Template support engineering structure integrity diagnosis method based on image recognition

Abstract

The invention relates to the technical field of engineering monitoring, in particular to a template support engineering structure integrity diagnosis method based on image recognition. Comprising the steps of obtaining a monitoring image; identifying the monitoring image through a convolutional neural network, and obtaining a plurality of support nodes in the monitoring image; deriving space coordinate distribution of all support nodes in the monitoring image; and comparing the space coordinates of the identified support nodes in the monitoring image with the space coordinates of all the support nodes in the monitoring image, and judging whether the local template support engineering structure displayed by the current monitoring image is complete or not. According to the method, structural integrity inspection of the template supporting project can be completed in an image analysis mode, manpower and material resources are saved, long-time monitoring can be achieved during use, and the use risk of the template supporting project is remarkably reduced. When integrity detection is carried out, the problem of sight shielding is considered, the image data of the shielding part can be complemented by monitoring the supplementary image, and the detection accuracy is improved.

Description

Technical Field

[0001] This invention relates to the field of engineering monitoring technology, and in particular to a method for diagnosing the integrity of template-supported engineering structures based on image recognition. Background Technology

[0002] During construction, temporary load-bearing structures are often required, so formwork support frames are frequently used. These frames are constructed from steel pipes arranged horizontally and vertically to provide temporary load-bearing support.

[0003] For construction safety reasons, formwork support frames must meet a series of specifications, such as the "Technical Specification for Safety of Highway Engineering Construction JTGF90-2015". One of these requirements is the integrity of the formwork support structure. Support nodes must be set up at the intersections of horizontal, vertical, and diagonal braces, using cup-lock, portal, or fastener methods to secure the horizontal, vertical, and diagonal braces. As key load-bearing nodes in the entire support structure, the integrity of these support nodes is crucial for safe construction.

[0004] Currently, construction supervisors typically inspect the integrity of support nodes. Existing automated inspection articles mostly focus on stress analysis of formwork support projects and do not address integrity testing. The shortcomings are: 1. Manual inspection is time-consuming and labor-intensive, and may result in missed inspections. 2. Manual inspection is generally conducted after the support frame is constructed, but necessary repeated inspections are lacking during construction and use. Summary of the Invention

[0005] This invention discloses a method for diagnosing the integrity of template-supported engineering structures based on image recognition. The specific method is as follows:

[0006] The template support project was photographed from preset angles to obtain monitoring images;

[0007] A convolutional neural network is used to identify monitoring images and obtain several supporting nodes in the monitoring images;

[0008] The distance between all identified support nodes and the shooting point in the monitoring image is obtained using a depth camera, and the spatial coordinates of all identified support nodes in the monitoring image are calculated based on the spatial coordinates of the shooting point.

[0009] Based on the spatial coordinate distribution of the identified support nodes in the monitoring images, deduce the spatial coordinate distribution of all support nodes in the monitoring images;

[0010] By comparing the spatial coordinates of the identified support nodes in the monitoring image with the spatial coordinates of all support nodes in the monitoring image, if the spatial coordinates of the identified support nodes in the monitoring image are consistent, it is determined that the local template support structure shown in the current monitoring image is complete.

[0011] The advantage of this embodiment is that the structural integrity inspection of the formwork support project can be completed by using image analysis, saving manpower and material resources, and long-term monitoring can be achieved during use, which significantly reduces the risk of using the formwork support project.

[0012] Furthermore, when comparing the spatial coordinates of the identified support nodes in the monitoring image with the spatial coordinates of all support nodes in the monitoring image, if the spatial coordinates of the identified support nodes in the monitoring image are not continuous, the viewing angle of the captured monitoring image is adjusted to obtain a supplementary monitoring image.

[0013] By combining the spatial coordinates of the identified support nodes in the combined monitoring image and the supplementary monitoring image, it is determined whether they are continuous. If they are continuous, the local formwork support structure is considered complete; otherwise, the local formwork support structure is considered incomplete.

[0014] The advantage of this embodiment is that it takes into account the problem of line-of-sight obstruction when performing integrity detection. By monitoring supplementary images, the image data of the obstructed parts can be completed, thereby improving the accuracy of detection.

[0015] Furthermore, there are several preset viewing angles, and each preset viewing angle acquires a monitoring image;

[0016] When the partial template support structure shown in several monitoring images is judged to be complete, the entire template support structure is judged to be complete.

