Structural deformation monitoring method based on beidou space-time data fusion and ai technology

Abstract

The present application relates to the technical field of image data processing, and particularly relates to a structure deformation monitoring method based on Beidou space-time data fusion and AI technology, comprising: acquiring a remote sensing image set of a structure under Beidou satellite time positioning and latitude and longitude positioning; performing contour edge preprocessing on each remote sensing image; confirming a key area in each structure contour map based on edge changes between adjacent structure contour maps in time sequence; acquiring a deformation reason of the key area corresponding to each historical time node; dividing the structure contour map and the key area corresponding to each remote sensing image to obtain a training group and a prediction group; training a preset neural network model based on the training group to obtain a structure deformation neural network; the present application can effectively improve the accuracy of the structure deformation neural network by extracting the structure contour to determine the real contour map, and determining the deformation key area after screening the vegetation area changes, which serves as the input information of the structure deformation neural network.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image data processing, and particularly relates to a structure deformation monitoring method based on Beidou space-time data fusion and AI technology. BACKGROUND

[0002] Remote sensing satellites can provide spatial information collection and data acquisition, and combined with the high-precision positioning technology and time service of Beidou navigation satellites, the information and content displayed by remote sensing images can be accurately positioned. On this basis, combined with the support of Beidou navigation satellites on remote sensing satellite data collection, the small changes of structures over time can be effectively detected and analyzed.

[0003] However, when detecting the structure, the existing method assumes that the structure is a rectangle or a combination of rectangles. Due to the shadow blocking and other situations of the obtained images, or the morphological characteristics of the top of the structure deviate and distort due to the errors of remote sensing images, it is difficult to directly obtain the area position of the structure using a rectangle. Therefore, when using remote sensing images in the later stage, it is impossible to accurately monitor whether the building or other structure has deformed.

[0004] Therefore, there is an urgent need for a method for effectively monitoring the deformation of structures using Beidou space-time data. SUMMARY

[0005] In order to solve the above technical problems, the purpose of the present application is to provide a structure deformation monitoring method based on Beidou space-time data fusion and AI technology to solve the above problems. In order to achieve the above purpose, the technical scheme adopted by the present application is as follows:

[0006] In a first aspect, the present application provides a structure deformation monitoring method based on Beidou space-time data fusion and AI technology, which comprises:

[0007] Obtaining a remote sensing image set of a structure under Beidou satellite time positioning and latitude and longitude positioning, the remote sensing image set comprising remote sensing images corresponding to a current time node and a plurality of historical time nodes; performing contour edge preprocessing on each remote sensing image to obtain a structure contour map corresponding to each remote sensing image; confirming a key area in each structure contour map based on edge changes between adjacent structure contour maps in time sequence, the key area being a region where the contour changes over time without being affected by vegetation; obtaining a deformation reason of the key area corresponding to each historical time node; dividing the structure contour map and the key area corresponding to each remote sensing image to obtain a training group and a prediction group; training a preset neural network model based on the training group to obtain a structure deformation neural network; and inputting the prediction group into the structure deformation neural network to obtain a structure deformation monitoring result.

[0008] In a possible implementation, the contour edge preprocessing is performed on each of the remote sensing images to obtain a structure contour map corresponding to each of the remote sensing images, including:

[0009] The contour extraction and preprocessing are performed on the remote sensing images based on a preset contour extraction algorithm to obtain the contour of the structure in the remote sensing images; the inflection point recognition is performed on the contour of the structure in the remote sensing images to obtain a plurality of structure feature points; the contour is segmented based on the structure feature points to obtain a plurality of contour edge line segments; the straight line analysis is performed on each of the contour edge line segments, and it is determined whether the contour edge line segment is a structure edge line based on the straight line analysis result; and the structure contour map is formed based on all the structure edge lines.

[0010] In a possible implementation, the preset contour extraction algorithm is a Canny operator edge detection algorithm.

[0011] In a possible implementation, the straight line analysis is performed on each of the contour edge line segments, and it is determined whether the contour edge line segment is a structure edge line based on the straight line analysis result, including:

[0012] The straight line identification is performed on the contour edge line segment based on a preset straight line identification algorithm to obtain a straight line identification result, the straight line identification result including a plurality of straight line segments and a plurality of curve segments; the slope similarity of each of the curve segments and its adjacent two straight line segments is calculated; the edge line occlusion possibility evaluation value is calculated based on the slope similarity corresponding to each of the curve segments and the proportion of the contour edge line segment where all the curve segments are located; and it is determined whether the contour edge line segment is a structure edge line based on the edge line occlusion possibility evaluation value and a preset first threshold value.

