Unmanned aerial vehicle image ground crack identification and extraction method based on machine learning

Abstract

The invention discloses an unmanned aerial vehicle image ground fracture identification and extraction method based on machine learning, and relates to the technical field of ground fracture identification. A monitoring range is acquired, and a monitoring model is constructed corresponding to the monitoring range; marking identification information corresponding to the monitoring range, marking the identification information in the monitoring model, and establishing an association relationship between the monitoring model and the monitoring range; acquiring an image in the monitoring range based on the unmanned aerial vehicle, positioning the image based on the identification information, and identifying a graph in the image to obtain a crack graph; and constructing the crack graph in the monitoring model based on machine learning to obtain a geographic model. According to the method, information such as the size of the crack can be determined according to identification of the geographic shape features, the ground crack can be extracted finally, and a good ground crack extraction effect is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of ground fissure identification, and in particular to a method for identifying and extracting ground fissures using drone images based on machine learning. Background Art

[0002] With the development of society, some ground surfaces, such as roads and some rocky ground, need road maintenance. Road maintenance includes ground crack treatment and obstacle treatment. At present, ground fissures are an aspect that needs attention in ground maintenance. At present, remote ground conditions can be collected through drones. However, it is difficult to obtain the actual conditions of the cracks during the collection process, and it is impossible to accurately obtain information such as the shape and size of the ground cracks, and there is no good ground crack information extraction function. Summary of the invention

[0003] The purpose of the present invention is to provide a method for identifying and extracting ground fissures from drone images based on machine learning to address the shortcomings of the background technology.

[0004] In order to achieve the above object, the present invention provides the following technical solution: a method for identifying and extracting ground fissures from drone images based on machine learning, comprising the following steps: Obtain the monitoring scope and build a monitoring model corresponding to the monitoring scope; Mark identification information corresponding to the monitoring scope, mark the identification information in the monitoring model and establish an association relationship between the monitoring model and the monitoring scope; Based on the UAV to collect images within the monitoring range, the images are located based on the identification information and the graphics in the images are identified to obtain the crack graphics; The crack graph is constructed into the monitoring model based on machine learning to obtain the geographic model.

[0005] In a preferred embodiment, the step of obtaining the monitoring range and building a monitoring model corresponding to the monitoring range includes: Define the monitoring scope and collect geographic information within the monitoring scope, where the geographic information includes surface shape features and corresponding dimensions; Construct a monitoring model based on the geographic information within the monitoring range.

[0006] In a preferred embodiment, the step of marking the corresponding monitoring range with identification information, marking the identification information in the monitoring model and establishing an association relationship between the monitoring model and the monitoring range includes: Select multiple identification points in the monitoring range, and connect the multiple identification points to obtain an identification network; Acquire a range with a preset diameter with the identification point as the center as the identification range, acquire identification geographic information within the identification range, acquire the position of the identification point in the identification geographic information, and record the identification geographic information within the identification range corresponding to the identification point and the position of the identification point in the identification geographic information as identification point information; Using multiple identification point information and identification network as identification information; The identification information is marked in the monitoring model and an association relationship is established between the monitoring model and the monitoring scope.

[0007] In a preferred embodiment, the step of marking the identification information in the monitoring model and establishing an association relationship between the monitoring model and the monitoring scope includes: Mark identifying information in the monitoring model; Select a starting point in the monitoring range and mark the starting point in the monitoring model accordingly, and establish a corresponding relationship between the identification points in the identification network within the monitoring range and the identification points in the identification network in the monitoring model based on the starting point one by one; A mapping line is established between the identification points in the identification network within the monitoring range and the identification points in the identification network in the monitoring model based on the corresponding relationship, wherein the mapping line is a top-to-bottom trend line in the virtual space between the identification network within the monitoring range and the monitoring model; The correlation between the monitoring model and the monitoring range is obtained based on the mapping line.

