Automatic monitoring and early warning method for deformation of offshore guide embankment based on remote sensing image

Abstract

The invention relates to the technical field of ocean engineering, in particular to a remote-sensing-image-based automatic monitoring and early warning method for deformation of an offshore guide dike, which comprises the following steps: acquiring remote sensing images and standard images of the offshore guide dike, and forming a remote sensing image set by the remote sensing images according to acquisition time and position; dividing the remote sensing image set into a plurality of image pairs, and performing image registration according to the positions of the plurality of image pairs and the standard image on the sea guide dike to obtain a target remote sensing image; carrying out image difference on the target remote sensing images at different time points to obtain deformation features, and determining deformation regions related to distribution of the deformation features; combining the deformation areas according to the distribution of the deformation characteristics, determining deformation trend directions corresponding to the deformation areas, and performing influence analysis on deformation quantities and deformation rates in the deformation trend directions to obtain influence paths corresponding to the deformation areas; based on the obtained influence path, determining an alarm type of the corresponding offshore guide embankment; and the accuracy and efficiency of offshore guide dike deformation identification are improved.

Description

Technical Field

[0001] The invention relates to the technical field of marine engineering, and in particular to an automatic monitoring and early warning method for deformation of an offshore embankment based on remote sensing images. Background Art

[0002] In marine engineering, offshore dikes are important infrastructures, and their safety is directly related to marine transportation, marine resource development and many other aspects. However, due to the complexity of the marine environment and the particularity of the dike structure, traditional deformation monitoring methods are often difficult to achieve real-time monitoring and comprehensive coverage. In recent years, with the rapid development of remote sensing technology, the use of remote sensing images for offshore dike deformation monitoring has gradually become a new trend. Remote sensing images have the advantages of wide coverage, high update frequency, and large amount of information, which can achieve real-time monitoring and comprehensive coverage of offshore dikes. Therefore, based on the background technology, the current scheme combines remote sensing imaging technology and automation algorithms to achieve automatic monitoring and early warning of offshore dike deformation, so as to improve monitoring efficiency and accuracy, and provide strong guarantees for the safe operation of marine engineering.

[0003] For example, Chinese patent publication number CN118730007A discloses a method and device for automatically monitoring multi-point deformation of a dam surface based on acoustic wave recognition, wherein the method for automatically monitoring multi-point deformation of a dam surface includes: installing acoustic wave receiving devices at three fixed points on the mountains on both sides of the dam that are not colinear; driving an unmanned vehicle on the dam surface, installing an acoustic wave transmitter on the unmanned vehicle, and continuously transmitting acoustic wave signals of a specific frequency into the air; the acoustic wave receiving device at the fixed point of the dam calculates the distance between the receiving device and the vehicle's transmitting point based on the time difference between the transmission and arrival of the received acoustic wave signal; the GNSS coordinates at the vehicle's transmitting point are calculated based on the three-point positioning method and the fixed point GNSS coordinates; the vehicle patrols regularly along the set route and calculates the instantaneous coordinates of the set points, compares them with the historical coordinates, and obtains the displacement changes of the fixed points on the dam surface.

[0004] For example, Chinese patent publication number CN117433480A discloses a high-precision real-time dynamic monitoring and early warning system and method based on remote sensing monitoring, including: a perception module for acquiring monitoring data of a monitored object; a communication module for transmitting the monitoring data to a security monitoring cloud platform; a security monitoring cloud platform for parsing the monitoring data, calculating the position and displacement information of the monitoring station, and providing real-time millimeter-level deformation data for engineering monitoring; a risk analysis and early warning module for performing statistical analysis on the monitoring data based on the security monitoring cloud platform, and outputting risk assessment and early warning information to the outside in the form of a combined early warning matrix.

[0005] The prior art describes a process of using acoustic wave reflection, flow detection, and pressure detection to treat facilities such as guide embankments on offshore projects. However, it ignores the distribution of deformation amounts on the guide embankments and their changing trends when the guide embankments are deformed, resulting in poor results in identifying guide embankment risks and failure to promptly take corresponding measures based on the deformation trends of the guide embankments. Summary of the invention

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is: an automatic monitoring and early warning method for offshore guide embankment deformation based on remote sensing images, including: S1, obtaining remote sensing images and standard images of the offshore guide embankment, and forming the remote sensing images into a remote sensing image set according to the time and location of acquisition; the remote sensing image set includes satellite remote sensing images, UAV remote sensing images and SAR images.

[0007] S2, dividing the remote sensing image set into a plurality of image pairs, performing image registration according to the positions of the plurality of image pairs and the standard image on the offshore guide embankment, and obtaining a registered target remote sensing image.

[0008] S3, performing image difference on the target remote sensing images at different time points, analyzing the deformation features related to the surface deformation of the offshore dike, and determining the deformation areas related to the distribution of each deformation feature.

[0009] S4, combining the deformation regions according to the distribution of deformation characteristics, determining the deformation trend direction corresponding to the deformation region, performing influence analysis on the deformation amount and deformation rate in the deformation trend direction, and obtaining the influence path corresponding to each deformation region.

