An image recognition-based low-altitude unmanned aerial vehicle multi-gradient fusion ground surface three-dimensional deformation monitoring method and system

Abstract

The embodiment of the application provides an image recognition-based low-altitude unmanned aerial vehicle multi-gradient fusion ground surface three-dimensional deformation monitoring method and system, which comprises the following steps: in the case that a large-area suspected deformation area is acquired, a periodic patrol mode is started, the large-area suspected deformation area is screened from a wide-area ground surface by a high-altitude satellite and sent to a low-altitude unmanned aerial vehicle; in a current patrol period, actual ground surface images of the large-area suspected deformation area are collected; the actual ground surface images are compared and analyzed with historical ground surface images collected in a previous patrol period, to determine each suspected deformation sub-area in the large-area suspected deformation area that may exist deformation; at least one suspected deformation sub-area in each suspected deformation sub-area is sent to a ground monitoring device, to determine whether the received suspected deformation sub-area has deformation through ground measurement of the ground monitoring device, so that the accuracy of detecting ground deformation can be improved, thereby solving the problems in the prior art.

Description

Technical Field

[0001] This application relates to the field of ground deformation monitoring technology, and in particular to a method and system for monitoring three-dimensional surface deformation by low-altitude unmanned aerial vehicles based on image recognition and multi-gradient fusion. Background Technology

[0002] Three-dimensional surface deformation is a core early warning indicator for geological disasters (such as landslides, collapses, and ground subsidence) and the safety of major infrastructure (such as bridges, tunnels, and high-rise buildings). Therefore, carrying out three-dimensional surface deformation monitoring is one of the key factors for achieving early detection, early warning, and early response to geological disasters.

[0003] Currently, the field of three-dimensional surface deformation monitoring has formed a mainstream technology system centered on satellite remote sensing. This system relies on high-altitude satellites with orbital altitudes of hundreds to thousands of kilometers to perceive three-dimensional surface deformation by carrying dedicated monitoring payloads. For example, spaceborne synthetic aperture radar interferometry (InSAR) technology inverts surface displacement by capturing the phase difference of radar reflection signals between satellites and surface targets; Global Navigation Satellite System Reflection (GNSS-R) technology uses the signal reflection characteristics of navigation satellites (BeiDou, GPS) to obtain information on vertical surface deformation; and high-resolution optical satellites assist in identifying deformation signs by comparing the texture and geometry of surface images from different time periods.

[0004] However, existing surface three-dimensional deformation monitoring technologies based on satellite remote sensing have a key bottleneck that limits the accuracy of monitoring results. From a technical perspective, the resolution of high-altitude satellite monitoring is generally negatively correlated with orbital altitude. The orbital altitude of current mainstream high-altitude satellites exceeds 700km. This high orbital altitude of high-altitude satellites directly limits their monitoring resolution, thus affecting the accuracy of monitoring results. Summary of the Invention

[0005] The purpose of this application is to propose a method and system for monitoring three-dimensional surface deformation of low-altitude unmanned aerial vehicles based on image recognition multi-gradient fusion, so as to solve the problem of low accuracy in the prior art.

[0006] To address the aforementioned technical problems, this application provides a method for monitoring three-dimensional surface deformation of low-altitude unmanned aerial vehicles (UAVs) based on image recognition and multi-gradient fusion. The method is applied to low-altitude UAVs and includes: Upon acquiring a large area of ​​suspected deformation, a periodic patrol mode is activated to periodically patrol the large area of ​​suspected deformation. The large area of ​​suspected deformation is selected from a wide area of ​​the Earth's surface by a high-altitude satellite and sent to the low-altitude unmanned aerial vehicle. During the current patrol cycle of the large area suspected of deformation, actual surface images of the large area suspected of deformation are collected. The actual surface image is compared and analyzed with the historical surface image collected by the low-altitude unmanned aerial vehicle during the previous patrol cycle when it patrolled the large area of ​​suspected deformation area, in order to determine the various suspected deformation sub-regions that may exist in the large area of ​​suspected deformation area. At least one suspected deformation sub-region is sent to a ground monitoring device for determination of whether the received suspected deformation sub-region has undergone deformation through ground-based measurements by the ground monitoring device.

[0007] Preferably, at least one suspected deformation sub-region is sent to a ground monitoring device for determination of whether the received suspected deformation sub-region has undergone deformation through ground-based measurements by the ground monitoring device. Specifically, this includes: Update the deformation record parameters corresponding to each suspected deformation sub-region. The deformation record parameters are used to record the number of times the corresponding suspected deformation sub-region was identified as a suspected deformation sub-region in the most recent N inspection cycles, including the current inspection cycle, where N is a positive integer greater than or equal to 3. The updated deformation record parameters corresponding to each suspected deformation sub-region are compared with a preset threshold to select target suspected deformation sub-regions whose deformation record parameters are greater than the preset threshold from each suspected deformation sub-region. The suspected deformation sub-region of the target is sent to the ground monitoring equipment so that the ground monitoring equipment can determine whether the suspected deformation sub-region of the target has undergone deformation through ground measurement.

[0008] Preferably, there are multiple suspected deformation sub-regions; and before sending the suspected deformation sub-regions to the ground monitoring equipment, the method further includes: Determine the average distance between each pair of suspected deformable sub-regions of each target; The suspected deformation sub-regions with an average distance less than a preset distance, along with the regions in between, are merged into a new suspected deformation sub-region.

[0009] Preferably, before acquiring actual surface images of the large-area suspected deformation area, the method further includes: Determine the orthophoto acquisition points for the large-area suspected deformation region; and, Acquiring actual surface images of the large-area suspected deformation region, specifically including: The low-altitude unmanned aerial vehicle is controlled to fly to the orthophoto acquisition point, and from the orthophoto acquisition point, it acquires the actual surface image of the large area suspected of deformation.

[0010] Preferably, determining the orthophoto acquisition points for the large-area suspected deformation region specifically includes: Determine the minimum bounding polygon of the large-area suspected deformation region; Identify the geometric centroid coordinates of the smallest circumscribed polygon and use them as the orthophoto acquisition point.

[0011] Preferably, the actual surface image is compared and analyzed with historical surface images collected by the low-altitude unmanned aerial vehicle during the previous patrol cycle when patrolling the large-area suspected deformation area, in order to determine the various suspected deformation sub-regions that may exist in the large-area suspected deformation area, specifically including: The actual land surface image and the historical land surface image are divided into multiple grid sub-regions; Extract image feature parameters for each grid sub-region, wherein the image feature parameters are specifically grayscale features and texture features; The feature parameter difference between each grid sub-region of the actual land surface image and the corresponding grid sub-region in the historical land surface image is calculated using the extracted image feature parameters. The grid sub-regions in the actual surface image whose feature parameter differences are greater than a preset threshold are obtained as the suspected deformation sub-regions.