[0017] The advantage of this embodiment is that by superimposing multiple local diagnostic structures from preset perspectives, the structural diagnostic results of the entire template support project can be obtained, which can significantly improve the diagnostic accuracy of the entire template support project.

[0018] Furthermore, the supporting node includes an intersection consisting of any two or more of the following structures:

[0019] Horizontal bars, vertical bars, and diagonal bars.

[0020] Furthermore, the convolutional neural network is trained using the following method:

[0021] Select a sample support node consisting of any number of uprights and crossbars;

[0022] The sample support nodes are occluded in a way that is identifiable to humans, and multi-angle shooting is performed to obtain sample training images;

[0023] Mix the training images with the non-supporting node images to form the training sample images;

[0024] Perform grayscale processing on the training sample images;

[0025] The convolutional neural network is trained using grayscale training sample images.

[0026] The advantage of this embodiment is that by partially occluding and switching the shooting angle, a large number of effective training samples can be generated artificially at low cost, which can significantly improve the optimization results of the convolutional neural network and improve the recognition accuracy.

[0027] Furthermore, the spatial coordinate distribution of all support nodes in the monitoring image is derived using the following method:

[0028] K-means clustering algorithm is used to analyze the spatial coordinates of known support nodes in the monitored image. Appearing in coordinate, coordinates and Clustering of coordinates separately;

[0029] After calculating the clusters respectively The distance between adjacent class centers of coordinates after clustering. The distance between adjacent class centers of the coordinates, and the distance after clustering. The distance between adjacent category centers of the coordinates;

[0030] Select the known spatial coordinates of the support nodes in the monitoring image The minimum coordinate point is used as the first reference point, and the known spatial coordinates of the support nodes in the monitoring image are selected. The coordinate point of the maximum value in the middle is used as the second reference point;

[0031] After adding clustering based on the first reference point Neighboring class center distance of coordinates, after clustering Neighboring class center distance and clustering results for coordinates The distance between adjacent category centers of the coordinates is used to obtain the adjacent reference points of the first reference point;

[0032] After adding clustering based on adjacent reference points Neighboring class center distance of coordinates, after clustering Neighboring class center distance and clustering results for coordinates The distance between adjacent class centers of coordinates, until... coordinate, coordinates or Coordinates reach the second reference point Coordinate values;

[0033] Record the spatial coordinates of all reference points as the spatial coordinate distribution of all support nodes in the monitoring image.

[0034] The advantage of this embodiment is that the support nodes of the template support project are generally regularly distributed in three dimensions. It is only necessary to find the distribution pattern of the rows, columns, and heights to deduce the distribution of the entire template support project. However, the support nodes in the same row, column, and height are likely to be tilted relative to the orthogonal coordinate system. Therefore, a method is needed to find the extension line formed by the support nodes in the same row, column, and height. Compared with the easily conceivable three-dimensional imaging method, the clustering algorithm can significantly reduce the amount of computation and is not affected by the imaging effect, resulting in higher accuracy.

[0035] Furthermore, the method for determining the continuity of the spatial coordinates of the identified support nodes in the monitoring image is as follows:

[0036] Select any known support node in the monitoring image and obtain the spatial coordinates of that known support node;

[0037] By calculating the Euclidean distance, the corresponding support node of the known support node is selected from the spatial coordinates of all support nodes in the monitoring image;

[0038] Select any remaining known support node in the monitoring image and obtain the spatial coordinates of the remaining known support node;

[0039] By calculating the Euclidean distance, the corresponding support node of the remaining known support node is selected from the spatial coordinates of all remaining support nodes in the monitoring image;

[0040] Repeat the above steps until all identified support nodes in all monitoring images correspond one-to-one with all support nodes in the monitoring images, and the Euclidean distance between any corresponding support nodes is less than the preset distance threshold. Then it is determined that the spatial coordinates of the identified support nodes in the monitoring images are continuous.

[0041] The advantage of this embodiment is that by comparing the points of two images, it is possible to verify whether the number of support nodes in the two images is consistent and whether there is a large error in their distribution. This allows it to determine whether the identified support nodes in the monitoring image completely correspond to the spatial coordinates of all support nodes. If they do not correspond, it proves that some of the support nodes are not among the identified support nodes and that there is no continuity.