[0010] S5, determining the alarm type corresponding to the offshore dike based on the acquired impact path.

[0011] The beneficial effects of the present invention are as follows: 1. The present invention obtains satellite remote sensing images, UAV remote sensing images and SAR images, and after determining the specific edge conditions on the guide dike, uses SAR images to determine the relative deformation amount generated, and sets the relative deformation area according to the deformation distribution, so that when the deformation features are identified, it can be determined whether the current deformation is caused by erosion or other forms of deformation caused by accumulation, etc., which is convenient for subsequent verification of the specific conditions on the guide dike and provides a basis for subsequent decision-making and processing of the guide dike.

[0012] 2. The present invention identifies the trend direction of the deformation amount in each deformation area, and after identification, projects and transforms these trend directions with the edge of the guide bank to obtain the deformation trend of the current guide bank at the edge, and finally completes the identification of the deformation trend direction of the guide bank according to the relative center trajectory. It can be understood that after identifying multiple small deformation features, the deformation features are combined to form the main trend of the deformation of the guide bank, which is convenient for subsequent personnel to make decisions based on this main trend. At the same time, the identification of relative deformation death at the macro and local levels can more comprehensively display the deformation trend of the guide bank, thereby improving the accuracy of subsequent deformation trend warnings.

[0013] 3. The present invention processes the path of the deformation on the current guide dike, so as to find out the main path of the current guide dike when it is deformed, as well as the points on the path where the deformation is prominent, so as to facilitate subsequent personnel to judge the stability of the guide dike structure based on these points. At the same time, the points on these paths are matched with relevant cases and treatment measures in the database, which can further assist personnel in decision-making and help reduce safety accidents of the guide dike structure. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0015] Figure 1 It is a flow chart of the automatic monitoring and early warning method of offshore dike deformation based on remote sensing images.

[0016] Figure 2 It is a flow chart of step S2 of the method for automatic monitoring and early warning of offshore dike deformation based on remote sensing images.

[0017] Figure 3 It is a flow chart of step S3 of the method for automatic monitoring and early warning of offshore dike deformation based on remote sensing images.

[0018] Figure 4 It is a flow chart of step S4 of the method for automatic monitoring and early warning of offshore dike deformation based on remote sensing images.

[0019] Figure 5 It is a flow chart of step S5 of the method for automatic monitoring and early warning of offshore dike deformation based on remote sensing images. DETAILED DESCRIPTION

[0020] The embodiments of the present invention are described in detail below. The embodiments described below are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention. If no specific techniques or conditions are specified in the embodiments, the techniques or conditions described in the literature in the art or the product specifications are used.

[0021] It should be noted that the guide dike is a dike built on one or both sides of the channel in the sandbar area of ​​the estuary. It is a water conservancy facility used to guide water flow, scour silt, and increase or maintain the water depth of the port channel. When the guide dike is deformed, it generally means that the rock or rock structure at the corresponding position is damaged or dust accumulates after the guide dike itself is scoured by sea water. When the relevant guide dike is identified using remote sensing images, relevant deformation occurs. That is, under the action of natural forces such as tides and waves, the water-facing surface of the guide dike will be strongly scoured. Long-term scouring will cause the foundation of the guide dike to collapse. Erosion and structural damage; the upstream side of the dike may be seriously silted up due to intercepted sediment, while the downstream side may be scoured due to insufficient sediment supply. This unbalanced sedimentation and scour will also cause deformation of the dike; these contents will eventually be identified to determine the main influencing factors of the dike in the process of guiding water flow, so as to accurately identify the hydrological information generated by soil erosion or water flow guidance in the corresponding area. By monitoring the changes in the dike, potential safety hazards can be discovered in time to ensure the safe navigation of ships in the waterway. At the same time, monitoring the changes in the dike can assess its impact on the marine environment, and take necessary measures to reduce adverse effects and ensure the long-term stable operation of the dike.

[0022] See also Figure 1 , an automatic monitoring and early warning method for deformation of an offshore embankment based on remote sensing images, comprising: S1, acquiring remote sensing images and standard images of the offshore embankment, and composing the remote sensing images into a remote sensing image set according to the time and location of acquisition; the remote sensing image set comprises satellite remote sensing images, UAV remote sensing images and SAR images.

[0023] S2, dividing the remote sensing image set into a plurality of image pairs, performing image registration according to the positions of the plurality of image pairs and the standard image on the offshore guide embankment, and obtaining a registered target remote sensing image.

[0024] S3, performing image difference on the target remote sensing images at different time points, analyzing the deformation features related to the surface deformation of the offshore dike, and determining the deformation areas related to the distribution of each deformation feature.

[0025] S4, combining the deformation regions according to the distribution of deformation characteristics, determining the deformation trend direction corresponding to the deformation region, performing influence analysis on the deformation amount and deformation rate in the deformation trend direction, and obtaining the influence path corresponding to each deformation region.

[0026] S5: Determine the alarm type corresponding to the offshore dike based on the acquired impact path.

