A bridge crack detection method

Abstract

The invention discloses a bridge crack detection method, which belongs to the field of image processing technology. The invention first converts an RGB image into an HSV color space, extracts a V channel and performs brightness suppression, and performs grayscale processing and grayscale suppression on the image. The invention then performs clustering processing on the RGB bridge image, finds suspected areas, calculates suspected coefficients for pixels on each suspected area, constructs a suspected coefficient matrix, uses a bridge crack detection model to process the V channel suppressed image and the grayscale suppressed image, and uses the suspected coefficient matrix to apply weights to obtain bridge cracks, providing accurate weights for the crack detection model, thereby significantly improving the accuracy and reliability of crack detection.

Description

Technical Field

[0001] The invention relates to the technical field of image processing, and in particular to a bridge crack detection method. Background Art

[0002] As an important infrastructure, the safety of bridges is directly related to transportation and public safety. Traditional bridge crack detection methods mainly rely on manual visual inspection and traditional detection equipment, which have many limitations. At present, manual detection is inefficient, highly subjective, prone to missed detection and misjudgment, and has a limited detection range; while traditional instrument detection equipment is expensive and complex to operate, making it difficult to achieve rapid and accurate detection of large-scale bridges. In recent years, image processing technology has been widely used in the field of bridge crack detection, but existing image processing methods still have serious technical bottlenecks. These methods usually face the problem of complex lighting and severe background interference, resulting in unclear crack subtle features and low detection accuracy. Summary of the invention

[0003] In view of the above-mentioned deficiencies in the prior art, a bridge crack detection method provided by the present invention solves the problem of low crack detection accuracy in the prior art.

[0004] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is: a bridge crack detection method, comprising the following steps:

[0005] S1, transform the RGB bridge image into HSV space, extract the V channel value, and construct the V channel image;

[0006] S2. Suppressing the V channel value in the V channel image according to the brightness suppression factor of the pixel point to obtain a V channel suppressed image;

[0007] S3, graying the RGB bridge image, and suppressing the gray value in the gray image according to the gray suppression factor of the pixel point to obtain a gray suppression image;

[0008] S4, clustering the RGB bridge image to find the suspected area, and calculating the suspected coefficient for each pixel in the suspected area to construct a suspected coefficient matrix;

[0009] S5. Use the bridge crack detection model to process the V channel suppression image and the grayscale suppression image, and use the suspected coefficient matrix to apply weights to obtain the bridge crack.

[0010] Further, S2 includes the following sub-steps:

[0011] S21, calculating a brightness suppression factor of the pixel for the V channel value of each pixel in the V channel image, wherein the brightness suppression factor is a ratio of the V channel value of the pixel to the maximum V channel value;

[0012] S22. According to the brightness suppression factor of the pixel point, the V channel value is suppressed to obtain the V channel suppression value of the pixel point: , where V v,i is the V channel suppression value of the i-th pixel, α v,i is the brightness suppression factor of the i-th pixel, i is the number of the pixel;

[0013] S23, replacing the V channel value of the same pixel with the V channel suppression value to obtain a V channel suppression image.

[0014] Furthermore, S3 includes the following sub-steps:

[0015] S31, graying the RGB bridge image to obtain a grayscale image, and calculating a grayscale suppression factor of each pixel in the grayscale image for the grayscale value of each pixel, wherein the grayscale suppression factor is the ratio of the grayscale value of the pixel to the maximum grayscale value;

[0016] S32. Suppress the grayscale value according to the grayscale suppression factor of the pixel to obtain the grayscale suppression value of the pixel: , where V G,i is the grayscale suppression value of the i-th pixel, α G,i is the grayscale suppression factor of the i-th pixel, i is the number of the pixel;

[0017] S33, replacing the grayscale value of the same pixel with the grayscale suppression value to obtain a grayscale suppression image.

