Intelligent concrete vibrating method based on support vector machine

Abstract

The application discloses a concrete intelligent vibration method based on a support vector machine, and comprises the following steps: collecting concrete images around a vibrating rod in the state of artificial vibration mode missing vibration, lack of vibration and vibration moderation, and taking the images as samples to perform support vector machine training; discharging concrete to a construction area through a pumping device; starting a vibrating motor and a camera at the same time, collecting concrete images in a vibrating process, and uploading the images to a computer in real time; inputting the concrete images into the trained support vector machine model to obtain the position of the real-time concrete vibrating image in the support vector machine training space; closing the vibrating motor, slowly pulling out the vibrating rod, and vibrating concrete in the next area. The application can solve the problems that the aggregate particle shaping effect depends on artificial operation, cannot be processed in batches, and is low in efficiency. The application combines image recognition gray level co-occurrence matrix calculation, image direction gradient feature extraction and support vector machine machine learning technology to control a concrete vibrating process.

Description

Technical Field

[0001] This invention relates to the technical field of building engineering, and in particular to a method for intelligent vibration compaction of concrete based on support vector machines. Background Technology

[0002] Concrete, an essential material in construction engineering, significantly impacts the overall strength of buildings through its production process and pouring quality. The concrete pouring process typically involves scaffolding erection, formwork support, concrete slump monitoring, concrete pouring, concrete vibration, concrete polishing, and concrete curing. Concrete vibration is a crucial component of the pouring process. Insufficient or inadequate vibration can lead to quality issues such as honeycomb texture and voids, while excessive vibration can cause localized segregation, separating coarse and fine aggregates and resulting in a loss of concrete strength.

[0003] Currently, concrete is typically vibrated manually using hand-held vibrators, relying on construction workers' experience to determine if the vibration is adequate, making it difficult to guarantee pouring quality. With the advancement of intelligent construction, using robots to replace manual labor has become a future trend. This necessitates an intelligent method to direct robots in concrete vibration operations. How to transfer human experience to robots has become a challenge for construction technicians in the intelligent era. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the technical problem to be solved by this invention is to provide a concrete intelligent vibration method based on support vector machines, which combines image recognition gray-level co-occurrence matrix calculation, image orientation gradient feature extraction, and support vector machine machine learning technology to control the concrete vibration process using information technology.

[0005] To achieve the above objectives, the present invention employs the following technical measures:

[0006] A method for intelligent concrete vibration compaction based on support vector machines includes the following steps:

[0007] S10. Collect images of the concrete around the vibrator under conditions of missed vibration, insufficient vibration, and moderate vibration in manual vibration mode, and use these images as samples for support vector machine training.

[0008] S20. Concrete is unloaded into the construction area using pumping equipment;

[0009] S30. Start the vibratory motor and simultaneously start the camera to capture images of the concrete during the vibration process and upload them to the computer in real time.

[0010] S40. Input the concrete image into the support vector machine model trained in step S10 to obtain the position of the real-time concrete vibration image in the support vector machine training space.

[0011] S50. Continue vibrating until the real-time concrete vibration image is located in the appropriate vibration area in the support vector machine training space. Then, turn off the vibration motor, slowly pull out the vibrator, and proceed with the next area of ​​concrete vibration.

[0012] Furthermore, the support vector machine training in step S10 includes the following steps:

[0013] S11. Divide the images collected in step S10 into a training group and a test group, accounting for 80% and 20% of the total number of images, respectively.

[0014] S12. Calculate the gray-level co-occurrence matrix of the training group and the test group images at 0°, 45°, 90° and 135° respectively, and normalize it to obtain the gray-level co-occurrence matrix feature vector;

[0015] S13. Calculate the normalized image orientation gradient feature vectors for the training group and the test group respectively.

[0016] S14. Combine the image orientation gradient feature vector and the gray-level co-occurrence matrix feature vector to obtain the support vector machine feature vector;

[0017] S15. Define the support vector machine feature values ​​for the training group images. The feature value is 1 for images with missing vibration or insufficient vibration, and -1 for images with moderate vibration.

[0018] S16. Construct a training space with the same dimension as the feature vectors of the support vector machine.

[0019] S17. Based on the support vector machine feature vector components, determine the position of the training group images in the training space. For example, the support vector machine feature vector corresponding to a certain image is (1, 2, 3, 4, 5, 6). T If so, the image is located at (1, 2, 3, 4, 5, 6) in the training space;

[0020] S18. In the training space, construct the hyperplane f(x) = w T x i +b, where w T Let x be the normal vector of the hyperplane, b be the offset of the hyperplane relative to the origin, and x be the normal vector of the hyperplane. i The independent variables are hyperplane variables, with the form [x1, x2, x3, ..., xn]. The number of independent variable components is the same as the dimension of the feature vectors of the support vector machine.

[0021] S19. Solve under constraint yi(w) T x i +b)≥1 below, The normal vector w of the hyperplane that reaches its maximum value T The hyperplane is offset from the origin by b, ||w|| is the second norm of the hyperplane normal vector, and yi is the feature value of the image support vector machine. The feature value is 1 for images with missing vibration or lack of vibration, and -1 for images with moderate vibration. The hyperplane obtained at this time can divide the training space into two parts. The positions of the images with feature values ​​of 1 and -1 in the training space are separated by the hyperplane.