[0042] Furthermore, the specific method for combining the spatial coordinates of the identified support nodes in the monitoring image and the supplementary monitoring image is as follows:

[0043] A convolutional neural network is used to identify and monitor supplementary images, and to obtain several supplementary support nodes in the monitoring supplementary images;

[0044] The distance between all identified supplementary support nodes and the shooting point in the monitoring supplementary image is obtained using a depth camera, and the spatial coordinates of the identified supplementary support nodes in the monitoring supplementary image are calculated based on the spatial coordinates of the shooting point.

[0045] A set of coordinates to be processed is constructed using the spatial coordinates of the identified support nodes in the monitored image and the spatial coordinates of the identified supplementary support nodes in the monitored supplementary image.

[0046] Based on the known spatial coordinates of the identified support nodes in the monitoring image, calculations are performed. The spatial coordinates of the identified supplementary support nodes in the monitoring supplementary image that are less than the deletion threshold are deleted, and the remaining spatial coordinates are used to construct a set of combined support node spatial coordinates.

[0047] The advantage of this embodiment is that when the support nodes are identified as discontinuous, it may be due to defects in the installation of the template support or obstruction. By adjusting the shooting angle multiple times to supplement the judgment, the specific reason for the discontinuity of the support nodes can be determined, which significantly reduces the possibility of misjudgment.

[0048] Furthermore, the viewing angle of the captured monitoring images is adjusted to obtain supplementary monitoring images. The specific method is as follows:

[0049] Draw a spherical space with the original captured monitoring image as the origin and a preset distance as the radius;

[0050] At the edge of the spherical space, multiple shooting angles were selected to capture several supplementary monitoring images.

[0051] Furthermore, there are four preset viewing angles, located on the four sides of the plate support project.

[0052] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0053] The accompanying drawings of this invention are described below.

[0054] Figure 1 This is a flowchart illustrating the process of this embodiment.

[0055] Figure 2 This is a schematic diagram of the formwork support structure.

[0056] Figure 3 This is a schematic diagram illustrating the working principle of a convolutional neural network.

[0057] Figure 4 A flowchart illustrating the process of deriving the spatial coordinates of all support nodes in the monitoring image.

[0058] Figure 5 This is a schematic diagram of the K-class mean algorithm. Detailed Implementation

[0059] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0060] A method for diagnosing the integrity of template-supported engineering structures based on image recognition, wherein the template-supported engineering structure is as follows: Figure 2 As shown, the specific steps are as follows: Figure 1 As shown:

[0061] S1. Take a picture of the template support project from the first preset perspective to obtain a monitoring image.

[0062] In step S1, the monitoring image can be captured by a human gesture-capturing device or by a drone carrying a capturing device. .

[0063] S2. Use a convolutional neural network to identify the monitoring image and obtain several support nodes in the monitoring image. ,in For supporting node numbering.

[0064] Support nodes can be several fixed joints between horizontal and vertical members, several fixed joints between horizontal and diagonal members, or several fixed joints between vertical and diagonal members.

[0065] The specific method for training a convolutional neural network is as follows:

[0066] S21. Select a sample support node consisting of any number of uprights and crossbars.

[0067] S22. Occlude the sample support nodes in a way that is identifiable to humans, and take pictures from multiple angles to obtain sample training images.

[0068] S23. Mix the sample training images with the non-sample support node images to form training sample images.

[0069] S24. Perform grayscale processing on the training sample images.

[0070] S25. Train the convolutional neural network using the grayscale training sample images.

[0071] In step S2, the convolutional neural network is as follows: Figure 3 As shown, it includes an input layer, a convolutional layer, a pooling layer, and a fully connected layer.

[0072] The convolution formula for a convolutional layer is as follows:

[0073]

[0074] In the formula, For the convolutional... OK Column pixel values, The input to the convolutional layer is the monitoring image. , This is a preset two-dimensional matrix.

[0075] The pooling layer uses average pooling for the output of the convolutional layer. Process it.

[0076] The output of the pooling layer is classified by the fully connected layer, and the classification result is output.

[0077] S3. Use a depth camera to acquire monitoring images and identify support nodes. spatial coordinates .

[0078] S4. Derive the spatial coordinates of all support nodes in the monitoring image based on the spatial coordinates of the identified support nodes in the monitoring image.