[0012] Preferably, the actual ground image is divided into multiple grid sub-regions in the following manner: Identify areas of dense and sparse texture in the actual land surface image; Small-area grids and large-area grids are used respectively to divide the texture-dense region and the texture-sparse region into multiple grid sub-regions.

[0013] Preferably, the method further includes: Extract edge feature parameters for each grid sub-region, wherein the edge feature parameters include edge gradient intensity, texture contrast, and feature continuity; Based on the edge feature parameters of each grid sub-region, the edge sensitivity score of each grid sub-region is calculated using the weighted fusion formula S=α×G+β×T+γ×C, where S is the calculated edge sensitivity score; G is the edge gradient intensity; T is the texture contrast; C is the feature similarity between adjacent regions; and α, β, and γ are weighting coefficients. Based on the edge sensitivity score of each grid sub-region, the boundary of each grid sub-region is adjusted. If the edge sensitivity score exceeds the preset value, the corresponding grid sub-region is split or merged to optimize the fit between the boundary of each grid sub-region and the natural features. The boundaries of each grid sub-region after boundary adjustment are smoothed to obtain new grid sub-regions.

[0014] Preferably, the method further includes: Determine the total area of ​​each suspected deformation sub-region and its proportion of the area in the large suspected deformation region; If the area ratio is greater than the preset ratio, the inspection cycle of the low-altitude unmanned aerial vehicle is reduced to increase the inspection frequency.

[0015] To address the aforementioned technical problems, this application also provides an image recognition-based multi-gradient fusion surface three-dimensional deformation monitoring system for low-altitude unmanned aerial vehicles (UAVs). The monitoring system includes a high-altitude satellite, a low-altitude UAV, and ground monitoring equipment. The low-altitude UAV includes: The mode switching unit is used to activate the periodic patrol mode when a large area of ​​suspected deformation is acquired, so as to periodically patrol the large area of ​​suspected deformation. The large area of ​​suspected deformation is selected from the wide-area surface by high-altitude satellites and sent to the low-altitude unmanned aerial vehicle. The acquisition unit is used to acquire actual surface images of the large-area suspected deformation area during the current inspection cycle of the large-area suspected deformation area; The comparison and determination unit is used to compare and analyze the actual surface image with the historical surface image collected by the low-altitude unmanned aerial vehicle during the previous patrol cycle when it patrolled the large area of ​​suspected deformation area, so as to determine the various suspected deformation sub-regions that may have deformation in the large area of ​​suspected deformation area. The transmitting unit is used to transmit at least one of the suspected deformation sub-regions to the ground monitoring equipment, so as to determine whether the received suspected deformation sub-region has undergone deformation through ground measurement by the ground monitoring equipment.

[0016] This application provides a method for monitoring three-dimensional surface deformation using a low-altitude unmanned aerial vehicle (UAV) based on image recognition. The method involves activating a periodic patrol mode when a large suspected deformation area is identified. This area is selected from a wide-area surface by a high-altitude satellite and transmitted to the UAV. During the current patrol cycle, an actual surface image of the suspected deformation area is acquired. This image is then compared with historical surface images acquired during the previous patrol cycle to identify potential sub-regions within the large suspected deformation area. At least one of these sub-regions is then transmitted to ground monitoring equipment for on-site measurement to determine whether deformation has occurred. This method uses high-altitude satellites to screen out large areas suspected of deformation. Then, a low-altitude unmanned aerial vehicle (UAV), which is more maneuverable and has higher identification accuracy, uses periodic patrols and image acquisition and comparison to further identify smaller suspected deformation sub-regions from the large area of ​​suspected deformation. Finally, ground monitoring equipment with higher accuracy conducts ground measurements to ultimately determine whether the suspected deformation sub-region has undergone deformation. This method only requires high-altitude satellites to screen out large areas of suspected deformation, without requiring precise area identification and final confirmation of deformation. Therefore, it avoids the problem of low identification accuracy caused by the high orbital altitude of high-altitude satellites and can improve the accuracy of surface deformation identification. Attached Figure Description

[0017] To more clearly illustrate the solutions in this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the structure of a low-altitude unmanned aerial vehicle multi-gradient fusion three-dimensional surface deformation monitoring system based on image recognition, provided in an embodiment of this application. Figure 2 This is a flowchart illustrating the implementation of a low-altitude unmanned aerial vehicle multi-gradient fusion method for monitoring three-dimensional surface deformation based on image recognition, provided in an embodiment of this application. Figure 3 This is a schematic diagram of the structure of the unmanned aerial vehicle in the method provided in the embodiments of this application; Figure 4 This is a schematic diagram of the structure of one embodiment of the computer device according to this application. Detailed Implementation

[0019] 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 this application pertains; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application, are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0020] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0021] As mentioned earlier, the field of three-dimensional surface deformation monitoring has formed a mainstream technology system with satellite remote sensing as its core. However, this technology has a key bottleneck that restricts the accuracy of monitoring results. From a technical perspective, the resolution of high-altitude satellite monitoring is generally negatively correlated with orbital altitude. The orbital altitude of current mainstream high-altitude satellites exceeds 700km. This high orbital altitude of high-altitude satellites directly limits their monitoring resolution, thus affecting the accuracy of monitoring results.

[0022] In view of this, embodiments of this application provide a method and system for monitoring three-dimensional surface deformation of low-altitude unmanned aerial vehicles based on image recognition multi-gradient fusion, which can be used to solve the problems in the prior art. To enable those skilled in the art to better understand the solutions of this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings.

[0023] like Figure 1 The diagram shows a specific structural schematic of the monitoring system provided in this application embodiment. The monitoring system includes a high-altitude satellite 11, a low-altitude unmanned aerial vehicle 12, and a ground monitoring device 13. The high-altitude satellite 11 performs a large-area preliminary screening of the surface (obtaining a large area of ​​suspected deformation regions). The low-altitude unmanned aerial vehicle 12 then receives the results of the large-area preliminary screening from the high-altitude satellite 11 and further identifies the suspected deformation sub-regions within the large area of ​​suspected deformation regions. The results are then sent to the ground monitoring device 13, which ultimately confirms whether the suspected deformation sub-region has deformed.

