Corrugated board delamination detection method and system based on infrared thermal imaging

Abstract

The present application relates to the technical field of image processing, and discloses a corrugated board delamination detection method and system based on infrared thermal imaging, which comprises collecting a sequence of infrared thermal imaging images during the cooling stage after the corrugated board is subjected to thermal excitation; calculating the multi-directional spatial gradient and the time dimension temperature decay rate of each pixel point, fusing through a preset tensor voting model to generate a space-time response feature map; dividing the space-time response feature map into a plurality of overlapping image sub-blocks, calculating the global complexity index of each overlapping image sub-block; determining a local binary threshold, adaptively binarizing the space-time response feature map according to the local binary threshold to generate an initial delamination candidate mask; and screening the connected domains in the initial delamination candidate mask to determine the delamination area. Through deep fusion of multi-directional gradient information and temperature decay rate information, the present application can accurately distinguish between defect signals and noise, and improve the accuracy of the corrugated board delamination detection result.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and more specifically to a method and system for detecting delamination of corrugated cardboard based on infrared thermal imaging. Background Technology

[0002] The bonding quality of corrugated cardboard is a key factor determining the pressure resistance and protective function of packaging boxes. During production, "delamination" is a common internal bonding defect, manifested as ineffective bonding between the corrugated core paper and the linerboard or inner liner, resulting in internal cavities. Traditional delamination detection methods, such as relying on subjective experience through manual tapping and listening, or destructive sampling inspection, are no longer sufficient to meet the efficiency and quality requirements of modern production. In contrast, infrared thermal imaging, as an advanced non-destructive testing technology, demonstrates enormous application potential due to its significant advantages such as non-contact operation, fast imaging speed, and large detection area. Its basic detection principle lies in the fact that after a brief thermal excitation is applied to the corrugated cardboard, the heat dissipation rate of the delaminated area is significantly slower than that of the well-bonded area due to the higher thermal resistance of the air layer inside, thus exhibiting a localized "hot spot" phenomenon during cooling. By capturing and analyzing this transient temperature difference using an infrared thermal imager, the identification and location of internal delamination defects can be achieved.

[0003] However, infrared thermal imaging technology faces a key challenge when applied to corrugated cardboard inspection: the periodic flute structure of the cardboard itself creates strong, parallel-flute-shaped background textures in the thermal image. The thermal contrast generated by this structural texture is often comparable to, or even stronger than, the weak temperature difference signal caused by delamination defects. This directly results in an extremely low signal-to-noise ratio (SNR) for the defect signal, making it easily obscured by the strong background texture. Existing image processing methods, such as global thresholding or simple edge-detection-based segmentation algorithms, struggle to accurately distinguish between signal and noise when faced with non-uniform heating and complex texture interference, easily leading to numerous false positives and false negatives, resulting in low accuracy in corrugated cardboard delamination detection. Summary of the Invention

[0004] This invention provides a method and system for detecting delamination of corrugated cardboard based on infrared thermal imaging to solve the problem in the prior art that it is difficult to accurately distinguish between defect signals and noise, resulting in low accuracy of the detection results for delamination of corrugated cardboard.

[0005] In a first aspect, the method for detecting delamination of corrugated cardboard based on infrared thermal imaging of the present invention includes the following steps: During the cooling stage after thermal excitation of corrugated cardboard, a sequence of infrared thermal imaging images is acquired at a preset frame rate; the multi-directional spatial gradient and time-dimensional temperature decay rate of each pixel in the sequence of infrared thermal imaging images are calculated, and fused through a preset tensor voting model to generate a spatiotemporal response feature map. The spatiotemporal response feature map is divided into multiple overlapping image sub-blocks, and the mean value of the feature entropy of each overlapping image sub-block is calculated as the global complexity index. The local binarization threshold is determined based on the feature entropy of each overlapping image sub-block and the normalized distance between the center of the overlapping image sub-block and the centroid of the spatiotemporal response feature map. The spatiotemporal response feature map is adaptively binarized based on the local binarization threshold to generate an initial debonding candidate mask. The connected components in the initial debonding candidate mask are screened to determine the debonding region. The screening includes: determining the corrugation principal axis by performing principal component analysis on the spatiotemporal response feature map. Based on the global complexity index, the morphological constraint threshold is determined through a preset nonlinear mapping function; When a connected region satisfies that the angle between the major axis of the minimum circumscribed rectangle and the principal axis of the angular direction is less than a first angle threshold and the compactness is less than a morphological constraint threshold, the connected region is determined to be a degummed region.