[0022] S110. Verify the hyperplane separation effect in the training space. Calculate the support vector machine feature vectors of the test group images and use them to determine their positions in the training space. The test group images include images with missing vibration, images with insufficient vibration, and images with moderate vibration. If the test group images are separated into two parts by the hyperplane in the training space, the support vector machine training is complete. If the test group images do not meet this condition, the training fails or the training accuracy is insufficient. In this case, the number of training and test group images should be increased, and the support vector machine training should be repeated.

[0023] Furthermore, the calculation of the image gray-level co-occurrence matrix eigenvector in step S12 includes the following steps:

[0024] S121. Read the values ​​of the red, green and blue channels of the image, and use them to convert the image to grayscale to obtain the grayscale value of each pixel.

[0025] S122. Set the binarization value of pixels with a gray value greater than 122 to 1, and the gray value of pixels with a gray value less than 122 to 0, to obtain the binarization matrix.

[0026] S123. Calculate the horizontal gray-level co-occurrence matrix. For a binary gray-level image, the horizontal gray-level co-occurrence matrix has an order of 2 and has four components: (0,0), (0,1), (1,0), and (1,1). Taking the (0,1) component as an example, its calculation method is as follows: If there are pixels in the binary matrix whose corresponding binary values ​​are 0 and 1 and are horizontally adjacent, then this pair of pixels is defined as a pair, and the number of pairs in the binary matrix that meet this definition is counted.

[0027] S124. Calculate the gray-level co-occurrence matrix in the 45° direction. For a binary gray-level image, the gray-level co-occurrence matrix in the 45° direction is of order 2 and has four components: (0,0), (0,1), (1,0), and (1,1). Taking the (0,1) component as an example, its calculation method is as follows: If there are pixels in the binary matrix whose corresponding binary values ​​are 0 and 1 and are adjacent at 45°, then this pair of pixels is defined as a pair, and the number of pairs in the binary matrix that meet this definition is counted.

[0028] S125. Calculate the gray-level co-occurrence matrix in the 90° direction. For a binary gray-level image, the gray-level co-occurrence matrix in the 90° direction is of order 2 and has four components: (0,0), (0,1), (1,0), and (1,1). Taking the (0,1) component as an example, its calculation method is as follows: If there are pixels in the binary matrix whose corresponding binary values ​​are 0 and 1 and are 90° vertically adjacent, then this pair of pixels is defined as a pair, and the number of pairs in the binary matrix that meet this definition is counted.

[0029] S126. Calculate the gray-level co-occurrence matrix in the 135° direction. For a binary gray-level image, the gray-level co-occurrence matrix in the 135° direction is of order 2 and has four components: (0,0), (0,1), (1,0), and (1,1). Taking the (0,1) component as an example, its calculation method is as follows: If there are pixels in the binary matrix whose corresponding binary values ​​are 0 and 1 and are adjacent in a 135° diagonal direction, then this pair of pixels is defined as a pair, and the number of pairs in the binary matrix that meet this definition is counted.

[0030] S127. Calculate the numerical contrast, correlation, angular second moment, and inverse difference moment of the four components of the gray-level co-occurrence matrix in each direction, and calculate the mean of the numerical contrast (gmean1), the mean of the correlation (gmean2), the mean of the angular second moment (gmean3), the mean of the inverse difference moment (gmean4), the standard deviation of the contrast (gstd1), the standard deviation of the correlation (gstd2), the standard deviation of the angular second moment (gstd3), and the standard deviation of the inverse difference moment (gstd4) of the four components of the gray-level co-occurrence matrix in each direction.

[0031] S128. Define the gray-level co-occurrence matrix eigenvector, which contains eight components: the mean contrast (gmean1), mean correlation (gmean2), mean angular second moment (gmean3), mean inverse difference moment (gmean4), standard deviation of contrast (gstd1), standard deviation of correlation (gstd2), standard deviation of angular second moment (gstd3), and standard deviation of inverse difference moment (gstd4). The eigenvector of the image gray-level co-occurrence matrix is ​​in the form of (gmean1, gmean2, gmean3, gmean4, gstd1, gstd2, gstd3, gstd4). T .

[0032] Furthermore, the calculation of the normalized image orientation gradient feature vector in step S13 includes the following steps:

[0033] S131. Read the values ​​of the red, green and blue channels of the image, and use them to convert the image to grayscale to obtain the grayscale value of each pixel.

[0034] S132. Set the binarization value of pixels with a gray value greater than 122 to 1, and the gray value of pixels with a gray value less than 122 to 0, to obtain the binarization matrix.

[0035] S133. Based on the binarization matrix, calculate the horizontal and vertical binarization gradients of each pixel in the image. The calculation method is as follows: T p =R(x, y+1) - R(x, y-1), T h =R(x+1,y)-R(x-1,y). In the formula, T h For the horizontal binarized gradient, T p R is the vertical binarized gradient, x is the number of rows in the binarized matrix, and y is the number of columns in the binarized matrix.

