A method for inspecting the size of prefabricated box girder reinforcement mesh

Abstract

This invention relates to the field of building construction technology, and in particular to a method for inspecting the dimensions of precast box girder reinforcement mesh, comprising the following steps: S1, acquiring color images of the precast box girder reinforcement mesh, and inputting the acquired images into a neural convolutional network. The neural convolutional network performs redundant wavelet transform on the image data to obtain low-frequency approximate components and high-frequency detail components in three directions. Then, the high-frequency detail components in the three directions are multiplied by wavelet coefficients of adjacent scales to obtain wavelet coefficient maps after multiplying wavelet coefficients in three different directions; S2, then performing linear processing on the wavelet coefficient maps, and after the processing is completed, thresholding processing is performed on the wavelet coefficient maps, and edge processing is performed on the low-frequency approximate components to obtain low-frequency subbands. This invention can effectively solve the problem that existing methods for inspecting the dimensions of precast box girder reinforcement mesh require the use of tools such as measuring tapes, which are time-consuming, labor-intensive, and have a low degree of intelligence.

Description

Technical Field

[0001] This invention relates to the field of building construction technology, specifically a method for inspecting the dimensions of steel mesh in precast box girders. Background Technology

[0002] Precast box girder reinforcement mesh is a device used in building construction for casting precast box girders. During the manufacturing process, the dimensions of the precast box girder reinforcement mesh need to be inspected, but the existing inspection methods still have shortcomings. Specifically, the existing methods for inspecting the dimensions of precast box girder reinforcement mesh require the use of tools such as measuring tapes, which is time-consuming, labor-intensive, and has a low level of automation.

[0003] Therefore, a method for inspecting the dimensions of the steel mesh in precast box girders is needed to address the problems raised in the background section. Summary of the Invention

[0004] The purpose of this invention is to provide a method for inspecting the dimensions of the reinforcing mesh of precast box girders, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A method for inspecting the dimensions of steel reinforcement mesh in precast box girders includes the following steps:

[0007] S1. Acquire color images of the precast box girder steel mesh and input the acquired images into a neural convolutional network. The neural convolutional network performs redundant wavelet transform on the image data to obtain low-frequency approximate components and high-frequency detail components in three directions. Then, multiply the high-frequency detail components in the three directions with the wavelet coefficients of adjacent scales to obtain wavelet coefficient images after multiplying the wavelet coefficients in three different directions.

[0008] S2, then perform linear processing on the wavelet coefficient map, after which threshold processing is performed on the wavelet coefficient map, and edge processing is performed on the low-frequency approximate components to obtain the low-frequency sub-band. The denoised wavelet coefficient map and the low-frequency sub-band are subjected to maxima processing. The pixel point at each position is taken as the point with the largest pixel value at the corresponding position of these four images as the edge pixel point of the image, thereby obtaining the edge image. The extracted edge image is enhanced to expand the grayscale range and obtain the precast box girder steel mesh feature image.

[0009] S3. Calculate the coordinates of the first contour vertices of the precast box girder reinforcement mesh feature image, obtain the actual vertex coordinates of the precast box girder reinforcement mesh, establish a perspective transformation matrix based on the first contour vertices and the actual vertex coordinates, map the precast box girder reinforcement mesh feature image using the perspective transformation matrix to obtain the restored image, extract the second contour of the precast box girder reinforcement mesh from the restored image, calculate the coordinates of the second contour vertices of the second contour, calculate the distance between the coordinates of the two second contour vertices on the diagonal using the distance formula and obtain the difference, verify the difference based on a preset deviation threshold, and output the size detection result.

[0010] As a preferred embodiment of the present invention, the specific steps of the redundant wavelet transform in S1 are as follows: each pixel of the image data is processed by a wavelet filter first by row and then by column to obtain a low-frequency approximate component with the same size as the original image and a high-frequency detail component in three directions.

[0011] As a preferred embodiment of the present invention, the redundant wavelet transform in S1 needs to be performed twice. The high-frequency detail components in the three directions correspond to the high-frequency sub-image that retains horizontal detail information, the high-frequency sub-image that contains vertical detail information, and the high-frequency sub-image that reflects diagonal details. The neural convolutional network is constructed based on the neuron model. The convolutional neural network has three layers, and each layer of the neural convolutional network includes an input layer, an intermediate layer, and an output layer.