[0008] In a preferred embodiment, the step of collecting images within the monitoring range based on a drone, locating the images based on the identification information, and identifying the graphics in the images to obtain crack graphics includes: The drone is passed through the starting point of the monitoring range to collect images within the monitoring range, and the height at which the drone collects images within the monitoring range based on the starting point is determined. During the image collection process, the geographic feature information and the corresponding size in the image are identified in real time as the image geographic information; Matching the identification geographic information in the identification point information with the image geographic information to obtain the identification point in the image as the image identification point; The image recognition points are matched with the corresponding recognition points in the recognition network within the monitoring range to obtain the corresponding relationship of the recognition points, and based on the corresponding relationship of the recognition points, the recognition network within the monitoring range is re-established as a real-time recognition network according to the image recognition points to locate the image; Based on the association between the monitoring model and the monitoring range, a real-time mapping line is established between the image recognition points in the real-time recognition network and the recognition points of the recognition network in the monitoring model, wherein the real-time mapping line is a top-to-bottom trend line in the virtual space between the real-time recognition network and the monitoring model, and the real-time mapping line is corresponded to the mapping line based on the correspondence between the recognition points; Extract the crack image in the image. If the real-time mapping line and the corresponding mapping line cannot coincide, obtain the distance information between the image recognition points around the crack image and the corresponding recognition points in the monitoring range, and mark the size of the crack image according to the distance information to obtain the crack graph.

[0009] In a preferred embodiment, the step of extracting the crack image from the image, if the real-time mapping line and the corresponding mapping line cannot coincide with each other, obtaining the distance information between the image recognition points around the crack image and the recognition points in the corresponding monitoring range, and marking the size of the crack image according to the distance information to obtain the crack graph includes: Extracting a crack image from the image, wherein the crack image is a shape image of the crack; Obtain image recognition points around the crack image. If the real-time mapping lines corresponding to the image recognition points around the crack image cannot coincide with the corresponding mapping lines, obtain the offset angle between the real-time mapping lines corresponding to the image recognition points around the crack image and the corresponding mapping lines. According to the offset angle and the height of the image collected within the monitoring range by the UAV, calculate the distance information between the image recognition points around the crack image and the corresponding recognition points in the monitoring range by using the Pythagorean theorem. The crack image is dimensioned according to the distance information, and a crack graph is constructed based on the dimensioned crack image to obtain a crack graph, wherein the distance information includes the distance and angle between the surrounding image recognition points in the crack image and the corresponding recognition points in the monitoring range.

[0010] In a preferred embodiment, the step of constructing the fracture graph in the monitoring model based on machine learning to obtain the geographic model includes: Based on machine learning, the monitoring model is updated according to the correspondence between the image recognition points in the real-time recognition network and the recognition points in the recognition network in the monitoring model to obtain an updated monitoring model; The fracture graph is constructed into the updated monitoring model to obtain a geographic model.

[0011] In the above technical solution, the technical effects and advantages provided by the present invention are: The present invention obtains the distance information between the image recognition points around the crack image and the corresponding recognition points in the monitoring range, and performs size marking on the crack image according to the distance information to obtain a crack graph, so as to determine the change size of the crack. The crack image is a direction of the crack. The actual size and width of the crack can be accurately understood through the distance between the image recognition points around the crack image and the corresponding recognition points in the monitoring range. Finally, the crack graph is obtained, and the size and other information of the crack can be determined according to the recognition of the geographical shape features. Finally, the ground crack can be extracted, and a good ground crack extraction effect is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0013] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0014] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0015] Example 1, please refer to Figure 1 As shown, the method for identifying and extracting ground fissures from drone images based on machine learning in this embodiment includes the following steps: S1. Obtain the monitoring scope and build a monitoring model corresponding to the monitoring scope; S2. Mark the identification information corresponding to the monitoring range, mark the identification information in the monitoring model and establish an association relationship between the monitoring model and the monitoring range; S3, collecting images within the monitoring range based on the drone, locating the images based on the identification information, and identifying the graphics in the images to obtain crack graphics; S4, constructing the crack graph in the monitoring model based on machine learning to obtain the geographic model; As described in the above steps S1-S4, with the development of society, some ground surfaces such as roads and some rocky ground surfaces need to be maintained. The maintenance of the road surface includes the treatment of ground cracks and obstacles. At present, ground cracks are an aspect that needs to be paid attention to in ground maintenance. At present, remote ground conditions can be collected by drones. However, it is difficult to obtain the actual situation of the cracks during the collection process, and it is impossible to accurately obtain information such as the shape and size of the ground cracks. There is no good ground crack information extraction effect. The present application obtains the distance information between the image recognition points around the crack image and the recognition points in the corresponding monitoring range, and marks the size of the crack image according to the distance information to obtain a crack graph, so as to determine the change size of the crack. The crack image is a direction of the crack. The distance between the image recognition points around the crack image and the recognition points in the corresponding monitoring range can accurately understand the actual size and width of the crack, and finally obtain a crack graph. The size of the crack and other information can be determined based on the recognition of geographical shape features. Finally, the ground cracks can be extracted, and a good ground crack extraction effect is achieved.