[0036] S134. Based on the horizontal and vertical binarization gradients of each pixel, calculate the direction and magnitude of the binarization gradient at each point. The calculation method is as follows: In the formula, T is the magnitude of the binarized gradient, and α T The direction of the binary gradient;

[0037] S135. Divide the image into square blocks with a side length of n pixels, so that each block contains n*n pixels, and remove the parts of the image that cannot be divided into whole blocks after the blocks are divided.

[0038] S136, α T The binarized gradient direction takes the value [0, 360°]. This interval is divided into m equal parts to obtain m directional gradient statistical angle intervals. The directional gradient statistical value of each block's gradient statistical angle interval is calculated. The specific method is as follows: based on the binarized gradient direction α of each pixel in the block... T The pixels are classified into gradient statistical angle intervals in each direction. The binarized gradient magnitude T of the pixels in each gradient statistical angle interval is summed in a counterclockwise direction to obtain the gradient statistical angle interval directional gradient statistical value.

[0039] S137. Divide the image into Q computational regions by taking four adjacent blocks as one computational region. Taking a resolution of 16*16 pixels and a block size of 4*4 as an example, the image is divided into (16 / 4-1)*(16 / 4-1)=9 computational regions in this way.

[0040] S138. Combine the statistical values ​​of the gradient angle intervals of each block within the calculation region to obtain the directional gradient feature vector of each calculation region. Taking the statistical values ​​of the gradient angle intervals of the four blocks within a certain calculation region as (1, 2, 3), (4, 5, 6), (7, 8, 9), and (10, 11, 12) as an example, the directional gradient feature vector of this calculation region is (1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12).

[0041] S139. Combine the directional gradient feature vectors of each calculation region to obtain the image directional gradient feature vector;

[0042] S1310. Calculate the sum of squares of each component of the image orientation gradient feature vector, take the square root of the sum of squares to obtain the orientation gradient normalization target value, divide each component of the image orientation gradient feature vector by the orientation gradient normalization target value to obtain the normalized image orientation gradient feature vector.

[0043] In summary, compared with existing traditional methods, the beneficial effects and advantages of the intelligent concrete vibration method based on support vector machines of the present invention are as follows:

[0044] By combining image recognition gray-level co-occurrence matrix calculation, image directional gradient feature extraction, and support vector machine (SVM) machine learning techniques, information technology is used to control the concrete vibration process. The use of image recognition and machine learning methods solves the problem of reliance on manual concrete vibration, providing an intelligent concrete vibration technology and laying the groundwork for robotic vibration operations. The SVM-based intelligent concrete vibration device is equipped with a 4G transmission module, enabling it to meet the requirements of long-distance image transmission. Computer and construction technicians do not need to be on-site, reducing management costs. The device is also equipped with LED lights to meet the needs of nighttime construction or under insufficient lighting conditions. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 This is a flowchart of the intelligent concrete vibration method based on support vector machine according to the present invention;

[0047] Figure 2 This is a flowchart of the image gray-level co-occurrence matrix eigenvector calculation process according to the present invention;

[0048] Figure 3 This is a flowchart of the image orientation gradient feature vector calculation process of the present invention;

[0049] Figure 4 This is a flowchart of the support vector machine training process of the present invention;

[0050] Figure 5 These are example images for image analysis of Examples 1 and 2;

[0051] Figure 6 This is a block diagram of Example 2;

[0052] Figure 7 Images of the training group in Example 3;

[0053] Figure 8 Images of the test group in Example 3;

[0054] Figure 9 Images of the initial state of the concrete in Example 4;

[0055] Figure 10 Images of the concrete after vibration were captured in Example 4.

[0056] Figure 11 This is a schematic diagram of a concrete intelligent vibration device based on support vector machines. Detailed Implementation

[0057] Example 1

[0058] This embodiment aims to illustrate the calculation steps of the image gray-level co-occurrence matrix eigenvector. To clearly describe the calculation steps, this embodiment uses a 16*16 resolution image as an example. The calculation object in this embodiment is as follows: Figure 5 As shown, the steps for calculating the eigenvectors of the image gray-level co-occurrence matrix are as follows:

[0059] G1. Read the values ​​of the red, green, and blue channels of the image, and convert the image to grayscale to obtain the grayscale value of each pixel. The grayscale values ​​of the image in this embodiment are shown in the table below:

[0060] Table 1. Image Gray Value Table of Example 1

[0061] 0 2 254 254 255 254 255 255 5 3 255 255 254 255 248 255 0 9 254 255 255 255 255 253 0 3 252 252 254 255 255 248 255 251 10 255 252 255 255 255 7 0 255 255 255 253 254 255 255 255 255 12 1 247 255 255 6 5 253 255 252 255 255 255 255 255 248 1 10 255 255 255 4 4 254 252 255 254 249 255 254 255 255 251 252 7 249 255 6 0 4 250 253 255 251 255 253 255 255 255 255 252 9 1 0 0 9 255 255 0 4 0 255 255 254 254 253 254 4 4 0 7 4 253 255 3 0 4 255 255 255 252 255 255 0 3 0 0 1 3 4 246 255 255 255 250 254 250 252 255 11 0 2 0 2 0 0 255 255 255 253 255 255 11 18 0 252 255 10 13 7 255 253 252 251 254 252 255 0 0 0 255 255 254 2 0 0 3 7 254 255 255 255 243 3 11 0 254 252 255 7 0 4 2 0 254 252 254 0 11 249 255 254 252 255 253 3 0 8 254 255 1 251 255 253 250 255 250 254 254 255 254 4 17 252 249 255 255 1 13 255 253 254 254 255 255 252 254 11 3 255 253 255 251 15 1

[0062] G2. Set the binarization value of pixels with a gray value greater than 122 to 1, and the gray value of pixels with a gray value less than 122 to 0, to obtain the binarization matrix.

