A bridge delamination disease infrared imaging extraction method based on morphology reconstruction

Abstract

The application discloses a bridge delamination disease infrared imaging extraction method based on morphological reconstruction and belongs to the technical field of bridge detection. A weight attenuation function is used to perform regularization processing on gray scale morphological reconstruction, regional maximum value extraction is realized, and the mean value and the variation coefficient of the delamination boundary temperature gradient can be used for discrimination. The technology can process the frequently-occurring non-uniform background in practice, thereby eliminating the need for the existing method to infer the background. In this framework, the gradient statistics of the thermal image delamination boundary are used as an effective criterion for improved segmentation. Through the method, the delamination can be segmented, so that the detection personnel can quickly locate the delamination position and size of the bridge deck slab, and the bridge deck slab can be maintained later.

Description

Technical Field

[0001] The invention belongs to the technical field of bridge detection, and in particular relates to an infrared imaging extraction method for bridge delamination defects based on morphological reconstruction. Background Art

[0002] my country boasts a vast number of bridges, and ensuring their safety is a daunting task. Various factors, including steel corrosion, thermal effects, continuous freeze-thaw cycles, and shrinkage, can cause concrete bridge decks to develop defects such as cracks, delamination, spalling, and voids, posing safety hazards. Compared to more common spalling or potholes, delamination is highly insidious and can develop over a large area. If delamination is not detected and addressed promptly, internal delamination can continue to develop, eventually forming large, open spalling areas. This can seriously impact traffic safety, shorten the service life of the bridge deck, and potentially lead to accidents. In reality, delamination often occurs within the bridge deck itself, remaining unnoticed until significant damage has occurred. Therefore, a thorough understanding of the current condition of the bridge deck, including the presence of voids, delamination, and other defects, is crucial. Identifying bridge deck delamination at an early stage is crucial to prevent damage and providing appropriate repair options for potentially damaged decks. This is crucial for the safe operation of bridges.

[0003] Infrared thermography detection of shallow delaminations (less than 10.2 cm or 4.0 inches) in bridge decks has been reported in several studies. The detection principle is based on the temperature difference between delamination and non-delamination areas during a solar heating cycle. Although the principle is intuitive, thermal imaging quality is often affected by unfavorable factors such as image capture time, unclear temperature differences, and surface inhomogeneity. Therefore, accurately segmenting delaminations from raw thermal images remains a challenge. To address this challenge, several quantitative methods based on temperature differences, temperature gradients, and temperature density distributions have been developed to process thermal images. Despite some progress, these methods perform poorly under conditions of spatial temperature variation, often referred to as an inhomogeneous thermal background. This issue has been observed in experimental setups using inhomogeneous excitation heating sources as well as in natural outdoor environments. The underlying assumption of these methods requires a relatively uniform background to represent "intact" concrete areas, so that the highest temperature areas are associated with delaminations. However, this assumption is not always met. As a result, delamination areas represented by regional maxima are often missed. To address this issue, new delamination segmentation methods are needed to segment regional maxima against an inhomogeneous background.

[0004] Traditional delamination segmentation methods are primarily based on a bottom-up framework that requires a predefined reference temperature. Some researchers have employed hard thresholding of temperature values ​​or temperature variation percentiles, converting raw thermal images into binary images. Delamination is indicated by assuming that low temperatures or the lowest percentile represent intact areas. Due to the large temperature variations observed across the entire bridge deck, the scanned area needs to be split into subimages to apply the discrimination criteria, making it impossible to use a single threshold as a global criterion for each subimage. Consequently, thresholding methods rely heavily on operator experience and are time-consuming. Omar and Nehdi introduced a clustering-based approach, developing a k-means clustering model that identifies delamination areas by examining the condition of the bridge deck (e.g., age, temperature variation, and the size of observed surface spalling). Based on the condition of the bridge deck, a k-value is determined to classify the thermal images into k groups. Since the k-means clustering algorithm is a density-based distance metric, it selects the group with the lowest mean as intact during the day and the group with the highest mean as intact at night, without considering spatial information. Abdel-Qader et al. developed an automated process for segmenting defect regions in thermal images using a region growing method. This method assumes that the delamination region has the highest temperature in the entire scene, using the difference in temperature deviation within a 9x9 pixel window as a criterion for filtering images. Ellenberg et al. extended this method by using the temperature gradient as a threshold criterion. Although region growing-based segmentation methods exhibit insensitivity to inhomogeneous backgrounds, limitations remain in their seeding and growth-stop mechanisms, which are still determined by the global maximum in temperature or temperature gradient of the delamination region. Overall, these methods rely on determining a reference point, whereas true delamination is more associated with local maxima and spatial characteristics. Summary of the Invention

