitations of existing technologies in dynamic feature capture, noise resistance and visualization effects, and can achieve accurate positioning of early corrosion areas and corrosion degree assessment with high sensitivity and low noise. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0054] Figure 1 is a flow chart of an early corrosion detection method based on eddy current thermal imaging and optical flow-PCA according to an embodiment of the present invention;

[0055] Figure 21 is a structural block diagram of a pulsed eddy current thermal imaging system in an early corrosion detection method based on eddy current thermal imaging and optical flow-PCA according to an embodiment of the present invention;

[0056] Figure 3 is an amplitude / phase image of a specific frequency in the early corrosion detection method based on eddy current thermal imaging and optical flow-PCA according to an embodiment of the present invention;

[0057] Figure 4 is a thermal response diagram of different regions in the early corrosion detection method based on eddy current thermal imaging and optical flow-PCA according to an embodiment of the present invention;

[0058] Figure 5 is a temperature curve diagram of different pixel points on the sample surface in the early corrosion detection method based on eddy current thermal imaging and optical flow-PCA according to an embodiment of the present invention;

[0059] Figure 6 is a thermal difference map between a non-corroded area and a corroded area in an early corrosion detection method based on eddy current thermal imaging and optical flow-PCA according to an embodiment of the present invention;

[0060] Figure 7 A flowchart of extracting dynamic heat flow features from infrared image sequences using the Lucas-Kanade optical flow method in an early corrosion detection method based on eddy current thermography and optical flow-PCA according to an embodiment of the present invention;

[0061] Figure 8 is a flow chart of a PCA feature extraction process in an early corrosion detection method based on eddy current thermal imaging and optical flow-PCA according to an embodiment of the present invention;

[0062] Figure 9 This is the best effect image of the corrosion area visualization enhancement on the 49th day in the early corrosion detection method based on eddy current thermal imaging and optical flow-PCA according to an embodiment of the present invention;

[0063] Figure 10 This is the best effect diagram in the visualization enhancement of the corrosion area on the 66th day in the early corrosion detection method based on eddy current thermal imaging and optical flow-PCA according to an embodiment of the present invention. DETAILED DESCRIPTION

[0064] To further illustrate each embodiment, the present invention provides drawings, which are part of the disclosure of the present invention. They are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. By referring to these contents, ordinary technicians in this field should be able to understand other possible implementation methods and advantages of the present invention. The components in the figures are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0065] According to an embodiment of the present invention, an early corrosion detection method based on eddy current thermal imaging and optical flow-PCA is provided.

[0066] The present invention adopts pulsed eddy current thermography (ECPT, Eddy Current Pulsed Thermalgrphy) technology, and the system structure is as follows Figure 2 As shown in Figure 2, a high-frequency pulsed current is applied to the sample surface via an induction coil, inducing eddy currents and generating Joule heating within the material, thereby heating the sample surface. After the heating process is complete, the sample cools in a natural environment. The dynamic changes in its surface temperature are recorded in real time by an infrared thermal imager as a thermal image sequence, which is then transmitted to a computer for subsequent processing. The computer applies specific image processing algorithms to extract parameters related to corrosion characteristics, thereby reconstructing images of the corroded area and quantitatively assessing the corrosion extent.

[0067] Currently, the main algorithm for corrosion detection using pulsed eddy current thermography technology is the fast Fourier transform method.

[0068] The steps of the fast Fourier transform method can be divided into:

[0069] ①Fourier transform;

[0070] The infrared image sequence is regarded as a sampling sequence of temperature changes over time. The temperature data of each pixel in the image is subjected to fast Fourier transform (FFT) and converted to the frequency domain for analysis.

[0071]

[0072] Where, T k Indicates the temperature value of a pixel in the infrared image at the kth frame, is a complex exponential factor, a rotation factor in Fourier transform, used to convert the time domain signal to the frequency domain, N represents the number of frames in the image sequence, F n Represents the complex frequency component after Fourier transform, Re n and Im n Respectively represent F n Get the real and imaginary parts.

[0073] ②Calculate the amplitude-frequency characteristics and phase-frequency characteristics;

[0074] The Fourier transform result F n Analyze and obtain the amplitude-frequency characteristics and phase-frequency characteristics in the frequency domain:

[0075]

[0076] Where A n Represents the amplitude-frequency characteristics of the pixel temperature sequence, that is, the amplitude of different frequency components; φn It represents the phase-frequency characteristic, that is, the phase information of the frequency component.