[0013] In a possible implementation, the preset first threshold value is 0.8.

[0014] In a possible implementation, the structure includes a plurality of structures, and the structure contour map is formed based on all the structure edge lines, including:

[0015] The slope of the straight line segment in each of the contour edge line segments is counted and the mean value is calculated to obtain the mean value of the straight line slope corresponding to each of the contour edge line segments; the authenticity evaluation value of each of the structures is calculated based on the perpendicularity between each pair of inflection point edge line segments in each of the structures, the pair of inflection point edge line segments being two contour edge line segments connected with one structure feature point; it is determined whether the contour edge line segment corresponding to the structure is a structure contour line based on the authenticity evaluation value of each of the structures and a preset second threshold value; and the structure contour map is formed based on all the structure contour lines.

[0016] In a possible implementation, the preset second threshold is 0.7.

[0017] In a possible implementation, identifying a key area in each structure outline image based on edge changes between the structure outline images that are adjacent in time sequence includes:

[0018] Based on the time series changes, at least one candidate edge affected by vegetation is confirmed in the structural outline map; the length of each candidate edge affected by vegetation in the time series and the tangent slope of the candidate edge affected by vegetation are counted; the evaluation index of each candidate edge affected by vegetation is calculated based on the length of each candidate edge affected by vegetation in the time series and the tangent slope of the candidate edge affected by vegetation; at least one deformed edge is obtained based on each evaluation index judgment and a third threshold judgment, and the evaluation index corresponding to the deformed edge is judged to be less than the third threshold; and a key area is constructed based on all the deformed edges.

[0019] In a possible implementation, the preset third threshold is 0.8.

[0020] In one possible implementation, the preset neural network model is a convolutional neural network model.

[0021] The present invention has the following beneficial effects:

[0022] The invention determines the true structure outline based on the straight lines and the positional relationships between them. It then removes vegetation from areas of the structure outline that have changed over time, using vegetation growth patterns to identify key structural regions where deformation has occurred. This, in turn, allows accurate identification of the deformed regions within the remote sensing image, effectively improving the accuracy of the structure deformation neural network when used as input. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0024] Figure 1 A flow chart of a method for monitoring structural deformation based on Beidou spatiotemporal data fusion and AI technology, provided by one embodiment of the present invention;

[0025] Figure 2 A schematic flow chart of step S2 provided in one embodiment of the present invention;

[0026] Figure 3 The profile edge line segment part recognition result schematic diagram shown in one embodiment of the present application;

[0027] Figure 4 The flowchart schematic diagram of step S3 provided by one embodiment of the present application. DETAILED DESCRIPTION

[0028] In order to further clarify the technical means and effects taken by the present application to achieve the predetermined object, the following describes in detail the specific implementation, structure, features and effects of the structure deformation monitoring method based on Beidou space-time data fusion and AI technology according to the present application, in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures or characteristics in one or more embodiments can be combined in any suitable form.

[0029] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs.

[0030] Embodiment:

[0031] The following specifically describes the specific scheme of the structure deformation monitoring method based on Beidou space-time data fusion and AI technology according to the present application in combination with the accompanying drawings.

[0032] Please refer to Figure 1 which shows the flowchart schematic diagram of the structure deformation monitoring method based on Beidou space-time data fusion and AI technology according to one embodiment of the present application, which includes steps S1-S7.

[0033] S1, acquiring a remote sensing image set of a structure under Beidou satellite time positioning and latitude and longitude positioning, the remote sensing image set including remote sensing images corresponding to a current time node and a plurality of historical time nodes.

[0034] It should be noted that the structure mentioned in the present embodiment includes but is not limited to buildings, bridges, water conservancy facilities or other structures. Further, in the present embodiment, the constituent parts in the structure are referred to as structural bodies. For example, exhaust ports, elevator shafts and other structural bodies on the top of a building. Meanwhile, in the present embodiment, the interval between the historical time nodes is preferably half a year or one year, and the time interval between the current time node and the last historical time node is the same as the interval of the historical time nodes.