[0005] The purpose of the embodiment of the present invention is to provide an infrared imaging extraction method for bridge delamination defects based on morphological reconstruction, which has the advantages of quickly extracting the regional maximum of the bridge deck and segmenting the delamination, so that the inspection personnel can quickly locate the position and size of the bridge deck delamination, so as to carry out later maintenance of the bridge deck, thereby solving the above-mentioned technical problems.

[0006] The embodiment of the present invention is achieved as follows:

[0007] An embodiment of the present invention provides a method for extracting bridge delamination defects by infrared imaging based on morphological reconstruction, comprising the following steps:

[0008] S1: The original image is smoothed by anisotropic diffusion process to obtain the smoothed image T s and the gradient map G s ;

[0009] S2: by initial offset hin The general grayscale morphological reconstruction is performed to obtain the initial image I0 to meet the minimum criterion of temperature contrast;

[0010] S3: Offset I through regularization n =I n-1 -Δ*λ moving image;

[0011] S4: By using I n-1 Reconstruction I n Get I n ', then detect the regional maximum value and get the regional maximum value R n ;

[0012] S5: By G s and R n Get each sub-region r i The precise area or area boundary information, and in r i Extract statistics from above;

[0013] S6: From the gradient graph G by clustering method s Estimate the delamination gradient information and then extract the statistic D g , use it as a feature template;

[0014] S7:r i By D g Matching, so that r i The boundaries of the regional maxima matched in the middle are classified as delaminations;

[0015] S8: Store the regional maximum values ​​R1', R2', R3'...R after each iteration n ', n is the number of iterations; when |R n -R n-1 When |<δ or the maximum number of steps is reached, the iteration stops and the final segmentation layer is R n The union of .

[0016] Preferably, the anisotropic diffusion process in step S1 is expressed as follows:

[0017]

[0018] Among them, I t represents the image at the tth iteration, is the gradient operator, is the diffusion coefficient, usually defined as:

[0019]

[0020] Among them, K is the control diffusion rate constant, and its value is generally 5 to 30, and can also be adjusted according to experiments.

[0021] Preferably, the image T after smoothing in step S1 s and the gradient map G s , which is expressed as follows:

[0022]

[0023] Preferably, the initial image I0 in step S2 is expressed as follows:

[0024] I0=Reconstruction(T s ,h in )

[0025] The morphological reconstruction operation is defined as:

[0026] Reconstruction(T s ,h in )=imreconstruct(T s -h in ,T s )

[0027] Among them, imreconstruct(·,·) is the morphological reconstruction function.

[0028] Preferably, the regularization offset in step S3 is expressed as follows:

[0029] I n =I n-1 -Δ·λ

[0030] Where Δ is the offset, calculated as follows:

[0031] Δ=I n-1 -Reconstruction(I n-1 ,h in )

[0032] Here, λ is the regularization parameter.

[0033] Preferably, the image is reconstructed in step S4, and its expression is as follows:

[0034] I′ n =Reconstruction(I n-1 ,I n )

[0035] The expression for detecting regional maximum value is as follows:

[0036] R n =RegionalMax(I' n )

[0037] Among them, RegionalMax(·) is the regional maximum detection function.

