Underwater concrete structure apparent defect detection method based on Dark-Retinex model

Abstract

The invention discloses an underwater concrete structure apparent defect detection method based on a Dark-Retinex model. The method comprises the following steps: acquiring an original image of an underwater concrete structure; processing the original image by using a dark channel defogging algorithm and a Retinex defogging algorithm to obtain a DR image; converting the DR image into a grayscale image and mapping a brightness value of the grayscale image to a new value image; carrying out image binarization on the new value image and reversing the image, so that white pixels in a binary image represent defect parts; marking a connected domain to remove interference information; and extracting apparent characteristic parameters of a white area in the binary image. The method is an image enhancement Dark-Retinex algorithm fusing a dark channel algorithm and a Retinex algorithm, image recovery of underwater concrete structure apparent defects and defect parameter quantitative evaluation are realized, and the apparent defect evaluation precision of underwater concrete structure images is improved.

Description

Technical Field

[0001] The present invention relates to the field of concrete defect detection, and in particular to a method for detecting surface defects of underwater concrete structures based on a Dark-Retinex model. Background Art

[0002] Due to the influence of external environment (such as wind and waves, corrosion, hydraulic scouring, and temperature stress, etc.) and other factors (such as human and geological disasters, etc.), underwater structures will suffer various degrees of damage during long-term service, such as hollowing, cracks, holes, corrosion pits, grooves, etc. The types of defects in water-related structures usually include: hollowing caused by water scouring, hollowing, insufficient pile foundation burial depth, surface weathering, peeling, and corrosion pits caused by steel bar corrosion, and insufficient foundation bearing capacity caused by cavities, pitting, and honeycombs. Accurate and effective damage analysis and assessment are essential to ensure the safety of underwater structures.

[0003] In recent years, with the rapid development of computer technology combined with traditional visual inspection methods for underwater concrete surface damage, computer vision technology has been applied to the safety and health inspection of underwater structures, which is objective, highly efficient and safe in practical applications. However, due to the refraction or scattering of light in water, the images of underwater concrete structures will produce ghosting and shadows, making the target unclear, which greatly affects the assessment accuracy of concrete structure surface defects.

[0004] Therefore, the art needs a new method for detecting surface defects of underwater concrete structures to solve the above problems. Summary of the invention

[0005] The purpose of the present invention is to provide a method for detecting apparent defects of underwater concrete structures based on the Dark-Retinex model, aiming to solve the problem of improving the accuracy of apparent defect assessment of underwater concrete structure images.

[0006] In order to achieve the above object, the present invention adopts the following technical solution:

[0007] A method for detecting surface defects of underwater concrete structures based on a Dark-Retinex model, the method comprising:

[0008] Get the original image of underwater concrete structure;

[0009] The original image is processed using a dark channel defogging algorithm and a Retinex defogging algorithm to obtain a DR image;

[0010] Converting the DR image to a grayscale image and mapping the brightness values ​​of the grayscale image to a new value image;

[0011] Binarizing the new value image and inverting the image so that white pixels in the binary image represent defective parts;

[0012] Marking connected areas to remove interference information;

[0013] The apparent characteristic parameters of the white area in the binary image are extracted.

[0014] Preferably, before the step of "processing the original image using a dark channel defogging algorithm and a Retinex defogging algorithm to obtain a DR image", the method further comprises:

[0015] Compensating the red channel and the blue channel of the original image;

[0016] Performing white balance processing on the compensated image;

[0017] Performing Gamma correction on the image after white balance processing;

[0018] Performing image sharpening processing on the Gamma-corrected image;

[0019] The gamma-corrected image and the image sharpened image are fused.

[0020] Preferably, the step of "mapping the brightness value of the grayscale image to the new value image" includes:

[0021] Map the brightness value of the grayscale image to the new value image according to the mapping relationship shown in the following formula:

[0022]

[0023] Among them, r represents the grayscale value of the grayscale image, S represents the grayscale value of the new value image, and c represents any RGB channel.

[0024] Preferably, the step of "processing the original image using a dark channel defogging algorithm and a Retinex defogging algorithm to obtain a DR image" includes:

[0025] The original image is processed using the dark channel dehazing algorithm shown below:

[0026]

[0027] Among them, Q1(x) represents the image output by the dark channel dehazing algorithm, I(x) represents the original image, A represents the light intensity in water, and T(x) represents the transmittance of the medium.