[0077] ③Draw the amplitude / phase image of a specific frequency;

[0078] Based on the calculated amplitude-frequency characteristics and phase-frequency characteristics, a specific frequency component is selected and its amplitude image and phase image are plotted to characterize the distribution range of the corrosion area, such as Figure 3 As shown, 3Hz corresponds to the amplitude image, (a) corrosion for 21 days, (b) corrosion for 35 days; 3Hz corresponds to the phase image, (c) corrosion for 21 days, (d) corrosion for 35 days.

[0079] Amplitude image: used to highlight the thermal response intensity of the corrosion area and reflect the significance of the corrosion characteristics.

[0080] Phase image: used to distinguish the thermal diffusion delay between corroded and non-corroded areas, further confirming the location and extent of corrosion.

[0081] To address the shortcomings of existing fast Fourier transform (FFT) methods, this paper proposes an infrared thermal image processing method based on a combination of Lucas-Kanade optical flow and principal component analysis (LK-OF-PCA). This method comprehensively considers the dynamic changes in the temperature field and the directional characteristics of the heat flow on the material surface under external eddy current excitation. It uses optical flow to capture the dynamic information of heat flow motion between pixels, and combines PCA to reduce the dimensionality and reconstruct the feature data. The method proposes a highly sensitive and high-contrast method for reconstructing images of early corrosion areas and a corrosion severity assessment strategy.

[0082] The present invention will now be further described with reference to the accompanying drawings and specific embodiments. Figure 1 As shown, according to an embodiment of the present invention, the early corrosion detection method based on eddy current thermal imaging and optical flow-PCA includes the following steps:

[0083] S1. Based on a sequence of infrared thermal images of an early-stage corrosion sample acquired in advance, analyze the temperature response curves of the corrosion area and the non-corrosion area of the early-stage corrosion sample, and select the time frame interval with the largest relative tangent slope in the temperature response curve;

[0084] As a preferred embodiment, the method of analyzing the temperature response curves of the corroded area and the non-corroded area of the early corrosion sample based on the infrared thermal image sequence acquired in advance, and selecting the time frame interval with the largest relative tangent slope in the temperature response curve includes the following steps:

[0085] S11. Acquire an infrared thermal image sequence of an early-stage corrosion sample pre-detected using eddy current pulse thermography technology, and record the temperature changes of the corroded area and the non-corroded area of the early-stage corrosion sample during the heating and cooling processes;

[0086] S12. Based on the temperature changes of the corrosion area and the non-corrosion area of the early corrosion sample during the heating and cooling processes, the temperature change curves of the corrosion area and the non-corrosion area are analyzed, and the time frame interval with the largest relative tangent slope is selected.

[0087] S2. Based on the Lucas-Kanade optical flow method, optical flow is calculated for the selected time frame interval, and an optical flow amplitude matrix is constructed based on the calculation results. Based on the optical flow amplitude matrix, an optical flow amplitude image sequence of consecutive frames is obtained, and the optical flow amplitude difference is calculated based on the optical flow amplitude image sequence to evaluate the corrosion degree of the early corrosion sample.

[0088] As a preferred embodiment, the method of performing optical flow calculation on a selected time frame interval based on the Lucas-Kanade optical flow method and constructing an optical flow amplitude matrix based on the calculation results; obtaining an optical flow amplitude image sequence of consecutive frames based on the optical flow amplitude matrix, and calculating the optical flow amplitude difference based on the optical flow amplitude image sequence to evaluate the corrosion degree of early-stage corrosion samples includes the following steps:

[0089] S21, using the Lucas-Kanade optical flow method, performing pixel displacement calculation on adjacent frames within the selected time frame interval, and obtaining an optical flow component of each pixel based on the calculation result;

[0090] As a preferred embodiment, the Lucas-Kanade optical flow method is used to calculate the pixel displacement of adjacent frames within the selected time frame interval, and the optical flow component of each pixel is obtained according to the calculation result, which includes the following steps:

[0091] S211, according to the pixel grayscale value at each position within the selected time frame interval, by setting the grayscale value change of the pixel within the preset interval to be negligible, a grayscale value conservation formula is obtained;

[0092] S212. Taylor expansion of the grayscale conservation formula is performed to obtain a linear approximation formula including pixel displacement; and the optical flow constraint equation is obtained in combination with the grayscale conservation formula.