[0006] Preferably, the calculation of the multi-directional spatial gradient and temporal temperature decay rate of each pixel in the sequence of infrared thermal imaging images includes: use The Sobel operator calculates the spatial gradient magnitude of each pixel in four directions: 0°, 45°, 90°, and 135°. A first-order linear regression was performed on the temperature value of each pixel over time, and the absolute value of the slope of the fitted line was taken as the temperature decay rate.

[0007] Preferably, dividing the spatiotemporal response feature map into multiple overlapping image sub-blocks includes: The spatiotemporal response feature map is divided into Overlapping image sub-blocks of pixels, with an overlap step of 8 pixels between them.

[0008] Preferably, the calculation of the mean value entropy of the feature values ​​of each overlapping image sub-block as a global complexity index includes: Calculate the gradient covariance matrix of all pixels within each overlapping image sub-block to obtain the eigenvalues. and , The eigenvalue entropy of overlapping image sub-blocks is calculated using the following formula. : ;in, ; The arithmetic mean of the eigenvalue entropies of all overlapping image sub-blocks is calculated as a global complexity metric.

[0009] Preferably, the local binarization threshold is calculated using the following formula: ; in, For local binarization threshold, The global mean of the spatiotemporal response feature map. The feature entropy of overlapping image sub-blocks. The normalized Euclidean distance from the center of the overlapping image sub-block to the centroid of the spatiotemporal response feature map is given by: and The preset non-negative modulation coefficient.

[0010] Preferably, the step of determining the corrugated principal axes by performing principal component analysis on the spatiotemporal response feature map includes: Select pixels in the spatiotemporal response feature map whose response values ​​are greater than the global mean, and construct a point set by the two-dimensional coordinates (x, y) of the pixels; Principal component analysis is performed on the point set to calculate the first eigenvector of the covariance matrix of the point set. The direction of the first eigenvector is the angular principal axis.

[0011] Preferably, the morphological constraint threshold is calculated using the following formula: ; in, For morphological constraint threshold, This serves as a preset lower bound for the morphological constraint threshold. Here, k is the preset upper bound for the morphological constraint threshold, k is the kurtosis of the function, and G is the global complexity index. The center point of the function.

[0012] Preferably, the first angle threshold is 15 degrees; the tightness is calculated using the following formula: ;in, Let A be the compactness, A be the area of ​​the connected region, and P be the perimeter of the connected region.

[0013] Preferably, the step of acquiring a sequence of infrared thermal imaging images at a preset frame rate during the cooling stage after thermal excitation of the corrugated cardboard includes: The corrugated cardboard surface is heated instantaneously and uniformly for 2 to 5 milliseconds using a halogen lamp. After heating stops, an infrared thermal imager is immediately activated to continuously acquire a sequence of infrared thermal images for 5 seconds at a frame rate of 50 frames per second.

[0014] Secondly, the corrugated cardboard delamination detection system based on infrared thermal imaging of the present invention includes a memory and a processor. The memory stores computer instructions, and when the processor executes the computer instructions, it implements the above-mentioned corrugated cardboard delamination detection method based on infrared thermal imaging.

[0015] The beneficial effects of this invention are as follows: By deeply fusing multi-directional gradient information in the spatial dimension and temperature decay rate information in the temporal dimension of infrared image sequences, this invention constructs a spatiotemporal response feature that enhances weak degumming signals while suppressing interference from inherent periodic stripe textures in corrugated structures, thereby improving the signal-to-noise ratio of defects. In the segmentation stage, the segmentation criteria are determined based on the information complexity and spatial location of different local regions in the feature map, reducing the inconsistency in overall image brightness and contrast caused by uneven thermal excitation, and enabling the extraction of candidate defects. Through a priori understanding of corrugated structure based on principal component analysis, combined with morphological indicators related to the global complexity of the image, a dual-constraint screening of candidate regions is performed, eliminating pseudo-defects formed by residual background textures and random noise, ensuring high accuracy and reliability of the detection results, and comprehensively solving the technical bottleneck of unstable identification of internal defects under strong background interference. Attached Figure Description

[0016] Figure 1 This is a schematic flowchart of the corrugated cardboard delamination detection method based on infrared thermal imaging provided in an embodiment of the present invention. Detailed Implementation

[0017] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0018] like Figure 1 As shown, an embodiment of the corrugated cardboard delamination detection method based on infrared thermal imaging provided by the present invention includes the following steps: S1. During the cooling stage after thermal excitation of the corrugated cardboard, a sequence of infrared thermal imaging images is acquired at a preset frame rate. The multi-directional spatial gradient and time-dimensional temperature decay rate of each pixel in the sequence of infrared thermal imaging images are calculated and fused through a preset tensor voting model to generate a spatiotemporal response feature map.