[0018] Further, S4 includes the following sub-steps:

[0019] S41, clustering the pixel values ​​of each pixel point on the RGB bridge image to obtain each clustering area;

[0020] S42, extracting the cluster area with the largest area as the bridge area;

[0021] S43, separating other clustering areas in the bridge area into multiple connected areas according to the positions of the pixel points;

[0022] S44, marking each connected area as a suspected area;

[0023] S45, calculating the shape complexity coefficient and the spatial dispersion coefficient according to the spatial distribution characteristics of the upper boundary pixel points of the suspected area;

[0024] S46, calculating the suspected coefficient of each pixel point on the suspected area according to the pixel value, shape complexity coefficient and spatial discrete coefficient of each suspected area, wherein the pixel value includes: R channel value, G channel value and B channel value;

[0025] S47, replacing the original pixel value with the suspected coefficient of each pixel point, assigning the pixel values ​​of other pixel points to 0, and constructing a suspected coefficient matrix with the same size as the RGB bridge image.

[0026] Furthermore, the formula for calculating the shape complexity coefficient in S45 is: , where S n is the shape complexity coefficient of the nth suspected area, x n,m is the horizontal coordinate of the mth boundary pixel point on the nth suspected area, x n,m+1 is the horizontal coordinate of the m+1th boundary pixel point on the nth suspected area, y n,m is the ordinate of the mth boundary pixel point on the nth suspected area, y n,m+1 is the ordinate of the m+1th boundary pixel point on the nth suspected area, m is the number of the boundary pixel point, n is the number of the suspected area, E n is the area of ​​the nth suspected region, and N is the number of pixels on the upper boundary of the suspected region.

[0027] Furthermore, the formula for calculating the spatial dispersion coefficient in S45 is: , where γ n is the spatial discrete coefficient of the nth suspected area, x n,m is the horizontal coordinate of the mth boundary pixel point on the nth suspected area, y n,m is the ordinate of the mth boundary pixel on the nth suspected area, | | is the absolute value operation, m is the number of the boundary pixel, n is the number of the suspected area, and N is the number of boundary pixels on the suspected area.

[0028] Furthermore, the formula for calculating the suspected coefficient in S46 is: , where ε n,i is the suspected coefficient of the ith pixel in the nth suspected area, S n is the shape complexity coefficient of the nth suspected area, γ n is the spatial discrete coefficient of the nth suspected area, R n,i is the R channel value of the ith pixel in the nth suspected area, G n,i is the G channel value of the i-th pixel in the n-th suspected area, B n,i is the B channel value of the i-th pixel in the n-th suspected area, i is the number of the pixel, and n is the number of the suspected area.

[0029] Further, the bridge crack detection model in S5 includes: a first convolutional layer, a second convolutional layer, a multiplier M1, a multiplier M2, a first residual neural network, a second residual neural network, an adder A1 and a multi-scale feature classification network;

[0030] The input end of the first convolutional layer is used to input the V channel suppressed image; the input end of the second convolutional layer is used to input the grayscale suppressed image;

[0031] The first input terminal of the multiplier M1 is connected to the output terminal of the first convolutional layer, the second input terminal thereof is used to input the suspected coefficient matrix, and the output terminal thereof is connected to the input terminal of the first residual neural network; the first input terminal of the multiplier M2 is connected to the output terminal of the second convolutional layer, the second input terminal thereof is used to input the suspected coefficient matrix, and the output terminal thereof is connected to the input terminal of the second residual neural network; the first input terminal of the adder A1 is connected to the output terminal of the first residual neural network, the second input terminal thereof is connected to the output terminal of the second residual neural network, and the output terminal thereof is connected to the input terminal of the multi-scale feature classification network; the output terminal of the multi-scale feature classification network serves as the output terminal of the bridge crack detection model.

[0032] Further, the multi-scale feature classification network includes: a third convolutional layer, a first maximum pooling layer, a second maximum pooling layer, a third maximum pooling layer, a first classification layer, a second classification layer, a third classification layer and a weighted output layer;

[0033] The input end of the third convolutional layer is used as the input end of the multi-scale feature classification network, and its output end is connected to the input end of the first maximum pooling layer, the input end of the second maximum pooling layer, and the input end of the third maximum pooling layer respectively;

[0034] The input end of the first classification layer is connected to the output end of the first maximum pooling layer; the input end of the second classification layer is connected to the output end of the second maximum pooling layer; the input end of the third classification layer is connected to the output end of the third maximum pooling layer; the input end of the weighted output layer is respectively connected to the output end of the first classification layer, the output end of the second classification layer and the output end of the third classification layer, and the output end of the weighted output layer serves as the output end of the multi-scale feature classification network.