[0027] In one embodiment of the present invention, multiple images are used to illustrate the deformation and specific conditions of the offshore guide dike; the satellite remote sensing image will directly display the macroscopic scene of the offshore guide dike, and a larger acquisition cycle is used to obtain the image; the UAV remote sensing image will use a camera or other sensors to collect images of the offshore guide dike according to the drone's route, and collect high-resolution images or video materials in real time; the SAR image uses microwave imaging technology to generate images by receiving signals transmitted back from the ground, and when the surface undergoes a slight deformation, the path length of the two SAR images will change, and it will be reflected in the phase represented by the image, and then the phase difference evolution is performed to obtain the deformation amount appearing on the ground, thereby knowing the deformation and change trend of the offshore guide dike; the standard image represents the standard image of the offshore guide dike under normal processing, which is used to differentiate the current remote sensing image.

[0028] When composing a remote sensing image set, the offshore dikes that need to be collected will be matched according to the positions corresponding to the satellite remote sensing images, UAV remote sensing images and SAR images at that time to obtain an image set of the offshore dike composed of multiple images. This image set will be enhanced and combined according to the time and path of acquisition to ensure that the corresponding images can correspond to each other during image calibration to obtain the multivariate data information in the current image.

[0029] At the same time, due to the different contents represented by SAR images, satellite remote sensing images and UAV remote sensing images, satellite remote sensing images and UAV remote sensing images show the color and texture of the offshore guide dike, and most of them are in the form of color images. SAR images are grayscale images, reflecting the characteristics of radar wave reflection of objects on the offshore guide dike. As a result, when calibrating the image, the satellite remote sensing images and UAV remote sensing images will be grayscaled. After obtaining the grayscale histogram, the contents of the satellite remote sensing images, UAV remote sensing images and SAR images will be calibrated to ensure that the current target remote sensing image can maintain the consistency and comparability of the offshore guide dike displayed in the image.

[0030] The set standard images will be set in the form of satellite remote sensing images, UAV remote sensing images and SAR images in sequence. These standard images can be image data collected in the past, or images processed from past collected data. These images can be compared with the currently acquired data to complete the deformation analysis of the offshore guide dike.

[0031] In one embodiment of the present invention, in step S2, the optical images corresponding to the SAR images, satellite remote sensing images, and UAV remote sensing images in the remote sensing image set are mainly aligned to determine whether the information after image comparison can assist in identifying enough deformed areas; these images are compared for spatial consistency to obtain a target remote sensing image of a combination of multiple images with spatial consistency, and the target remote sensing image includes satellite remote sensing images, UAV remote sensing images, and SAR images with consistent position representations.

[0032] The remote sensing image set is divided into several image pairs by dividing the remote sensing image set according to the time of acquisition. For example, images at the same time are combined into image pairs. This image pair represents a pair of images that are adjacent in time. The purpose of dividing into multiple image pairs here is to ensure that the impact of acquisition can cover the guide embankment, and the spatial range displayed by the image is consistent, and no other erroneous images will appear.

[0033] like Figure 2 As shown, the implementation method of step S2 includes: S21, obtaining the acquisition time of the remote sensing image set, taking the adjacent frame images of the remote sensing image set within the adjacent acquisition time as an image pair, and obtaining a plurality of image pairs; for adjacent frame images, it means that at the corresponding acquisition time, continuous acquisition is performed for a certain position in an analog sense. For example, the drone remote sensing image in the remote sensing image set will be represented as video data when it is collected, which is to extract image frames from the video data; if it is a satellite remote sensing image or a SAR image, it is adjacent in an analog sense. Generally, images with close time are selected as image pairs. The contents represented by such image pairs themselves are almost the same, and the difference is small, which is convenient for image calibration with standard images.

[0034] S22, preprocessing each image pair, including geometric correction, radiation normalization and other processing, extracting feature points from the preprocessed image pair and the standard image, and performing the nearest neighbor algorithm on the feature points to obtain feature point pairs of the matching image pair and the standard image. Geometric correction is a method of changing the original coordinates of each image pair using affine coefficients, which is described in the prior art and will not be described in detail here; as for radiation normalization, it is used to eliminate the difference in lighting conditions between the satellite remote sensing image and the drone remote sensing image in the image pair, and the original value of the pixel in the satellite remote sensing image and the drone remote sensing image is subtracted from the average value of the pixel and divided by the standard deviation of the pixel to perform normalization processing.

[0035] The nearest neighbor algorithm is then implemented by calculating the distance between pixels of the extracted feature points, that is, calculating the distance between the pixels corresponding to the feature points in the image pair and the standard image to determine whether the corresponding positions in the image are the same, and retaining the points with high confidence as the implementation of the matching method at this time, and finally obtaining multiple feature point pairs.

[0036] S23, after transforming the feature point pairs of the matching image pair and the standard image, the image pair satisfying the mean square error is output as the target remote sensing image.

[0037] The changes here refer to affine changes, projection changes, etc. At this time, there is no restriction on the processing methods of matching images and standard images. Only restrictions are imposed on image pairs that satisfy the mean square error. The RMSE of satellite remote sensing images and UAV remote sensing images is required to be less than 1 pixel, and the RMSE of SAR images is required to be less than 0.2 pixels. The contents that are not restricted are all explained in the prior art and will not be explained in detail here.