[0093] S213. Construct an overdetermined system of equations based on the optical flow constraint equation, and use the least squares method to solve the overdetermined system of equations to obtain the optical flow component of each pixel.

[0094] S22. Calculate the optical flow amplitude according to the optical flow component of each pixel using the optical flow amplitude calculation formula;

[0095] As a preferred embodiment, the optical flow amplitude calculation formula is:

[0096]

[0097] Where F represents the optical flow amplitude, u represents the optical flow component of the pixel in the x-axis direction, and v represents the optical flow component of the pixel in the y-axis direction.

[0098] S23, normalizing the optical flow amplitude using a normalization method to obtain a normalized optical flow amplitude, and constructing an optical flow amplitude matrix based on the normalized optical flow amplitude;

[0099] As a preferred embodiment, the calculation formula for normalizing the optical flow amplitude using the normalization method is:

[0100]

[0101] Where, F norm Represents the normalized optical flow amplitude, F min Indicates the minimum value of the optical flow amplitude, F max Indicates the maximum value of the optical flow amplitude.

[0102] S24, converting the normalized optical flow amplitude into a color image using a Jet color mapping scheme to obtain a sequence of optical flow amplitude images of consecutive frames;

[0103] S25. Based on the normalized optical flow amplitude, the optical flow amplitudes of the corroded area and the non-corroded area are selected, and the average optical flow amplitudes of the corroded area and the non-corroded area are calculated respectively. The corrosion degree of the early corrosion sample is evaluated according to the difference between the average optical flow amplitudes of the corroded area and the non-corroded area.

[0104] S3. Reconstruct the optical flow amplitude image sequence of continuous frames to obtain a reconstruction matrix, and use the principal component analysis method to extract features of the reconstruction matrix to obtain principal component features. By performing weighted reconstruction on the principal component features, the heat flow features of the corrosion area are obtained.

[0105] As a preferred embodiment, the method of reconstructing a sequence of optical flow amplitude images of consecutive frames to obtain a reconstructed matrix, extracting features from the reconstructed matrix using a principal component analysis method to obtain principal component features, and performing weighted reconstruction on the principal component features to obtain heat flow features of the corrosion area includes the following steps:

[0106] S31, reconstructing the optical flow amplitude image sequence of consecutive frames into a two-dimensional matrix, and determining the mean matrix of the row vectors of the two-dimensional matrix;

[0107] S32. Constructing a covariance matrix of the two-dimensional matrix according to the mean matrix of the row vectors of the two-dimensional matrix and the two-dimensional matrix;

[0108] S33, decomposing the covariance matrix by singular value decomposition method to obtain an eigenvector matrix and an eigenvalue matrix;

[0109] S34. Sort the eigenvectors according to the eigenvalue matrix, select several principal components, and use the several principal components to perform image weighted reconstruction to obtain the heat flow characteristics of the corrosion area.

[0110] As a preferred embodiment, the calculation formula for constructing the covariance matrix of the two-dimensional matrix based on the mean matrix of the row vectors of the two-dimensional matrix and the two-dimensional matrix is:

[0111] C=(RR mean )(RR mean ) T

[0112] Where C represents the covariance matrix, R mean is the mean matrix of the row vectors of the two-dimensional matrix R, where R represents a two-dimensional matrix and T represents a transposed matrix.

[0113] As a preferred embodiment, the calculation formula for decomposing the covariance matrix by singular value decomposition is:

[0114] C=QΛQ T

[0115] Where C represents the covariance matrix, Q is the matrix consisting of the eigenvectors of the covariance matrix, Λ is the corresponding diagonal eigenvalue matrix, and T represents the transposed matrix.

[0116] As a preferred embodiment, the heat flow characteristics of the corrosion area are calculated by using several principal components to perform weighted image reconstruction:

[0117]

[0118] Where, represents the reconstructed data matrix, represents the reconstruction vector, μ represents the mean vector of the data, y i represents the coefficient of the i-th principal component, w i represents the eigenvector corresponding to the i-th principal component, k represents the number of principal components, N x Indicates the width of the image, N y Indicates the height of the image.

[0119] In order to better understand the above technical solutions of the present invention, the above technical solutions of the present invention are further described below through specific implementation methods.