[0016] In one embodiment, the step S1 of obtaining a monitoring range and building a monitoring model corresponding to the monitoring range includes: S11, defining a monitoring range, and collecting geographic information within the monitoring range, wherein the geographic information includes surface shape features and corresponding dimensions; S12. Construct a monitoring model based on geographic information within the monitoring range; As described in the above steps S11 and S12, in the identification and monitoring of ground fissures, it is necessary to determine the scope of monitoring as the monitoring scope, and then collect the surface shape characteristics and corresponding dimensions within the monitoring scope as geographic information. Here, a monitoring model within the monitoring scope can be constructed to facilitate the subsequent marking and display of ground fissures, wherein the monitoring model is a three-dimensional model that displays the surface shape characteristics and dimensions within the monitoring scope. Subsequently, images collected by drones are correspondingly displayed in real time within the monitoring scope to display the surface shape characteristics within the monitoring scope in real time, so that ground fissures in the ground can be identified and extracted to display the ground conditions; In one embodiment, the step S2 of marking the corresponding monitoring range with identification information, marking the identification information in the monitoring model and establishing an association relationship between the monitoring model and the monitoring range includes: S21, selecting a plurality of identification points in the monitoring range, and connecting the plurality of identification points to obtain an identification network; S22, obtaining a range with a preset diameter with the identification point as the center as the identification range, obtaining identification geographic information within the identification range, obtaining the position of the identification point in the identification geographic information, and recording the identification geographic information within the identification range corresponding to the identification point and the position of the identification point in the identification geographic information as identification point information; S23, using the multiple identification point information and the identification network as identification information; S24, marking the identification information in the monitoring model and establishing an association relationship between the monitoring model and the monitoring scope; As described in the above steps S21-S24, multiple identification points are selected within the monitoring range, and the positions of the selected identification points are random. Multiple identification points are set on the edge line of the monitoring range and inside the monitoring range. Then, a range with the identification point as the center and under a preset diameter is obtained as the identification range, and identification geographic information within the identification range is obtained. The identification geographic information includes surface shape features and corresponding sizes within the identification range. The position of the identification point in the identification geographic information is obtained, and the identification geographic information within the identification range corresponding to the identification point and the position of the identification point in the identification geographic information are recorded as the identification point information. Here, the setting position of the identification point is determined by the identification point information, and the identification point can be confirmed when the image is collected by the drone in the future. Then, the multiple identification point information and the identification network are used as identification information, and the identification information is marked in the monitoring model and an association relationship between the monitoring model and the monitoring range is established, so that the images collected by the drone in the future can correspond to the monitoring model; In one embodiment, the step S24 of marking the identification information in the monitoring model and establishing an association relationship between the monitoring model and the monitoring scope includes: S241, marking the identification information in the monitoring model; S242, selecting a starting point in the monitoring range and marking the starting point in the monitoring model accordingly, and establishing a correspondence between the identification points in the identification network within the monitoring range and the identification points in the identification network in the monitoring model one by one based on the starting point; S243, establishing a mapping line between the identification points in the identification network within the monitoring range and the identification points in the identification network in the monitoring model based on the corresponding relationship, wherein the mapping line is a top-to-bottom trend line in the virtual space between the identification network within the monitoring range and the monitoring model; S244, obtaining the correlation relationship between the monitoring model and the monitoring range based on the mapping line; As described in the above steps S241-S244, the identification information within the monitoring range is marked in the monitoring model. The identification information is set in the same way as in the monitoring range. The monitoring model is the same as the ground conditions within the monitoring range. Then, a starting point is selected in the monitoring range and the starting point is marked in the monitoring model accordingly. Based on the starting point, a correspondence is established one by one between the identification points in the identification network within the monitoring range and the identification points in the identification network in the monitoring model. The starting point here also serves as the starting point for the subsequent collection of ground images by the drone. Here, the identification network in the monitoring model and the identification network within the monitoring range can be corresponded according to the starting point. In this way, the identification points in the monitoring model and the identification points within the monitoring range can also be corresponded to obtain a corresponding relationship. Then, a mapping line is established between the identification points in the identification network within the monitoring range and the identification points in the identification network in the monitoring model based on the corresponding relationship, wherein the mapping line is a top-down virtual space between the identification network within the monitoring range and the monitoring model. Trend line: the identification points in the identification network within the monitoring range and the identification points in the identification network in the monitoring model are in an upper and lower layer relationship, and the mapping line between them is the mapping line in the space within the upper and lower layers. Therefore, the mapping line and the subsequent real-time mapping line are not a distance comparison on the plane, but a comparison of the passing line in a space. For example, there is an offset in the comparison of points. If the offset is small, it is difficult to understand whether it is cheap through a distance comparison on the ground plane. However, the trend line that is from top to bottom in space can magnify the offset degree through the spatial comparison of the mapping line and the real-time mapping line, so as to achieve better offset data comparison. The monitoring model is on the top, and the identification network within the monitoring range is on the bottom. The mapping line here is the corresponding relationship between the identification points in the established monitoring model and the identification points within the monitoring range. Then, the correlation relationship between the monitoring model and the monitoring range is obtained based on the mapping line under this operation, which can facilitate a subsequent analysis of the ground conditions within the monitoring range, and thus better identify and extract ground fissures, and has better operation. In one embodiment, the step S3 of collecting images within the monitoring range based on the drone, locating the images based on the identification information, and identifying the graphics in the images to obtain the crack graphics includes: S31, collecting images within the monitoring range by passing the drone through the starting point of the monitoring range, determining the height at which the drone collects images within the monitoring range based on the starting point, and identifying geographic feature information and corresponding dimensions in the image in real time during the image collection process as image geographic information; S32, matching the identification geographic information in the identification point information with the image geographic information to obtain the identification point in the image as the image identification point; S33, matching the image recognition points with the corresponding recognition points in the recognition network within the monitoring range to obtain a corresponding relationship between the recognition points, and re-establishing the recognition network within