[0028] The Retinex algorithm is used to calculate the image output by the dark channel defogging algorithm to obtain a DR image.

[0029] Preferably, the step of "extracting apparent characteristic parameters of the white area in the binary image" includes:

[0030] The geometric parameters of the white area are counted, and the actual geometric parameters of the defective part are evaluated based on the geometric parameters, where the actual geometric parameters include at least one of width, length, average width, average length, and area.

[0031] A method for detecting surface defects of underwater concrete structures based on a Dark-Retinex model, the method comprising:

[0032] Get the original image of underwater concrete structure;

[0033] The original image is processed using a dark channel defogging algorithm and a Retinex defogging algorithm to obtain a DR image;

[0034] Converting the DR image into a grayscale image;

[0035] Performing frequency domain denoising on the grayscale image;

[0036] Performing edge detection on the frequency domain denoised image;

[0037] Connect pixels with the same gray value to form a connected area;

[0038] Invert and edge filter the image;

[0039] The apparent feature parameters of the white area in the image are extracted.

[0040] Preferably, before the step of "processing the original image using a dark channel defogging algorithm and a Retinex defogging algorithm to obtain a DR image", the method further comprises:

[0041] Compensating the red channel and the blue channel of the original image;

[0042] Performing white balance processing on the compensated image;

[0043] Performing Gamma correction on the image after white balance processing;

[0044] Performing image sharpening processing on the Gamma-corrected image;

[0045] The gamma-corrected image and the image sharpened image are fused.

[0046] Preferably, a multi-scale fusion algorithm is used to fuse the image after the gamma correction and the image after the image sharpening processing.

[0047] Preferably, the step of “performing frequency domain denoising on the grayscale image” comprises:

[0048] The grayscale image is denoised in the frequency domain by manually selecting a specific position and setting the value of the selected position to the minimum value in the spectrum graph.

[0049] Preferably, the step of "processing the original image using a dark channel defogging algorithm and a Retinex defogging algorithm to obtain a DR image" includes:

[0050] The original image is processed using the dark channel dehazing algorithm shown below:

[0051]

[0052] Among them, Q1(x) represents the image output by the dark channel dehazing algorithm, I(x) represents the original image, A represents the light intensity in water, and T(x) represents the transmittance of the medium.

[0053] The Retinex algorithm is used to calculate the image output by the dark channel defogging algorithm to obtain a DR image.

[0054] Preferably, the step of "extracting apparent characteristic parameters of the white area in the binary image" includes:

[0055] The geometric parameters of the white area are counted, and the actual geometric parameters of the defective part are evaluated based on the geometric parameters, where the actual geometric parameters include at least one of width, length, average width, average length, and area.

[0056] The advantages of the present invention are:

[0057] The method for detecting apparent defects of underwater concrete structures based on the Dark-Retinex model provided by the present invention adopts an image enhancement Dark-Retinex algorithm that fuses a dark channel algorithm and a Retinex algorithm, thereby realizing image restoration of apparent defects of underwater concrete structures and quantitative evaluation of defect parameters, and improving the accuracy of apparent defect evaluation of underwater concrete structure images. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 Schematic diagram of main steps of a method for detecting surface defects of underwater concrete structures based on a Dark-Retinex model in an embodiment of the present invention;

[0059] Figure 2 is a schematic diagram of a main process of restoring an original image in an embodiment of the present invention;

[0060] Figures 3 to 9 1 is a schematic diagram of a processing effect of restoring an original image in an embodiment of the present invention; wherein (a) is the original image, (b) is red and blue channel compensation; (c) is white balance; (d) is gamma correction; (e) is image sharpening; and (f) is image fusion;

[0061] Fig.10 A schematic diagram of selecting pixels layer by layer using a pyramid method in an embodiment of the present invention;

[0062] Fig.11 Schematic diagram of the processing effect of a method for detecting surface defects of underwater concrete structures based on the Dark-Retinex model in an embodiment of the present invention; wherein (a) is a DR image, (b) is a grayscale image, (c) is a binary image, and (d) is image inversion and edge filtering;