[0120] First, from the perspective of ECPT technology principles, a function generator under computer control outputs a control signal to enable the infrared thermal imager and induction heater to work together. The induction heater receives the signal from the function generator, generates a high-frequency alternating current, and transmits it to the induction coil. Due to the electromagnetic induction effect, eddy currents are generated in the sample. According to Joule's law, when eddy currents pass through conductive materials, due to the material's resistance, part of the electrical energy is converted into heat energy, specifically Joule heat Q. The generation process can be expressed by the Joule heat generation formula, which is:

[0121]

[0122] Where, J s represents the eddy current density, E represents the electric field strength, and σ represents the electrical conductivity of the material. The heat generated by the eddy current will continue to diffuse inside the sample. This heat diffusion process can be described by the heat diffusion formula, which is:

[0123]

[0124] Where z represents the depth from the surface, T(z,t) represents the temperature at time t at position z, ρ is the material density, and C p is the heat capacity, and k is the thermal conductivity of the material. For the early corrosion area, the eddy current density distribution in the sample will undergo non-uniform changes. Since the local corrosion of the material leads to changes in the electrical conductivity and thermal conductivity, eddy currents are more likely to concentrate in the corrosion area. The resistance of the corrosion area increases, and more Joule heat is generated, causing the temperature rise rate in the corrosion area to be significantly faster than that in the uncorroded area, thereby forming a high-temperature area in the infrared thermal image. Subsequently, the induction coil stops heating and the sample enters the natural cooling stage. During this process, the thermal diffusion in the corrosion area is affected, and the material structure inhomogeneity caused by local corrosion will hinder the conduction of heat, resulting in a difference in the temperature reduction rate in the corrosion area and the uncorroded area, which ultimately manifests as abnormal temperature distribution characteristics in the infrared thermal image.

[0125] The infrared thermal imager can capture the temperature difference between the early corrosion area and the non-corroded area and store it in the computer in the form of a thermal image sequence for subsequent feature extraction and corrosion degree analysis.

[0126] According to the above technical principles, since the temperature response curve of the early corrosion area will show obvious dynamic changes during the heating and cooling stages, in order to more effectively capture this difference, the present invention selects the time interval with the most significant temperature change as the solution window.

[0127] In order to select the frame with the largest relative temperature change difference to optimize the optical flow estimation effect, such as Figure 4-5 As shown in Figure 2, the temperature difference between area A and area B is calculated and obtained as Figure 6The curve shown in the figure shows that during the heating phase, the maximum temperature difference between areas A and B reaches 2.9°C, then rapidly decreases and remains stable at 0.78°C during the cooling phase. Therefore, based on the significance of this temperature difference, the study selected time period II (10-60 frames) as the optimal time period for optical flow analysis, as this is when the temperature change is most pronounced and the optical flow algorithm is most effective. This selection mechanism effectively reduces interference from background thermal reflections and random noise, minimizing the effects of noise and heating unevenness.

[0128] Specifically, to characterize the temperature change characteristics of each point in the infrared image, such as Figure 7 As shown, the present invention chooses to use the Lucas-Kanade optical flow method to extract dynamic heat flow features in infrared image sequences. Specifically, optical flow refers to the velocity field formed by the change of pixel position of a moving object on the observation plane in an image sequence. By analyzing the change of pixel brightness in the image over time and the correlation between adjacent frames, the optical flow method can find the correspondence between the previous frame and the current frame, thereby calculating the motion information of the object. In a thermal image sequence, the temperature distribution is closely related to the thermal diffusion and heat flow inside the material, and this thermal flow characteristic can be characterized by the pixel motion field (i.e., the optical flow field). Therefore, the optical flow can be described as the heat flow movement between two adjacent frames of thermal images, and the change law of the heat flow distribution inside the sample can be analyzed by calculating the optical flow. Based on optical flow analysis, the visualization and quantification of the heat flow distribution in the spatial and temporal dimensions can be achieved, which can then be used to reveal the damage characteristics and thermal response behavior inside the material.

[0129] In the LK optical flow method, assume that at time t, the pixel at position (x, y) has a grayscale value I(x, y, t). After a time interval Δt, the pixel undergoes a displacement (Δx, Δy) in the next frame. If the change in the pixel's grayscale value during this process is negligible, the grayscale conservation formula can be obtained. Taylor expansion of the grayscale conservation formula yields a linear approximation formula that includes pixel displacement. The grayscale conservation formula and the linear approximation formula that includes pixel displacement are:

[0130] I(x,y,t)=I(x+Δx,y+Δy,t+Δt)

[0131]

[0132] After ignoring high-order infinitesimals, the most basic optical flow constraint equation can be derived from the grayscale conservation formula and the linear approximation formula including pixel displacement:

[0133]

[0134] make The most basic optical flow constraint equation can be expressed as follows:

[0135] I x u+I y v+I t =0

[0136] Where u and v represent the moving speed in two directions (x-axis and y-axis), I x , I y , I t Represents the partial derivative (gradient) of brightness along the three axes. Because it involves two unknowns, u and v, additional constraints on the optical flow are required. In the Lucas-Kanade optical flow method, the optical flow is assumed to be constant in the neighborhood of a pixel, and the optical flow is solved by minimizing the error in the local area.