[0035] Furthermore, the convolution kernel sizes of the first and second convolution layers are , the pooling kernel size of the first maximum pooling layer is , the pooling kernel size of the second maximum pooling layer is , the pooling kernel size of the third maximum pooling layer is .

[0036] In summary, the beneficial effects of the present invention are:

[0037] 1. The present invention can effectively enhance the visibility of cracks by converting RGB images into HSV space and extracting V channel. It combines brightness and grayscale suppression processing to highlight subtle crack features, reduce interference from other pixels, and significantly improve detection accuracy.

[0038] 2. The present invention automatically identifies suspected crack areas through clustering processing of RGB images, calculates suspected coefficients, constructs a suspected coefficient matrix, and applies the suspected coefficient matrix to the bridge crack detection model, so that the detection model can be more focused on potential crack areas, thereby improving the accuracy of crack detection.

[0039] 3. Starting from two images, namely, V-channel suppressed image and grayscale suppressed image, the present invention provides multi-dimensional input features, thereby enhancing the detection accuracy of the bridge crack detection model. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 is a flow chart of a bridge crack detection method;

[0041] Figure 2 It is the structural schematic diagram of the bridge crack detection model;

[0042] Figure 3 Schematic diagram of the structure of the multi-scale feature classification network. DETAILED DESCRIPTION

[0043] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0044] like Figure 1 As shown, a bridge crack detection method comprises the following steps:

[0045] S1, transform the RGB bridge image into HSV space, extract the V channel value, and construct the V channel image;

[0046] S2. Suppressing the V channel value in the V channel image according to the brightness suppression factor of the pixel point to obtain a V channel suppressed image;

[0047] S3, graying the RGB bridge image, and suppressing the gray value in the gray image according to the gray suppression factor of the pixel point to obtain a gray suppression image;

[0048] S4, clustering the RGB bridge image to find the suspected area, and calculating the suspected coefficient for each pixel in the suspected area to construct a suspected coefficient matrix;

[0049] S5. Use the bridge crack detection model to process the V channel suppression image and the grayscale suppression image, and use the suspected coefficient matrix to apply weights to obtain the bridge crack.

[0050] In this embodiment, S2 includes the following sub-steps:

[0051] S21, calculating a brightness suppression factor of the pixel for the V channel value of each pixel in the V channel image, wherein the brightness suppression factor is a ratio of the V channel value of the pixel to the maximum V channel value in the V channel image;

[0052] S22. According to the brightness suppression factor of the pixel point, the V channel value is suppressed to obtain the V channel suppression value of the pixel point: , where V v,i is the V channel suppression value of the i-th pixel, α v,i is the brightness suppression factor of the i-th pixel, i is the number of the pixel;

[0053] S23, replacing the V channel value of the same pixel with the V channel suppression value to obtain a V channel suppression image.

[0054] After converting the RGB image to the HSV color space, the V channel (brightness channel) can more effectively represent the brightness information of the image. By focusing on the V channel, subtle features in the image, such as cracks, can be more clearly extracted. This conversion helps to reduce the interference of color information on crack detection. By calculating the brightness suppression factor and adjusting the V channel value, the present invention can effectively enhance the contrast of cracks in the image, making subtle cracks more obvious and improving the accuracy of subsequent detection.

[0055] In this embodiment, S3 includes the following sub-steps:

[0056] S31, graying the RGB bridge image to obtain a gray image, and calculating a gray suppression factor of each pixel in the gray image for the gray value of each pixel, wherein the gray suppression factor is the ratio of the gray value of the pixel to the maximum gray value in the gray image;

[0057] S32. Suppress the grayscale value according to the grayscale suppression factor of the pixel to obtain the grayscale suppression value of the pixel: , where V G,i is the grayscale suppression value of the i-th pixel, α G,i is the grayscale suppression factor of the i-th pixel, i is the number of the pixel;

[0058] S33, replacing the grayscale value of the same pixel with the grayscale suppression value to obtain a grayscale suppression image.