[0038] After completing the image registration, it is also necessary to verify the distribution of these images on the offshore dike to determine whether the position of the image after calibration is the same as the distribution of the offshore dike.

[0039] When the image pair that satisfies the mean square error is output as the target remote sensing image, the following implementation method is also included: according to the images corresponding to the multiple image pairs, the first image area and the second image area are obtained in sequence, the first image area represents the area represented by the initial frame of the remote sensing image set in the image pair, and the second image area represents the area represented by the middle frame of the remote sensing image set in the image pair. The initial frame represents an image that is the earliest in time in the corresponding data for a certain area when the images are aligned on an area after the image pairs are formed according to time proximity; the middle frame represents the area represented by the middle time in these image pairs.

[0040] Detect the similarity between the first image area and the second image area, and judge the registration degree of the target remote sensing image according to the similarity between the first image area and the second image area. Here, the image pairs within a short time interval are compared, that is, to evaluate whether the relevant images are aligned after the image registration at this time. The similarity at this time is calculated in the form of average pixel distance or mean square error.

[0041] When the similarity between the first image area and the second image area is greater than the similarity threshold, and the registration of the target remote sensing image satisfies the mean square error, the corresponding target remote sensing image is output. For the similarity threshold, the average value of the similarity between the first image area and the second image area that are normally used as the target remote sensing image in the historical data can be used to obtain an image set suitable for differential processing.

[0042] Geographic features such as offshore dikes may change over a long period of time due to natural factors (such as tides, wind and waves) or human activities (such as construction and maintenance). By comparing the similarity between the initial frame and the intermediate frame, not only can the registration effect be verified, but these changes can also be detected indirectly. If the similarity is significantly reduced, it may mean that some parts of the offshore dike have changed significantly during the time interval of image acquisition. This comparison still emphasizes the temporal continuity of the currently acquired remote sensing image set to determine whether the images at the corresponding positions can maintain a large similarity in time after calibration in spatial position, so as to reduce the influence of natural factors, acquisition time and other conditions on the image.

[0043] In one embodiment of the present invention, when performing image difference in step S3, it is necessary to identify the actual volume and area of ​​the dike under the condition of image difference to verify the deformation generated at this time and the specific situation of the current dike. At the same time, according to the differences between satellite remote sensing images, drone remote sensing images and SAR images in the target remote sensing image, it is necessary to perform difference processing in different dimensions, and then superimpose the features on the difference areas of these images to display the specific situation of the current target remote sensing image. For example, satellite remote sensing images can identify the changes in the large-scale coverage of the offshore dike, drone remote sensing images can identify local deformation or damage, and SAR images can identify large-scale deformation. At this time, the features of image difference will identify whether there is obvious deformation on the main body of the offshore dike from the coverage of the offshore dike, and then use SAR images to identify the deformation generated in the large-scale situation, and finally use drone remote sensing images to identify the deformation and changes generated locally to determine the overall relative deformation trend, and finally obtain the deformation area to be identified, or find the relationship and similarities between the deformation areas.

[0044] like Figure 3 As shown, the implementation method of step S3 includes: S31, performing image difference on the target remote sensing images at different time points to obtain a morphological difference image and an intensity difference image respectively.

[0045] During image differentiation, the satellite remote sensing images and the UAV remote sensing images are differentiated on the pixel values ​​to obtain the texture features of the guide bank regarding the texture and morphological changes, so as to determine the differences in texture features and identify whether there are relevant morphological changes in the guide bank. Then, SAR images are used to identify whether there are deformations in these areas with morphological changes. The deformation of the deformed area can be identified in the form of phase difference, and then the features are superimposed to obtain the features required for differentiation.

[0046] The morphological difference image represents the morphological changes of the guide dike in satellite remote sensing images and UAV remote sensing images, and is described using pixel value differences. The intensity difference image is a SAR image that uses the standard deviation of pixels at different time points to describe the intensity value difference.

[0047] S32, extracting edge features of the guide bank edge in the morphological difference image, and identifying the morphological change area of ​​the guide bank surface.

[0048] When extracting edge features, Canny detection is performed on the edge of the guide bank, and then the part with edge difference is identified, and the difference in the area is determined, such as edge breakage or area with difference due to deformation. These areas will be marked in the form of pixel difference, and these morphological change areas will be displayed using images.

[0049] S33, identifying the deformation amount and deformation rate of the guide bank according to the intensity difference image, and generating a deformation field of the intensity difference image according to the deformation amount and deformation rate.

[0050] The deformation of the intensity difference image can be calculated by obtaining the phase difference of the intensity difference image and the radar wavelength. For example, the deformation is expressed as: ;in, represents the shape variable, represents the radar wavelength, represents pi, It represents the phase difference. The phase difference is obtained by differential interference and phase unwrapping. After removing the terrain phase, the phase difference about the phase of the dike is obtained. This phase difference will indicate the slight deformation of the dike at different time points. The deformation amount can then be directly calculated.