[0019] Specifically, a high-power halogen lamp was used to heat the surface of the corrugated cardboard instantaneously and uniformly for 2 to 5 milliseconds. After heating was stopped, an infrared thermal imager was immediately activated to continuously acquire a sequence of infrared thermal images for 5 seconds at a frame rate of 50 frames per second, for a total of 250 images.

[0020] For each frame of the image sequence, the gradient magnitude of each pixel is calculated using the Kirsch operator in eight directions. Then, along the time axis, the maximum gradient magnitude of each pixel in all frames across the eight directions is extracted, generating a maximum spatial gradient map in eight directions. Simultaneously, a first-order exponential decay fit is applied to the temperature data of each pixel over the time series to obtain a temperature decay rate map.

[0021] Next, the principal direction (i.e., the direction with the largest gradient magnitude) and spatial saliency (i.e., the maximum gradient magnitude) of a pixel are determined based on the maximum spatial gradient values ​​in eight directions. The temperature decay rate of this point is defined as the temporal saliency, and it is multiplied by the spatial saliency to obtain an initial saliency value that incorporates spatiotemporal information. Subsequently, a second-order rod-shaped tensor is constructed for each pixel, with its principal eigenvector aligned with the principal direction and its principal eigenvalue equal to the fused saliency value. Each tensor acts as a voter, voting for other pixels in its neighborhood, and the voting information is attenuated based on the consistency of distance and direction. After voting, the final tensor field is subjected to eigenvalue decomposition, and the spatiotemporal response feature map is finally generated by calculating the saliency metric of each pixel's tensor (i.e., the difference between the maximum and minimum eigenvalues).

[0022] In an optional embodiment, calculating the multi-directional spatial gradient and temporal temperature decay rate of each pixel in the sequence of infrared thermal imaging images includes: use The Sobel operator calculates the spatial gradient magnitude of each pixel in four directions: 0°, 45°, 90°, and 135°. A first-order linear regression was performed on the temperature value of each pixel over time, and the absolute value of the slope of the fitted line was taken as the temperature decay rate.

[0023] For example, taking any pixel in an image as an example, the pixel and its eight neighboring pixels are selected as the basis for the image. The Sobel operator used to calculate the gradient in the 0° direction is convolved with the temperature value of the region to obtain the spatial gradient magnitude of the pixel in the 0° direction. Similarly, the Sobel operators for the corresponding 45°, 90°, and 135° directions are used to calculate the spatial gradient magnitudes in the other three directions. In this way, each pixel will obtain a feature consisting of four gradient magnitudes to represent the degree of temperature change in different directions.

[0024] For the same pixel, its temperature values ​​across multiple consecutive frames are extracted to form a time series dataset. For example, in 10 consecutive frames, the temperature sequence for a pixel might be 50.5℃, 48.2℃, 46.1℃, 44.0℃, etc. These temperature values ​​are then subjected to a first-order linear regression analysis with their corresponding time points to find the best-fitting straight line that describes the temperature change over time. The slope of this fitted line reflects the rate of temperature change, and its absolute value is defined as the temperature decay rate of that pixel, characterizing how quickly the pixel dissipates heat.

[0025] S2, the spatiotemporal response feature map is divided into multiple overlapping image sub-blocks, and the mean value of the feature entropy of each overlapping image sub-block is calculated as the global complexity index; the local binarization threshold is determined based on the feature entropy of each overlapping image sub-block and the normalized distance between the center of the overlapping image sub-block and the centroid of the spatiotemporal response feature map, and the spatiotemporal response feature map is adaptively binarized based on the local binarization threshold to generate the initial debonding candidate mask.