[0024] Therefore, this monitoring system employs a three-tiered monitoring structure consisting of a high-altitude satellite 11, a low-altitude unmanned aerial vehicle (UAV) 12, and ground monitoring equipment 13. This structure fully leverages the high altitude of the high-altitude satellite 11 for initial screening of a wide area of ​​the Earth's surface, without relying on it for detailed identification and final confirmation. This effectively utilizes the high-altitude satellite's limited resolution, thus avoiding the low accuracy issue caused by its low resolution. The low-altitude UAV 12 then performs detailed identification based on the initial screening results, fully utilizing its relatively low altitude, resulting in higher resolution and more accurate identification. Furthermore, the low-altitude UAV 12 offers greater flexibility than the high-altitude satellite 11 (which has a certain orbital period), facilitating periodic patrols to increase monitoring frequency and further improve accuracy. After the low-altitude UAV 12 performs detailed identification (screening suspected deformation sub-regions from a large area of ​​suspected deformation), the ground monitoring equipment 13 conducts ground measurements to confirm whether deformation has occurred in the suspected sub-regions. Therefore, this method, through this three-layer monitoring structure, forms a hierarchical system of initial screening by high-altitude satellites, detailed investigation by low-altitude UAVs, and confirmation by ground monitoring equipment.

[0025] like Figure 2 The diagram shown is a schematic representation of the process for a low-altitude unmanned aerial vehicle (UAV) multi-gradient fusion method for monitoring three-dimensional surface deformation based on image recognition, as provided in this embodiment of the application. This method can be applied to... Figure 1 The monitoring system shown includes a low-altitude unmanned aerial vehicle, and the method comprises the following steps: Step S21: When a large area of ​​suspected deformation is obtained, the periodic inspection mode is activated to periodically inspect the large area of ​​suspected deformation.

[0026] The large suspected deformation area was selected from a wide area of ​​the Earth's surface by a high-altitude satellite and transmitted to the low-altitude unmanned aerial vehicle. In practical applications, high-altitude satellites can use current remote sensing methods (such as InSAR differential interferometry and optical image change detection) to identify large geographical areas with signs of surface displacement as suspected deformation areas. The identification of this large suspected deformation area by the high-altitude satellite using existing remote sensing methods is not specifically limited here. Furthermore, the high-altitude satellite's identification of this large suspected deformation area leverages its high altitude to select the area from a wide area of ​​the Earth's surface.

[0027] After the high-altitude satellite identifies the large area of ​​suspected deformation, it is not necessary for the high-altitude satellite to accurately distinguish and confirm the deformation area. Instead, the large area of ​​suspected deformation can be sent to the low-altitude unmanned aerial vehicle (UAV). For example, it can be sent to the low-altitude UAV through a wireless communication link (which can be a 4G communication network, 5G communication network, satellite communication network, or dedicated communication network, etc.). In this way, in step S21, the low-altitude UAV obtains the large area of ​​suspected deformation.

[0028] It should be noted that the low-altitude unmanned aerial vehicle is equipped with a periodic patrol mode. When the periodic patrol mode is activated, the low-altitude unmanned aerial vehicle can periodically patrol a designated area, thereby continuously tracking the evolution of the surface state of the designated area. In practical applications, the periodic patrol mode can be automatically activated by the aircraft control system of the low-altitude unmanned aerial vehicle when a large area suspected of deformation is detected.

[0029] The patrol cycle in the periodic patrol mode can be a fixed set cycle (e.g., 24 hours, 48 ​​hours, 120 hours, etc.) or it can be dynamically adjusted as needed. In step S21 of this application, when the low-altitude unmanned aerial vehicle acquires a large area suspected of deformation, it can activate the periodic patrol mode and designate the large area suspected of deformation as the designated area, thereby conducting periodic patrols of the large area suspected of deformation.

[0030] For example, after a high-altitude satellite identifies a large area suspected of deformation, it sends the large area suspected of deformation to a low-altitude unmanned aerial vehicle (UAV). Once the low-altitude UAV receives the large area suspected of deformation, it can start a periodic inspection mode and conduct periodic inspections of the large area suspected of deformation.

[0031] Step S22: During the current patrol cycle of the large area suspected of deformation, acquire actual surface images of the large area suspected of deformation.

[0032] The current patrol cycle can be any patrol cycle during which the low-altitude UAV begins its periodic patrol mode and conducts periodic patrols of the large area suspected of deformation. Thus, in step S22, the low-altitude UAV can acquire actual surface images of the large area suspected of deformation during the current patrol cycle. For example, it can use its onboard image acquisition equipment (such as a high-definition camera) to acquire these images.

[0033] In practical applications, the method for acquiring actual surface images of the large area suspected of deformation can be to control the low-altitude unmanned aerial vehicle to fly to the airspace above or around the large area suspected of deformation, thereby acquiring images of the large area suspected of deformation and obtaining actual surface images of the large area suspected of deformation.

[0034] It is important to note that, since the acquired actual surface images will be used to further identify potentially deformed sub-regions from a large area of ​​suspected deformation, in order to further improve the accuracy of the identification results, during the acquisition of the actual surface images, the orthophoto acquisition point of the large area of ​​suspected deformation can be controlled to acquire the orthophoto of the large area of ​​suspected deformation as the actual surface image. Therefore, before performing step S22 to acquire the actual surface image of the large area of ​​suspected deformation, the method may further include first determining the orthophoto acquisition point of the large area of ​​suspected deformation, then controlling the low-altitude unmanned aerial vehicle to fly to the orthophoto acquisition point, and acquiring the orthophoto of the large area of ​​suspected deformation from the orthophoto acquisition point as the actual surface image.

[0035] In this method, since the low-altitude UAV acquires orthophotos of the large area suspected of deformation from the orthophoto acquisition point, the UAV performs vertical downward imaging at this orthophoto acquisition point, which reduces the possibility of perspective distortion in the acquired actual ground image and ensures uniform proportions, making it easier to identify the actual ground image from the orthophoto.

[0036] In this image acquisition method, determining the orthorectified image acquisition point of the large-area suspected deformation region is crucial. In this application, the orthorectified image acquisition point can be determined as follows: First, the smallest bounding polygon of the large-area suspected deformation region can be determined. For example, the boundary contour of the large-area suspected deformation region can be parsed first. Then, methods such as computational geometry algorithms can be used to determine the smallest bounding polygon of the boundary contour (e.g., a minimum-area rectangle or a polygonal convex hull). After obtaining the smallest bounding polygon of the boundary contour, the geometric centroid of the smallest bounding polygon can be further identified, and the coordinates of the geometric centroid (i.e., geometric centroid coordinates) can be obtained. Then, the geometric centroid coordinates can be used as the orthorectified image acquisition point. In this way, after determining the orthorectified image acquisition point of the large-area suspected deformation region, a low-altitude unmanned aerial vehicle can be controlled to fly to the orthorectified image acquisition point to acquire images.