[0059] The present invention can dynamically adjust the grayscale values ​​of different areas in the image by calculating the grayscale suppression factor of each pixel. This method can highlight the crack characteristics, reduce the influence of background noise, and make the crack more prominent.

[0060] The larger the pixel value, V channel value and grayscale value, the brighter the color and the greater the brightness of the imaged object. Therefore, the brightness suppression factor and grayscale suppression factor of the present invention make the larger the V channel value of the same pixel point, the smaller its V channel suppression value, and the larger the grayscale value, the smaller the grayscale suppression value. Since the V channel value and grayscale value of the crack are usually close to 0 when imaging, the suppression processing of steps S2 and S3 can highlight the crack characteristics of the bridge.

[0061] In this embodiment, S4 includes the following sub-steps:

[0062] S41, clustering the pixel values ​​of each pixel point on the RGB bridge image to obtain each clustering area;

[0063] S42, extracting the cluster area with the largest area as the bridge area;

[0064] S43, separating other clustering areas in the bridge area into multiple connected areas according to the positions of the pixel points;

[0065] S44, marking each connected area as a suspected area;

[0066] S45, calculating the shape complexity coefficient and the spatial dispersion coefficient according to the spatial distribution characteristics of the upper boundary pixel points of the suspected area;

[0067] S46, calculating the suspected coefficient of each pixel point on the suspected area according to the pixel value, shape complexity coefficient and spatial discrete coefficient of each suspected area, wherein the pixel value includes: R channel value, G channel value and B channel value;

[0068] S47, replacing the original pixel value with the suspected coefficient of each pixel point, assigning the pixel values ​​of other pixel points to 0, and constructing a suspected coefficient matrix with the same size as the RGB bridge image.

[0069] In the present invention, the boundary pixel point is the edge pixel point on the suspected area.

[0070] Step S41 of the present invention performs clustering processing on each pixel point, so as to classify the pixel points with similar pixel values ​​into one category, and use the cluster area with the largest area as the bridge area. Since clustering processing can only classify the pixel points with similar pixel values ​​into one area, there are cases where the pixel points with similar pixel values ​​are separated by the pixel points in other areas. Therefore, the other cluster areas are separated into multiple connected areas (a connected area is an area composed of pixel points with similar pixel values ​​and adjacent positions in the image. A connected area is a complete area that is not separated by the pixel points in other areas). That is, after clustering to obtain multiple cluster areas, the cluster areas separated by pixel points are split into multiple connected areas. Since the crack is located on the bridge and is located inside the bridge in the image, the other cluster areas in the bridge area are processed to reduce data interference.

[0071] According to the spatial distribution characteristics of the boundary pixels on the suspected area, the shape complexity coefficient and spatial dispersion coefficient are calculated to characterize the morphological characteristics of the suspected area. Then, based on the pixel values ​​of the pixels on the suspected area, the suspected coefficient of each pixel is calculated. The larger the suspected coefficient, the greater the possibility that the pixel is a pixel on the crack.

[0072] In this embodiment, the formula for calculating the shape complexity coefficient in S45 is: , where S n is the shape complexity coefficient of the nth suspected area, x n,m is the horizontal coordinate of the mth boundary pixel point on the nth suspected area, x n,m+1 is the horizontal coordinate of the m+1th boundary pixel point on the nth suspected area, y n,m is the ordinate of the mth boundary pixel point on the nth suspected area, y n,m+1 is the ordinate of the m+1th boundary pixel point on the nth suspected area, m is the number of the boundary pixel point, n is the number of the suspected area, E n is the area of ​​the nth suspected region, and N is the number of pixels on the upper boundary of the suspected region.

[0073] The present invention reflects the irregularity of the suspected area by calculating the ratio of the perimeter to the area of ​​the suspected area. The larger the shape complexity coefficient is, the more irregular the shape is, which reflects the degree of tortuosity of the edge of the area and is used to measure the complexity of the edge of the area.

[0074] In this embodiment, the formula for calculating the spatial discrete coefficient in S45 is: , where γ n is the spatial discrete coefficient of the nth suspected area, x n,m is the horizontal coordinate of the mth boundary pixel point on the nth suspected area, y n,mis the ordinate of the mth boundary pixel on the nth suspected area, | | is the absolute value operation, m is the number of the boundary pixel, n is the number of the suspected area, and N is the number of boundary pixels on the suspected area.