[0063] Table 2 Binarization Matrix of Example 1

[0064]

[0065]

[0066] G3. Calculate the horizontal gray-level co-occurrence matrix. For a binary gray-level image, the horizontal gray-level co-occurrence matrix is ​​of order 2 and has four components: (0,0), (0,1), (1,0), and (1,1). Taking the (0,1) component as an example, its calculation method is as follows: If there are pixels in the binary matrix whose corresponding binary values ​​are 0 and 1 and are horizontally adjacent, then this pair of pixels is defined as a pair. Count the number of pairs in the binary matrix that meet this definition. In this embodiment, the four horizontal components are 58, 27, 28, and 127, respectively.

[0067] G4. Calculate the 45° direction gray-level co-occurrence matrix. For a binary grayscale image, the 45° direction gray-level co-occurrence matrix is ​​of order 2, with four components: (0, 0), (0, 1), (1, 0), and (1, 1). Taking the (0, 1) component as an example, its calculation method is as follows: If there are pixels in the binary matrix whose corresponding binary values ​​are 0 and 1 and are diagonally adjacent at 45°, then this pair of pixels is defined as a pair, and the number of pairs in the binary matrix that meet this definition is counted. In this embodiment, the four components of the 45° direction are 50, 31, 32, and 112, respectively.

[0068] G5. Calculate the 90° direction gray-level co-occurrence matrix. For a binary grayscale image, the 90° direction gray-level co-occurrence matrix is ​​of order 2, with four components: (0, 0), (0, 1), (1, 0), and (1, 1). Taking the (0, 1) component as an example, its calculation method is as follows: If there are pixels in the binary matrix whose corresponding binary values ​​are 0 and 1 and are 90° vertically adjacent, then this pair of pixels is defined as a pair, and the number of pairs in the binary matrix that meet this definition is counted. In this embodiment, the four components of the 90° direction are 62, 23, 23, and 132, respectively.

[0069] G6. Calculate the gray-level co-occurrence matrix (GLCM) in the 135° direction. For a binary grayscale image, the GLCM in the 135° direction is of order 2 and has four components: (0, 0), (0, 1), (1, 0), and (1, 1). Taking the (0, 1) component as an example, its calculation method is as follows: If there are pixels in the binary matrix whose corresponding binary values ​​are 0 and 1 and are diagonally adjacent at 135°, then this pair of pixels is defined as a pair. Count the number of pairs in the binary matrix that meet this definition. The four components in the 135° direction are 51, 32, 31, and 111.

[0070] G7. Calculate the numerical contrast, correlation, angular second moment, and inverse difference moment of the four components of the gray-level co-occurrence matrix for each direction, and calculate the mean of the numerical contrast (gmean1), the mean of the correlation (gmean2), the mean of the angular second moment (gmean3), the mean of the inverse difference moment (gmean4), the standard deviation of the contrast (gstd1), the standard deviation of the correlation (gstd2), the standard deviation of the angular second moment (gstd3), and the standard deviation of the inverse difference moment (gstd4) of the four components of the gray-level co-occurrence matrix for each direction. In this embodiment, the four components of the gray-level co-occurrence matrix in each direction have the following numerical values: the mean contrast (gmean1) is 0.245, the mean correlation (gmean2) is 0.468, the mean angular second moment (gmean3) is 0.356, the mean inverse difference moment (gmean4) is 0.877, the standard deviation of contrast (gstd1) is 0.043, the standard deviation of correlation (gstd2) is 0.090, the standard deviation of angular second moment (gstd3) is 0.025, and the standard deviation of inverse difference moment (gstd4) is 0.021.

[0071] G8. Define the gray-level co-occurrence matrix eigenvector, which contains eight components: the mean contrast (gmean1), mean correlation (gmean2), mean angular second moment (gmean3), mean inverse difference moment (gmean4), and standard deviations of contrast (gstd1, gstd2, gstd3, and gstd4) for each direction. In this embodiment, the image gray-level co-occurrence matrix eigenvector is in the form of (0.245, 0.468, 0.356, 0.877, 0.043, 0.090, 0.025, 0.021). T .

[0072] Example 2

[0073] This embodiment aims to illustrate the calculation steps of the directional gradient normalized eigenvector. To clearly describe the calculation steps, this embodiment uses a 16*16 resolution image as an example. The calculation object in this embodiment is as follows: Figure 5 As shown.

[0074] H1. Read the values ​​of the red, green, and blue channels of the image and convert the image to grayscale to obtain the grayscale value of each pixel. The grayscale value of each pixel is shown in Table 1 of Example 1.