[0063] Fig.12 It is a schematic diagram of the main steps of another method for detecting surface defects of underwater concrete structures based on the Dark-Retinex model in an embodiment of the present invention;

[0064] Fig.13 1 is a schematic diagram of underwater concrete apparent crack detection results in an embodiment of the present invention; wherein (a) is a DR image, (b) is a grayscale image, (c) is frequency domain denoising, (d) is edge detection, (e) is regional connectivity; (f) is image inversion and edge filtering;

[0065] Fig.14 1 is a schematic diagram of the detection result of the surface roughness of underwater concrete in an embodiment of the present invention; wherein (a) is a DR image, (b) is a grayscale image, (c) is frequency domain denoising, (d) is edge detection, (e) is regional connectivity, and (f) is image inversion and edge filtering;

[0066] Fig.15 1 is a schematic diagram of underwater concrete surface spalling detection results in an embodiment of the present invention; wherein, (a) is a DR image, (b) is a grayscale image, (c) is frequency domain denoising, (d) is edge detection, (e) is regional connectivity; and (f) is image inversion and edge filtering. DETAILED DESCRIPTION

[0067] The present invention is further described in detail below with reference to the accompanying drawings and embodiments.

[0068] See attached Figure 1 , Figure 1 The main steps of a method for detecting surface defects of underwater concrete structures based on the Dark-Retinex model are as follows. The method for detecting surface defects of underwater concrete structures based on the Dark-Retinex model provided in this embodiment includes:

[0069] Step S11: acquiring an original image of the underwater concrete structure;

[0070] Specifically, taking Fengman Reservoir as an example, the normal water level of the reservoir is 263.50m, the flood limit water level is 260.50m, the dead water level is 242.00m, and the verified flood level is 267.70m. The maximum dam height of the barrage is 91.7m, the dam top length is 1080m, and the dam top elevation is 267.70m. This embodiment uses a high-definition camera carried by an underwater robot for image acquisition.

[0071] See attached Figure 3 (a) and Figure 4 (a) shows the apparent cracks and corroded exposed aggregates of underwater concrete under low illumination. The original image has the characteristics of low contrast, strong blur, uneven color brightness, color distortion, and difficult to identify the target. Therefore, it is necessary to restore the original image so that the apparent cracks, spalling, roughness, honeycomb and other defects of the underwater concrete structure can be clearly shown.

[0072] See attached Figure 2 , the process of restoring the original image in this embodiment is as follows:

[0073] Step Sa1: Compensate the red channel and blue channel of the original image.

[0074] Specifically, compensation by red and blue channels can make the color distribution of the image balanced. Figures 3 to 9 (b).

[0075] Step Sa2: Perform white balance processing on the compensated image.

[0076] Specifically, the basic concept of white balance is to restore white objects to white regardless of the light source, and to compensate for the color cast that occurs when shooting under a specific light source by enhancing the corresponding complementary color. In underwater shooting, you will encounter a variety of light sources, and different light sources have different color temperatures. Sometimes the captured image will have a color cast, and white balance can solve this problem. This step enhances the appearance of the image by removing the extra color cast caused by various light sources. After white balancing, the image color changes from yellowish to more natural. See also Figures 3 to 9 (c).

[0077] Step Sa3: Perform gamma correction on the image after white balance processing.

[0078] Specifically, Gamma correction performs nonlinear operations on the grayscale values ​​of the input image, so that the grayscale values ​​of the output image are exponentially related to the grayscale values ​​of the input image. The human eye's sensitivity to external light sources is not linearly related to the input light intensity, but exponentially related. Under low illumination, the human eye can more easily distinguish changes in brightness. As the illumination increases, the human eye is less likely to distinguish changes in brightness. However, the camera's sensitivity is linearly related to the input light intensity. In order to adapt to the characteristics of the human eye, it is necessary to perform Gamma correction on the original image captured by the camera. See also Figures 3 to 9 (d).

[0079] Step Sa4: Perform image sharpening processing on the gamma-corrected image.

