A waterline detection method based on historical waterline and image information-driven random walk

Abstract

This invention relates to the field of coastal waterline detection technology, specifically a waterline detection method based on historical waterline and image information driven by random walk. The method includes: acquiring historical waterline and historical SAR images; generating prior information using a change detection method; constructing a waterline detection model; obtaining posterior probability class labels by maximizing the likelihood probability; combining the class labels with a random walk algorithm to infer the current class attribute; and using the inferred current class attribute as a reference to solve for the current waterline, thus achieving random walk waterline detection. This invention addresses the problem that traditional methods struggle to effectively adapt to image differences under different imaging conditions at different times due to the exceptionally complex and dynamic coastal environment; and how to effectively utilize historical information to guide efficient waterline detection in current images. This invention can fully utilize historical waterline information, has better applicability to the extraction of various waterlines, and improves detection efficiency and accuracy.

Description

Technical Field

[0001] This invention relates to the field of coastal waterline detection technology, and in particular to a waterline detection method based on historical waterline and image information-driven random walk. Background Technology

[0002] Traditional methods for detecting waterline in SAR images are mainly based on manual interpretation, but these methods have limitations in terms of efficiency, labor costs, and subjective interpretation, driving an urgent need for automated, high-precision solutions.

[0003] Automated waterline detection methods primarily rely on image processing techniques. These techniques mainly include edge detection algorithms such as the Sobel, Roberts, and Prewitt operators, as well as the more complex Canny algorithm, which identify edges based on grayscale value variations and directional derivatives. Another common approach is region-based segmentation, which divides an image into different regions based on shared attributes such as color, texture, or intensity, including region growing (with or without seeds) and region splitting and merging. Otsu thresholding is also frequently used for automatic threshold selection in segmentation tasks. The emergence of deep learning (DL) and advanced computer vision techniques has brought significant potential for improving the accuracy and efficiency of waterline detection. These methods typically treat waterline detection as a semantic segmentation or instance segmentation problem, classifying each pixel as water or non-water.

[0004] However, the actual coastal environment is exceptionally complex and dynamic. Remote sensing images of some waterline areas obtained under different imaging conditions may show significant differences, posing a challenge to the versatility and adaptability of algorithms. In marine monitoring and management, historical waterline locations and image data are typically available. Despite temporal differences, due to the persistence of land-sea properties over a certain time scale, most areas in historical data often retain the same properties in current images. How to utilize this prior knowledge to guide the efficient extraction of waterlines from current images has become a new research direction.

[0005] Compared to blindly detecting from the current image, fusing historical information promises to reduce detection difficulty and improve accuracy. However, since time differences may cause changes in image pixel values ​​or structure, effectively quantifying and utilizing this change information is crucial for guiding waterline detection. Furthermore, exploring new methods is needed to integrate historical waterline locations as known labels with image information. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a waterline detection method based on historical waterline information and image-driven random walks. This invention primarily employs an unsupervised change detection method based on graph structure consistency, identifying and acquiring two types of information: firstly, the degree of change in the strip regions extending to both sides of the historical waterline; and secondly, the binary information of these strips, the latter obtainable through a threshold method. A novel prior is constructed based on change metrics and historical waterline markings, and this prior is incorporated into the RWR model. This invention utilizes a "change detection-prior-walk" model architecture, which can fully leverage historical waterline information and effectively reduce detection difficulty.

[0007] The technical means employed in this invention are as follows:

[0008] A waterline detection method based on historical waterline and image information-driven random walk includes: acquiring historical waterline and historical SAR images, generating priors using a change detection method; constructing a waterline detection model, obtaining posterior probability category labels by maximizing the likelihood probability; combining the category labels with a random walk algorithm to infer the current category attribute; and using the inferred current category attribute as a reference to solve for the current waterline, thereby realizing random walk waterline detection.

[0009] Furthermore, the method of generating priors using change detection specifically includes: representing historical SAR images as... The current SAR image is represented as ,right and The nonlinear mapping for change detection is as follows The change detection result image is represented as ;

[0010] Assume the size of the expansion to both sides of the waterline in the historical SAR image is... The region of change becomes An adaptive threshold was obtained using Otsu's method. Perform segmentation to obtain the segmentation results. .