[0079] Specifically, the spatial coordinates of all support nodes in the monitoring image are derived, such as... Figure 4 As shown, the specific method is as follows:

[0080] S41. Extract the spatial coordinates of the identified support nodes from the monitoring image. All of them Coordinates, construct a set ;

[0081] S42. Use the K-class mean algorithm to analyze the set. Perform clustering and divide into Each category is recorded as follows , No. The center record for each category is Calculate according to the following formula Inter-class distance of coordinates:

[0082]

[0083] S43, targeting coordinates and Coordinates, repeat step S42, and calculate. and .

[0084] S44. Select the known spatial coordinates of the support nodes in the monitoring image. The minimum coordinate point is used as the first reference point, i.e. Select the known spatial coordinates of the support nodes in the monitoring image. The coordinates of the maximum value in the middle are used as the second reference point, that is... .

[0085] S45, After adding clustering based on the first reference point... Neighboring class center distance of coordinates, after clustering Neighboring class center distance and clustering results for coordinates The distance between adjacent category centers of the coordinates is obtained by acquiring the adjacent reference points of the first reference point. The formula for calculating adjacent reference points is as follows:

[0086]

[0087]

[0088]

[0089] S46. Repeat the calculation of the three adjacent reference points for each adjacent reference point until the coordinates of any reference in the three systems reach the coordinates of the second reference point, i.e. Stop calculation Coordinates of adjacent reference points in the direction Stop calculation Coordinates of adjacent reference points in the direction Stop calculation Coordinates of adjacent reference points in the direction.

[0090] S47. Record the coordinates of all reference points and adjacent reference points, which constitutes the spatial coordinate set of all support nodes in the monitoring image. .

[0091] In step S4, the k-means clustering algorithm is used, such as... Figure 5 As shown:

[0092] Set the number of categories k and the number of iterations p;

[0093] Randomly select k samples from the sample set as initial cluster centers. The cluster set is... express;

[0094] The entire sample is partially traversed, and the distance between each sample and k samples is calculated. The sample is classified into a class based on the distance between the samples and the k samples.

[0095] The mean value method is used to recalculate the center in each cluster, and the above steps are repeated until the iteration is complete. Specifically, the Minkowski distance formula is used.

[0096]

[0097] in, For the i-th sample, For the j-th sample, Let u be the u-th feature of the i-th sample. For the u-th feature of the j-th sample, u=1 in this embodiment, The time is the Euclidean distance formula. The time is the Manhattan distance, in this embodiment .

[0098] S5. First, determine whether the spatial coordinates of the identified support nodes in the monitoring image under the first preset shooting angle are continuous. If they are not continuous, proceed to step S6; if they are continuous, proceed to step S8.

[0099] To determine whether a sequence is coherent, the specific steps are as follows:

[0100] S51, The set of support nodes identified in the monitored images is as follows: The set of all support nodes in the monitored image is Select known support nodes .

[0101] S52, Calculation and The Euclidean distance of each support node is calculated, and the support node with the closest Euclidean distance is selected for cancellation. The set of remaining identified support nodes in the monitored image is then calculated. The set of spatial coordinates of all remaining support nodes in the monitoring image is .

[0102] S53. Repeat step S52 until all support nodes in the monitored image are set. Clear and reset to zero.

[0103] S54, when And the distance between each pair of cancelling support nodes is less than the distance threshold. At that time, it is determined that the spatial coordinates of the identified support nodes in the monitoring image are consistent.

[0104] S6. Based on the first preset viewpoint, adjust the image to obtain one or more supplementary monitoring images, and identify several supporting nodes in the supplementary monitoring images. .

[0105] S7. Combine the spatial coordinates of the identified support nodes of the monitoring image under the first preset view and several supplementary monitoring images. Make a second judgment on whether the support nodes in the monitoring image under the first preset shooting view are continuous. If they are not continuous, it is judged that the template support structure is incomplete. If they are continuous, proceed to step S8.

[0106] The spatial coordinates of the identified support nodes in the combined monitoring image and the supplementary monitoring image are determined through the following steps:

[0107] S71. Use a depth camera to obtain the distance between all identified supplementary support nodes and the shooting point in the monitoring supplementary image, and calculate the spatial coordinates of the identified supplementary support nodes in the monitoring supplementary image based on the spatial coordinates of the shooting point.

[0108] S72. Construct a set of coordinates to be processed using the spatial coordinates of the identified support nodes in the monitored image and the spatial coordinates of the identified supplementary support nodes in the monitored supplementary image. ;

[0109] S73. Using the known spatial coordinates of the identified support nodes in the monitoring image as a reference, calculate and delete the spatial coordinates of the identified supplementary support nodes in the monitoring supplementary image whose distance is less than the deletion threshold. The remaining spatial coordinates are used to construct a set of combined support node spatial coordinates.