[0075] The present invention calculates the spatial dispersion coefficient through the difference between the horizontal and vertical coordinates of the boundary pixel points and the mean value, reflecting the discrete situation of the distribution of the boundary pixel points.

[0076] In this embodiment, the formula for calculating the suspected coefficient in S46 is: , where ε n,i is the suspected coefficient of the ith pixel in the nth suspected area, S n is the shape complexity coefficient of the nth suspected area, γ n is the spatial discrete coefficient of the nth suspected area, R n,i is the R channel value of the ith pixel in the nth suspected area, G n,i is the G channel value of the i-th pixel in the n-th suspected area, B n,i is the B channel value of the i-th pixel in the n-th suspected area, i is the number of the pixel, and n is the number of the suspected area.

[0077] The present invention calculates the suspected coefficient of each pixel point according to the pixel value of each pixel point on the suspected area and the boundary shape of the suspected area. The larger the suspected coefficient is, the greater the possibility that the pixel point is a crack pixel point is.

[0078] like Figure 2 As shown, the bridge crack detection model in S5 includes: a first convolutional layer, a second convolutional layer, a multiplier M1, a multiplier M2, a first residual neural network, a second residual neural network, an adder A1 and a multi-scale feature classification network;

[0079] The input end of the first convolutional layer is used to input the V channel suppressed image; the input end of the second convolutional layer is used to input the grayscale suppressed image;

[0080] The first input terminal of the multiplier M1 is connected to the output terminal of the first convolutional layer, the second input terminal thereof is used to input the suspected coefficient matrix, and the output terminal thereof is connected to the input terminal of the first residual neural network; the first input terminal of the multiplier M2 is connected to the output terminal of the second convolutional layer, the second input terminal thereof is used to input the suspected coefficient matrix, and the output terminal thereof is connected to the input terminal of the second residual neural network; the first input terminal of the adder A1 is connected to the output terminal of the first residual neural network, the second input terminal thereof is connected to the output terminal of the second residual neural network, and the output terminal thereof is connected to the input terminal of the multi-scale feature classification network; the output terminal of the multi-scale feature classification network serves as the output terminal of the bridge crack detection model.

[0081] The present invention uses a first convolutional layer to process a V-channel suppressed image, and uses a second convolutional layer to process a grayscale suppressed image, so as to combine image information of two dimensions, and applies a suspected coefficient matrix to corresponding features through multipliers M1 and M2, so as to make the features belonging to cracks more significant and reduce non-crack features, and then uses a first residual neural network and a second residual neural network to reduce the problem of gradient vanishing, and adds corresponding features through adder A1 to further highlight crack features, and uses a multi-scale feature classification network to classify the features output by adder A1 to realize the identification of bridge cracks.

[0082] In the present invention, both the multiplier M1 and the multiplier M2 perform element-by-element multiplication, and the adder A1 performs element-by-element addition.

[0083] like Figure 3 As shown, the multi-scale feature classification network includes: a third convolution layer, a first maximum pooling layer, a second maximum pooling layer, a third maximum pooling layer, a first classification layer, a second classification layer, a third classification layer and a weighted output layer;

[0084] The input end of the third convolutional layer is used as the input end of the multi-scale feature classification network, and its output end is connected to the input end of the first maximum pooling layer, the input end of the second maximum pooling layer, and the input end of the third maximum pooling layer respectively;

[0085] The input end of the first classification layer is connected to the output end of the first maximum pooling layer; the input end of the second classification layer is connected to the output end of the second maximum pooling layer; the input end of the third classification layer is connected to the output end of the third maximum pooling layer; the input end of the weighted output layer is respectively connected to the output end of the first classification layer, the output end of the second classification layer and the output end of the third classification layer, and the output end of the weighted output layer serves as the output end of the multi-scale feature classification network.

[0086] In this embodiment, the convolution kernel sizes of the first convolution layer and the second convolution layer are , the pooling kernel size of the first maximum pooling layer is , the pooling kernel size of the second maximum pooling layer is , the pooling kernel size of the third maximum pooling layer is The pooling kernel is the window size selected by the pooling layer.