[0137] In a window containing N pixels (for example, an n×n window), the following overdetermined system of equations can be constructed:

[0138]

[0139] Where, I xi , I yi , I ti is the spatial and temporal gradient of the i-th pixel, and the system of equations is expressed in matrix form:

[0140]

[0141] Where A is an N×2 matrix, b is an N×1 vector, where I ti Represents the temporal gradient of the i-th pixel, representing the temporal gradient change of the pixel brightness in the image sequence. Using the least squares method, the goal is to minimize the error function E(u,v):

[0142]

[0143] Taking partial derivatives of u and v and setting the error function to zero, we get the equation:

[0144]

[0145] The optical flow component can be solved:

[0146]

[0147] After the calculation is completed, the optical flow components u(x,y) and v(x,y) of each pixel can be obtained.

[0148] The optical flow in a thermal image can be used as a mapping of heat flow, reflecting the direction and speed of heat diffusion. The optical flow amplitude F can represent the intensity of the heat flow movement, and the optical flow amplitude can be calculated using the optical flow amplitude calculation formula.

[0149] The formula for calculating the optical flow amplitude is:

[0150]

[0151] Where F represents the optical flow amplitude, u represents the optical flow component of the pixel in the x-axis direction, and v represents the optical flow component of the pixel in the y-axis direction.

[0152] In order to map the optical flow amplitude into the range of the visualized image, the optical flow amplitude is normalized using the normalization method to obtain the normalized optical flow amplitude. The calculation formula for normalizing the optical flow amplitude using the normalization method is:

[0153]

[0154] Where, F norm Represents the normalized optical flow amplitude, F min Indicates the minimum value of the optical flow amplitude, F max It represents the maximum value of the optical flow amplitude. Normalization maps the amplitude to the interval [0, 1]. The normalized optical flow amplitude is then converted into a color image through the Jet color mapping scheme to make the intensity distribution of the optical flow clearer. The present invention uses the Lucas-Kanade optical flow method to capture the pixel displacement between adjacent frames, quantitatively characterizes the dynamic heat flow change characteristics between the corroded area and the non-corroded area, and effectively improves the sensitivity to the tiny heat flow characteristics of the early corrosion area.

[0155] Furthermore, in order to quantitatively evaluate the degree of early corrosion, the optical flow amplitudes of the corroded area and the non-corroded area were selected, and the average optical flow amplitudes of the corroded area and the non-corroded area were calculated respectively. and The corrosion degree of the early corrosion samples was evaluated based on the difference in the average optical flow amplitude between the corroded area and the non-corroded area.

[0156]

[0157] Where ΔF represents the difference between the average optical flow amplitudes of the corroded area and the non-corroded area.

[0158] By calculating the average optical flow amplitude of the early corrosion area and the non-corrosion area through the optical flow method and obtaining the difference, the dynamic heat flow anomaly characteristics caused by the initial changes in physical properties can be effectively captured, thereby achieving a quantitative assessment of the degree of early corrosion.

[0159] The physical properties of materials undergo subtle but significant changes during the early stages of corrosion, directly impacting the conduction characteristics of heat flow within the material. Specifically, during the early stages of corrosion, the thermal conductivity, resistivity, and heat capacity of the corroded area gradually change. This is due to the formation of early corrosion products (such as iron oxide and ferrous oxide) on the material surface, leading to localized inhomogeneities in the material's microstructure and composition. The thermal conductivity of the early corrosion area begins to decrease slightly because these corrosion products have lower thermal conductivity than the bulk metal, locally hindering the diffusion of heat flow. Simultaneously, the resistivity begins to increase, and the insulating or semiconducting properties of the corrosion products lead to more pronounced localized Joule heating caused by eddy currents. Although these initial physical property changes are relatively subtle, they still cause slight delays and retardations in the heat flow conduction process, manifesting as a slight slowing of heat flow diffusion and a localized increase in temperature. This subtle heat flow anomaly causes the optical flow amplitude in the early corrosion area to be slightly higher than that in the non-corroded area, creating a detectable characteristic difference that provides a scientific basis and technical support for the highly sensitive identification and quantitative assessment of early corrosion.