the monitoring range as a real-time recognition network based on the corresponding relationship between the recognition points and the image recognition points to locate the image; S34, based on the association between the monitoring model and the monitoring range, a real-time mapping line is established between the image recognition points in the real-time recognition network and the recognition points of the recognition network in the monitoring model, wherein the real-time mapping line is a top-to-bottom trend line in the virtual space between the real-time recognition network and the monitoring model, and the real-time mapping line is corresponded to the mapping line based on the correspondence between the recognition points; S35, extracting the crack image in the image, if the real-time mapping line and the corresponding mapping line cannot overlap, obtaining the distance information between the image recognition points around the crack image and the recognition points in the corresponding monitoring range, and marking the size of the crack image according to the distance information to obtain the crack graph; As described in the above steps S31-S35, the drone collects images within the monitoring range through the starting point of the monitoring range, determines the height at which the drone collects images within the monitoring range based on the starting point, and identifies the geographic feature information and the corresponding size in the image in real time during the image collection process as the image geographic information. The image geographic information here is the subsequent collection of geographic information within the monitoring range. Since the collected image geographic information may be different from the previous geographic information in the case of cracks, the recognition geographic information in the recognition point information is matched with the image geographic information to obtain the recognition point in the image as the image recognition point. It is necessary to match the recognition geographic information in the recognition point information with the image geographic information in the collected image. The image recognition points are obtained by matching the shapes of the image recognition points with the corresponding recognition points in the recognition network within the monitoring range to obtain the corresponding relationship between the recognition points. Based on the corresponding relationship between the recognition points, the recognition network within the monitoring range is re-established as the real-time recognition network according to the image recognition points to locate the image. In this way, the corresponding relationship between the image recognition points in the images collected by the drone and the recognition points in the recognition network within the monitoring range can be established. Based on the correlation between the monitoring model and the monitoring range, a real-time mapping line is established between the image recognition points in the real-time recognition network and the recognition points of the recognition network in the monitoring model. The real-time mapping line is a top-to-bottom trend line in the virtual space between the real-time recognition network and the monitoring model. The image recognition point and the recognition point of the recognition network in the monitoring model are in a top-to-bottom relationship, and the real-time recognition network, the monitoring model and the recognition network within the monitoring range are in a three-layer relationship. Finally, the real-time mapping line spatial movement is corresponded to the mapping line. The mapping line is the guiding direction of the upper and lower layers in the space. The real-time mapping line is corresponded to the mapping line based on the corresponding relationship of the recognition point. Here, since the mapping line between the recognition point in the recognition network within the monitoring range and the recognition point in the recognition network in the monitoring model is established according to the corresponding relationship, it is necessary to compare the real-time mapping line with the mapping line to obtain whether the mapping recognition point has changed compared with the corresponding recognition point in the recognition network within the monitoring range. If the real-time mapping line and the corresponding mapping line cannot coincide (that is, the mapping The position of the identification point has changed compared with the identification point in the identification network within the corresponding monitoring range), extract the crack image in the image, obtain the distance information between the image identification points around the crack image and the identification points in the corresponding monitoring range, and mark the size of the crack image according to the distance information to obtain the crack graph, so as to determine the change size of the crack. The crack image is a direction of the crack. The distance between the image identification points around the crack image and the identification points in the corresponding monitoring range can accurately understand the actual size and width of the crack, and finally obtain the crack graph, which can determine the size of the crack and other information according to the identification of geographical shape features, and finally extract the ground cracks, which has a good ground crack extraction effect; In one embodiment, the step S35 of extracting the crack image from the image, if the real-time mapping line and the corresponding mapping line cannot coincide with each other, obtaining the distance information between the image recognition points around the crack image and the recognition points in the corresponding monitoring range, and marking the size of the crack image according to the distance information to obtain the crack graph includes: S351, extracting a crack image from the image, wherein the crack image is a shape image of the crack; S352, obtaining image recognition points around the crack image, if the real-time mapping line corresponding to the image recognition points around the crack image cannot coincide with the corresponding mapping line, obtaining the offset angle between the real-time mapping line corresponding to the image recognition points around the crack image and the corresponding mapping line, and calculating the distance information between the image recognition points around the crack image and the corresponding recognition points in the monitoring range by the Pythagorean theorem according to the offset angle and the height of the image collected by the drone within the monitoring range; S353, marking the size of the crack image according to the distance information, and constructing a graph according to the size-marked crack image to obtain a crack graph, wherein the distance information includes the distance and angle between the surrounding image recognition points in the crack image and the corresponding recognition points in the monitoring range; As described in the above steps S351-S353, the crack image here is only an image of the shape of the crack and the width of the crack is unknown. Then, the image recognition points around the crack image are obtained. If the real-time mapping lines corresponding to the image recognition points around the crack image cannot coincide with the corresponding mapping lines, the offset angle between the real-time mapping lines corresponding to the image recognition points around the crack image and the corresponding mapping lines is obtained. According to the offset angle and the height of the image collected by the drone within the monitoring range, the distance information between the image recognition points around the crack image and the corresponding recognition points in the monitoring range is calculated by the Pythagorean theorem. The distance information includes the distance and angle between the image recognition points around the crack image and the corresponding recognition points in the monitoring range. The crack image is marked with a size according to the distance information. A crack graph is constructed according to the crack image after the size marking. The image can be given a size according to the distance information to obtain the crack graph of the extracted ground crack. In one embodiment, the step S4 of constructing the fracture graph in the monitoring model based on machine learning to obtain the geographic model includes: S41, based on machine learning, updating the monitoring model according to the correspondence between the image recognition points in the real-time recognition network and the recognition points in the recognition network in the monitoring model to obtain an updated monitoring model; S42, constructing the crack graph in the updated monitoring model to obtain a geographic model; As described in the above steps S41 and S42, the correspondence between the image recognition points in the real-time recognition network and the recognition points in the recognition network in the monitoring model is recorded based on machine learning, and the geographic information in the image is constructed in the monitoring model according to the correspondence as an updated monitoring model, and the crack graphics are constructed in the updated monitoring model to obtain a geographic model, which can clearly display the ground cracks in the geographic model to facilitate personnel's understanding of the ground cracks.