[0051] That is, step S33 also includes: interfering the intensity difference images at multiple different time points to generate an interference pattern corresponding to the intensity difference image, removing the terrain phase from the part of the interference pattern that produces a phase difference, and generating multiple interference pairs, and identifying the deformation amount and deformation rate of the guide embankment in turn according to the phase difference on the interference pairs.

[0052] At this time, the terrain phase will be processed according to the different phase values ​​generated by the radar mapping to the seawater, and the generated terrain phase will be removed. Then, an algorithm such as SNAPHU will be used to restore the phase to a continuous phase. Places where there is a phase difference will be regarded as interference pairs, and then the phase difference and the corresponding deformation can be calculated.

[0053] The deformation rate needs to be set according to the number of interference pairs used at this time, for example ;in, represents the deformation rate, Indicates a time span, usually expressed in years; represents the phase difference of the i-th interference pair, Represents the number of interference pairs, and the value of i ranges from 1 to N.

[0054] Then, the positions corresponding to the deformation amount and deformation rate are marked on the guide bank, so as to obtain a deformation field that can represent the spatial distribution of the guide bank.

[0055] S34, superimposing the morphological change region and the deformation field to obtain a superimposed deformation feature.

[0056] The morphological change area and the deformation field are registered according to the points on the image. After determining that the expressed positions are consistent, the information on the morphological change area is fused with the information on the corresponding position of the deformation field. For example, weighted superposition is used to superimpose the weight of the texture at the corresponding position and the weight of the deformation to illustrate what other forms of deformation are involved when the guide bank is deformed. Is the deformation caused by corrosion of the guide bank, or is the guide bank itself undergoing slight deformation while remaining basically intact? Combining this information will represent the deformation features currently required.

[0057] In the subsequent processing of the deformation area, a neural network will be used to predict or segment the deformation features based on the deformation features obtained at this time, so as to obtain deformation areas with different deformation degrees, such as areas with fast deformation rate, large deformation and other different forms.

[0058] The implementation method for determining the deformation area related to the distribution of each deformation feature also includes the following: taking the position of any deformation feature as the center position, comparing the difference between adjacent deformation features at the center position, the difference between adjacent deformation features is calculated based on the deformation variable, and the difference between adjacent deformation features can identify whether the deformation feature has changed significantly in the local range, and whether the change in the deformation variable maintains continuity in the local area or has a jump interruption phenomenon. The value calculated at this time is to use data to represent these existing phenomena, and then cluster them to obtain areas with relatively consistent deformation and areas with local anomalies.

[0059] According to the horizontal offset of the difference between adjacent deformation features, the deformation features are clustered and combined into multiple deformation regions.

[0060] At this time, clustering is performed by taking the difference values ​​of adjacent deformation features to identify whether the area represented by the deformation feature has undergone overall lateral displacement, local torsional deformation, and tensile or compressive deformation. These values ​​are clustered or analyzed to find points with the same pattern and combine them into the required deformation area.

[0061] When calculating the difference between adjacent deformation features, the Euclidean distance measurement between deformation features is used for identification. For example, the deformation features are clustered and combined into multiple deformation areas. The implementation method includes: converting the deformation features into horizontal displacement vectors and corresponding spatial coordinates, calculating the spatial distance between each deformation feature, and judging whether there are adjacent deformation features that meet the spatial adjacency at the center position. When the spatial adjacency is met and the difference between adjacent deformation features is less than the deformation difference threshold, the adjacent deformation features and the center position are considered to be in the same area; then all deformation features that meet the spatial adjacency and deformation difference threshold are recursively calculated to obtain the deformation area; for spatial adjacency, it is judged whether the distance between adjacent deformation features is less than the preset adjacency distance. At this time, the preset adjacency distance set will be based on the impact of wave erosion on the guide embankment. For example, if the erosion is serious, it is set to a few centimeters, and if the erosion is not serious, the identified deformation feature is enlarged to a few meters to determine the identification processing of relatively local and larger areas; at the same time, the deformation difference threshold uses the average value of the difference between adjacent deformation features in the historical data to identify obvious and relatively average areas; finally, the areas with consistent changes are combined, and for points where the difference between adjacent deformation features is greater than or equal to the deformation difference threshold, when this happens, it means that the points adjacent to the current deformation feature are interrupted or are not points of the same deformation. At this time, a new central position will be selected for clustering, and finally all deformation features will be combined together in the form of spatial proximity and small difference, which is convenient for subsequent analysis of the distribution of deformation features in different deformation areas.

[0062] In one embodiment of the present invention, step S4 mainly identifies the deformation trend direction based on the deformation amount existing in the deformation area. After fitting the points where the deformation occurs, the direction of the deformation and the corresponding values ​​can be known. Then, the directions on all the deformation areas are combined to fit a trend line, which represents the deformation trend direction of the currently identified guide embankment.

[0063] The influence path indicates that the appearance of the deformation variable is affected by the environment, or in the case of global distribution, the path of the overall change of the deformation variable, as well as the weights between multiple nodes on the path, to obtain the influence path that expresses the correlation between the deformation areas.