[0080] Specifically, image sharpening compensates for the contour of an image, enhances the edge of the image and the grayscale jump part, and makes the image clear. It includes two types of processing: spatial domain processing and frequency domain processing. Image sharpening can highlight the edge, contour, or characteristics of certain linear target elements of the image target. Accordingly, the contrast between the target edge and the background pixel is improved. Figures 3 to 9 (e).

[0081] Step Sa5: Fusing the gamma-corrected image and the image sharpened image.

[0082] Specifically, the standard multi-scale fusion algorithm is used to fuse the gamma correction result and the sharpening result of the original image. The contrast of the fused image is significantly enhanced, the color brightness of the image is evenly distributed, and the target is easier to identify. Figures 3 to 9 (f).

[0083] Figure 5 (a) is an underwater crack picture taken under exposure. The exposed underwater image is also one of the difficult problems in image restoration. The contrast of the underwater image after exposure is significantly reduced, the blur is also strong, and the image color brightness is uneven, resulting in poor enhancement effect. Figure 6 (a) shows the apparent cracks of underwater concrete in turbid water environment. Turbid water weakens the image of underwater cracks, which becomes more blurred and the contrast is greatly reduced. Figure 7 (a) is a picture of an underwater crack taken under dim light. The overall image features are dark and the contrast is low. For low-light underwater image processing, the most common problem is color distortion. Figure 8 (a) is the original soil interference image. Due to the absorption and scattering of light, the original image is blurred, the contrast between the cracks and the background image is not high, and there is noise interference. Fig. 9(a) is a highly blurred underwater crack image. Blurred underwater images are a common phenomenon in image restoration, with characteristics such as low image contrast and uneven image color brightness. After being processed by the image enhancement algorithm, the final fused image has significantly improved in brightness, contrast, and image detail expression, and is more in line with the visual observation characteristics of the human eye. The final fused image can well suppress noise interference and solve the color cast problem of the image. The image contrast is significantly enhanced, the image color brightness is evenly distributed, the target is easier to identify, and the defects such as apparent cracks, peeling, pockmarks, and honeycombs can be clearly displayed.

[0084] Step S12: Process the original image using the dark channel defogging algorithm and the Retinex defogging algorithm to obtain a DR image.

[0085] It should be noted that the original image here is the image after the above image restoration processing. First, the original image is processed by the dark channel defogging algorithm shown in the following formula:

[0086]

[0087] Among them, O1(x) represents the image output by the dark channel dehazing algorithm, I(x) represents the original image, A represents the light intensity in water, and T(x) represents the transmittance of the medium.

[0088] Since the underwater scene contains white scenes, the image processed by the dark channel defogging algorithm is prone to block phenomenon and halo effect. Therefore, the present invention adds the Retinex defogging algorithm to enhance the image. That is, the image is transformed into the logarithmic domain. Assuming that the essential attributes of the image are the root cause of the image block phenomenon and halo effect, the image model can be expressed as:

[0089] O1(x,y)=O2(x,y)*L(x,y) (2)

[0090] Logarithmic transformation of both sides of equation (2) yields

[0091] O1=O2+L (3)

[0092] From this equation (3), O2 and L can be obtained.

[0093] Using the pyramid method to select pixels layer by layer, a top-down layer-by-layer iteration is established to improve efficiency, such as Fig.10 shown.

[0094] 1) Perform logarithmic transformation on each RGB channel of the color image to obtain the initialization image L0(x,y) and determine the number of pyramid layers;

[0095] 2) The order of the pyramid layers from top to bottom is defined as: 11 , O12 , O 13 ...., O 1m ;

[0096] 3) From the top layer to the bottom layer, eight neighborhood comparison operations are performed. The operation rules are as follows:

[0097] I n =L n +O 1n+1 -O 1n (4)

[0098]

[0099]

[0100] 4) After the calculation of the mth layer is completed, its results are interpolated. Therefore, the mth layer becomes twice the original size and the same as the m+1th layer; the final reflection coefficient is:

[0101]

[0102] When the underlying calculations are completed, the final enhanced image, namely the DR image, can be obtained.

[0103] Step S13: Convert the DR image into a grayscale image and map the brightness value of the grayscale image to a new value image.