[0037] It should also be noted that the acquisition height of the actual ground surface image, that is, the height of the orthophoto acquisition point above the ground, is also a factor affecting image quality. When the acquisition height is too high, it will also affect the clarity of the actual ground surface image. In this application, the acquisition height can be determined comprehensively based on the graphic parameters of the minimum bounding polygon (such as the longest side length or coverage area) and the performance parameters of the image acquisition equipment in the low-altitude unmanned aerial vehicle.

[0038] Specifically, the longest side length L and coverage area S of the minimum bounding polygon can be obtained, as well as the performance parameters of the image acquisition device, which may include: the number of horizontal pixels P of the sensor. x Sensor vertical pixel count P y Horizontal field of view (FOV) of the lens x Vertical field of view (FOV) of the lens y Lens focal length f, sensor horizontal physical size W x Sensor vertical physical dimension W y wait.

[0039] Then, based on the requirement of full coverage of this large suspected deformed area during image acquisition, the minimum height H can be determined. min (That is, the minimum value of the acquisition height), specifically in the form of H. min =K1×L / (2×tan(FOV x ÷2), where K1 is the coverage redundancy coefficient, which can range from 1.1 to 1.3, used to ensure that the minimum circumscribed polygon falls completely within the effective area of ​​image acquisition. Therefore, the longest side length L of the minimum circumscribed polygon and the number of horizontal pixels P of the sensor can be used as the basis for determining the coverage redundancy coefficient. x Substituting these values ​​into the formula, the minimum height H is calculated. min .

[0040] Furthermore, the maximum height H can be calculated based on image resolution requirements. max (That is, the maximum value of the acquisition height), specifically, it can be H max =GSD×f×10 / W x In this formula, GSD is the ground sampling distance (GSD) required for deformation monitoring. In practical applications, the value of GSD can be about 0.5 cm / pixel.

[0041] The minimum sampling height H can be calculated using the method described above. min and maximum value H maxTherefore, the average of the two values ​​or a value in between can be used as the acquisition altitude, so that the low-altitude unmanned aerial vehicle can acquire images at the orthophoto acquisition point at this acquisition altitude, thereby simultaneously meeting the requirements of full coverage and high precision for large areas of suspected deformation.

[0042] Step S23: Compare and analyze the actual surface image with the historical surface image collected by the low-altitude unmanned aerial vehicle during the previous patrol cycle when patrolling the large area of ​​suspected deformation, in order to determine the various suspected deformation sub-regions that may exist in the large area of ​​suspected deformation.

[0043] The historical surface image refers to an image acquired in the previous patrol cycle (i.e., the patrol cycle preceding the current one) by the same or similar low-altitude unmanned aerial vehicle (UAV) during a patrol of the large area suspected of deformation, using the same methods. This historical surface image serves as the benchmark for comparison and analysis with the actual surface image. Of course, before comparing the actual surface image with the historical surface image, geometric registration and radiometric normalization can be performed to eliminate discrepancies caused by factors such as flight altitude errors, angle deviations, and differences in illumination intensity.

[0044] The suspected deformation sub-region refers to a local area within the large suspected deformation region that may be deformed. Obviously, the area of ​​the suspected deformation sub-region is much smaller than the large suspected deformation region. Therefore, by identifying the suspected deformation sub-region through step S23, the deformation region can be focused, which is convenient for further confirmation.

[0045] In practical applications, step S23 can be implemented by first dividing the actual surface image and the historical surface image into multiple grid sub-regions to ensure spatial comparability for subsequent feature comparisons. This division process can typically be aligned based on a unified geographic coordinate system or image pixel coordinate system to avoid misalignment due to image registration errors. Furthermore, the grid sub-regions can be divided into regular grids (such as square or rectangular grids) or irregular grids adaptively generated based on terrain features. Regular grids are suitable for large areas with relatively flat terrain and uniformly distributed features, offering simplicity and high computational efficiency; while irregular grids are more suitable for complex terrain environments, such as mountainous or valley areas, where image segmentation algorithms (such as superpixel segmentation and watershed algorithms) can generate spatial units that better fit natural boundaries.

[0046] Furthermore, to improve detection sensitivity, differentiated grid granularity can be used in different regions. For example, it can identify texture-dense and texture-sparse regions in actual surface images. Obviously, texture-dense regions exhibit frequent structural changes, while texture-sparse regions show sparse structural changes. Small-area and large-area grids are then used respectively to divide these regions into multiple grid sub-regions. In other words, texture-dense regions are divided into multiple grid sub-regions using small-area grids, while texture-sparse regions are divided into multiple grid sub-regions using large-area grids. Similarly, historical surface images can be divided into multiple grid sub-regions in the same way. This division method differentiates between texture-dense and texture-sparse regions, increasing the number of grid sub-regions in texture-dense regions, thereby improving the accuracy of subsequent identification; and reducing the number of grid sub-regions in texture-sparse regions, reducing the computational load during subsequent identification.

[0047] After dividing the actual land surface image and the historical land surface image into multiple grid sub-regions, image feature parameters of each grid sub-region can be extracted. These image feature parameters are specifically grayscale features and texture features. Then, using the extracted image feature parameters, the feature parameter difference between each grid sub-region of the actual land surface image and the corresponding grid sub-region in the historical land surface image is calculated. Finally, the grid sub-regions in the actual land surface image whose feature parameter difference is greater than a preset threshold are selected as the suspected deformation sub-regions.

[0048] Among them, the grayscale feature can be a statistical measure such as the average grayscale value, the root mean square error of grayscale values, and the difference between the maximum and minimum grayscale values ​​of each pixel in the corresponding grid sub-region, which reflects the change in the reflectivity of the surface material. The texture feature reflects the surface roughness and spatial arrangement pattern of the surface. In practical applications, the contrast of each pixel in the corresponding grid sub-region can be extracted through the gray-level co-occurrence matrix (GLCM) as the texture feature of that grid sub-region.

[0049] For example, in practical applications, the average gray value and contrast of each pixel in each grid sub-region can be extracted and used as its gray-level feature and texture feature, respectively. Then, the feature parameter difference degree between each grid sub-region of the actual land surface image and the corresponding grid sub-region in the historical land surface image can be calculated. This feature parameter difference degree reflects the overall difference between the grid sub-region of the actual land surface image and the corresponding grid sub-region in the historical land surface image from the perspective of pixel gray-level and texture.