[0110] S74. Construct a set of combined support node spatial coordinates using the remaining spatial coordinates. To determine if the sequence is continuous, follow the same procedure as in step S5.

[0111] S8. Switch to the second, third and fourth preset views, take monitoring images respectively, return to step S2, and when all the support nodes under the preset views are judged to be continuous, the template support structure is judged to be complete.

[0112] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A template support engineering structure integrity diagnosis method based on image recognition, characterized by, The specific method is as follows: The template support project is photographed at a preset visual angle to obtain a monitoring image; The monitoring image is recognized by a convolutional neural network to obtain a plurality of support nodes in the monitoring image; The distance between all the recognized support nodes in the monitoring image and the shooting point is obtained by a depth camera, and the spatial coordinates of all the recognized support nodes in the monitoring image are calculated according to the spatial coordinates of the shooting point; The spatial coordinate distribution of all the support nodes in the monitoring image is derived according to the spatial coordinate distribution of the recognized support nodes in the monitoring image; The spatial coordinates of the recognized support nodes in the monitoring image are compared with the spatial coordinates of all the support nodes in the monitoring image, and if the spatial coordinates of the recognized support nodes in the monitoring image have continuity, it is judged that the local template support project structure displayed in the monitoring image is complete.

2. The image recognition based template support engineering structure integrity diagnostic method of claim 1, wherein, When the spatial coordinates of the recognized support nodes in the monitoring image are compared with the spatial coordinates of all the support nodes in the monitoring image, if the spatial coordinates of the recognized support nodes in the monitoring image do not have continuity, the visual angle for shooting the monitoring image is adjusted to obtain a monitoring supplementary image; The spatial coordinates of the recognized support nodes in the monitoring image and the monitoring supplementary image are combined to determine whether they have continuity, and if they have continuity, it is judged that the local template support project structure is complete, otherwise it is judged that the local template support project structure is not complete.

3. The image recognition based template support engineering structure integrity diagnostic method of claim 1, wherein, The preset visual angle is a plurality of visual angles, and each preset visual angle obtains one monitoring image; When the local template support projects displayed in a plurality of monitoring images are all judged to be complete, it is judged that the entire template support project structure is complete.

4. The image recognition based template support engineering structure integrity diagnostic method of claim 1, wherein, The support nodes include the intersection of any two or more of the following structures: horizontal rods, vertical rods and inclined rods.

5. The image recognition based template support engineering structure integrity diagnostic method of claim 1, wherein, The convolutional neural network is trained as follows: A sample support node composed of any number of vertical rods and horizontal rods is selected; The sample support node is occluded under human recognizable conditions, multi-angle shooting is completed, and sample training pictures are obtained; The sample training pictures and non-sample support node images are mixed into training sample pictures; The training sample images are subjected to grayscale processing; The convolutional neural network is trained using the grayscale processed training sample images.

6. The image recognition based template support engineering structure integrity diagnostic method of claim 2, wherein, The spatial coordinate distribution of all the support nodes in the monitoring image is derived as follows: The K-means clustering algorithm is adopted to cluster the known support node spatial coordinates in the monitoring image coordinates, coordinates and coordinates respectively​ Respectively calculate the adjacent cluster center distance of coordinates after clustering Adjacent cluster center distance of coordinates after clustering Adjacent cluster center distance of coordinates after clustering Adjacent cluster center distance of coordinates after clustering Select the known support node spatial coordinates of the monitoring image Select the known support node spatial coordinates of the monitoring image Select the known support node spatial coordinates of the monitoring image After respectively increasing the clustering of the first reference point The adjacent category center distance of the coordinate, the clustering of the coordinate The adjacent category center distance of the coordinate, the clustering of the coordinate The adjacent category center distance of the coordinate, respectively obtain the first reference point adjacent reference point; After clustering, the distance between the adjacent reference points is increased by the adjacent cluster center distance of the coordinates, after clustering the adjacent cluster center distance of the coordinates and after clustering the adjacent cluster center distance of the coordinates, until the coordinates, the coordinates or the coordinates reach the second reference point the coordinate value; The spatial coordinates of all the reference points are recorded as the spatial coordinate distribution of all the support nodes in the monitoring image.