[0104] Specifically, the hue and saturation information of the DR image is removed and the brightness value is kept to convert the RGB image into a grayscale image. The weighted average algorithm of the R, G, and B components is

[0105] Gray DR =0.2989R+05870G+0.1140B. (8)

[0106] Adjust the brightness of the grayscale image. Map the brightness value of the grayscale image to the new value image J:

[0107]

[0108] Among them, r represents the grayscale value of the grayscale image, S represents the grayscale value of the new value image, and c represents any RGB channel.

[0109] Step S14: binarize the new value image and invert the image so that the white pixels in the binary image represent defective parts.

[0110] Specifically, the maximum inter-class variance method is used to search for the optimal threshold of J in the interval [0,1]. After obtaining the optimal threshold, image binarization is performed. After inversion, white pixels represent defective parts.

[0111] Step S15: Mark the connected domain to remove interference information.

[0112] Specifically, the concept of connected domains is used for region segmentation. Small regions connected to the connected domains in the binary graph are deleted to remove interference information, thereby revealing the final defect region.

[0113] Step S16: extracting the apparent characteristic parameters of the white area in the binary image.

[0114] Specifically, geometric parameters of the white area are counted, and actual geometric parameters of the defective part are evaluated based on the geometric parameters, where the actual geometric parameters include at least one of width, length, average width, average length, and area.

[0115] The binary image obtained after the above image processing, in which white pixels represent the apparent defect area of ​​the underwater concrete structure, and black pixels represent the background area of ​​the underwater concrete structure. Therefore, by performing infinite skeleton extraction on the apparent defects of the underwater concrete structure, the length and width of the cracks, as well as the defect areas such as peeling, pitting, and honeycomb can be estimated by counting the number of pixels. The resolution of the camera used when taking the image is used as a calibration, and the actual length value μ corresponding to each pixel in the image can be estimated. The number of statistical pixels is multiplied by the μ value to obtain the actual length and width of the target crack and the area of ​​defects such as peeling, pitting, and honeycomb. So far, this method can effectively identify the apparent defects of underwater concrete structures and can quantitatively estimate the characteristic parameters of the apparent defects.

[0116] See attached Fig.11 , this embodiment exemplarily displays 6 groups of images to show the apparent defects of underwater concrete structures. Fig.11 The first column (a) is the DR image, the second column (b) is the corresponding grayscale image, the third column (c) is the corresponding binary image; the fourth column (d) is image inversion and edge filtering. The binary images Image1-4(c) and Image 6(c) can obtain the interference information of black pixels, and these interference information will cause the statistical results of the defect area ratio to be too large. The details in Image 5(c) are richer, and the white pixels in the leaking reinforcement area caused by the scouring of the bridge pile foundation cause the statistical results of the defect area ratio to be too small. In addition, Image1-6(d) is the result of removing the interference pixels in the orange rectangular box. The direction of the crack in Image 1-4(d) is consistent with that in Image 1-4(a), the crack edge is clear, and the image processing effect is good. Although there are still "burrs" on the crack edge in Image 4(d), overall, the improvement effect is significant. The present invention believes that slight differences in image brightness reflect the degree of damage to the underwater structure. Here, the image brightness values ​​225-235, 235-245 and 245-255 are defined as white, very white and pure white, respectively. According to Fig.11The processing result of (d) can realize the detection of defective areas.

[0117] See attached Figures 12 to 15 The embodiment of the present invention also provides a method for detecting surface defects of underwater concrete structures based on a Dark-Retinex model, comprising:

[0118] Step S21: Acquire the original image of the underwater concrete structure. For details, please refer to step S11, which will not be described here.

[0119] Step S22: The original image is processed using a dark channel defogging algorithm and a Retinex defogging algorithm to obtain a DR image. For details, please refer to step S21, which will not be described here. Figures 13 to 15 (a) is the DR image.

[0120] Step S23: Convert the DR image into a grayscale image.

[0121] Specifically, Figures 13 to 15 (b) is a grayscale image. The DR image is in RGB color mode. When grayscale processing is performed, the three components of RGB need to be processed separately. In fact, RGB cannot reflect the morphological characteristics of the image, but only adjusts the color based on the optical principle. Image grayscale processing is a preprocessing step for the image, preparing for the subsequent upper-level operations such as image segmentation, image recognition and image analysis.