[0035] S2, performing contour edge preprocessing on each of the remote sensing images to obtain a structure contour map corresponding to each of the remote sensing images.

[0036] In the present embodiment, edge extraction of the remote sensing image can be performed using a Canny operator edge detection algorithm. However, in the present embodiment, it is considered that the contour edges obtained by the Canny operator edge detection algorithm can have some edges as noise and some edges distorted due to the influence of cloud layers. Therefore, referring to Figure 2 In the present embodiment, step S2 can further include steps S21-S25 to take a single contour edge segment as the most basic recognition unit and optimize recognition of the distorted contour edge segment.

[0037] First, in the present embodiment, the Canny operator edge detection algorithm is used as a preset contour extraction algorithm for contour extraction, and then the contour chain code tracking algorithm is used to remove the influence of noise or non-structural body edges. Details are shown in step S21.

[0038] S21, performing contour extraction and preprocessing on the remote sensing image based on a preset contour extraction algorithm to obtain the contour of the structure in the remote sensing image.

[0039] Specifically, the contour chain code tracking algorithm can be the Ren Mingwu contour chain code tracking algorithm, which selects the chain code structure of the 8-neighborhood chain code.

[0040] S22, identifying the inflection points of the contour of the structure in the remote sensing image to obtain a plurality of structure feature points.

[0041] That is, in the present embodiment, the inflection points are used as the feature points for segmenting the contour. Specifically, the inflection points can be identified by an angle point detection algorithm, which is a prior art and will not be described in the present embodiment.

[0042] S23, segmenting the contour based on the structure feature points to obtain a plurality of contour edge segments.

[0043] S24, performing straight line analysis on each contour edge segment respectively, and determining whether the contour edge segment is a structural body edge based on the straight line analysis result.

[0044] In the present embodiment, it is considered that the structural body edge is generally a straight line spliced together. Therefore, in the present embodiment, it can be directly determined whether the contour edge segment is a straight line to determine whether the line is a structural body edge. The determination of the straight line can be performed using a straight line detection algorithm in image processing, such as Hough Transform or Radon Transform. These algorithms can effectively identify the straight line features in the image, and can maintain a high accuracy even in the presence of noise or slight distortion. The specific implementation process is a prior art and will not be described in the present embodiment.

[0045] Meanwhile, due to the cloud cover in the remote sensing image, part of the contour edge line segment is distorted, so that the originally straight edge line segment may be incorrectly identified as a combination of multiple straight line segments and curve segments, as shown in the partial identification result of the contour edge line segment, which is straight on both sides and curved between the two straight lines. Therefore, in order to solve the above phenomenon, steps S241 to S244 are included in the embodiment to solve the above phenomenon. Figure 3

[0046] S241, performing straight line identification on the contour edge line segment based on a preset straight line identification algorithm to obtain a straight line identification result, the straight line identification result including multiple straight line segments and multiple curve segments.

[0047] Specifically, the preset straight line identification algorithm in the embodiment is Hough Transform or Radon Transform.

[0048] S242, calculating the slope similarity of each curve segment and its adjacent two straight line segments, respectively.

[0049] In the embodiment, it is considered that the curve segment may be a straight line portion distorted into a curve due to the dynamic drift of the cloud cover. Therefore, in order to accurately identify and process these curve segments caused by the obstruction, the embodiment proposes a slope similarity calculation formula:

[0050]

[0051] wherein, p u,v represents the slope similarity of the curve segment between the u-th straight line segment and the v-th straight line segment; k u represents the slope of the u-th straight line segment; k v represents the slope of the v-th straight line segment; k u,v represents the slope between the two endpoints of the curve segment between the u-th straight line segment and the v-th straight line segment; ε represents a constant to prevent the denominator from being zero, which can be selected by those skilled in the art according to the actual situation, and is preferably 0.01 in the embodiment.

[0052] In the above calculation formula, |k u -k v | represents the slope difference between the u-th straight line segment and the v-th straight line segment. represents the slope difference between the slope of the straight line segment on both sides of the curve segment and the slope of the line connecting the endpoints of the curve segment, and the greater the difference, the more the curve segment is a distorted straight line region due to the obstruction.

[0053] ​S243, calculating an edge line occlusion possibility evaluation value based on the slope similarity of each curve segment and the proportion of the contour edge line segment in which all the curve segments are located.