[0160] Finally, in order to extract the corrosion information from the optical flow magnitude image sequence, PCA is applied to the optical flow magnitude, which involves mapping the data of the entire physical process from a high-dimensional space to a low-dimensional space, with the goal of maximizing the variance of the projection data. The optical flow magnitude sequence can be represented as a three-dimensional matrix n x ×n y × N. In order to obtain a two-dimensional matrix that can be processed by PCA, the optical flow amplitude sequence is reconstructed to generate a new matrix R, such as Figure 8 As shown. The new matrix R contains N rows and n x ×n y Columns, where each row corresponds to a frame of data in the optical flow amplitude matrix. PCA needs to reconstruct the covariance matrix of the matrix R, which can be calculated by the following formula:

[0161] C=(RR mean )(RR mean ) T

[0162] Where R mean Representing the mean matrix of the row vectors of the matrix R, the covariance matrix C can be decomposed into:

[0163] C=QΛQ T

[0164] Where Q represents the matrix composed of the eigenvectors of the covariance matrix, and Λ represents the corresponding diagonal eigenvalue matrix. These eigenvectors are linearly independent, and each eigenvector corresponds to an eigenvalue. The size of the eigenvalue reflects the degree of variance of the original data in the direction of the eigenvector. By sorting the eigenvalues in descending order, the first k principal components corresponding to the main features can be extracted. The principal component M of PCA can be expressed as:

[0165] M=Q T R

[0166] In the above formula of principal component M, M represents the data matrix after dimensionality reduction, Q T Represents the transposed eigenvector matrix, and R represents the original data matrix. The data matrix M only contains the scores of the first k principal components. In order to reconstruct the image, the first k principal components are used to reconstruct and reshape it into a two-dimensional image matrix

[0167]

[0168] Where, Represents the reconstructed data matrix, which has the same size as the original matrix M. represents the reconstruction vector, which is the result of adding the weighted first k principal components to the mean, μ represents the mean vector of the data, which is the average value of all the original data in each dimension, y i represents the coefficient of the i-th principal component, that is, the projection size of the original data on the i-th principal component, w i Represents the eigenvector corresponding to the i-th principal component, that is, the direction of the principal component, N x Indicates the width of the image, N y Indicates the height of the image.

[0169] like Figure 8 As shown in the figure, the first k principal components extracted by PCA are weighted reconstructed to further suppress background information, improve the contrast between the corroded area and the non-corroded area, and solve the problem of recognition difficulty caused by unclear local temperature difference.

[0170] This paper innovatively combines the LK optical flow method with PCA to characterize the dynamic temperature distribution characteristics of corrosion areas. Specifically, variations in the electrical conductivity, magnetic permeability, and thermal conductivity of the material lead to significant differences in the heat diffusion process between corroded and non-corroded areas, manifesting as uneven directionality and intensity of temperature changes in infrared images. The LK optical flow method captures these dynamic changes in the temperature field and extracts pixel motion information; PCA further reduces the dimensionality of the optical flow features and reconstructs them, highlighting the key heat flow distribution characteristics.

[0171] According to the above analysis, the corroded area and the non-corroded area show obvious differences in the optical flow characteristics and PCA principal component results. Through the LK-PCA calculation process, the present invention can effectively reconstruct the heat flow distribution characteristics of the corroded area, thereby accurately locating the early corrosion area.

[0172] In addition, optical flow estimation and feature dimensionality reduction methods based on other principles can replace some of the algorithm steps in the present invention, but the present invention has obvious advantages in terms of method innovation and pertinence. At present, the method of optical flow feature extraction is not limited to the Lucas-Kanade optical flow method, and other methods such as the Horn-Schunck optical flow method and the Farneback optical flow method can also be used; similarly, the dimensionality reduction and reconstruction steps can also be replaced by independent component analysis (ICA), singular value decomposition (SVD) or convolutional autoencoder (CAE) to perform feature extraction instead of PCA. However, the present invention uses a combination of optical flow amplitude and PCA to capture the dynamic heat flow direction and intensity differences caused by early corrosion. This method is groundbreaking, and no other alternatives have yet been found to achieve the same effect in early corrosion detection.