[0026] Specifically, the generated spatiotemporal response feature map is divided into regions of size 16 pixels with a step size of 16 pixels. Overlapping image sub-blocks of pixels. For each sub-block, calculate the gradients of all pixels within it in the spatiotemporal response feature map, and construct a [database / feature map] based on these gradients. The covariance matrix is ​​then calculated. Next, eigenvalue decomposition is performed on this covariance matrix to obtain two eigenvalues. These two eigenvalues ​​are normalized so that their sum is 1, and then the eigenvalue entropy of the image sub-block is calculated using the Shannon entropy formula. Finally, the arithmetic mean of the eigenvalue entropies of all sub-blocks is taken, and the result is the global complexity index representing the overall texture complexity of the feature map.

[0027] Next, the centroid coordinates of the entire spatiotemporal response feature map are calculated. For each image sub-block, the Euclidean distance from its center point to the centroid is calculated, and this distance is normalized by dividing it by half the length of the feature map's diagonal, thus obtaining the normalized distance. The local binarization threshold for each image sub-block is a linear combination of a base threshold, the eigenvalue entropy of the sub-block, and the normalized distance. For any pixel in the feature map, its final threshold is obtained by bilinear interpolation of the local thresholds of all overlapping sub-blocks covering that pixel. Finally, the value of each pixel in the feature map is compared with its corresponding final threshold; pixels greater than the threshold are set to 1, otherwise to 0, thus generating the initial debonding candidate mask.

[0028] S3, the connected components in the initial debonding candidate mask are screened to determine the debonding region. The screening includes: determining the corrugated principal axis by performing principal component analysis on the spatiotemporal response feature map; determining the morphological constraint threshold by using a preset nonlinear mapping function based on the global complexity index; when a connected component satisfies that the angle between the major axis of the minimum bounding rectangle and the corrugated principal axis is less than the first angle threshold and the compactness is less than the morphological constraint threshold, the connected component is determined to be the debonding region.

[0029] Specifically, the position coordinates of all pixels in the spatiotemporal response feature map are used as the sample point set, and a weighted covariance matrix is ​​constructed using the pixel values ​​corresponding to each point as weights. Subsequently, eigenvalue decomposition is performed on this weighted covariance matrix, where the eigenvector corresponding to the largest eigenvalue is defined as the direction of the corrugated principal axis.

[0030] The preset nonlinear mapping function is an S-shaped decay function. The function's purpose is to: when the global complexity index is low, the output morphological constraint threshold is high, thus allowing relatively irregular regions to pass the screening; while when the global complexity index is high, the threshold will decrease rapidly, imposing stricter morphological constraints.

[0031] Connectivity analysis is performed on the generated initial debonding candidate mask to extract all independent candidate regions. For each connected region, its minimum bounding rectangle is first calculated, and the direction vector of its major axis is determined. Next, the angle between this major axis direction vector and the angular principal axis direction vector determined in the previous step is calculated. Simultaneously, the area and perimeter of the connected region are calculated, and its compactness is calculated using a formula. A first angle threshold of 20 degrees is set. Finally, a judgment is made: if the calculated angle of a connected region is less than 20 degrees, and its calculated compactness is also less than the morphological constraint threshold determined in the previous step, then the connected region is identified as a debonding defect region and marked in the result image; otherwise, the connected region is considered a false defect and is discarded.

[0032] In an optional embodiment, dividing the spatiotemporal response feature map into multiple overlapping image sub-blocks includes: The spatiotemporal response feature map is divided into Overlapping image sub-blocks of pixels, with an overlap step of 8 pixels between them.

[0033] The calculation of the mean value entropy of the feature values ​​of each overlapping image sub-block is used as a global complexity index, including: Calculate the gradient covariance matrix of all pixels within each overlapping image sub-block to obtain the eigenvalues. and , The eigenvalue entropy of overlapping image sub-blocks is calculated using the following formula. : ;in, ; The arithmetic mean of the eigenvalue entropies of all overlapping image sub-blocks is calculated as a global complexity metric.

[0034] For example, with a single Taking the spatiotemporal response feature map of a pixel as an example, we perform gridded segmentation on it. The operation starts from the top left corner of the image, first selecting one... The image sub-block is processed by pixels. After completing the calculation for this sub-block, the processing window shifts 8 pixels to the right to select the next pixel. The window is then divided into sub-blocks, each with an 8-pixel overlap with the previous sub-block. This process is repeated until an entire row is processed. The window then shifts down 8 pixels to begin processing a new row, continuing until the entire feature map has been traversed. This overlapping partitioning method ensures a smooth transition and complete utilization of the feature map information.