[0054] Specifically, the edge line occlusion possibility evaluation value calculation formula in this embodiment is as follows:

[0055]

[0056] wherein n represents the total number of straight line segments in a contour edge line segment; li represents the length of the i-th straight line segment; L represents the length of the contour edge line segment of the whole contour; pi represents the proportion of the i-th straight line segment in the contour edge line segment; and p represents the proportion of the j-th curve segment in the contour edge line segment. i li represents the length of the i-th straight line segment; L represents the length of the contour edge line segment of the whole contour; pi represents the proportion of the i-th straight line segment in the contour edge line segment; and p represents the proportion of the j-th curve segment in the contour edge line segment. j-1,j pi-1j represents the slope similarity of the curve segment between the j-1-th straight line segment and the j-th straight line segment; and softmax() represents an activation function.

[0057] In the above calculation formula, pi-1j represents the slope similarity of the curve segment between the j-1-th straight line segment and the j-th straight line segment; and softmax() represents an activation function.

[0058] S244, determining whether the contour edge line segment is a structure edge line based on the edge line occlusion possibility evaluation value and a preset first threshold value.

[0059] Specifically, the preset first threshold value in this embodiment is 0.8. That is, when the edge line occlusion possibility evaluation value is greater than 0.8, the contour edge line segment is actually a structure edge line.

[0060] In summary, this embodiment effectively solves the problem of the partially twisted contour edge line segment in the structure edge line recognition process by introducing the way of analyzing the slope relationship between a curve segment and its adjacent straight line segment.

[0061] Meanwhile, in practice, the corner point detection algorithm can only detect the points with significant changes. That is, one end or both ends of a contour edge line segment may be curved. For this situation, the skilled person in the art can discard this part of the curve segment.

[0062] S25, constructing a structure contour map based on all the structure edge lines.

[0063] Further, in the present embodiment, it is also considered that, based on the basic principles of architectural design and the stability requirements of physical structure, the profiles of the structures in the general structure are mostly in a state of perpendicular to each other, and only a small part of the angle between the lines may be other angles due to design factors or other factors. Therefore, in the present embodiment, it is considered that most of the adjacent lines in the remote sensing image are not the profiles of the structures. Therefore, in the present embodiment, the perpendicular relationship between the adjacent profile edge line segments in the structure is also used as an important basis for constructing the structure profile map. That is, in the present embodiment, step S25 can also include steps S251-S254.

[0064] S251, respectively, calculate the slope of each straight line segment in the profile edge line segment, and perform mean value calculation to obtain the mean value of the straight line slope corresponding to each profile edge line segment.

[0065] S252, based on the perpendicularity between each pair of corner edge line segments in each structure, calculate the authenticity evaluation value of each structure. A pair of corner edge line segments is two profile edge line segments connected with a structure feature point.

[0066] Specifically, in the present embodiment, the calculation formula of the authenticity evaluation value is as follows:

[0067]

[0068] Wherein, y represents the authenticity evaluation value, represents the angle value corresponding to the mean value of the straight line slope of the i-th profile edge line segment in the structure; represents the angle value corresponding to the mean value of the straight line slope of the i+1-th profile edge line segment in the structure; m represents the number of profile edge lines of the structure divided by the structure feature point.

[0069] In the above calculation formula, represents whether the angle between the two profile edge line segments is 90 degrees, if it is 90 degrees, the value of sin is 1, and if it is close to 90 degrees, the value of sin is also close to 1.

[0070] S253, based on the authenticity evaluation value of each structure and the preset second threshold value, determine whether the profile edge line segment corresponding to the structure is a structure profile line.

[0071] Specifically, in the present embodiment, the preset second threshold value is 0.7. That is, when the authenticity evaluation value of the structure is greater than 0.8, the profile edge line segment corresponding to the structure is actually a structure profile line.

[0072] S254, based on all the structure profile lines, construct a structure profile map.

[0073] That is, each structure is analyzed and judged one by one through the above method. Finally, the contour edge segments corresponding to multiple structures are used to form the structure outline. This reduces the impact of remote sensing image distortion.

[0074] Through the above steps, when extracting the outline of a structure not affected by vegetation, this embodiment first extracts the outline of the structure, and then determines the straight lines of the edge lines of the inner contour of the structure and the positional relationship between the straight lines to determine the true outline of the structure.