[0173] In addition, if Figure 9-10 As shown in the figure, the LK optical flow method is combined with PCA to process the corrosion detection image, extract the first four principal components, and generate a reconstructed image that only retains key information through weighted combination. Figure 9 and Figure 10 The best results of the early corrosion detection method based on eddy current thermal imaging and optical flow-PCA proposed in the present invention in the visualization enhancement of the corrosion area on the 49th and 66th days are respectively demonstrated. From the results shown in the figure, it can be observed that the proposed method is significantly better than the traditional method in defect visualization enhancement of thermal image sequences. The contrast between the corrosion area and the non-corrosion area is significantly improved, and the influence of the background area is greatly reduced. At the same time, due to the ability to accurately capture the subtle differences in heat flow direction and intensity between the corrosion area and the non-corrosion area, the outline of the corrosion area is better highlighted, making its boundary clearer. These results show that the proposed method has significant advantages in enhancing the visualization effect of early corrosion.

[0174] In summary, with the help of the above technical solutions of the present invention, the present invention uses the LK optical flow method to capture the dynamic motion information of pixel points in the infrared image sequence, and converts the spatiotemporal changes of the temperature field into optical flow amplitude and direction characteristics, thereby making up for the deficiency of the FFT method in ignoring the dynamic evolution characteristics of temperature. This process can accurately reflect the subtle differences in heat flow direction and intensity between the corrosion area and the non-corrosion area, greatly improving the sensitivity of the early corrosion signal. The present invention calculates the temperature difference between the corrosion area and the non-corrosion area and selects the key frame with the most significant change in optical flow value for analysis. This adaptive frame selection mechanism effectively reduces the interference of background noise and heating unevenness, and improves the stability and accuracy of feature extraction. The present invention combines PCA dimensionality reduction and reconstruction, extracts the main components of the optical flow field, and performs weighted reconstruction on the corrosion image, highlighting the heat flow characteristics of the corrosion area, while suppressing the redundant information of the background area, significantly enhancing the contrast and boundary clarity between the corrosion area and the background area. This method avoids the subjectivity of FFT frequency selection and ensures the objectivity and consistency of the reconstruction results. The present invention combines the optical flow method with PCA, breaking through the limitations of existing technologies in dynamic feature capture, noise resistance and visualization effects, and can achieve accurate positioning of early corrosion areas and corrosion degree assessment with high sensitivity and low noise.

[0175] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, optical storage, etc.) containing computer-usable program code.

[0176] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An early corrosion detection method based on eddy current thermal imaging and optical flow-PCA, characterized in that: The method comprises the following steps: S1. Based on a sequence of infrared thermal images of an early-stage corrosion sample acquired in advance, analyze the temperature response curves of the corrosion area and the non-corrosion area of the early-stage corrosion sample, and select the time frame interval with the largest relative tangent slope in the temperature response curve; S2. Based on the Lucas-Kanade optical flow method, optical flow is calculated for the selected time frame interval, and an optical flow amplitude matrix is constructed based on the calculation results. Based on the optical flow amplitude matrix, an optical flow amplitude image sequence of consecutive frames is obtained, and the optical flow amplitude difference is calculated based on the optical flow amplitude image sequence to evaluate the corrosion degree of the early corrosion sample. S3. Reconstruct the optical flow amplitude image sequence of continuous frames into a two-dimensional matrix, and determine the mean matrix of the row vectors of the two-dimensional matrix; construct the covariance matrix of the two-dimensional matrix based on the mean matrix of the row vectors of the two-dimensional matrix and the two-dimensional matrix; decompose the covariance matrix by singular value decomposition to obtain an eigenvector matrix and an eigenvalue matrix; sort the eigenvectors according to the eigenvalue matrix, select several principal components, and use the several principal components to perform image weighted reconstruction to obtain the heat flow characteristics of the corrosion area; The calculation formula for constructing the covariance matrix of the two-dimensional matrix based on the mean matrix of the row vectors of the two-dimensional matrix and the two-dimensional matrix is: C=(R-R mena )-R mean ) T Where C represents the covariance matrix, R mean is the mean matrix of the row vectors of the two-dimensional matrix R, where R represents a two-dimensional matrix and T represents a transposed matrix.