[0017] There is a premise that cracks are monitored from scratch and crack identification and extraction is carried out. For example, as mentioned in the background technology, "highways and some rocky ground conditions", there are no cracks on the highway and rocky ground in the early stage. Under this condition, through the collection of a series of geographic information within the monitoring range and the subsequent selection of identification points, after cracks appear in the monitoring range, the distance between the identification points around the cracks will change. Since the changes of cracks are usually slow and the size changes are small, the real-time mapping line is used here as the top-down trend line in the virtual space between the real-time identification network and the monitoring model to accurately obtain the deviation, improve the accurate collection of the appearance and changes of cracks, and identify the changes and appearance of cracks. After determining that the cracks have appeared or changed, "the starting point here is The point is also used as the starting point for the subsequent collection of ground images by the drone. Here, the identification network in the monitoring model can be matched with the identification network within the monitoring range according to the starting point, so that the identification points in the monitoring model and the identification points within the monitoring range can also be matched. It can be known that the position and height of the ground image collected by the drone are consistent. Therefore, after obtaining the position and height of the ground image collected by the drone, it is also known that as stated in the manual, "extract the crack image in the image. If the real-time mapping line and the corresponding mapping line cannot coincide with each other, obtain the distance information between the image identification points around the crack image and the corresponding identification points in the monitoring range." The real-time mapping line and the corresponding mapping line cannot coincide with each other, and the deviation angle between the real-time mapping line and the corresponding mapping line can be obtained. According to the deviation angle between the real-time mapping line and the corresponding mapping line And the location and altitude of the drone collecting ground images , the offset distance L between the identification point and the original identification point can be obtained by the Pythagorean theorem (obtaining the distance information between the image identification points around the crack image and the identification points in the corresponding monitoring range). Here, the real-time mapping line is the top-to-bottom trend line in the virtual space between the real-time identification network and the monitoring model, which is vertical. Therefore, by the Pythagorean theorem , the premise is that the offset distance L of the crack from non-existence to existence is the size of the crack. Based on this, the crack image can be marked with size according to the distance information, and the crack graph can be constructed based on the crack image after size marking.