[0087] Since the convolution kernel sizes of the first convolution layer and the second convolution layer of the present invention are Therefore, after the first and second convolutional layers are processed, the size of the image features does not change. The corresponding features are multiplied by the suspected coefficient matrix, that is, weights are applied to make the crack features more prominent.

[0088] The present invention uses three maximum pooling layers of different scales to extract the features of the output of the third convolutional layer, and then uses three classification layers to classify to obtain the classification value of each channel, and then uses the weighted output layer to weight the three classification values ​​to obtain the final output, thereby improving the accuracy of crack detection.

[0089] In this embodiment, the first classification layer, the second classification layer, the third classification layer, and the weighted output layer may use a Sigmoid function or a fully connected layer. When the classification value output by the weighted output layer is greater than a threshold, it can be determined that there is a crack on the crack image.

[0090] In this embodiment, the bridge crack detection model is trained using the gradient descent method.

[0091] The present invention can effectively enhance the visibility of cracks by converting RGB images into HSV space and extracting V channel. It combines brightness and grayscale suppression processing to highlight subtle crack features, reduce interference from other pixels, and significantly improve detection accuracy.

[0092] The present invention automatically identifies suspected crack areas through clustering processing of RGB images, calculates suspected coefficients, constructs a suspected coefficient matrix, and applies the suspected coefficient matrix to a bridge crack detection model, so that the detection model can be more focused on potential crack areas, thereby improving the accuracy of crack detection.

[0093] The present invention starts from two images, namely, a V-channel suppressed image and a grayscale suppressed image, provides multi-dimensional input features, and enhances the detection accuracy of the bridge crack detection model.