2. The early corrosion detection method based on eddy current thermal imaging and optical flow-PCA according to claim 1 is characterized in that: The method of analyzing the temperature response curves of the corroded area and the non-corroded area of the early corrosion sample based on the infrared thermal image sequence acquired in advance and selecting the time frame interval with the largest relative tangent slope in the temperature response curve comprises the following steps: S11. Acquire an infrared thermal image sequence of an early-stage corrosion sample pre-detected using eddy current pulse thermography technology, and record the temperature changes of the corroded area and the non-corroded area of the early-stage corrosion sample during the heating and cooling processes; S12. Based on the temperature changes of the corrosion area and the non-corrosion area of the early corrosion sample during the heating and cooling processes, the temperature change curves of the corrosion area and the non-corrosion area are analyzed, and the time frame interval with the largest relative tangent slope is selected.

3. The early corrosion detection method based on eddy current thermal imaging and optical flow-PCA according to claim 1 is characterized in that: The Lucas-Kanade optical flow method is used to calculate the optical flow of the selected time frame interval and construct an optical flow amplitude matrix based on the calculation results; According to the optical flow amplitude matrix, an optical flow amplitude image sequence of consecutive frames is obtained, and the optical flow amplitude difference is calculated based on the optical flow amplitude image sequence. The corrosion degree of the early corrosion sample is evaluated, which includes the following steps: S21. Calculate pixel displacement of adjacent frames within the selected time frame interval using the Lucas-Kanade optical flow method, and obtain an optical flow component for each pixel based on the calculation result. S22. Calculate the optical flow amplitude according to the optical flow component of each pixel using the optical flow amplitude calculation formula; S23, normalizing the optical flow amplitude using a normalization method to obtain a normalized optical flow amplitude, and constructing an optical flow amplitude matrix based on the normalized optical flow amplitude; S24, converting the normalized optical flow amplitude into a color image using a Jet color mapping scheme to obtain a sequence of optical flow amplitude images of consecutive frames; S25. Based on the normalized optical flow amplitude, the optical flow amplitudes of the corroded area and the non-corroded area are selected, and the average optical flow amplitudes of the corroded area and the non-corroded area are calculated respectively. The corrosion degree of the early corrosion sample is evaluated according to the difference between the average optical flow amplitudes of the corroded area and the non-corroded area.

4. The early corrosion detection method based on eddy current thermal imaging and optical flow-PCA according to claim 3 is characterized in that: The Lucas-Kanade optical flow method is used to calculate the pixel displacement of adjacent frames within the selected time frame interval, and the optical flow component of each pixel is obtained according to the calculation result, which includes the following steps: S211, according to the pixel grayscale value at each position within the selected time frame interval, by setting the grayscale value change of the pixel within the preset interval to be negligible, a grayscale value conservation formula is obtained; S212. Taylor expansion of the grayscale conservation formula is performed to obtain a linear approximation formula including pixel displacement; and the optical flow constraint equation is obtained in combination with the grayscale conservation formula. S213. Construct an overdetermined system of equations based on the optical flow constraint equation, and use the least squares method to solve the overdetermined system of equations to obtain the optical flow component of each pixel.

5. The early corrosion detection method based on eddy current thermal imaging and optical flow-PCA according to claim 3 is characterized in that: The optical flow amplitude calculation formula is: Where F represents the optical flow amplitude, u represents the optical flow component of the pixel in the x-axis direction, and v represents the optical flow component of the pixel in the y-axis direction.

6. The early corrosion detection method based on eddy current thermal imaging and optical flow-PCA according to claim 3 is characterized in that: The calculation formula for normalizing the optical flow amplitude using the normalization method is: Where, F norm Represents the normalized optical flow amplitude, F min Indicates the minimum value of the optical flow amplitude, F max Indicates the maximum value of the optical flow amplitude.

7. The early corrosion detection method based on eddy current thermal imaging and optical flow-PCA according to claim 1 is characterized in that: The calculation formula for decomposing the covariance matrix by singular value decomposition is: C=QΛQ T Where C represents the covariance matrix, Q is the matrix consisting of the eigenvectors of the covariance matrix, Λ is the corresponding diagonal eigenvalue matrix, and T represents the transposed matrix.

8. The early corrosion detection method based on eddy current thermal imaging and optical flow-PCA according to claim 7 is characterized in that: The calculation formula for the heat flow characteristics of the corrosion area obtained by performing image weighted reconstruction using several principal components is: Where, represents the reconstructed data matrix, represents the reconstruction vector, μ represents the mean vector of the data, y i represents the coefficient of the i-th principal component, w i represents the eigenvector corresponding to the i-th principal component, k represents the number of principal components, N x Indicates the width of the image, N y Indicates the height of the image.

Images