[0064] like Figure 4As shown, the implementation method of step S4 includes: S41, identifying the deformation amount and deformation rate on each deformation area, and determining the trend direction of each deformation area; the trend direction of each deformation area will identify the deformation generated by the deformation area, and express it in the form of a vector, such as using an arrow to indicate the direction corresponding to the position before and after deformation, and the length of the arrow indicates the size of the deformation amount, so as to obtain the trend direction existing in each deformation area. It should be noted that these trend directions represent the direction of movement of the points where displacement occurs in this area, and do not represent the average trend of the entire area; even if each deformation area is obtained by aggregation through spatial continuity and deformation similarity, there are still some points in the deformation area whose deformation trend direction is different from that of other points. At this time, the trend direction generated by all points is displayed to facilitate the subsequent combination of these points.

[0065] S42, fitting the trend direction of each deformation area according to the edge of the guide bank, identifying the trend direction of the guide bank at the left and right edges, and using the trend direction of the guide bank at the left and right edges and the center trajectory of the guide bank during deformation to obtain the deformation trend direction corresponding to the deformation area.

[0066] The center trajectory represents the center lines of all deformation points when deformation occurs. The coordinates of the center lines of all time points are fitted by weighted averaging to obtain the center trajectory currently in use.

[0067] When fitting the trend direction of each deformation area according to the edge of the guide bank, the points represented by these trend direction arrows are mainly connected according to the edge of the guide bank under normal circumstances to determine the trend that can be displayed under the current deformation amount, and the main trend direction obtained is used as the deformation trend direction.

[0068] When the deformation trend direction corresponding to the deformation area is obtained, the processing method for each deformation area can be through identifying the deformation point on each deformation area, converting the deformation variable of the deformation point into a displacement vector, that is, expressing its relative displacement on the horizontal coordinate X-axis and Y-axis according to the displacement generated by the deformation point, and then constructing a covariance matrix for all deformation points in the deformation area, and then solving the eigenvector of the covariance matrix, and using the inverse tangent function to express the ratio of the eigenvector of the covariance matrix on the Y-axis and the X-axis corresponding to the data, to obtain the trend direction corresponding to each deformation area, and this trend direction represents an angle value.

[0069] Then the trend direction is projected to the left and right edges of the guide embankment, and the projected edge points are used to calculate the projected angle. The calculation method of the projected angle value is the same as the calculation method of the trend direction corresponding to the deformation area, and multiple adjacent edge points are calculated together; then the angles of all edge points on the left and right edges are weighted averaged to obtain the angle values ​​corresponding to the left and right edges.

[0070] Then, the weighted average of the angles corresponding to the center points on the center line of adjacent time points is calculated as the center trajectory at this time. Finally, the angle value of the center trajectory is weighted and summed with the angle values ​​corresponding to the left edge and the right edge to obtain an overall angle value, which represents the deformation trend direction corresponding to the deformation area.

[0071] After obtaining the deformation trend direction, we can know the main direction of the current guide bank when it is deformed, and analyze the deformation amount and deformation rate in this direction to find out the main factors that mainly affect the deformation of the guide bank under the influence of multiple points.

[0072] Therefore, the implementation method of obtaining the influence path corresponding to each deformation area in step S4 also includes: using the Moran index to extract the relevant points corresponding to the deformation amount and deformation rate in the deformation trend direction, and taking the shortest path of the relevant points as the output influence path.

[0073] The Moran index is to calculate the spatial correlation of the positions corresponding to the deformation variable and deformation rate in the deformation trend direction after setting the weights. The weight at each position is set according to the ratio of its deformation variable to the total deformation variable. When using the Moran index for calculation, the deformation variable and deformation rate are calculated separately; for example, the deformation variable at a single position is compared with the average value of all deformation variables to obtain the Moran index of the deformation variable. At this time, the Moran index is calculated for all the deformation positions in the deformation trend direction, and the deformation variables and deformation rates at these positions are calculated; then the single The deformation rate of the position is compared with the average value of all deformation rates to obtain the Moran index of the deformation rate. It should be noted that the weight used for the deformation rate is the ratio of its deformation variable to the total deformation variable; then the deformation variable and the Moran index calculated by the deformation rate are added to determine whether the added value is greater than the preset value. The preset value is the average value of the deformation variable and the Moran index calculated by the deformation rate measured in the area with large gradient changes, large settlement deformation and surface cracks in the historical data; when it is greater than the preset value, the position corresponding to the deformation variable and deformation rate is set as the relevant point. Finally, the relevant points are calculated by the shortest path method, and the Dijkstra algorithm is used to connect the relevant points. The weight of each relevant point is the weight used when calculating the Moran index. The value of each relevant point is the sum of the Moran index calculated by its deformation variable and deformation rate. Then, according to the distance and weight of each relevant point in the spatial position, the shortest path is obtained, and the shortest path obtained at this time is used as the influencing path for subsequent output.

[0074] The shortest path obtained at this time will represent the path connecting the areas with obvious deformation in the deformation trend direction, which will represent the part of the guide bank with obvious displacement and how this part produces the relevant displacement.