[0035] For each division For each pixel sub-block, the gradient features of its 256 internal pixels are first statistically analyzed, and a gradient is calculated based on these gradient features. The gradient covariance matrix. Find the two eigenvalues ​​of this matrix, for example... It equals 5.2. It equals 1.3. Calculate the eigenvalue entropy based on these two eigenvalues: first calculate the proportion of each eigenvalue to the sum of all eigenvalues, for example... Approximately 0.8 The value is approximately 0.2. Then, the eigenvalue entropy H is calculated to be approximately 0.72 according to the formula. This entropy value reflects the complexity and directionality of the texture within this local region. Finally, the eigenvalue entropies calculated from all image sub-blocks are summed, and then divided by the total number of sub-blocks to obtain the global complexity index.

[0036] In an optional embodiment, the local binarization threshold is calculated using the following formula: ; in, For local binarization threshold, The global mean of the spatiotemporal response feature map. The feature entropy of overlapping image sub-blocks. The normalized Euclidean distance from the center of the overlapping image sub-block to the centroid of the spatiotemporal response feature map is given by: and The preset non-negative modulation coefficient.

[0037] For example, to set an adaptive threshold for each image sub-block, the first step is to calculate the required basic parameters. Assuming the average pixel value of the entire spatiotemporal response feature map is 128, then... The value is 128. Simultaneously, the centroid coordinates of the entire feature map are calculated, for example, at point (130, 145). Furthermore, two modulation coefficients need to be pre-set, for example... It is 0.5. The coefficients are 0.3, and these two coefficients are used to adjust the weights of the influence of eigenvalue entropy and normalized distance on the final threshold, respectively.

[0038] Subsequently, calculations are performed for each image sub-block. Taking a sub-block with center coordinates (50, 60) as an example, assuming its eigenvalue entropy has already been calculated... The value is 0.65. First, the Euclidean distance between the center of the sub-block and the centroid of the feature map (130, 145) is calculated and normalized to obtain the normalized distance. The threshold is 0.4. Substituting the value into the given formula, the local binarization threshold for the sub-block is approximately 149.2. Repeating this process for all sub-blocks generates segmentation thresholds for different regions of the image.

[0039] In an optional embodiment, determining the corrugated principal axes by performing principal component analysis on the spatiotemporal response feature map includes: Select pixels in the spatiotemporal response feature map whose response values ​​are greater than the global mean, and construct a point set by the two-dimensional coordinates (x, y) of the pixels; Principal component analysis is performed on the point set to calculate the first eigenvector of the covariance matrix of the point set. The direction of the first eigenvector is the angular principal axis.

[0040] For example, this step aims to identify high-response regions in the feature map, which typically correspond to the flute peaks of corrugated cardboard. First, the average value of all pixel values ​​in the entire spatiotemporal response feature map is calculated; let's assume this average is 110. Next, each pixel in the map is iterated over, and if a pixel's response value is greater than 110, its two-dimensional coordinates (x, y) are recorded. For instance, if the response values ​​of pixels (25, 30), (26, 31), and (27, 32) are 120, 150, and 135 respectively, since they are all greater than 110, these three coordinate points are added to a set of points. After the iteration is complete, a set consisting of the coordinates of all high-response pixels is obtained.

[0041] Next, principal component analysis is performed on the two-dimensional coordinate point set obtained in the previous step to find the most predominant direction of the data point distribution. Specifically, first, the mean of all x-coordinates and the mean of all y-coordinates in the point set are calculated, and then a principal component analysis is constructed for these points. The covariance matrix. By solving for the eigenvalues ​​and eigenvectors of this covariance matrix, two mutually orthogonal principal component directions can be obtained. The eigenvector corresponding to the larger eigenvalue (i.e., the first eigenvector) represents the direction with the largest variance in the data point set. In a corrugated cardboard image, this direction corresponds to the overall orientation of the corrugations (i.e., the corrugation principal axis). For example, if the calculated first eigenvector is (0.707, 0.707), it represents a vector pointing from the origin to the northeast (i.e., a 45° angle), and this direction is determined as the corrugation principal axis.

[0042] In an optional embodiment, the morphological constraint threshold is calculated using the following formula: ; in, For morphological constraint threshold, This serves as a preset lower bound for the morphological constraint threshold. Here, k is the preset upper bound for the morphological constraint threshold, k is the kurtosis of the function, and G is the global complexity index. The center point of the function.