[0075] S3. Confirming a key area in each structure outline image based on edge changes between the structure outline images that are adjacent in time sequence, wherein the key area is an area where the outline changes with time without being affected by vegetation.

[0076] In this embodiment, it is considered that the difference between different time nodes is not only the result of the deformation of the structure, but also the influence of the accompanying growth of vegetation. Therefore, this embodiment includes a method for further identifying key areas. For details, see Figure 4 , which shows that step S3 includes steps S31 to S35.

[0077] S31. Confirm and obtain at least one candidate vegetation impact edge in the structure outline map based on temporal changes.

[0078] In this embodiment, the method for confirming the candidate edge affected by vegetation is to compare the contour edge segments at the same position at different time nodes one by one to see whether they have changed. If there is a change, it is a candidate edge affected by vegetation.

[0079] S32: Count the length of each candidate vegetation impact edge in the time series and the maximum tangent slope of the candidate vegetation impact edge.

[0080] It should be noted that the length of each vegetation impact candidate edge in the time series mentioned in this step refers to the length of each vegetation impact candidate edge at each historical time node. Furthermore, the maximum tangent slope of a vegetation impact candidate edge refers to the maximum tangent slope of the curve portion of the vegetation impact candidate edge.

[0081] S33 . Calculate an evaluation index of each candidate vegetation impact edge based on the length of each candidate vegetation impact edge in the time series and the tangent slope of the candidate vegetation impact edge.

[0082] It should be noted that the evaluation index calculation formula is as follows:

[0083]

[0084] Among them, w represents the evaluation index; s represents the total number of time nodes; lq,q+# represents the length of the vegetation-affected candidate edge changed from the qth historical time node to the q+1th historical time node; θ q,q+# represents the angle difference corresponding to the maximum tangent slope of the qth historical time node and the maximum tangent slope of the q+1th historical time node, which can also be referred to as the angle difference changed from the qth historical time node to the q+1th historical time node; represents the average angle corresponding to the maximum tangent slopes of the plurality of adjacent differences of the edge, The smaller the value is, the more the vegetation-affected candidate edge changes in a straight line. ε represents a constant to prevent the denominator from being zero, which can be selected by a person skilled in the art according to actual conditions, and is preferably 0.01 in the embodiment.

[0085] S34, at least one deformation edge is obtained based on each of the evaluation index judgment and the third threshold judgment, the deformation edge corresponding to the evaluation index judgment being less than the third threshold.

[0086] It should be noted that the third threshold is preferably 0.8 in the embodiment.

[0087] S35, a key region is constructed based on each of the deformation edges.

[0088] It should be noted that the key region mentioned in the step is surrounded by a plurality of small boxes, each key region includes a key region, and the size of each small box is determined according to the scale of the remote sensing image, which is not specifically limited in the embodiment.

[0089] S4, a deformation reason of each key region corresponding to the historical time node is obtained.

[0090] Specifically, in the embodiment, the deformation reason is marked by artificial. The deformation reason can be displacement, settlement, inclination, no change, etc.

[0091] S5, a training group and a prediction group are obtained by dividing the structure contour map and the key region corresponding to each of the remote sensing images.

[0092] S6, a structure deformation neural network is obtained by training a preset neural network model based on the training group.

[0093] Specifically, in the embodiment, the preset neural network model is a convolutional neural network model, and other network models can also be selected by a person skilled in the art, which is not specifically limited in the embodiment.

[0094] Meanwhile, the training process mentioned in this step is to divide the training group into training samples and verification samples in a ratio of 7:3 or 8:2, and then train and verify the preset neural network model. In the embodiment, the prediction group includes the structure contour map and the key area corresponding to the current time node and at least one historical time node. The number of time nodes included in the prediction group is determined by the logical computing ability of the mode implementation carrier by the person skilled in the art, and is not specifically limited in the embodiment. Correspondingly, each sample in the training sample and the verification sample is composed of the same number of structure contour maps and key areas as the time nodes in the prediction group, and the time nodes in each sample are adjacent.

[0095] S7, inputting the prediction group into the structure deformation neural network to obtain the deformation monitoring result of the structure.