[0018] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for identifying and extracting ground fissures from drone images based on machine learning, characterized in that: The following steps are involved: Obtain the monitoring scope and build a monitoring model corresponding to the monitoring scope; Mark identification information corresponding to the monitoring scope, mark the identification information in the monitoring model and establish an association relationship between the monitoring model and the monitoring scope; Based on the UAV to collect images within the monitoring range, the images are located based on the identification information and the graphics in the images are identified to obtain the crack graphics; The fracture graph is built into the monitoring model based on machine learning to obtain a geographic model; The step of collecting images within the monitoring range based on the drone, locating the images based on the identification information, and identifying the graphics in the images to obtain the crack graphics includes: The drone is passed through the starting point of the monitoring range to collect images within the monitoring range, and the height at which the drone collects images within the monitoring range based on the starting point is determined. During the image collection process, the geographic feature information and the corresponding size in the image are identified in real time as the image geographic information; Matching the identification geographic information in the identification point information with the image geographic information to obtain the identification point in the image as the image identification point; The image recognition points are matched with the corresponding recognition points in the recognition network within the monitoring range to obtain the corresponding relationship of the recognition points, and based on the corresponding relationship of the recognition points, the recognition network within the monitoring range is re-established as a real-time recognition network according to the image recognition points to locate the image; Based on the association between the monitoring model and the monitoring range, a real-time mapping line is established between the image recognition points in the real-time recognition network and the recognition points of the recognition network in the monitoring model, wherein the real-time mapping line is a top-to-bottom trend line in the virtual space between the real-time recognition network and the monitoring model, and the real-time mapping line is corresponded to the mapping line based on the correspondence between the recognition points; Extract the crack image in the image. If the real-time mapping line and the corresponding mapping line cannot coincide, obtain the distance information between the image recognition points around the crack image and the corresponding recognition points in the monitoring range, and mark the size of the crack image according to the distance information to obtain the crack graph.