[0038] Preferably, the precise region or region boundary information of the extracted sub-region in step S5 is expressed as follows:

[0039] r i =RegionProps(G s ,R n )

[0040] Among them, RegionProps(·,·) is the region attribute extraction function used to extract the sub-region r i The precise area or area boundary information of the

[0041] Preferably, in step S5, i The statistics are extracted from the above, and the expression is as follows:

[0042] Stats(r i )={mean(r i ),std(r i )}

[0043] Here, mean(·) and std(·) represent the mean and standard deviation, respectively.

[0044] Preferably, the clustering method in step S6 is expressed as follows:

[0045] {C1,C2,...,C k}=k-means(G s ,k)

[0046] Here, k-means(·,k) is the k-means clustering algorithm, and k is the number of clusters.

[0047] Preferably, in step S6, g The statistics are extracted from the above, and the expression is as follows:

[0048] D g ={mean(G s ),std(G s )}

[0049] Preferably, in step S7, g Statistics and subregion r i The statistics of r are compared to i The boundaries of the regional maxima matched in are classified as delaminations, and their expressions are as follows:

[0050] Match(r i ,Dg )

[0051] The beneficial effects of the present invention are:

[0052] The infrared imaging extraction method for bridge delamination defects based on morphological reconstruction provided by the embodiment of the present invention uses a weight attenuation function to regularize the grayscale morphological reconstruction to achieve regional maximum extraction, and can use the mean and coefficient of variation of the temperature gradient at the delamination boundary for discrimination. This technology can handle the situation of uneven background that often occurs in practice, thereby eliminating the need for existing methods to infer the background. Within this framework, the gradient statistics of the thermal image delamination boundary are used as an effective criterion for improving segmentation. Therefore, compared with traditional methods, better performance is obtained. This method can segment the delamination, so that inspection personnel can quickly locate the position and size of the bridge deck delamination, so as to carry out later maintenance of the bridge deck. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following is a brief introduction to the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0054] Figure 1 A flowchart of an iterative method provided for an embodiment of the present invention;

[0055] Figure 2 Schematic diagram of the process of concrete delamination and segmentation according to the present invention. DETAILED DESCRIPTION

[0056] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. It should be noted that the embodiments of the present invention and the features of the embodiments can be combined with each other unless there is a conflict.

[0057] refer to Figure 1 As shown, an embodiment of the present invention provides a method for extracting bridge delamination defects by infrared imaging based on morphological reconstruction, comprising the following steps:

[0058] S1: The original image is smoothed by anisotropic diffusion process to obtain the smoothed image T s and the gradient map G s .

[0059] The anisotropic diffusion process in this step is expressed as follows:

[0060]

[0061] Among them, I t represents the image at the tth iteration, is the gradient operator, is the diffusion coefficient, usually defined as:

[0062]

[0063] Among them, K is the control diffusion rate constant, and its value is generally 5 to 30, and can also be adjusted according to experiments.

[0064] The smoothed image T in this step s and the gradient map G s , which is expressed as follows:

[0065]

[0066] S2: by initial offset h in The general grayscale morphological reconstruction is performed to obtain the initial image I0 to meet the minimum criterion of temperature contrast.

[0067] The initial image I0 in this step is expressed as follows:

[0068] I0=Reconstruction(T s ,h in )

[0069] The morphological reconstruction operation is defined as:

[0070] Reconstruction(T s ,h in )=imreconstruct(T s -h in ,T s )

[0071] Among them, imreconstruct(·,·) is the morphological reconstruction function.

[0072] S3: Offset I through regularization n =I n-1 -Δ*λ moves the image.

[0073] The regularization offset in this step is expressed as follows:

[0074] I n =I n-1 -Δ·λ

[0075] Where Δ is the offset, calculated as follows:

[0076] Δ=I n-1 -Reconstruction(I n-1 ,h in )

[0077] Among them, λ is the regularization parameter and n is the number of iterations.

[0078] S4: By using I n-1 Reconstruction I n Get I n ', then detect the regional maximum value and get the regional maximum value R n .