[0096] In the embodiment, the contour of the structure is first extracted by the Canny operator edge detection, and then the real structure contour map is determined according to the straight line condition of the contour and the change state of the curve segment between the straight line segments, which can effectively reduce the dynamic influence of cloud cover. Finally, by comparing the changes of the structure contour map at different historical time nodes, and based on the growth of the vegetation over time, the vegetation area in the structure contour map that changes is screened out to determine the key area of the structure due to deformation. Finally, input into the structure deformation neural network to effectively improve the prediction accuracy.

[0097] It should be noted that the above-mentioned order of the embodiments of the present application is only for description, and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are also possible or may be advantageous.

[0098] Each embodiment in the specification is described in a progressive manner, and the same and similar parts between each embodiment can be referred to each other. Each embodiment focuses on the difference from other embodiments.

Claims

1. A structural deformation monitoring method based on Beidou spatiotemporal data fusion and AI technology, characterized in that: The method comprises: Obtain a remote sensing image set of the structure under Beidou satellite time positioning and longitude and latitude positioning. The remote sensing image set includes remote sensing images corresponding to the current time node and multiple historical time nodes; Perform contour edge preprocessing on each remote sensing image to obtain the contour map of the structure corresponding to each remote sensing image, including: extracting and preprocessing the remote sensing image based on the Canny operator edge detection algorithm to obtain the contour of the structure in the remote sensing image, identifying the inflection points of the contour of the structure in the remote sensing image to obtain multiple structural feature points, and segmenting the contour based on the structural feature points to obtain multiple contour edge segments. Based on a preset straight line recognition algorithm, the contour edge line segment is recognized to obtain a straight line recognition result. The straight line recognition result includes multiple straight line segments and multiple curved segments. The slope similarity of each curved segment and its two adjacent straight line segments is calculated respectively. Based on the slope similarity corresponding to each curved segment and the proportion of the contour edge segments where all the curved segments are located, an edge line occlusion probability evaluation value is calculated. Based on the edge line occlusion probability evaluation value and a preset first threshold, it is determined whether the contour edge line segment is a structure edge line. The structure includes multiple structures, and the slopes of the straight line segments in each contour edge segment are counted respectively, and the average is calculated to obtain the average slope of the straight line corresponding to each contour edge segment. The authenticity evaluation value of each structure is calculated based on the verticality between each pair of inflection point edge segments in each structure. A pair of inflection point edge segments is two contour edge segments connected to a structural feature point. Based on the authenticity evaluation value of each structure and a preset second threshold, it is determined whether the contour edge segment corresponding to the structure is a structure contour line, and a structure contour map is formed based on all structure contour lines. Confirming a key area in each structure contour map based on edge changes between adjacent structure contour maps in a time series, where the key area is an area where the contour is not affected by vegetation but changes with time, including: confirming at least one vegetation-affected candidate edge in the structure contour map based on the time series change, counting the length of each vegetation-affected candidate edge in the time series and the maximum tangent slope of the vegetation-affected candidate edge, calculating an evaluation index of each vegetation-affected candidate edge based on the length of each vegetation-affected candidate edge in the time series and the tangent slope of the vegetation-affected candidate edge, determining at least one deformed edge based on each evaluation index and a preset third threshold, determining that the evaluation index corresponding to the deformed edge is less than the preset third threshold, and constructing a key area based on each deformed edge; Obtain the deformation cause of the key area corresponding to each historical time node; Based on the structure outline and key areas corresponding to each remote sensing image, the training group and prediction group are obtained; Based on the training group, a preset neural network model is trained to obtain a structural deformation neural network; The prediction group is input into the structural deformation neural network to obtain the deformation monitoring results of the structure.

2. The method for monitoring structural deformation based on BeiDou spatiotemporal data fusion and AI technology according to claim 1 is characterized in that: The first threshold is preset to 0.

8.

3. The method for monitoring structural deformation based on BeiDou spatiotemporal data fusion and AI technology according to claim 1 is characterized in that: The second threshold is preset to 0.

7.

4. The method for monitoring structural deformation based on BeiDou spatiotemporal data fusion and AI technology according to claim 1 is characterized in that: The third threshold is preset to 0.

8.

5. The method for monitoring structural deformation based on BeiDou spatiotemporal data fusion and AI technology according to claim 1 is characterized in that: The default neural network model is a convolutional neural network model.

Images