[0075] In one embodiment of the present invention, step S5 mainly performs rule matching on the influence path based on the acquired influence path, takes the deformation amount, deformation rate, path direction and the position corresponding to the influence path on the influence path as input features, and then matches the type of alarm required at this time to determine whether the current guide dike needs subsequent processing.

[0076] like Figure 5 As shown, the implementation method of step S5 includes: S51, separating the features of the influencing path according to the deformation variable and deformation rate on the influencing path, and matching the separated features with the preset database for rule matching; when performing feature separation, it is mainly divided according to the values ​​of the deformation variable and the deformation rate to separate the parts with high deformation variable and low deformation variable for rule matching, and when performing rule matching, these feature values ​​are matched with the features set in the preset database to calculate their similarity. The similarity can be in the form of Pearson correlation coefficient, cosine similarity, confidence and support. After the similarity of these separated features is calculated, the part with a similarity greater than 0.6 is regarded as having completed the match, that is, when calculating the current Pearson correlation coefficient, cosine similarity, confidence and support values, under the condition of specific required parameters, it is required that a value greater than 0.6 can be calculated before it is considered that there are features in the preset database that can match the current influencing path.

[0077] S52, if there is a matching feature, the alarm type represented by the feature is used as the subtype of the current influencing path. After obtaining the matching feature, the subtype of the current influencing path is determined according to the distribution of each subtype on the influencing path, the connection relationship of the subtypes is determined, and the path operation of the alarm type is executed in sequence according to the connection relationship of the subtypes.

[0078] The connection relationship of the subtypes is represented in the guide bank deformation scenario. According to the order of appearance of the alarm types corresponding to the guide bank deformation, the alarm types are connected according to the time of appearance, and the operations that should be performed after the alarm type appears are executed in sequence. If the executed operations are repeated, the repeated parts will be deleted.

[0079] S53, if there is no matching feature, the deformation trend direction corresponding to the influencing path and the position corresponding to the influencing path are taken as matching objects, the alarm type with the largest similarity value after matching is taken as the alarm type corresponding to the current influencing path, and the operation event corresponding to the alarm type is executed. When no matching feature is found, the angle corresponding to the change trend direction and the position where the deformation occurs on the influencing path are calculated with the content in the preset database to find the similar parts. At this time, the calculation method is consistent with the content during the matching calculation, and can be expressed in the form of Pearson correlation coefficient, cosine similarity, confidence and support, etc., and then the alarm type with the largest value is selected. This alarm type will contain data related to the deformation trend direction corresponding to the influencing path and the position corresponding to the influencing path in the preset database to obtain a calculated similarity, and then the current guide embankment is alarmed or other processing methods are performed according to the operation event recorded on the alarm type.

[0080] For example, there is a monitoring system for an offshore embankment that collects deformation and deformation rate data for each month of the past year. In a particular month, a section of the embankment about 50 meters long was detected to have abnormal deformation, with the deformation in the center significantly higher than at the ends. After analysis, it was found that the deformation in the center of this section of the embankment reached 10 cm / month, while the deformation at the ends was only 2 cm / month.

[0081] At this time, according to the deformation amount and deformation rate that affect the path, the guide bank is divided into three parts: the two ends are low deformation areas, and the middle is high deformation area.

[0082] By calculating the similarity with the historical events recorded in the preset database, it was found that the deformation pattern of the middle part was closest to the "local structural failure" alarm type in the historical records, with a Pearson correlation coefficient of 0.75.

[0083] After confirming it as a "local structural failure", the system automatically initiated emergency repair procedures and increased the frequency of real-time monitoring of the area.

[0084] If no directly matching features are found, the deformation trend direction of the area will be further analyzed (for example, whether it is tilting or sinking in a specific direction), and the closest historical case will be found in combination with the location information. The treatment of relevant situations in the historical case will then be used to deal with the treatment of the offshore dike at this time.

[0085] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations on the present invention. A person skilled in the art may make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention and they are still covered by the protection scope of the present invention.

Claims

1. An automatic monitoring and early warning method for offshore embankment deformation based on remote sensing images, characterized in that: include: S1, obtaining remote sensing images and standard images of the offshore dike, and composing the remote sensing images into a remote sensing image set according to the time and location of acquisition; the remote sensing image set includes satellite remote sensing images, UAV remote sensing images and SAR images; S2, dividing the remote sensing image set into a plurality of image pairs, performing image registration according to the positions of the plurality of image pairs and the standard image at the offshore guide embankment, and obtaining a registered target remote sensing image; S3, performing image difference on the target remote sensing images at different time points, analyzing deformation features related to the surface deformation of the offshore dike, and determining deformation areas related to the distribution of each deformation feature; S4, combining the deformation regions according to the distribution of deformation characteristics, determining the deformation trend direction corresponding to the deformation region, performing influence analysis on the deformation amount and deformation rate in the deformation trend direction, and obtaining the influence path corresponding to each deformation region; S5, determining the alarm type corresponding to the offshore dike based on the acquired impact path.