[0075] H2. Set the binarization value of pixels with a gray value greater than 122 to 1, and the gray value of pixels with a gray value less than 122 to 0, to obtain the binarization matrix. The binarization matrix is ​​shown in Table 2 of Example 1.

[0076] H3. Based on the binarization matrix, calculate the horizontal and vertical binarization gradients of each pixel in the image. The calculation method is as follows: T p =R(x, y+1) - R(x, y-1), T h =R(x+1,y)-R(x-1,y). In the formula T h For the horizontal binarized gradient, T p R represents the vertical binarization gradient, x represents the number of rows in the binarization matrix, and y represents the number of columns in the binarization matrix. The horizontal and vertical binarization gradients for each pixel in this embodiment are shown in Tables 3 and 4.

[0077] Table 3 Horizontal Binarization Gradient of Example 2

[0078]

[0079]

[0080] Table 4. Vertical Binarization Gradient in Example 2

[0081] 0 0 1 1 1 1 1 1 0 0 1 1 1 1 1 1 1 1 -1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 0 -1 -1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 -1 -1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 -1 0 0 0 0 -1 0 0 0 0 0 0 0 0 1 1 0 -1 -1 0 0 -1 0 0 -1 -1 -1 0 0 0 0 0 1 -1 -1 0 0 0 0 0 -1 -1 -1 0 0 0 0 0 0 0 0 0 0 0 -1 -1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 -1 -1 1 1 1 0 0 0 -1 -1 -1 1 1 0 0 0 1 1 0 0 0 0 0 -1 -1 -1 0 1 1 0 0 0 0 0 0 0 0 0 0 -1 0 0 1 0 0 0 0 0 -1 -1 0 0 0 -1 -1 1 1 1 0 0 0 0 0 0 1 1 -1 0 0 0 0 1 1 1 0 0 0 0 0 1 1 1 0 -1 -1 1 1 0 0 0 0 0 0 0 0 1 0 0 1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 0 0 -1 -1 -1 -1 0 0

[0082] H4. Based on the horizontal and vertical binarized gradients of each pixel, calculate the direction and magnitude of the binarized gradient at each point. The calculation method is as follows: In the formula, T is the magnitude of the binarized gradient, and α T The direction of the binarized gradient is shown in Tables 5 and 6. The binarized gradient direction and magnitude at each point in this embodiment are shown in Tables 5 and 6.

[0083] Table 5 Binarized gradient direction in Example 2

[0084]

[0085]

[0086] Table 6 Binarization gradient magnitude in Example 2

[0087] 0.00 1.00 1.41 1.00 1.00 1.00 1.00 1.41 1.00 1.00 1.41 1.00 1.00 1.00 1.00 1.41 1.00 1.41 1.41 0.00 0.00 0.00 0.00 1.00 1.00 1.00 1.00 0.00 0.00 0.00 0.00 1.00 1.41 1.41 0.00 1.41 1.00 0.00 0.00 1.00 1.00 1.00 1.00 0.00 0.00 0.00 0.00 1.00 1.00 0.00 1.41 1.41 1.41 1.00 0.00 1.00 1.00 1.00 1.00 0.00 0.00 0.00 0.00 1.00 1.00 0.00 1.00 1.41 1.41 1.41 0.00 1.00 1.00 1.00 1.41 0.00 0.00 0.00 0.00 1.00 1.00 0.00 0.00 1.00 1.41 0.00 1.41 1.41 1.00 0.00 1.41 1.00 0.00 1.00 1.00 1.41 1.00 0.00 0.00 0.00 0.00 1.41 1.41 1.00 0.00 0.00 1.00 1.00 1.00 1.41 1.00 1.00 1.00 0.00 0.00 0.00 0.00 1.00 1.00 0.00 0.00 0.00 1.00 1.41 1.41 1.41 1.00 1.00 1.00 0.00 0.00 0.00 0.00 1.00 1.00 0.00 0.00 0.00 0.00 1.00 1.41 1.41 1.00 1.41 1.00 0.00 0.00 1.00 1.00 1.41 1.41 1.00 0.00 0.00 0.00 1.00 1.41 1.00 0.00 1.00 1.00 0.00 1.41 1.41 1.00 1.00 1.41 1.41 1.00 0.00 1.00 1.00 0.00 0.00 0.00 1.00 1.00 1.00 1.41 0.00 1.00 1.41 0.00 1.00 1.00 0.00 0.00 1.00 1.41 1.00 0.00 1.00 1.41 1.41 1.41 1.00 1.41 1.00 0.00 1.00 1.00 0.00 0.00 1.00 1.41 1.41 0.00 1.00 0.00 1.00 1.41 1.00 1.00 0.00 0.00 1.00 1.00 0.00 1.41 1.41 1.41 0.00 1.41 1.41 1.41 1.00 0.00 0.00 0.00 0.00 0.00 1.00 1.00 1.00 1.41 0.00 0.00 1.41 1.41 1.00 1.41 1.00 1.00 1.00 1.00 1.00 1.00 1.41 1.00 1.00 1.41 1.00 1.00 1.41 1.00 0.00

[0088] H5. Divide the image into square blocks with sides of 4 pixels, so that each block contains 4*4 pixels. Block numbering is as follows: Figure 6 As shown.