[0122] Step S24: performing frequency domain denoising on the grayscale image.

[0123] Specifically, Figures 13 to 15 (c) is the result of frequency domain denoising. The image is displayed in the frequency domain through Fourier transform. By observing the spectrum graph, if the relationship between the spectrum graph and the original image is understood, the characteristics of the noise in the spectrum graph can be easily found. This embodiment performs frequency domain denoising on the grayscale image by manually circling a specific position and setting the value of the selected position to the minimum value in the spectrum graph. Accordingly, it is possible to remove regular grid, reticulated, and line-shaped noise that appears in the image. For example, Fig.13 In (c), the horizontal stripes are eliminated.

[0124] Step S25: Perform edge detection on the frequency domain denoised image. Specifically, Figures 13 to 15 (d) is the edge detection result. In this embodiment, the edge detection is realized by using convolution operation.

[0125] Step S26: Connect pixels with the same grayscale value to form a connected area.

[0126] Specifically, defect extraction and segmentation are crucial for the detection of apparent defects in underwater concrete structures. The present invention uses adaptive threshold segmentation combined with morphological methods to extract the apparent defects of underwater concrete structures. Adaptive threshold segmentation is sensitive to noise. Therefore, morphological opening operation is used for processing. After the morphological opening operation, a threshold is set to remove noise points. Figures 13 to 15 (e) is the regional connectivity result.

[0127] Step S27: Invert and edge filter the image.

[0128] The purpose of this step is to highlight the defective area with white pixels and eliminate the burr edges and interfering pixels in the defective area. In this way, the apparent defects of the underwater concrete structure can be effectively extracted from the complex background. The apparent defect detection method adopted in this embodiment shows strong performance, which not only controls the influence of image noise to a certain extent, but also can more accurately extract the morphology of concrete apparent cracks, peeling, pockmarks, honeycombs and other defects. Figures 13 to 15 (f) is the result of binary grayscale image inversion and edge filtering.

[0129] Step S28: extracting the apparent characteristic parameters of the white area in the image. For details, please refer to the above step S16, which will not be described here.

[0130] In summary, the image enhancement Dark-Retinex algorithm proposed in the present invention, which integrates the Dark Channel algorithm and the Retinex algorithm, realizes the image restoration of the surface defects of underwater concrete structures and the quantitative evaluation of defect parameters.

[0131] In actual projects, the present invention uses the Dark-Retinex algorithm and edge detection method to process and analyze typical underwater structural defect images of dams and bridges, as well as underwater image data of the Fengman Old Dam test. The results show that the Dark-Retinex algorithm can effectively suppress image background noise, solve the problem of image color cast, significantly enhance the image contrast, and make the image color brightness distribution more uniform, which can more clearly highlight the concrete structure defects such as cracks, spalling, pockmarks, and honeycombs. The brightness, contrast, and detail expression capabilities of the final fused enhanced image are significantly improved. Based on the enhanced image obtained by the Dark-Retinex algorithm, the edge detection method realizes the image segmentation and feature parameter extraction of the underwater concrete structure surface defects, and can quantitatively evaluate the length and width of the cracks, as well as the area of ​​defects such as spalling, pockmarks, and honeycombs.

[0132] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0133] The term "comprising" or any other similar term is intended to cover a non-exclusive inclusion such that a parameter or method that includes a list of elements includes not only those elements but also other elements not expressly listed or inherent to those parameters or methods.

[0134] The above are preferred embodiments of the present invention and the technical principles used therein. For those skilled in the art, any obvious changes such as equivalent transformations, simple replacements, etc. based on the technical solution of the present invention, without departing from the spirit and scope of the present invention, are within the scope of protection of the present invention.

Claims

1. A method for detecting surface defects of underwater concrete structures based on the Dark-Retinex model, characterized in that: The method comprises: Get the original image of underwater concrete structure; The original image is processed using a dark channel defogging algorithm and a Retinex defogging algorithm to obtain a DR image; Converting the DR image to a grayscale image and mapping the brightness values ​​of the grayscale image to a new value image; Binarizing the new value image and inverting the image so that white pixels in the binary image represent defective parts; Marking connected areas to remove interference information; The apparent characteristic parameters of the white area in the binary image are extracted.