[0012] As a preferred embodiment of the present invention, the specific steps of threshold processing in S2 are as follows: using Photoshop to determine the threshold, setting the wavelet coefficients in the wavelet coefficient image that are less than the threshold to zero, thereby removing noise points from the image.

[0013] As a preferred embodiment of the present invention, the method for obtaining the coordinates of the first contour vertices in S2 is as follows: the four sides of the first contour are fitted using a straight line fitting method, and the coordinates of the four first contour vertices are calculated.

[0014] As a preferred embodiment of the present invention, the enhancement process in S2 is to perform grayscale enhancement processing on the image using the IMAQCastImage2 and IMAQThreshold functions.

[0015] As a preferred embodiment of the present invention, the neural convolutional network in S1 reads and stores the acquired images through the IMAQCreate and IMAQReadFile2 functions of LabVIEW.

[0016] As a preferred embodiment of the present invention, the wavelet filter includes a low-pass filter and a high-pass filter.

[0017] Compared with the prior art, the beneficial effects of the present invention are:

[0018] 1. In this invention, color images of the precast box girder reinforcement mesh are acquired and input into a neural convolutional network. The neural convolutional network performs redundant wavelet transform on the image data to obtain low-frequency approximation components and high-frequency detail components in three directions. Then, the high-frequency detail components in the three directions are multiplied by wavelet coefficients of adjacent scales to obtain three wavelet coefficient maps multiplied in different directions. The wavelet coefficient maps are then linearly processed, and after processing, thresholding is performed. Edge processing is then applied to the low-frequency approximation components to obtain a low-frequency sub-band. The denoised wavelet coefficient maps and low-frequency sub-bands are then subjected to maxima processing. For each pixel location, the pixel with the largest value at the corresponding position in the four images is taken as the edge pixel, thus obtaining the edge image. The extracted edge image is enhanced to expand the grayscale range, resulting in a feature image of the precast box girder reinforcement mesh. The coordinates of the first contour vertex of the feature image are calculated, and the actual vertex coordinates of the precast box girder reinforcement mesh are obtained. A perspective transformation matrix is ​​established based on the first contour vertex coordinates and the actual vertex coordinates. The perspective transformation matrix is ​​used to map the feature image of the precast box girder reinforcement mesh to obtain a restored image. The second contour of the precast box girder reinforcement mesh in the restored image is extracted, and the coordinates of the second contour vertex are calculated. The distance between the coordinates of two second contour vertices on the diagonal is calculated using a distance formula, and the difference is obtained. The difference is verified based on a preset deviation threshold, and the size detection result is output. The size is checked by using the beam reinforcement mesh image, which is time-saving, labor-saving, and highly intelligent. Detailed Implementation

[0019] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0020] In this embodiment, the present invention provides a technical solution:

[0021] A method for inspecting the dimensions of steel reinforcement mesh in precast box girders includes the following steps:

[0022] S1. Acquire color images of the precast box girder steel mesh and input the acquired images into a neural convolutional network. The neural convolutional network performs redundant wavelet transform on the image data to obtain low-frequency approximate components and high-frequency detail components in three directions. Then, multiply the high-frequency detail components in the three directions with the wavelet coefficients of adjacent scales to obtain wavelet coefficient images after multiplying the wavelet coefficients in three different directions.

[0023] S2, then perform linear processing on the wavelet coefficient map, after which threshold processing is performed on the wavelet coefficient map, and edge processing is performed on the low-frequency approximate components to obtain the low-frequency sub-band. The denoised wavelet coefficient map and the low-frequency sub-band are subjected to maxima processing. The pixel point at each position is taken as the point with the largest pixel value at the corresponding position of these four images as the edge pixel point of the image, thereby obtaining the edge image. The extracted edge image is enhanced to expand the grayscale range and obtain the precast box girder steel mesh feature image.