7. The image recognition based template support engineering structure integrity diagnostic method of claim 6, wherein, The spatial coordinates of the recognized support nodes in the monitoring image are compared with the spatial coordinates of all the support nodes in the monitoring image, and if the spatial coordinates of the recognized support nodes in the monitoring image have continuity, it is judged that the local template support project structure displayed in the monitoring image is complete. When the spatial coordinates of the recognized support nodes in the monitoring image are compared with the spatial coordinates of all the support nodes in the monitoring image, if the spatial coordinates of the recognized support nodes in the monitoring image do not have continuity, the visual angle for shooting the monitoring image is adjusted to obtain a monitoring supplementary image; The spatial coordinates of the recognized support nodes in the monitoring image and the monitoring supplementary image are combined to determine whether they have continuity, and if they have continuity, it is judged that the local template support project structure is complete, otherwise it is judged that the local template support project structure is not complete. The preset visual angle is a plurality of visual angles, and each preset visual angle obtains one monitoring image; When the local template support projects displayed in a plurality of monitoring images are all judged to be complete, it is judged that the entire template support project structure is complete. The support nodes include the intersection of any two or more of the following structures: horizontal rods, vertical rods and inclined rods. The convolutional neural network is trained as follows: A sample support node composed of any number of vertical rods and horizontal rods is selected; The sample support node is occluded under human recognizable conditions, multi-angle shooting is completed, and sample training pictures are obtained; The sample training pictures and non-sample support node images are mixed into training sample pictures; The training sample images are subjected to grayscale processing; The convolutional neural network is trained using the grayscale processed training sample images. The spatial coordinate distribution of all the support nodes in the monitoring image is derived as follows: The spatial coordinates of all the reference points are recorded as the spatial coordinate distribution of all the support nodes in the monitoring image. The spatial coordinates of the recognized support nodes in the monitoring image are compared with the spatial coordinates of all the support nodes in the monitoring image, and if the spatial coordinates of the recognized support nodes in the monitoring image have continuity, it is judged that the local template support project structure displayed in the monitoring image is complete. When the spatial coordinates of the recognized support nodes in the monitoring image are compared with the spatial coordinates of all the support nodes in the monitoring image, if the spatial coordinates of the recognized support nodes in the monitoring image do not have continuity, the visual angle for shooting the monitoring image is adjusted to obtain a monitoring supplementary image; The spatial coordinates of the recognized support nodes in the monitoring image and the monitoring supplementary image are combined to determine whether they have continuity, and if they have continuity, it is judged that the local template support project structure is complete, otherwise it is judged that the local template support project structure is not complete. Repeat the above steps until all the identified support nodes in the monitoring image are one-to-one corresponding to all the support nodes in the monitoring image, and the Euclidean distance between any corresponding support nodes is less than the preset distance threshold, then it is judged that the spatial coordinates of the identified support nodes in the monitoring image have continuity.

8. The image recognition based template support engineering structure integrity diagnostic method of claim 7, wherein, The specific method of combining the spatial coordinates of the identified support nodes in the monitoring image and the monitoring supplementary image is as follows: The convolutional neural network is used to identify the monitoring supplementary image to obtain a plurality of supplementary support nodes in the monitoring supplementary image. The depth camera is used to obtain the distance between all the identified supplementary support nodes in the monitoring supplementary image and the shooting point, and the spatial coordinates of the identified supplementary support nodes in the monitoring supplementary image are calculated according to the spatial coordinates of the shooting point. The spatial coordinates of the identified support nodes in the monitoring image and the spatial coordinates of the identified supplementary support nodes in the monitoring supplementary image are used to construct a to-be-processed coordinate set. The spatial coordinates of the identified support nodes in the monitoring image are used as a reference for calculation, and the spatial coordinates of the identified supplementary support nodes in the monitoring supplementary image with a distance less than the deletion threshold are deleted, and the remaining spatial coordinates construct a combined support node spatial coordinate set.

9. The image recognition based template support engineering structure integrity diagnostic method of claim 2, wherein, Adjust the shooting angle of the monitoring image to obtain the monitoring supplementary image, and the specific method is as follows: Take the original shooting angle of the monitoring image as the origin and a preset distance as the radius to draw a spherical space. On the edge of the spherical space, multiple shooting angles are selected to obtain a plurality of monitoring supplementary images.

10. The image recognition based template support engineering structure integrity diagnostic method of claim 3, wherein, The preset angle is four, which is located in the direction of the four sides of the plate support project.

Images