2. The underwater concrete structure apparent defect detection method based on the Dark-Retinex model according to claim 1, characterized in that: Before the step of "processing the original image using a dark channel defogging algorithm and a Retinex defogging algorithm to obtain a DR image", the method further includes: Compensating the red channel and the blue channel of the original image; Performing white balance processing on the compensated image; Performing Gamma correction on the image after white balance processing; Performing image sharpening processing on the Gamma-corrected image; The gamma-corrected image and the image sharpened image are fused.

3. The underwater concrete structure apparent defect detection method based on the Dark-Retinex model according to claim 2, characterized in that: The steps of "mapping the brightness value of the grayscale image to a new value image" include: Map the brightness value of the grayscale image to the new value image according to the mapping relationship shown in the following formula: Among them, r represents the grayscale value of the grayscale image, S represents the grayscale value of the new value image, and c represents any RGB channel.

4. The underwater concrete structure apparent defect detection method based on the Dark-Retinex model according to claim 2, characterized in that: The step of "processing the original image using a dark channel defogging algorithm and a Retinex defogging algorithm to obtain a DR image" includes: The original image is processed using the dark channel dehazing algorithm shown below: Among them, O1(x) represents the image output by the dark channel dehazing algorithm, I(x) represents the original image, A represents the light intensity in water, and T(x) represents the transmittance of the medium; The Retinex algorithm is used to calculate the image output by the dark channel defogging algorithm to obtain a DR image.

5. The underwater concrete structure apparent defect detection method based on the Dark-Retinex model according to claim 1, characterized in that: The step of "extracting apparent characteristic parameters of the white area in the binary image" includes: The geometric parameters of the white area are counted, and the actual geometric parameters of the defective part are evaluated based on the geometric parameters, where the actual geometric parameters include at least one of width, length, average width, average length, and area.

6. A method for detecting surface defects of underwater concrete structures based on the Dark-Retinex model, characterized in that: The method comprises: Get the original image of underwater concrete structure; The original image is processed using a dark channel defogging algorithm and a Retinex defogging algorithm to obtain a DR image; Converting the DR image into a grayscale image; Performing frequency domain denoising on the grayscale image; Performing edge detection on the frequency domain denoised image; Connect pixels with the same gray value to form a connected area; Invert and edge filter the image; The apparent feature parameters of the white area in the image are extracted.

7. The method for detecting surface defects of underwater concrete structures based on the Dark-Retinex model according to claim 6, characterized in that: Before the step of "processing the original image using a dark channel defogging algorithm and a Retinex defogging algorithm to obtain a DR image", the method further includes: Compensating the red channel and the blue channel of the original image; Performing white balance processing on the compensated image; Performing Gamma correction on the image after white balance processing; Performing image sharpening processing on the Gamma-corrected image; The gamma-corrected image and the image sharpened image are fused.

8. The method for detecting surface defects of underwater concrete structures based on the Dark-Retinex model according to claim 6, characterized in that: Using a multi-scale fusion algorithm to fuse the gamma-corrected image and the image after image sharpening; The step of "performing frequency domain denoising on the grayscale image" comprises: The grayscale image is denoised in the frequency domain by manually selecting a specific position and setting the value of the selected position to the minimum value in the spectrum graph.

9. The method for detecting surface defects of underwater concrete structures based on the Dark-Retinex model according to claim 6, characterized in that: The step of "processing the original image using a dark channel defogging algorithm and a Retinex defogging algorithm to obtain a DR image" includes: The original image is processed using the dark channel dehazing algorithm shown below: Among them, O1(x) represents the image output by the dark channel dehazing algorithm, I(x) represents the original image, A represents the light intensity in water, and T(x) represents the transmittance of the medium; The Retinex algorithm is used to calculate the image output by the dark channel defogging algorithm to obtain a DR image.

10. The underwater concrete structure surface defect detection method based on the Dark-Retinex model according to claim 6, characterized in that: The step of "extracting apparent characteristic parameters of the white area in the binary image" includes: The geometric parameters of the white area are counted, and the actual geometric parameters of the defective part are evaluated based on the geometric parameters, where the actual geometric parameters include at least one of width, length, average width, average length, and area.

Images