[0094] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A bridge crack detection method, characterized in that: The following steps are involved: S1, transform the RGB bridge image into HSV space, extract the V channel value, and construct the V channel image; S2. Suppressing the V channel value in the V channel image according to the brightness suppression factor of the pixel point to obtain a V channel suppressed image; S3, graying the RGB bridge image, and suppressing the gray value in the gray image according to the gray suppression factor of the pixel point to obtain a gray suppression image; S4, clustering the RGB bridge image to find the suspected area, and calculating the suspected coefficient for each pixel in the suspected area to construct a suspected coefficient matrix; S5, using the bridge crack detection model to process the V channel suppression image and the grayscale suppression image, and using the suspected coefficient matrix to apply weights to obtain the bridge crack; The S2 comprises the following sub-steps: S21, calculating a brightness suppression factor of the pixel for the V channel value of each pixel in the V channel image, wherein the brightness suppression factor is a ratio of the V channel value of the pixel to the maximum V channel value; S22. According to the brightness suppression factor of the pixel point, the V channel value is suppressed to obtain the V channel suppression value of the pixel point: , where V v,i is the V channel suppression value of the i-th pixel, α v,i is the brightness suppression factor of the i-th pixel, i is the number of the pixel; S23, replacing the V channel value of the same pixel with the V channel suppression value to obtain a V channel suppression image; The S3 comprises the following sub-steps: S31, graying the RGB bridge image to obtain a grayscale image, and calculating a grayscale suppression factor of each pixel in the grayscale image for the grayscale value of each pixel, wherein the grayscale suppression factor is the ratio of the grayscale value of the pixel to the maximum grayscale value; S32. Suppress the grayscale value according to the grayscale suppression factor of the pixel to obtain the grayscale suppression value of the pixel: , where V G,i is the grayscale suppression value of the i-th pixel, α G,i is the grayscale suppression factor of the i-th pixel, i is the number of the pixel; S33, replacing the grayscale value of the same pixel with the grayscale suppression value to obtain a grayscale suppression image; The S4 comprises the following sub-steps: S41, clustering the pixel values ​​of each pixel point on the RGB bridge image to obtain each clustering area; S42, extracting the cluster area with the largest area as the bridge area; S43, separating other clustering areas in the bridge area into multiple connected areas according to the positions of the pixel points; S44, marking each connected area as a suspected area; S45, calculating the shape complexity coefficient and the spatial dispersion coefficient according to the spatial distribution characteristics of the upper boundary pixel points of the suspected area; S46, calculating the suspected coefficient of each pixel point on the suspected area according to the pixel value, shape complexity coefficient and spatial discrete coefficient of each suspected area, wherein the pixel value includes: R channel value, G channel value and B channel value; S47, replacing the original pixel value with the suspected coefficient of each pixel point, assigning the pixel values ​​of other pixel points to 0, and constructing a suspected coefficient matrix with the same size as the RGB bridge image; The formula for calculating the shape complexity coefficient in S45 is: , where S n is the shape complexity coefficient of the nth suspected area, x n,m is the horizontal coordinate of the mth boundary pixel point on the nth suspected area, x n,m+1 is the horizontal coordinate of the m+1th boundary pixel point on the nth suspected area, y n,m is the ordinate of the mth boundary pixel point on the nth suspected area, y n,m+1 is the ordinate of the m+1th boundary pixel point on the nth suspected area, m is the number of the boundary pixel point, n is the number of the suspected area, E n is the area of ​​the nth suspected region, and N is the number of pixels on the upper boundary of the suspected region; The formula for calculating the spatial dispersion coefficient in S45 is: , where γ n is the spatial discrete coefficient of the nth suspected area, x n,m is the horizontal coordinate of the mth boundary pixel point on the nth suspected area, y n,m is the ordinate of the mth boundary pixel on the nth suspected region, | | is the absolute value operation, m is the number of the boundary pixel, n is the number of the suspected region, and N is the number of boundary pixels on the suspected region; The formula for calculating the suspected coefficient in S46 is: , where ε n,i is the suspected coefficient of the ith pixel in the nth suspected area, S n is the shape complexity coefficient of the nth suspected area, γ n is the spatial discrete coefficient of the nth suspected area, R n,i is the R channel value of the ith pixel in the nth suspected area, G n,i is the G channel value of the i-th pixel in the n-th suspected area, B n,i is the B channel value of the i-th pixel on the n-th suspected area, i is the number of the pixel, and n is the number of the suspected area; The bridge crack detection model in S5 includes: a first convolutional layer, a second convolutional layer, a multiplier M1, a multiplier M2, a first residual neural network, a second residual neural network, an adder A1 and a multi-scale feature classification network; The input end of the first convolutional layer is used to input the V channel suppressed image; the input end of the second convolutional layer is used to input the grayscale suppressed image; The first input end of the multiplier M1 is connected to the output end of the first convolutional layer, the second input end thereof is used to input the suspected coefficient matrix, and the output end thereof is connected to the input end of the first residual neural network; The first input end of the multiplier M2 is connected to the output end of the second convolutional layer, the second input end is used to input the suspected coefficient matrix, and the output end is connected to the input end of the second residual neural network; The first input end of the adder A1 is connected to the output end of the first residual neural network, the second input end is connected to the output end of the second residual neural network, and the output end is connected to the input end of the multi-scale feature classification network; The output end of the multi-scale feature classification network is used as the output end of the bridge crack detection model.

2. The bridge crack detection method according to claim 1, characterized in that: The multi-scale feature classification network includes: a third convolution layer, a first maximum pooling layer, a second maximum pooling layer, a third maximum pooling layer, a first classification layer, a second classification layer, a third classification layer and a weighted output layer; The input end of the third convolutional layer serves as the input end of the multi-scale feature classification network, and the output end thereof is respectively connected to the input end of the first maximum pooling layer, the input end of the second maximum pooling layer, and the input end of the third maximum pooling layer; The input end of the first classification layer is connected to the output end of the first maximum pooling layer; the input end of the second classification layer is connected to the output end of the second maximum pooling layer; the input end of the third classification layer is connected to the output end of the third maximum pooling layer; The input end of the weighted output layer is respectively connected to the output end of the first classification layer, the output end of the second classification layer and the output end of the third classification layer, and its output end serves as the output end of the multi-scale feature classification network.

3. The bridge crack detection method according to claim 2, characterized in that: The convolution kernel sizes of the first and second convolution layers are , the pooling kernel size of the first maximum pooling layer is , the pooling kernel size of the second maximum pooling layer is , the pooling kernel size of the third maximum pooling layer is .

Images