[0050] For example, in this application, one method is to divide or subtract the grayscale features (e.g., average grayscale value) of the grid sub-region of the actual surface image from the grayscale features of the corresponding grid sub-region in the historical surface image, and use the calculated quotient or difference as the grayscale difference. Similarly, the texture features of the grid sub-region of the actual surface image can be divided or subtracted from the texture features of the corresponding grid sub-region in the historical surface image, and use the calculated quotient or difference as the texture difference. Then, the magnitudes of the grayscale difference and the texture difference are compared, and the larger one is used as the feature parameter difference. That is, the feature parameter difference is calculated by the formula feature parameter difference = max{grayscale difference, texture difference}.

[0051] Of course, in this application, another approach is to separately target each grid sub-region, combining the grayscale features and texture features of the same grid sub-region into a feature vector; then calculate the cosine similarity between the feature vector of the grid sub-region in the actual land surface image and the feature vector of the corresponding grid sub-region in the historical land surface image, and use this cosine similarity as the feature parameter difference.

[0052] After obtaining the feature parameter differences between each grid sub-region of the actual land surface image and the corresponding grid sub-region in the historical land surface image through the above method, the feature parameter differences can be further compared with a preset threshold, and the grid sub-regions in the actual land surface image with feature parameter differences greater than the preset threshold can be selected as the suspected deformation sub-regions.

[0053] It should be further explained that, as mentioned above, the actual surface image and historical surface image are divided into multiple grid sub-regions. The grid sub-regions can be divided into regular grids (such as square or rectangular grids) or irregular grids that are adaptively generated based on terrain features. Here, we can further explain the method of adaptively adjusting the boundaries of the grid sub-regions. Specifically, we can first extract the edge feature parameters of each grid sub-region. These edge feature parameters are used as multi-dimensional quantitative indicators to characterize the saliency of local regional boundaries in the image. Their role is to identify areas of abrupt changes in ground structure, thereby providing a basis for the dynamic reconstruction of the grid boundaries. In practical applications, these edge feature parameters can represent the edge gradient intensity, texture contrast, and feature continuity of the edge regions in the grid sub-regions. Among them, edge gradient intensity refers to the maximum response value of the spatial rate of change of pixel gray values ​​within a grid sub-region, which can usually be detected by Sobel, Prewitt, or Canny operators; texture contrast is used to measure the degree of brightness difference between texture elements in the edge region of the grid sub-region, and is therefore an important visual feature for distinguishing landform types. In practical applications, it can be calculated by histogram variance after encoding by Local Binary Pattern (LBP); feature continuity reflects the spatial coherence and integrity of the edge region in the grid sub-region, and can be used to determine whether a certain region constitutes a potential land feature boundary. In practical applications, feature continuity can be calculated by detecting the length of line segments and the curvature continuity of edge chain codes using Hough transform.

[0054] After extracting the edge feature parameters of each grid sub-region, the edge sensitivity score of each grid sub-region can be calculated based on the edge feature parameters of each grid sub-region using the weighted fusion formula S=α×G+β×T+γ×C. Here, S is the calculated edge sensitivity score; G is the edge gradient intensity; T is the texture contrast; C is the feature similarity between adjacent regions; and α, β, and γ are weight coefficients, which can be configured as needed in practical applications.

[0055] Therefore, this formula can be used to calculate the edge sensitivity score of each grid sub-region. Based on this score, the boundaries of each grid sub-region can be adjusted. If the edge sensitivity score exceeds a preset value, the corresponding grid sub-region is split or merged to optimize the fit between the boundaries of each grid sub-region and natural features. In this method, the edge sensitivity score exceeding the preset value is used as a trigger condition to determine whether the grid sub-region is located in a potential feature boundary area (i.e., the boundary needs adjustment). This preset value can be determined based on historical data statistical distribution; for example, the 75th percentile of the edge sensitivity scores of all grid sub-regions can be used as the preset value.

[0056] In this application, when the S-value of a certain grid sub-region exceeds the preset score, it indicates that it contains strong boundary signals. In this case, the boundary adjustment mechanism should be activated. The splitting operation is suitable for situations where there are multiple independent edge structures within a large-scale grid sub-region. For example, if a large-area grid sub-region spans the boundary between a landslide and stable bedrock, subdividing it into several smaller sub-grids can avoid cross-regional feature mixing and improve deformation detection resolution. The merging operation is suitable for situations where multiple adjacent small grid sub-regions exhibit low edge responses and consistent features. The purpose is to reduce redundant calculations and improve processing efficiency. Thus, through this splitting or merging method, the boundaries of the grid sub-regions are adaptively adjusted.

[0057] Of course, for each grid sub-region after boundary adjustment, its boundaries can be further smoothed to obtain new grid sub-regions. This boundary smoothing process can eliminate jagged and irregular contours introduced by splitting or merging operations, improve the visual continuity and geometric stability of the region boundaries, and reduce the risk of mismatches in subsequent image analysis. In practical applications, Gaussian filtering can be used to perform convolution smoothing on the boundary point coordinate sequence to achieve this smoothing process.

[0058] Step S24: Send at least one suspected deformation sub-region from each suspected deformation sub-region to the ground monitoring equipment for use in ground-based measurements by the ground monitoring equipment to determine whether the received suspected deformation sub-region has undergone deformation.

[0059] The ground monitoring equipment refers to a professional measuring device deployed on-site or accessible to the site. In practical applications, the ground monitoring equipment can specifically be a total station, GNSS receiver, laser scanner (LiDAR), crack gauge, or portable InSAR device, etc. These ground monitoring devices usually have measurement accuracy of millimeters or higher, so they can be used to conduct ground measurements on suspected deformation sub-areas to ultimately determine whether the suspected deformation sub-area has undergone deformation, that is, whether there is physical displacement or structural damage.

[0060] The advantage of these ground monitoring devices is their high measurement accuracy, making them highly accurate for final confirmation of whether deformation has occurred. However, due to the relatively poor mobility of these ground monitoring devices, after obtaining a large area of ​​suspected deformation region via high-altitude satellites, this application does not directly send the large area of ​​suspected deformation region to the ground monitoring devices. Instead, a low-altitude unmanned aerial vehicle obtains a relatively smaller suspected deformation sub-region, and then the ground monitoring devices conduct ground measurements on this relatively smaller suspected deformation sub-region to finally confirm whether deformation has occurred.

[0061] In step S24, the low-altitude unmanned aerial vehicle (UAV) sends at least one suspected deformation sub-region from each suspected deformation sub-region to the ground monitoring equipment. For example, it can send one suspected deformation sub-region to the ground monitoring equipment, or it can send all suspected deformation sub-regions to the ground monitoring equipment. In a preferred embodiment of this application, the deformation recording parameters corresponding to each suspected deformation sub-region can be updated first. These deformation recording parameters can be used to record the number of times the corresponding suspected deformation sub-region was identified as a suspected deformation sub-region in the most recent N inspection cycles, including the current inspection cycle, where N is a positive integer greater than or equal to 3, such as 3, 4, 5, or other values.