[0079] The image is reconstructed in this step, and its expression is as follows:

[0080] I′ n =Reconstruction(I n-1 ,I n )

[0081] In this step, the regional maximum value is detected, and its expression is as follows:

[0082] R n =RegionalMax(I' n )

[0083] Among them, RegionalMax(·) is the regional maximum detection function.

[0084] S5: By G s and R n Get each sub-region r i The precise area or area boundary information, and in r i Extract statistics on .

[0085] The precise region or region boundary information of the extracted sub-region in this step is expressed as follows:

[0086] r i =RegionProps(G s ,R n )

[0087] Among them, RegionProps(·,·) is the region attribute extraction function used to extract the sub-region r i The precise area or area boundary information of the

[0088] In this step, i The statistics are extracted from the above, and the expression is as follows:

[0089] Stats(r i )={mean(r i ),std(r i )}

[0090] Here, mean(·) and std(·) represent the mean and standard deviation, respectively.

[0091] S6: From the gradient graph G by clustering method s Estimate the delamination gradient information and then extract the statistic D g , and use it as a feature template.

[0092] The clustering method in this step is expressed as follows:

[0093] {C1,C2,...,C k}=k-means(G s ,k)

[0094] Here, k-means(·,k) is the k-means clustering algorithm, and k is the number of clusters.

[0095] In this step, D g The statistics are extracted from the above, and the expression is as follows:

[0096] D g ={mean(G s ),std(G s )}

[0097] S7:r i By D g Matching, so that r i The boundaries of the matching regional maxima are classified as delaminations.

[0098] In this step, according to D g Statistics and subregion r i The statistics of r are compared to i The boundaries of the regional maxima matched in are classified as delaminations, and their expressions are as follows:

[0099] Match(r i ,D g )

[0100] S8: Store the regional maximum values ​​R1', R2', R3'...R after each iteration n ', when |R n -R n-1 When |<δ or the maximum number of steps is reached, the iteration stops and the final segmentation layer is R n The union of .

[0101] Figure 2 The process of delamination segmentation is shown in Fig. 1. First, the original image T is anisotropically smoothed and the formula Get the smoothed image T s , using the formula to get Gradient map G s Then the smoothed image T s Perform the morphological reconstruction iteration process, using the formula I0=Reconstruction(T s ,h in ) to get the original image I0, and then use formula I n =I n-1 -Δ·λ is used for regularization offset, and then formula I′ is used n =Reconstruction(I n-1 ,I n ) to reconstruct the image and then use the formula R n =RegionalMax(I' n ) Identify the regional maximum of the reconstructed image, and finally use the formula r i =RegionProps(G s ,R n ) Extract sub-region r i The precise region or region boundary information. At the same time, the gradient map G s Using the formula {C1,C2,...,C k}=k-means(G s ,k) Estimate the gradient information and then use formula D g ={mean(G s ),std(G s )} Extract statistics D g Finally, using the formula Match(r i ,D g ) will r i With D g Matching, so that r i The boundaries of the matched regional maxima are classified as delaminations, and the results of each iteration are stored until the iteration ends to obtain the delamination boundary map S.

[0102] The infrared imaging extraction method for bridge delamination defects based on morphological reconstruction provided by the embodiment of the present invention uses a weight attenuation function to regularize the grayscale morphological reconstruction to achieve regional maximum extraction, and can use the mean and coefficient of variation of the temperature gradient at the delamination boundary for discrimination. This technology can handle the situation of uneven background that often occurs in practice, thereby eliminating the need for existing methods to infer the background. Within this framework, the gradient statistics of the thermal image delamination boundary are used as an effective criterion for improving segmentation. Therefore, compared with traditional methods, better performance is obtained. This method can segment the delamination, so that inspection personnel can quickly locate the position and size of the bridge deck delamination, so as to carry out later maintenance of the bridge deck.

[0103] The present invention is not limited to the above-mentioned optional implementation modes. Anyone can derive other forms of products under the inspiration of the present invention. However, no matter what changes are made in the shape or structure, any technical solution that falls within the scope defined by the claims of the present invention falls within the scope of protection of the present invention.