[0043] For example, in order to adaptively obtain a threshold for morphological judgment based on the overall texture complexity of the image (i.e., the global complexity index), several key parameters of the sigmoid function need to be pre-defined. For example, setting... It is 0.2. The value is set to 0.8, which ensures that the output threshold always remains between 0.2 and 0.8. Simultaneously, the center point of the function is set. The value is set to 0.5, representing a baseline or average level of complexity. Finally, the kurtosis coefficient k is set to 10, which determines the degree of abrupt change in the function curve near the center point. After setting the parameters, the global complexity index G calculated in the previous step is used as input. Assuming that for the current image, the calculated G value is 0.6, G and the preset parameters are substituted into the formula to calculate the morphological constraint threshold. It is approximately 0.638. The morphological constraint threshold will adaptively adjust as the global complexity changes.

[0044] In an optional embodiment, the first angle threshold is 15 degrees; the tightness is calculated using the following formula: ;in, Let A be the compactness, A be the area of ​​the connected region, and P be the perimeter of the connected region.

[0045] After binarization, several independent white regions, or connected components, will form on the image; these are potential defect areas. To identify true delamination defects, each connected component needs to be examined for two geometric features. The first criterion is angle, with a fixed preset threshold of 15 degrees. The second criterion is compactness, with a threshold dynamically calculated based on image complexity in the previous step, representing a morphological constraint threshold. .

[0046] When analyzing a connected component, the first step is to calculate the smallest rectangle that can perfectly enclose it. The major axis of this rectangle represents the overall orientation of the connected component; let's assume the calculated direction is 50 degrees. This direction is compared to the previously determined principal axis direction, for example, 45 degrees, and the angle between them is calculated to be 5 degrees. Next, the compactness of the connected component is calculated: its area A is measured as 200 pixels, and its perimeter P as 80 pixels. The compactness C calculated using the formula is approximately 0.39. Finally, a judgment is made: since the calculated angle of 5 degrees is less than the first angle threshold of 15 degrees, and the calculated compactness of 0.39 is also less than the morphological constraint threshold of 0.638, both conditions are met, and the connected component is determined to be a debonded region.

[0047] The implementation principle of the corrugated cardboard delamination detection method based on infrared thermal imaging in this invention is as follows: This invention deeply integrates multi-directional gradient information and temperature decay rate information of infrared image sequences. Through this fusion, a spatiotemporal response feature is constructed, which can significantly enhance weak delamination signals while effectively suppressing interference from the inherent periodic stripe texture of corrugated cardboard, thereby greatly improving the signal-to-noise ratio of defects. Secondly, in the segmentation stage, this invention adopts an adaptive strategy to dynamically determine the segmentation criteria based on the information complexity and spatial location of different local regions in the feature map. This approach reduces the problem of inconsistent overall image brightness and contrast caused by uneven thermal excitation, and achieves accurate extraction of candidate defects. Finally, in the screening stage, this invention introduces a dual constraint mechanism: on the one hand, the corrugated principal axes are extracted through principal component analysis and used as structural priors; on the other hand, a morphological index dynamically correlated with the global complexity of the image is combined. Using this dual constraint of direction and morphology, candidate regions are rigorously screened, which can effectively eliminate false defects formed by residual background texture and random noise, ultimately ensuring high accuracy and high reliability of the detection results.

[0048] An embodiment of the corrugated cardboard delamination detection system based on infrared thermal imaging provided by the present invention includes a memory and a processor. The memory stores computer instructions, and when the processor executes the computer instructions, it implements the corrugated cardboard delamination detection method based on infrared thermal imaging in the above embodiment.