[0062] After the update, it's clear that the larger the value of the deformation record parameter, the more times the corresponding suspected deformation sub-region has been identified as a suspected deformation sub-region in the most recent N inspection cycles, indicating a lower probability of misjudgment. Therefore, the updated deformation record parameters corresponding to each suspected deformation sub-region can be compared with a preset threshold (e.g., 2 or 3). This allows for the selection of target suspected deformation sub-regions whose deformation record parameters are greater than the preset threshold. This indicates that the target suspected deformation sub-region has been identified as a suspected deformation sub-region more times in the most recent N inspection cycles. The target suspected deformation sub-region can then be sent to ground monitoring equipment for on-the-ground measurements to determine whether deformation has occurred. This method of identifying target suspected deformation sub-regions, sending them to ground monitoring equipment, and using on-the-ground measurements for final confirmation of deformation reduces the probability of misjudgment.

[0063] Furthermore, if there is only one suspected deformation sub-region, it can be directly sent to the ground monitoring equipment. If there are multiple suspected deformation sub-regions, before sending them to the ground monitoring equipment, the method can further include determining the average distance between each pair of suspected deformation sub-regions, and merging the suspected deformation sub-regions with an average distance less than a preset distance, along with the areas between them, into a new suspected deformation sub-region. This method, by merging the suspected deformation sub-regions with an average distance less than a preset distance, can reduce the number of suspected deformation sub-regions. To determine the number of potential deformation sub-regions and prevent misjudgment of the area between them, since the average distance between two suspected deformation sub-regions is less than a preset distance, it indicates that the distance between them is relatively close. Therefore, if deformation occurs in these two suspected deformation sub-regions, the area between them is also more likely to deform. Thus, these two suspected deformation sub-regions and the area between them can be merged into a new suspected deformation sub-region. This new suspected deformation sub-region is then sent to ground monitoring equipment for ground-based measurement to determine whether deformation has occurred. The average distance between two suspected deformation sub-regions can be the distance between their geometric centers; the specific value of this preset distance can be set according to actual needs.

[0064] The present application provides a method for monitoring three-dimensional surface deformation using a low-altitude unmanned aerial vehicle (UAV) based on image recognition. This method, applied to a low-altitude UAV, includes activating a periodic patrol mode upon acquiring a large area suspected of deformation. This large area is selected from a wide-area surface by a high-altitude satellite and transmitted to the low-altitude UAV. During the current patrol cycle, an actual surface image of the large area is acquired. This image is then compared with historical surface images acquired during the previous patrol cycle to identify potential sub-regions within the large area suspected of deformation. At least one of these sub-regions is then transmitted to ground monitoring equipment for on-site measurement to determine whether deformation has occurred. This method uses high-altitude satellites to screen out large areas suspected of deformation. Then, a low-altitude unmanned aerial vehicle (UAV), which is more maneuverable and has higher identification accuracy, uses periodic patrols and image acquisition and comparison to further identify smaller suspected deformation sub-regions from the large area of ​​suspected deformation. Finally, ground monitoring equipment with higher accuracy conducts ground measurements to ultimately determine whether the suspected deformation sub-region has undergone deformation. This method only requires high-altitude satellites to screen out large areas of suspected deformation, without requiring precise area identification and final confirmation of deformation. Therefore, it avoids the problem of low identification accuracy caused by the high orbital altitude of high-altitude satellites and can improve the accuracy of surface deformation identification.

[0065] It should be further explained that after determining each suspected deformation sub-region through the above step S23, the method can further include determining the total area of ​​each suspected deformation sub-region and the area ratio in the large area of ​​suspected deformation region. For example, the total area of ​​each suspected deformation sub-region can be calculated first, and then the total area can be divided by the area in the large area of ​​suspected deformation region to obtain the area ratio. If the area ratio is greater than the preset ratio, it indicates that a large area in the large area of ​​suspected deformation region may be at risk of deformation. Therefore, the inspection cycle of the low-altitude UAV can be reduced. For example, the inspection cycle of the low-altitude UAV can be reduced to 0.8 times the current inspection cycle to increase the inspection frequency of the low-altitude UAV.

[0066] Of course, the inspection cycle can also be adjusted in another way. For example, in step S24 above, a target suspected deformation sub-region was mentioned, and the target suspected deformation sub-region has a higher probability of deformation. At this time, the total area of ​​each target suspected deformation sub-region can be calculated. If the total area of ​​each target suspected deformation sub-region is greater than the preset area, it means that a large area of ​​the large area suspected deformation region is likely to have deformed. Therefore, the inspection cycle of the low-altitude UAV can also be reduced based on this to increase the inspection frequency of the low-altitude UAV.

[0067] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing related hardware through computer-readable instructions. These computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. The aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, optical disk, or read-only memory (ROM), or random access memory (RAM).

[0068] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0069] Based on the inventive concept of the image recognition-based multi-gradient fusion method for monitoring three-dimensional surface deformation of low-altitude unmanned aerial vehicles (UAVs) provided in the embodiments of this application, the embodiments of this application also provide an image recognition-based multi-gradient fusion system for monitoring three-dimensional surface deformation of low-altitude UAVs, combined with... Figure 1 As shown, the monitoring system includes a high-altitude satellite 11, a low-altitude unmanned aerial vehicle 12, and ground monitoring equipment 13. For any unclear points regarding the content of this monitoring system embodiment, please refer to the corresponding content in the method embodiment. For example... Figure 3 The diagram shows the specific structure of the low-altitude unmanned aerial vehicle 12 in the monitoring system. The low-altitude unmanned aerial vehicle 12 includes: a mode switching unit 121, a data acquisition unit 122, a comparison and determination unit 123, and a transmission unit 124, wherein: The mode switching unit 121 is used to activate the periodic patrol mode when a large area of ​​suspected deformation is acquired, so as to periodically patrol the large area of ​​suspected deformation. The large area of ​​suspected deformation is selected from the wide-area surface by high-altitude satellites and sent to the low-altitude unmanned aerial vehicle. The acquisition unit 122 is used to acquire actual surface images of the large-area suspected deformation area during the current inspection cycle of the large-area suspected deformation area; The comparison and determination unit 123 is used to compare and analyze the actual surface image with the historical surface image collected by the low-altitude unmanned aerial vehicle during the previous patrol cycle when it patrolled the large area of ​​suspected deformation area, so as to determine each suspected deformation sub-region that may exist in the large area of ​​suspected deformation area. The transmitting unit 124 is used to transmit at least one of the suspected deformation sub-regions to the ground monitoring equipment, so as to determine whether the received suspected deformation sub-region has undergone deformation through ground measurement by the ground monitoring equipment.