[0089] H6, α T The binarized gradient direction takes the value [0, 360°]. This interval is divided into 9 equal parts to obtain 9 directional gradient statistical angle intervals. The directional gradient statistical value of each block's gradient statistical angle interval is calculated. The specific method is as follows: based on the binarized gradient direction α of each pixel in the block... T Pixels are categorized into gradient statistical angle intervals for each direction. The binarized gradient magnitude T of each pixel within each gradient statistical angle interval is summed in a counterclockwise direction to obtain the gradient statistical angle interval directional gradient statistical value. The gradient statistical angle interval directional gradient statistical values ​​of each block of the image in Example 2 are shown in Table 7.

[0090] Table 7. Gradient statistics of each block in the image of Example 2, angular interval directional gradient statistics.

[0091]

[0092] H7. Divide the image into 9 computation regions, using four adjacent blocks as one computation region.

[0093] H8. By simultaneously calculating the gradient statistics of each block within the computational region, the directional gradient feature vectors of each computational region are obtained. Taking computational region 1, composed of blocks numbered 1, 2, 3, and 4, as an example, the gradient statistics of the angle intervals of the four blocks are (2, 3.41, 3, 3.41, 0, 3.41, 0, 3.41, 0), (2, 3, 0, 4, 4.41, 4, 0, 4, 4.41, 0), and (3, 0, 4, 4.41, 4, 0, 4, 4) respectively. .41,0)(1,1.41,2,0,0,0,2,0,0)(3,0,0,0,0,04,4.41,0), and simultaneously obtain the gradient eigenvector of region 1 (2,3.41,3,3.41,0,3.41,0,3.41,0,2,3,0,4,4.41,4,0,4,4.41,0,3,0,4,4.41,4,0,4,4.41,0,1,1.41,2,0,0,0,2,0,0).

[0094] H9. Combine the directional gradient feature vectors of each calculation region to obtain the image directional gradient feature vector.

[0095] H10. Calculate the sum of squares of each component of the image orientation gradient feature vector, take the square root of the sum of squares to obtain the orientation gradient normalization target value, divide each component of the image orientation gradient feature vector by the orientation gradient normalization target value to obtain the normalized image orientation gradient feature vector.

[0096] Example 3

[0097] This embodiment aims to illustrate the training steps of support vector machines. The steps for calculating the directional gradient normalized feature vector and the image gray-level co-occurrence matrix feature vector are described in Embodiments 1 and 2, and will not be described in detail in this embodiment.

[0098] V1. The images acquired in step S1 are divided into a training group and a test group, accounting for 80% and 20% of the total images, respectively. In this embodiment, both the training group and the test group images are 256*256 pixel images. The training group images consist of 17 images of under-vibrated concrete and 8 images of moderately vibrated concrete. The test group images consist of 4 images of under-vibrated concrete and 2 images of moderately vibrated concrete. See [link to training group images] for details. Figure 7 and Figure 8 .

[0099] V2. Calculate the gray-level co-occurrence matrix of the training group and the test group images at 0°, 45°, 90° and 135° respectively, and normalize it to obtain the gray-level co-occurrence matrix feature vector.

[0100] V3. Calculate the directional gradient feature vectors of the training group and the test group respectively. In this embodiment, the 256*256 resolution image is divided into 4096 4*4 blocks, so there are 3969 calculation regions.

[0101] V4. Combine the image orientation gradient feature vector with the gray-level co-occurrence matrix feature vector to obtain the support vector machine feature vector.

[0102] V5. Define the support vector machine feature values ​​for the training group images. The feature value is 1 for images with missing vibration or insufficient vibration, and -1 for images with moderate vibration.

[0103] V6. Construct the training space. The dimension of the training space is the same as the dimension of the feature vectors of the support vector machine. In this embodiment, the training space is a 142892-dimensional space.

[0104] V7. Based on the support vector machine feature vector components, determine the position of the training group images in the training space. For example, the support vector machine feature vector corresponding to a certain image is (1, 2, 3, 4, 5, 6). T If the image is located at (1, 2, 3, 4, 5, 6) in the training space.

[0105] V8. Construct a hyperplane f(x) = w in the training space. T x i +b, where w T Let be the normal vector of the hyperplane, b be the bias of the hyperplane relative to the origin, and xi be the independent variable of the hyperplane. The independent variable is in the form of [x1,x2,x3……xn]. The number of independent variable components is the same as the dimension of the feature vector of the support vector machine, with 142892 independent variable components.

[0106] V9. Solve under constraint yi(w) T x i +b)≥1 below, The normal vector w of the hyperplane that reaches its maximum value TLet b be the offset of the hyperplane relative to the origin, ||w|| be the second norm of the hyperplane normal vector, and yi be the eigenvalue of the image support vector machine. This eigenvalue is 1 for images with missing or incomplete vibration, and -1 for images with moderate vibration. The hyperplane obtained at this time can divide the training space into two parts, and the positions of the images with eigenvalues ​​of 1 and -1 in the training space are separated by the hyperplane. In this embodiment, the hyperplane normal vector has 142,892 components. Only the first 20 components are shown here: (0.000478486, -3.67E-06, -8.18E-05, 0, -0.000423967, 0, -0.000356181, -0.000305274, 0.000478486, 0.000203475, 7.44E-05, 0.000324927, 0, -0.000602289, 0, -9.11E-05, -9.01E-05, 0.000203475, 0.000103618, 0.000106364, ...) T In this embodiment, the offset b of the hyperplane relative to the origin is 1.1846.