2. The method for identifying and extracting ground fissures from drone images based on machine learning according to claim 1 is characterized in that: The step of obtaining the monitoring range and building a monitoring model corresponding to the monitoring range includes: Define the monitoring scope and collect geographic information within the monitoring scope, where the geographic information includes surface shape features and corresponding dimensions; Construct a monitoring model based on the geographic information within the monitoring range.

3. The method for identifying and extracting ground fissures from drone images based on machine learning according to claim 1 is characterized in that: The step of marking the identification information corresponding to the monitoring range, marking the identification information in the monitoring model and establishing an association relationship between the monitoring model and the monitoring range includes: Select multiple identification points in the monitoring range, and connect the multiple identification points to obtain an identification network; Acquire a range with a preset diameter with the identification point as the center as the identification range, acquire identification geographic information within the identification range, acquire the position of the identification point in the identification geographic information, and record the identification geographic information within the identification range corresponding to the identification point and the position of the identification point in the identification geographic information as identification point information; Using multiple identification point information and identification network as identification information; The identification information is marked in the monitoring model and an association relationship is established between the monitoring model and the monitoring scope.

4. The method for identifying and extracting ground fissures from drone images based on machine learning according to claim 3 is characterized by: The step of marking the identification information in the monitoring model and establishing an association relationship between the monitoring model and the monitoring scope includes: Mark identifying information in the monitoring model; Select a starting point in the monitoring range and mark the starting point in the monitoring model accordingly, and establish a corresponding relationship between the identification points in the identification network within the monitoring range and the identification points in the identification network in the monitoring model based on the starting point one by one; A mapping line is established between the identification points in the identification network within the monitoring range and the identification points in the identification network in the monitoring model based on the corresponding relationship, wherein the mapping line is a top-to-bottom trend line in the virtual space between the identification network within the monitoring range and the monitoring model; The correlation between the monitoring model and the monitoring range is obtained based on the mapping line.

5. The method for identifying and extracting ground fissures from drone images based on machine learning according to claim 1, characterized in that: The step of extracting the crack image from the image, and if the real-time mapping line and the corresponding mapping line cannot coincide with each other, obtaining distance information between the image recognition points around the crack image and the recognition points in the corresponding monitoring range, and marking the size of the crack image according to the distance information to obtain the crack graph includes: Extracting a crack image from the image, wherein the crack image is a shape image of the crack; Obtain image recognition points around the crack image. If the real-time mapping lines corresponding to the image recognition points around the crack image cannot coincide with the corresponding mapping lines, obtain the offset angle between the real-time mapping lines corresponding to the image recognition points around the crack image and the corresponding mapping lines. According to the offset angle and the height of the image collected within the monitoring range by the UAV, calculate the distance information between the image recognition points around the crack image and the corresponding recognition points in the monitoring range by using the Pythagorean theorem. The crack image is dimensioned according to the distance information, and a crack graph is constructed based on the dimensioned crack image to obtain a crack graph, wherein the distance information includes the distance and angle between the surrounding image recognition points in the crack image and the corresponding recognition points in the monitoring range.

6. The method for identifying and extracting ground fissures from drone images based on machine learning according to claim 1, characterized in that: The step of constructing the fracture graph in the monitoring model based on machine learning to obtain the geographic model includes: Based on machine learning, the monitoring model is updated according to the correspondence between the image recognition points in the real-time recognition network and the recognition points in the recognition network in the monitoring model to obtain an updated monitoring model; The fracture graph is constructed into the updated monitoring model to obtain a geographic model.

7. A computer readable medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images