[0070] Since this monitoring system adopts the same inventive concept as the low-altitude unmanned aerial vehicle multi-gradient fusion three-dimensional deformation monitoring method based on image recognition provided in the embodiments of this application, it can also solve the problems in the prior art, which will not be elaborated here.

[0071] Specifically, sending at least one suspected deformation sub-region from each suspected deformation sub-region to a ground monitoring device for determination of whether the received suspected deformation sub-region has undergone deformation through ground-based measurements by the ground monitoring device can include: Update the deformation record parameters corresponding to each suspected deformation sub-region. The deformation record parameters are used to record the number of times the corresponding suspected deformation sub-region was identified as a suspected deformation sub-region in the most recent N inspection cycles, including the current inspection cycle, where N is a positive integer greater than or equal to 3. The updated deformation record parameters corresponding to each suspected deformation sub-region are compared with a preset threshold to select target suspected deformation sub-regions whose deformation record parameters are greater than the preset threshold from each suspected deformation sub-region. The suspected deformation sub-region of the target is sent to the ground monitoring equipment so that the ground monitoring equipment can determine whether the suspected deformation sub-region of the target has undergone deformation through ground measurement.

[0072] The number of the suspected target deformation sub-regions is multiple; and before sending the suspected target deformation sub-regions to the ground monitoring equipment, the system may also include: Determine the average distance between each pair of suspected deformable sub-regions of each target; The suspected deformation sub-regions with an average distance less than a preset distance, along with the regions in between, are merged into a new suspected deformation sub-region.

[0073] Prior to acquiring actual surface images of the large-area suspected deformation region, the process may further include: Determine the orthophoto acquisition points for the large-area suspected deformation region; and, Acquiring actual surface images of the large-area suspected deformation region, specifically including: The low-altitude unmanned aerial vehicle is controlled to fly to the orthophoto acquisition point, and from the orthophoto acquisition point, it acquires the actual surface image of the large area suspected of deformation.

[0074] Specifically, determining the orthophoto acquisition points for the large-area suspected deformation region may include: Determine the minimum bounding polygon of the large-area suspected deformation region; Identify the geometric centroid coordinates of the smallest circumscribed polygon and use them as the orthophoto acquisition point.

[0075] Specifically, comparing and analyzing the actual surface images with historical surface images collected by the low-altitude unmanned aerial vehicle during the previous patrol cycle when patrolling the large-area suspected deformation area, in order to determine the various suspected deformation sub-regions that may exist within the large-area suspected deformation area, can include: The actual land surface image and the historical land surface image are divided into multiple grid sub-regions; Extract image feature parameters for each grid sub-region, wherein the image feature parameters are specifically grayscale features, texture features, and edge density features; The feature parameter difference between each grid sub-region of the actual land surface image and the corresponding grid sub-region in the historical land surface image is calculated using the extracted image feature parameters. The grid sub-regions in the actual surface image whose feature parameter differences are greater than a preset threshold are obtained as the suspected deformation sub-regions.

[0076] The actual surface image can be divided into multiple grid sub-regions in the following way: Identify areas of dense and sparse texture in the actual land surface image; Small-area grids and large-area grids are used respectively to divide the texture-dense region and the texture-sparse region into multiple grid sub-regions.

[0077] The unmanned aerial vehicle 12 may also include: Extract edge feature parameters for each grid sub-region, wherein the edge feature parameters include edge gradient intensity, texture contrast, and feature continuity; Based on the edge feature parameters of each grid sub-region, the edge sensitivity score of each grid sub-region is calculated using the weighted fusion formula S=α×G+β×T+γ×C, where S is the calculated edge sensitivity score; G is the edge gradient intensity; T is the texture contrast; C is the feature similarity between adjacent regions; and α, β, and γ are weighting coefficients. Based on the edge sensitivity score of each grid sub-region, the boundary of each grid sub-region is adjusted. If the edge sensitivity score exceeds the preset value, the corresponding grid sub-region is split or merged to optimize the fit between the boundary of each grid sub-region and the natural features. The boundaries of each grid sub-region after boundary adjustment are smoothed to obtain new grid sub-regions.

[0078] The unmanned aerial vehicle 12 may also include: Determine the total area of ​​each suspected deformation sub-region and its proportion of the area in the large suspected deformation region; If the area ratio is greater than the preset ratio, the inspection cycle of the low-altitude unmanned aerial vehicle is reduced to increase the inspection frequency.

[0079] To address the aforementioned technical problems, embodiments of this application also provide a computer device. Please refer to [link / reference needed]. Figure 4 , Figure 4 This is a basic structural block diagram of a computer device according to an embodiment of this application.

[0080] The computer device 400 includes a memory 410, a processor 420, and a network interface 430 that are interconnected via a system bus. It should be noted that only the computer device 400 with components 410-430 is shown in the figure; however, it should be understood that it is not required to implement all the shown components, and more or fewer components can be implemented alternatively. Those skilled in the art will understand that the computer device described here is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0081] The computer device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device can interact with the user via a keyboard, mouse, remote control, touchpad, or voice control.

[0082] The memory 410 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 410 may be an internal storage unit of the computer device 400, such as the hard disk or memory of the computer device 400. In other embodiments, the memory 410 may also be an external storage device of the computer device 400, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. Of course, the memory 410 may also include both internal storage units and external storage devices of the computer device 400. In this embodiment, the memory 410 is typically used to store the operating system and various application software installed on the computer device 400, such as computer-readable instructions of the method provided in this embodiment. Furthermore, the memory 410 can also be used to temporarily store various types of data that have been output or will be output.

[0083] In some embodiments, the processor 420 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip. The processor 420 is typically used to control the overall operation of the computer device 400. In this embodiment, the processor 420 is used to execute computer-readable instructions stored in the memory 410 or to process data, for example, to execute computer-readable instructions of the methods provided in this embodiment.

[0084] The network interface 430 may include a wireless network interface or a wired network interface, which is typically used to establish communication connections between the computer device 400 and other electronic devices.

[0085] This application also provides another embodiment, namely, providing a computer-readable storage medium storing computer-readable instructions that can be executed by at least one processor to cause the at least one processor to perform the steps of the method described above.

[0086] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0087] Obviously, the embodiments described above are only some embodiments of this application, not all embodiments. The accompanying drawings show preferred embodiments of this application, but do not limit the patent scope of this application. This application can be implemented in many different forms; rather, the purpose of providing these embodiments is to provide a more thorough and comprehensive understanding of the disclosure of this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this application's specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of patent protection of this application.