[0107] V10. Verify the hyperplane separation effect in the training space, calculate the support vector machine feature vectors of the test group images, and use them to determine their positions in the training space. The test group images are separated into two parts by the hyperplane in the training space, thus completing the support vector machine training.

[0108] Example 4

[0109] This embodiment aims to illustrate the overall steps of the present invention: the steps of support vector machine training, directional gradient normalization feature vector calculation, and image gray-level co-occurrence matrix feature vector calculation are described in Embodiments 1, 2, and 3, and will not be described in detail in this embodiment.

[0110] S1. Collect images of the concrete around the vibrator under conditions of missed vibration, insufficient vibration, and moderate vibration in manual vibration mode, and use these images as samples for support vector machine training.

[0111] S2. Concrete is unloaded to the construction area using pumping equipment.

[0112] S3. Start the vibrator, and simultaneously activate the camera to capture images of the concrete during vibration, such as... Figure 9 As shown, the image is uploaded to the computer in real time to determine its position in the support vector machine training space. The hyperplane divides the training space into two parts: a moderately vibrated area and a region with insufficient or missing vibration. This image is located in the region with insufficient or missing vibration.

[0113] S4. Continue concrete vibration operation, collect images in real time, and determine its position in the training space.

[0114] S5. Vibrate the concrete until the concrete image is as shown. Figure 10 As shown, the image is located in the moderately vibrated area of ​​the training space.

[0115] S6. Turn off the vibration motor, slowly pull out the vibrator, and proceed to the next area for vibration.

[0116] like Figure 11 As shown, the intelligent concrete vibration device based on support vector machine of the present invention consists of a power supply 01, a 4G transmission module 02, a vibration motor 03, a vibration motor handle 04, a transmission metal rod 05, a mechanical anti-shake camera platform 06, a camera 07, an LED bulb 08, a vibrator front connector 09, a rubber hose 010, a flexible shaft pin 011, and a metal vibration end 012.

[0117] The connection between the power supply 01, the 4G transmission module 02, and the vibration motor 03 is a bolt connection, while the connection between the vibration motor handle 04 and the vibration motor 03 is a weld connection. The transmission metal rod 05 is welded to the vibration motor 03. The mechanical image stabilization camera platform 06 is welded to the transmission metal rod 05. The camera 07, the LED bulb 08, and the mechanical image stabilization camera platform 06 are bolted connections. The transmission metal rod 05 and the rubber hose 010 are connected via the vibrating rod front connector 09. The rubber hose 010 and the metal vibration end 012 are connected via a flexible shaft pin 011.

[0118] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any transformations or substitutions that can be understood by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of the present invention.

Claims

1. A method for intelligent vibration compaction of concrete based on support vector machines, characterized in that, Includes the following steps: S10. Collect images of the concrete around the vibrator under conditions of missed vibration, insufficient vibration, and moderate vibration in manual vibration mode, and use these images as samples for support vector machine training. S20. Concrete is unloaded into the construction area using pumping equipment; S30. Start the vibratory motor and simultaneously start the camera to capture images of the concrete during the vibration process and upload them to the computer in real time. S40. Input the concrete image into the support vector machine model trained in step S10 to obtain the position of the real-time concrete vibration image in the support vector machine training space. S50. Continue vibrating until the real-time concrete vibration image is located in the appropriate vibration area in the support vector machine training space. Turn off the vibration motor, slowly pull out the vibrator, and proceed with the next area of ​​concrete vibration. The support vector machine training in step S10 includes the following steps: S11. Divide the images collected in step S10 into a training group and a test group, accounting for 80% and 20% of the total number of images, respectively. S12. Calculate the gray-level co-occurrence matrix of the training group and the test group images at 0°, 45°, 90° and 135° respectively, and normalize it to obtain the gray-level co-occurrence matrix feature vector; S13. Calculate the normalized image orientation gradient feature vectors for the training group and the test group respectively. S14. Combine the image orientation gradient feature vector and the gray-level co-occurrence matrix feature vector to obtain the support vector machine feature vector; S15. Define the support vector machine feature values ​​for the training group images. The feature value is 1 for images with missing vibration or insufficient vibration, and -1 for images with moderate vibration. S16. Construct a training space with the same dimension as the feature vectors of the support vector machine. S17. Based on the support vector machine feature vector components, determine the position of the training group images in the training space. The support vector machine feature vector corresponding to a certain image is (1, 2, 3, 4, 5, 6). T If so, the image is located at (1, 2, 3, 4, 5, 6) in the training space; S18. Construct a hyperplane f(x) = w in the training space. T x i +b, where w T Let x be the normal vector of the hyperplane, b be the offset of the hyperplane relative to the origin, and x be the normal vector of the hyperplane. i The independent variables are hyperplane variables, with the form [x1, x2, x3, ..., xn]. The number of independent variable components is the same as the dimension of the feature vectors of the support vector machine. S19. Solve under constraint yi(w) T x i +b)≥1 below, The normal vector w of the hyperplane that reaches its maximum value T The hyperplane is offset from the origin by b, ‖w‖ is the second norm of the hyperplane normal vector, and yi is the feature value of the image support vector machine. The feature value is 1 for images with missing vibration or lack of vibration, and -1 for images with moderate vibration. The hyperplane obtained at this time can divide the training space into two parts. The positions of the images with feature values ​​of 1 and -1 in the training space are separated by the hyperplane. S110. Verify the hyperplane separation effect in the training space. Calculate the support vector machine feature vectors of the test group images and use them to determine their positions in the training space. The test group images include images with missing vibration, images with insufficient vibration, and images with moderate vibration. If the test group images are separated into two parts by the hyperplane in the training space, the support vector machine training is complete. If the test group images do not meet this condition, the training fails or the training accuracy is insufficient. In this case, the number of training and test group images should be increased, and the support vector machine training should be repeated.