[0049] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for detecting delamination of corrugated cardboard based on infrared thermal imaging, characterized in that, The process includes the following steps: during the cooling stage after thermal excitation of the corrugated cardboard, a sequence of infrared thermal imaging images is acquired at a preset frame rate; the multi-directional spatial gradient and time-dimensional temperature decay rate of each pixel in the sequence of infrared thermal imaging images are calculated, and the images are fused using a preset tensor voting model to generate a spatiotemporal response feature map. The spatiotemporal response feature map is divided into multiple overlapping image sub-blocks, and the mean value of the feature entropy of each overlapping image sub-block is calculated as the global complexity index. The local binarization threshold is determined based on the feature entropy of each overlapping image sub-block and the normalized distance between the center of the overlapping image sub-block and the centroid of the spatiotemporal response feature map. The spatiotemporal response feature map is adaptively binarized based on the local binarization threshold to generate an initial debonding candidate mask. The connected components in the initial debonding candidate mask are screened to determine the debonding region. The screening includes: determining the corrugation principal axis by performing principal component analysis on the spatiotemporal response feature map. Based on the global complexity index, the morphological constraint threshold is determined through a preset nonlinear mapping function, satisfying the following relationship: , For morphological constraint threshold, This serves as a preset lower bound for the morphological constraint threshold. This serves as a preset upper bound for the morphological constraint threshold. For the steepness of the function, As a global complexity metric, The center point of the function; A connected region is identified as a debonded region when the angle between the major axis of the minimum bounding rectangle and the principal axis of the ridge is less than the first angle threshold and the compactness is less than the morphological constraint threshold.

2. The method for detecting delamination of corrugated cardboard based on infrared thermal imaging according to claim 1, characterized in that, The calculation of the multi-directional spatial gradient and temporal temperature decay rate of each pixel in the sequence of infrared thermal imaging images includes: use The Sobel operator calculates the spatial gradient magnitude of each pixel in four directions: 0°, 45°, 90°, and 135°. A first-order linear regression was performed on the temperature value of each pixel over time, and the absolute value of the slope of the fitted line was taken as the temperature decay rate.

3. The method for detecting delamination of corrugated cardboard based on infrared thermal imaging according to claim 1, characterized in that, The process of dividing the spatiotemporal response feature map into multiple overlapping image sub-blocks includes: The spatiotemporal response feature map is divided into Overlapping image sub-blocks of pixels, with an overlap step of 8 pixels between them.

4. The method for detecting delamination of corrugated cardboard based on infrared thermal imaging according to claim 3, characterized in that, The calculation of the mean value entropy of the feature values ​​of each overlapping image sub-block is used as a global complexity index, including: Calculate the gradient covariance matrix of all pixels within each overlapping image sub-block to obtain the eigenvalues. and ; The eigenvalue entropy of overlapping image sub-blocks is calculated using the following formula. : ;in, ; The arithmetic mean of the eigenvalue entropies of all overlapping image sub-blocks is calculated as a global complexity metric.

5. The method for detecting delamination of corrugated cardboard based on infrared thermal imaging according to claim 1, characterized in that, The local binarization threshold is calculated using the following formula: ； in, For local binarization threshold, The global mean of the spatiotemporal response feature map. The feature entropy of overlapping image sub-blocks. The normalized Euclidean distance from the center of the overlapping image sub-block to the centroid of the spatiotemporal response feature map is given by: and The preset non-negative modulation coefficient.

6. The method for detecting delamination of corrugated cardboard based on infrared thermal imaging according to claim 1, characterized in that, The step of determining the corrugated principal axes by performing principal component analysis on the spatiotemporal response feature map includes: Select pixels in the spatiotemporal response feature map whose response values ​​are greater than the global mean, and then set the two-dimensional coordinates of these pixels. This constitutes a point set; Principal component analysis is performed on the point set to calculate the first eigenvector of the covariance matrix of the point set. The direction of the first eigenvector is the angular principal axis.

7. The method for detecting delamination of corrugated cardboard based on infrared thermal imaging according to claim 1, characterized in that, The first angle threshold is 15 degrees; the tightness is calculated using the following formula: ;in, For firmness, Let the area of ​​the connected region be . Let be the perimeter of the connected region.

8. The method for detecting delamination of corrugated cardboard based on infrared thermal imaging according to claim 1, characterized in that, The cooling stage following thermal excitation of the corrugated cardboard involves acquiring a sequence of infrared thermal imaging images at a preset frame rate, including: The corrugated cardboard surface is heated instantaneously and uniformly for 2 to 5 milliseconds using a halogen lamp. After heating stops, an infrared thermal imager is immediately activated to continuously acquire a sequence of infrared thermal images for 5 seconds at a frame rate of 50 frames per second.

9. A corrugated cardboard delamination detection system based on infrared thermal imaging, characterized in that, It includes a memory and a processor. The memory stores computer instructions. When the processor executes the computer instructions, it implements the corrugated cardboard delamination detection method based on infrared thermal imaging as described in any one of claims 1-8.

Images