Claims

1. A method for monitoring three-dimensional surface deformation using low-altitude unmanned aerial vehicles based on image recognition and multi-gradient fusion, characterized in that, The method is applied to low-altitude unmanned aerial vehicles, and the method includes: Upon acquiring a large area of ​​suspected deformation, a periodic patrol mode is activated to periodically patrol the large area of ​​suspected deformation. The large area of ​​suspected deformation is selected from a wide area of ​​the Earth's surface by a high-altitude satellite and sent to the low-altitude unmanned aerial vehicle. During the current patrol cycle of the large area suspected of deformation, actual surface images of the large area suspected of deformation are collected. The actual surface image is compared and analyzed with the historical surface image collected by the low-altitude unmanned aerial vehicle during the previous patrol cycle when it patrolled the large area of ​​suspected deformation area, in order to determine the various suspected deformation sub-regions that may exist in the large area of ​​suspected deformation area. At least one suspected deformation sub-region is sent to a ground monitoring device for determination of whether the received suspected deformation sub-region has undergone deformation through ground-based measurements by the ground monitoring device.

2. The method according to claim 1, characterized in that, At least one suspected deformation sub-region from each suspected deformation sub-region is sent to a ground monitoring device for determination of whether the received suspected deformation sub-region has undergone deformation through ground-based measurements by the ground monitoring device. Specifically, this includes: Update the deformation record parameters corresponding to each suspected deformation sub-region. The deformation record parameters are used to record the number of times the corresponding suspected deformation sub-region was identified as a suspected deformation sub-region in the most recent N inspection cycles, including the current inspection cycle, where N is a positive integer greater than or equal to 3. The updated deformation record parameters corresponding to each suspected deformation sub-region are compared with a preset threshold to select target suspected deformation sub-regions whose deformation record parameters are greater than the preset threshold from each suspected deformation sub-region. The suspected deformation sub-region of the target is sent to the ground monitoring equipment so that the ground monitoring equipment can determine whether the suspected deformation sub-region of the target has undergone deformation through ground measurement.

3. The method according to claim 2, characterized in that, There are multiple suspected deformation sub-regions; and, before sending the suspected deformation sub-regions to the ground monitoring equipment, the method further includes: Determine the average distance between each pair of suspected deformable sub-regions of each target; The suspected deformable sub-regions with an average distance less than a preset distance, along with the regions in between, are merged into a new suspected deformable sub-region.

4. The method according to claim 1, characterized in that, Before acquiring actual surface images of the large-area suspected deformation region, the method further includes: Determine the orthophoto acquisition points for the large-area suspected deformation region; and, Acquiring actual surface images of the large-area suspected deformation region, specifically including: The low-altitude unmanned aerial vehicle is controlled to fly to the orthophoto acquisition point, and from the orthophoto acquisition point, it acquires the actual surface image of the large area suspected of deformation.

5. The method according to claim 4, characterized in that, Determining the orthophoto acquisition points for the large-area suspected deformation region specifically includes: Determine the minimum bounding polygon of the large-area suspected deformation region; Identify the geometric centroid coordinates of the smallest circumscribed polygon and use them as the orthophoto acquisition point.

6. The method according to claim 1, characterized in that, The actual surface images are compared and analyzed with historical surface images collected by the low-altitude unmanned aerial vehicle during the previous patrol cycle when patrolling the large-area suspected deformation area, in order to determine the various suspected deformation sub-regions that may exist in the large-area suspected deformation area, specifically including: The actual land surface image and the historical land surface image are divided into multiple grid sub-regions; Extract image feature parameters for each grid sub-region, wherein the image feature parameters are specifically grayscale features and texture features; The feature parameter difference between each grid sub-region of the actual land surface image and the corresponding grid sub-region in the historical land surface image is calculated using the extracted image feature parameters. The grid sub-regions in the actual surface image whose feature parameter differences are greater than a preset threshold are obtained as the suspected deformation sub-regions.

7. The method according to claim 6, characterized in that, The actual land surface image is divided into multiple grid sub-regions in the following manner: Identify areas of dense and sparse texture in the actual land surface image; Small-area grids and large-area grids are used respectively to divide the texture-dense region and the texture-sparse region into multiple grid sub-regions.

8. The method according to claim 7, characterized in that, The method further includes: Extract edge feature parameters for each grid sub-region, wherein the edge feature parameters include edge gradient intensity, texture contrast, and feature continuity; Based on the edge feature parameters of each grid sub-region, the edge sensitivity score of each grid sub-region is calculated using the weighted fusion formula S=α×G+β×T+γ×C, where S is the calculated edge sensitivity score; G is the edge gradient intensity; T is the texture contrast; C is the feature similarity between adjacent regions; and α, β, and γ are weighting coefficients. Based on the edge sensitivity score of each grid sub-region, the boundary of each grid sub-region is adjusted. If the edge sensitivity score exceeds the preset value, the corresponding grid sub-region is split or merged to optimize the fit between the boundary of each grid sub-region and the natural features. The boundaries of each grid sub-region after boundary adjustment are smoothed to obtain new grid sub-regions.

9. The method according to claim 1, characterized in that, The method further includes: Determine the total area of ​​each suspected deformation sub-region and its proportion of the area in the large suspected deformation region; If the area ratio is greater than the preset ratio, the inspection cycle of the low-altitude unmanned aerial vehicle is reduced to increase the inspection frequency.

10. A low-altitude unmanned aerial vehicle (UAV) multi-gradient fusion surface three-dimensional deformation monitoring system based on image recognition, characterized in that, The monitoring system includes high-altitude satellites, low-altitude unmanned aerial vehicles (UAVs), and ground monitoring equipment. The low-altitude UAVs include: The mode switching unit is used to activate the periodic patrol mode when a large area of ​​suspected deformation is acquired, so as to periodically patrol the large area of ​​suspected deformation. The large area of ​​suspected deformation is selected from the wide-area surface by high-altitude satellites and sent to the low-altitude unmanned aerial vehicle. The acquisition unit is used to acquire actual surface images of the large-area suspected deformation area during the current inspection cycle of the large-area suspected deformation area; The comparison and determination unit is used to compare and analyze the actual surface image with the historical surface image collected by the low-altitude unmanned aerial vehicle during the previous patrol cycle when it patrolled the large area of ​​suspected deformation area, so as to determine the various suspected deformation sub-regions that may have deformation in the large area of ​​suspected deformation area. The transmitting unit is used to transmit at least one of the suspected deformation sub-regions to the ground monitoring equipment, so as to determine whether the received suspected deformation sub-region has undergone deformation through ground measurement by the ground monitoring equipment.

Images