2. The method for automatic monitoring and early warning of offshore embankment deformation based on remote sensing images according to claim 1 is characterized in that: The implementation of step S2 includes: S21, obtaining the acquisition time of the remote sensing image set, taking adjacent frame images of the remote sensing image set within adjacent acquisition times as an image pair, and obtaining a plurality of image pairs; S22, preprocessing each image pair, extracting feature points from the preprocessed image pair and the standard image, and performing a nearest neighbor algorithm on the feature points to obtain feature point pairs of the matching image pair and the standard image; S23, after transforming the feature point pairs of the matching image pair and the standard image, the image pair satisfying the mean square error is output as the target remote sensing image.

3. The method for automatic monitoring and early warning of offshore embankment deformation based on remote sensing images according to claim 2 is characterized in that: When the image pair that satisfies the mean square error is output as the target remote sensing image, the following implementation methods are also included: According to the images corresponding to the plurality of image pairs, a first image region and a second image region are sequentially acquired, wherein the first image region represents the region represented by the initial frame of the remote sensing image set in the image pair, and the second image region represents the region represented by the intermediate frame of the remote sensing image set in the image pair; Detecting the similarity between the first image region and the second image region, and determining the registration degree of the target remote sensing image according to the similarity between the first image region and the second image region; When the similarity between the first image area and the second image area is greater than the similarity threshold, and the registration degree of the target remote sensing image satisfies the mean square error, the corresponding target remote sensing image is output.

4. The method for automatic monitoring and early warning of offshore embankment deformation based on remote sensing images according to claim 1 is characterized in that: The implementation of step S3 includes: S31, performing image difference on the target remote sensing images at different time points to obtain a morphological difference image and an intensity difference image respectively; S32, extracting edge features of the guide bank edge in the morphological difference image to identify the morphological change area on the guide bank surface; S33, identifying the deformation amount and deformation rate of the guide bank according to the intensity difference image, and generating a deformation field of the intensity difference image according to the deformation amount and deformation rate; S34, superimposing the morphological change region and the deformation field to obtain a superimposed deformation feature.

5. The method for automatic monitoring and early warning of offshore embankment deformation based on remote sensing images according to claim 4 is characterized in that: Step S33 also includes: interfering the intensity difference images at multiple different time points to generate an interference pattern corresponding to the intensity difference image, removing the terrain phase from the part of the interference pattern that produces a phase difference, and generating multiple interference pairs, and identifying the deformation amount and deformation rate of the guide embankment in turn according to the phase difference on the interference pairs.

6. The method for automatic monitoring and early warning of offshore embankment deformation based on remote sensing images according to claim 1 is characterized in that: The implementation method of determining the deformation area related to the distribution of each deformation feature also includes the following: Taking the position of any deformation feature as the center position, the difference between adjacent deformation features at the center position is compared; According to the horizontal offset of the difference between adjacent deformation features, the deformation features are clustered and combined into multiple deformation regions.

7. The method for automatic monitoring and early warning of offshore embankment deformation based on remote sensing images according to claim 6 is characterized in that: The implementation methods of clustering deformation features and combining them into multiple deformation regions include: The deformation features are converted into horizontal displacement vectors and corresponding spatial coordinates, the spatial distance between each deformation feature is calculated, and it is determined whether there are adjacent deformation features that meet the spatial adjacency at the center position. When the spatial adjacency is met and the difference between adjacent deformation features is less than the deformation difference threshold, the adjacent deformation features and the center position are considered to be in the same area; then all deformation features that meet the spatial adjacency and deformation difference thresholds are recursively calculated to obtain the deformation area.

8. The method for automatic monitoring and early warning of offshore embankment deformation based on remote sensing images according to claim 1 is characterized in that: The implementation of step S4 includes: S41, identifying the deformation amount and deformation rate in each deformation area, and determining the trend direction of each deformation area; S42, fitting the trend direction of each deformation area according to the edge of the guide bank, identifying the trend direction of the guide bank at the left and right edges, and using the trend direction of the guide bank at the left and right edges and the center trajectory of the guide bank during deformation to obtain the deformation trend direction corresponding to the deformation area.

9. The method for automatic monitoring and early warning of offshore embankment deformation based on remote sensing images according to claim 8 is characterized in that: The implementation method of obtaining the influence path corresponding to each deformation area in step S4 also includes: The Moran index is used to extract the relevant points corresponding to the deformation amount and deformation rate in the deformation trend direction, and the shortest path of the relevant points is used as the output influence path.

10. The method for automatic monitoring and early warning of offshore embankment deformation based on remote sensing images according to claim 1, characterized in that: The implementation of step S5 includes: S51, separating the features of the influencing path according to the deformation amount and deformation rate on the influencing path, and matching the separated features with the preset database by rules; S52, if there is a matching feature, the alarm type represented by the feature is used as a subtype of the current influencing path, the subtypes of the current influencing path are distributed according to the distribution of each subtype on the influencing path, the connection relationship of the subtypes is determined, and the path operation of the alarm type is performed in sequence according to the connection relationship of the subtypes; S53, if there is no matching feature, the deformation trend direction and the position corresponding to the influencing path are taken as matching objects, the alarm type with the largest similarity value after matching is taken as the alarm type corresponding to the current influencing path, and the operation event corresponding to the alarm type is executed.

Images