[0024] S3. Calculate the coordinates of the first contour vertices of the precast box girder reinforcement mesh feature image, obtain the actual vertex coordinates of the precast box girder reinforcement mesh, establish a perspective transformation matrix based on the first contour vertices and the actual vertex coordinates, map the precast box girder reinforcement mesh feature image using the perspective transformation matrix to obtain the restored image, extract the second contour of the precast box girder reinforcement mesh from the restored image, calculate the coordinates of the second contour vertices of the second contour, calculate the distance between the coordinates of the two second contour vertices on the diagonal using the distance formula and obtain the difference, verify the difference based on a preset deviation threshold, and output the size detection result.

[0025] Furthermore, the specific steps of the redundant wavelet transform in S1 are as follows: each pixel of the image data is processed by the wavelet filter first by row and then by column to obtain the low-frequency approximate component with the same size as the original image and the high-frequency detail components in three directions.

[0026] Furthermore, the redundant wavelet transform in S1 needs to be performed twice. The high-frequency detail components in the three directions correspond to high-frequency sub-images that retain horizontal detail information, high-frequency sub-images that contain vertical detail information, and high-frequency sub-images that reflect diagonal details. The neural convolutional network is constructed based on a neuron model. The convolutional neural network has three layers, and each layer of the neural convolutional network includes an input layer, an intermediate layer, and an output layer.

[0027] Furthermore, the specific steps of threshold processing in S2 are as follows: using Photoshop to determine the threshold, and setting the wavelet coefficients in the wavelet coefficient image that are less than the threshold to zero, thereby removing noise points from the image.

[0028] Furthermore, the method for obtaining the coordinates of the first contour vertices in S2 is as follows: the four sides of the first contour are fitted using a straight line fitting method, and the coordinates of the four first contour vertices are calculated.

[0029] Furthermore, the enhancement process in S2 involves performing grayscale enhancement on the image using the IMAQCastImage2 and IMAQThreshold functions.

[0030] Furthermore, in S1, the neural convolutional network reads and stores the acquired images using the IMAQCreate and IMAQReadFile2 functions of LabVIEW.

[0031] Furthermore, the wavelet filter includes a low-pass filter and a high-pass filter.

[0032] Color images of the precast box girder reinforcement mesh are acquired and input into a neural convolutional network. The neural convolutional network reads and stores the acquired images using the IMAQCreate and IMAQReadFile2 functions of LabVIEW. The neural convolutional network processes the image data using wavelet filters, first row-wise and then column-wise, for each pixel of the image data. This yields low-frequency approximate components of the same size as the original image, as well as high-frequency detail components in three directions. Redundant wavelet transform is performed twice. The high-frequency detail components in the three directions correspond to high-frequency sub-images that retain horizontal detail information, high-frequency sub-images that contain vertical detail information, and high-frequency sub-images that reflect diagonal details. The neural convolutional network is constructed based on a neuron model. The convolutional neural network has three layers, each including an input layer, an intermediate layer, and an output layer. Then, the high-frequency detail components in the three directions are multiplied with wavelet coefficients of adjacent scales to obtain three wavelet coefficient images after multiplying wavelet coefficients in different directions.

[0033] Then, the wavelet coefficient image is linearly processed. After processing, a threshold is determined using Photoshop, and wavelet coefficients in the wavelet coefficient image that are less than the threshold are set to zero, thereby removing noise points in the image. Edge processing is performed on the low-frequency approximation components to obtain the low-frequency sub-band. The denoised wavelet coefficient image and the low-frequency sub-band are then subjected to maxima processing. The pixel with the largest pixel value at the corresponding position in the four images is taken as the edge pixel of the image, thereby obtaining the edge image. The extracted edge image is enhanced by the IMAQCastImage2 and IMAQThreshold functions to expand the grayscale range and obtain the feature image of the precast box girder steel mesh.

[0034] The coordinates of the first contour vertices of the precast box girder reinforcement mesh feature image are calculated. The four sides of the first contour are fitted using a straight line fitting method to obtain the coordinates of the four first contour vertices. The actual vertex coordinates of the precast box girder reinforcement mesh are obtained. A perspective transformation matrix is ​​established based on the first contour vertex coordinates and the actual vertex coordinates. The precast box girder reinforcement mesh feature image is mapped using the perspective transformation matrix to obtain a restored image. The second contour of the precast box girder reinforcement mesh in the restored image is extracted. The coordinates of the second contour vertices of the second contour are calculated. The distance between the coordinates of two second contour vertices on the diagonal is calculated using a distance formula, and the difference is obtained. The difference is verified based on a preset deviation threshold, and the size detection result is output.