2. The intelligent concrete vibration method based on support vector machine according to claim 1, characterized in that, The calculation of the image gray-level co-occurrence matrix eigenvector in step S12 includes the following steps: S121. Read the values ​​of the red, green and blue channels of the image, and use them to convert the image to grayscale to obtain the grayscale value of each pixel. S122. Set the binarization value of pixels with a gray value greater than 122 to 1, and the gray value of pixels with a gray value less than 122 to 0, to obtain the binarization matrix. S123. Calculate the horizontal gray-level co-occurrence matrix. For a binary gray-level image, the horizontal gray-level co-occurrence matrix has an order of 2 and has four components: (0, 0), (0, 1), (1, 0), and (1, 1). S124. Calculate the gray-level co-occurrence matrix in the 45° direction. For a binary gray-level image, the gray-level co-occurrence matrix in the 45° direction is of order 2 and has four components: (0, 0)(0, 1)(1, 0)(1, 1). S125. Calculate the gray-level co-occurrence matrix in the 90° direction. For a binary gray-level image, the gray-level co-occurrence matrix in the 90° direction is of order 2 and has four components: (0, 0)(0, 1)(1, 0)(1, 1). S126. Calculate the gray-level co-occurrence matrix in the 135° direction. For a binary gray-level image, the gray-level co-occurrence matrix in the 135° direction is of order 2 and has four components: (0, 0)(0, 1)(1, 0)(1, 1). S127. Calculate the numerical contrast, correlation, angular second moment, and inverse difference moment of the four components of the gray-level co-occurrence matrix in each direction, and calculate the mean of the numerical contrast (gmean1), the mean of the correlation (gmean2), the mean of the angular second moment (gmean3), the mean of the inverse difference moment (gmean4), the standard deviation of the contrast (gstd1), the standard deviation of the correlation (gstd2), the standard deviation of the angular second moment (gstd3), and the standard deviation of the inverse difference moment (gstd4) of the four components of the gray-level co-occurrence matrix in each direction. S128. Define the gray-level co-occurrence matrix eigenvector, which contains eight components: the mean contrast (gmean1), mean correlation (gmean2), mean angular second moment (gmean3), mean inverse difference moment (gmean4), standard deviation of contrast (gstd1), standard deviation of correlation (gstd2), standard deviation of angular second moment (gstd3), and standard deviation of inverse difference moment (gstd4) for each direction. The eigenvector of the image gray-level co-occurrence matrix is ​​in the form of (gmean1, gmean2, gmean3, gmean4, gstd1, gstd2, gstd3, gstd4). T .

3. The intelligent concrete vibration method based on support vector machine according to claim 1, characterized in that, The calculation of the normalized image orientation gradient feature vector in step S13 includes the following steps: S131. Read the values ​​of the red, green and blue channels of the image, and use them to convert the image to grayscale to obtain the grayscale value of each pixel. S132. Set the binarization value of pixels with a gray value greater than 122 to 1, and the gray value of pixels with a gray value less than 122 to 0, to obtain the binarization matrix. S133. Based on the binarization matrix, calculate the horizontal and vertical binarization gradients of each pixel in the image. S134. Based on the horizontal and vertical binarization gradients of each pixel, calculate the direction and magnitude of the binarization gradient at each point. S135. Divide the image into square blocks with a side length of n pixels, so that each block contains n*n pixels, and remove the parts of the image that cannot be divided into whole blocks after the blocks are divided. S136. The value of the binarized gradient direction is [0, 360°]. Divide this interval into m equal parts to obtain m directional gradient statistical angle intervals. Calculate the directional gradient statistical value of each block gradient statistical angle interval. S137. Divide the image into Q computational regions, taking four adjacent blocks as one computational region; S138. Combine the statistical values ​​of the gradient angle intervals of each block within the calculation region to obtain the directional gradient feature vector of each calculation region. S139. Combine the directional gradient feature vectors of each calculation region to obtain the image directional gradient feature vector; S1310. Calculate the sum of squares of each component of the image orientation gradient feature vector, take the square root of the sum of squares to obtain the orientation gradient normalization target value, divide each component of the image orientation gradient feature vector by the orientation gradient normalization target value to obtain the normalized image orientation gradient feature vector.

Images