Claims

1. A method for extracting bridge delamination defects by infrared imaging based on morphological reconstruction, characterized by: The following steps are included: S1: The original image is smoothed by anisotropic diffusion process to obtain the smoothed image T s and the gradient map G s ; S2: by initial offset h in The general grayscale morphological reconstruction is performed to obtain the initial image I0 to meet the minimum criterion of temperature contrast; S3: Offset I through regularization n =I n-1 -Δ*λ moves the image, where n is the number of iterations; S4: By using I n-1 Reconstruction I n Get I n ', then detect the regional maximum value and get the regional maximum value R n ; S5: By G s and R n Get each sub-region r i The precise area or area boundary information, and in r i Extract statistics from above; S6: From the gradient graph G by clustering method s Estimate the delamination gradient information and then extract the statistic D g , use it as a feature template; S7:r i By D g Matching, so that r i The boundaries of the regional maxima matched in the middle are classified as delaminations; S8: Store the regional maximum values ​​R1', R2', R3'...R after each iteration n ', n is the number of iterations; when |R n -R n-1 When |<δ or the maximum number of steps is reached, the iteration stops and the final segmentation layer is n The union of .

2. The infrared imaging extraction method for bridge delamination defects based on morphological reconstruction according to claim 1 is characterized by: The anisotropic diffusion process in step S1 is expressed as follows: Among them, I t represents the image at the tth iteration, is the gradient operator, is the diffusion coefficient, usually defined as: where K is the controlling diffusion rate constant.

3. The infrared imaging extraction method for bridge delamination defects based on morphological reconstruction according to claim 2 is characterized by: The smoothed image T in step S1 s and the gradient map G s , which is expressed as follows:

4. The method for extracting bridge delamination defects by infrared imaging based on morphological reconstruction according to claim 1, characterized in that: The initial image I0 in step S2 is expressed as follows: I0=Reconstruction(T s ,h in ) The morphological reconstruction operation is defined as: Reconstruction(T s ,h in )=imreconstruct(T s -h in ,T s ) Among them, imreconstruct(·,·) is the morphological reconstruction function.

5. The infrared imaging extraction method for bridge delamination defects based on morphological reconstruction according to claim 1 is characterized by: The regularization offset in the S3 step is expressed as follows: I n =I n-1 -D·l Where Δ is the offset, calculated as follows: Δ=I n-1 -Reconstruction(I n-1 ,h in ) Here, λ is the regularization parameter.

6. The infrared imaging extraction method for bridge delamination defects based on morphological reconstruction according to claim 1 is characterized by: The image is reconstructed in step S4, and its expression is as follows: I′ n =Reconstruction(I n-1 ,I n ) The expression for detecting regional maximum value is as follows: R n =RegionalMax(I' n ) Among them, RegionalMax(·) is the regional maximum detection function.

7. The infrared imaging extraction method for bridge delamination defects based on morphological reconstruction according to claim 1 is characterized by: The precise region or region boundary information of the extracted sub-region in step S5 is expressed as follows: r i =RegionProps(G s ,R n ) Among them, RegionProps(·,·) is the region attribute extraction function used to extract the sub-region r i Precise area or area boundary information; In r i The statistics are extracted from the above, and the expression is as follows: Stats(r i )={mean(r i ),std(r i )} Here, mean(·) and std(·) represent the mean and standard deviation, respectively.

8. The infrared imaging extraction method for bridge delamination defects based on morphological reconstruction according to claim 1 is characterized by: The clustering method in step S6 is expressed as follows: {C1,C2,...,C k }=k-means(G s ,k) Where k-means(·,k) is the k-means clustering algorithm, and k is the number of clusters; In D g The statistics are extracted from the above, and the expression is as follows: D g ={mean(G s ),std(G s )}。 9. The infrared imaging extraction method for bridge delamination defects based on morphological reconstruction according to claim 1 is characterized by: According to D in step S7 g Statistics and subregion r i The statistics of r are compared to i The boundaries of the regional maxima matched in are classified as delaminations, and their expressions are as follows: Match(r i ,D g )。

Images