[0035] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for inspecting the dimensions of reinforcing mesh in precast box girders, characterized in that, Includes the following steps: S1. Acquire color images of the precast box girder steel mesh and input the acquired images into a neural convolutional network. The neural convolutional network performs redundant wavelet transform on the image data to obtain low-frequency approximate components and high-frequency detail components in three directions. Then, multiply the high-frequency detail components in the three directions with the wavelet coefficients of adjacent scales to obtain wavelet coefficient images after multiplying the wavelet coefficients in three different directions. S2, then perform linear processing on the wavelet coefficient map, after which threshold processing is performed on the wavelet coefficient map, and edge processing is performed on the low-frequency approximate components to obtain the low-frequency sub-band. The denoised wavelet coefficient map and the low-frequency sub-band are subjected to maxima processing. The pixel point at each position is taken as the point with the largest pixel value at the corresponding position of these four images as the edge pixel point of the image, thereby obtaining the edge image. The extracted edge image is enhanced to expand the grayscale range and obtain the precast box girder steel mesh feature image. S3. Calculate the coordinates of the first contour vertices of the precast box girder reinforcement mesh feature image, obtain the actual vertex coordinates of the precast box girder reinforcement mesh, establish a perspective transformation matrix based on the first contour vertices and the actual vertex coordinates, map the precast box girder reinforcement mesh feature image using the perspective transformation matrix to obtain the restored image, extract the second contour of the precast box girder reinforcement mesh from the restored image, calculate the coordinates of the second contour vertices of the second contour, calculate the distance between the coordinates of the two second contour vertices on the diagonal using the distance formula and obtain the difference, verify the difference based on a preset deviation threshold, and output the size detection result.

2. The method for inspecting the dimensions of the reinforcing mesh of a precast box girder according to claim 1, characterized in that: The specific steps of the redundant wavelet transform in S1 are as follows: each pixel of the image data is processed by the wavelet filter first by row and then by column to obtain the low-frequency approximate component with the same size as the original image and the high-frequency detail component in three directions.

3. The method for inspecting the dimensions of the reinforcing mesh of a precast box girder according to claim 1, characterized in that: The redundant wavelet transform in S1 needs to be performed twice. The high-frequency detail components in the three directions correspond to the high-frequency sub-image that retains horizontal detail information, the high-frequency sub-image that contains vertical detail information, and the high-frequency sub-image that reflects diagonal details. The neural convolutional network is constructed based on the neuron model. The convolutional neural network has three layers, and each layer of the neural convolutional network includes an input layer, an intermediate layer, and an output layer.

4. The method for inspecting the dimensions of the reinforcing mesh of a precast box girder according to claim 1, characterized in that: The specific steps of threshold processing in S2 are as follows: using Photoshop to determine the threshold, assigning zero to wavelet coefficients in the wavelet coefficient image that are less than the threshold, thereby removing noise points from the image.

5. The method for inspecting the dimensions of the reinforcing mesh of a precast box girder according to claim 1, characterized in that: The method for obtaining the coordinates of the first contour vertices in S2 is as follows: the four sides of the first contour are fitted using a straight line fitting method, and the coordinates of the four first contour vertices are calculated.

6. The method for inspecting the dimensions of the reinforcing mesh of a precast box girder according to claim 1, characterized in that: The enhancement process in S2 involves performing grayscale enhancement on the image using the IMAQCastImage2 and IMAQThreshold functions.

7. The method for inspecting the dimensions of the reinforcing mesh of a precast box girder according to claim 1, characterized in that: In S1, the neural convolutional network reads and stores the acquired images using the IMAQCreate and IMAQReadFile2 functions of LabVIEW.

8. The method for inspecting the dimensions of the reinforcing mesh of a precast box girder according to claim 2, characterized in that: The wavelet filter includes a low-pass filter and a high-pass filter.