[0011] Furthermore, the waterline detection model specifically includes: for the current SAR image Represented by image pixels , Indicates the first There are 10 pixels, and the label for each pixel is 1. , Indicates the sea surface. Representing land; if the maximum posterior probability of a pixel belonging to the labels of sea and land is known, then it should belong to the category label with the larger maximum posterior probability. The posterior probability is expressed as:

[0012]

[0013] in, Let be the likelihood probability. Let be the prior probability of the label, with a constant denominator. Assuming the prior probability follows a uniform distribution, maximizing the posterior probability is equivalent to maximizing the likelihood probability. have If there are seed points, the likelihood probability is expressed as:

[0014]

[0015] in, It is a normalized constant. The pixel distribution describes the degree of association with the label; Let the label probability be . If the distribution is uniform, then the likelihood probability is expressed as:

[0016]

[0017] in, Indicates that the label is The m-th seed point.

[0018] Furthermore, the inference of the current category attribute specifically includes: assuming that it has VV polarimetric SAR image of 1 pixel It can generate with A weighted graph with nodes, represented as: ,in It is a set of nodes. It is a set of edges. The weight matrix has edge connections; in the node set any node Representing an image The Middle 1 pixel, in the edge set any edge in Indicates at node In the four-neighbor or eight-neighbor domains of the node The connection; in middle The definition of is:

[0019]

[0020] in, and They are nodes and nodes The mean of the block it belongs to. These are control parameters;

[0021] Using assumptions similar to Kim's, the summation term of the likelihood probability is expressed as:

[0022]

[0023] in, Let the steady-state probability be... Represented as a vector The steady-state probability of a random walk based on change detection is then defined as:

[0024]

[0025] Where c is a weighting factor between 0 and 1. It is the category label for the area outside the waterline on both sides of the historical SAR image data, and U is the current node to and The transition probability of the nodes. The row-normalized transition matrix is ​​expressed as: ;

[0026] in, It is a diagonal matrix. , , , Including two types of markers, used in addition to expanding on both sides of the waterline in historical SAR image data. , Each element in the text has a corresponding tag of 1, otherwise it is 0; for Mapping of regional change detection information Defined as: ;

[0027] in, For normalization operations, An indicator related to historical SAR imagery, changes, and change segmentation is defined as:

[0028]

[0029] Strip area based on historical waterfront extension The current category attributes are inferred from the historical land and sea labels and the degree of change detection within the area.

[0030] Furthermore, the solution for the current waterline specifically includes: [details of the solution and its implications]. After mathematical processing, we get:

[0031]

[0032] in, ,get :

[0033]

[0034] in, Represents image pixel nodes Belongs to the label Likelihood probability A set;

[0035] The current solution for the waterline is as follows:

[0036]

[0037] By label The category is used to obtain the current image. The pixel category is used to complete the waterline detection.

[0038] Compared with the prior art, the present invention has the following advantages:

[0039] This invention provides a waterline detection method based on historical waterline and image information-driven random walk. It acquires historical waterline data and historical SAR images, generates prior information using a change detection method, constructs a waterline detection model, and obtains posterior probability category labels by maximizing the likelihood probability. The category labels are combined with the random walk algorithm to infer the current category attribute. Using the inferred current category attribute as a reference, the current waterline is solved, achieving random walk-based waterline detection. This invention can fully utilize historical waterline information, has better applicability to the extraction of various waterlines, and improves detection efficiency and accuracy. This invention is not only applicable to coastal waterline monitoring but can also theoretically be extended to other SAR image information extraction tasks that require the use of multi-temporal information and the fusion of historical knowledge.

[0040] This invention addresses the problem that traditional methods struggle to adapt to image differences under varying time and imaging conditions due to the exceptionally complex and dynamic coastal environment; and how to effectively utilize historical information to guide efficient detection of waterline in current images.

[0041] Based on the above reasons, this invention can be widely applied in fields such as coastal waterline detection. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 This is a diagram illustrating the architecture of the waterline detection method based on historical waterline and image information-driven random walk, as presented in this invention.

[0044] Figure 2 This is a comparison image of waterline extraction in an embodiment of the present invention. Detailed Implementation

[0045] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0046] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the present invention or its application or use. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0047] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0048] Unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps described in these embodiments do not limit the scope of the invention. It should also be understood that, for ease of description, the dimensions of the various parts shown in the drawings are not drawn to actual scale. Techniques, methods, and devices known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and devices should be considered part of the specification. In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar reference numerals and letters in the following figures denote similar items; therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.

[0049] like Figure 1As shown, this invention provides a waterline detection method based on historical waterline and image information-driven random walk. How to utilize change detection information in the current SAR image is crucial for subsequent extraction of the waterline from the current data. First, it is assumed that two single-polarization images have been registered. The backscattering coefficients and structures of the land and sea near the waterline are key to change detection and extraction. Since SAR image imaging is affected by polarization, incident angle, speckle pattern, and the complex coastal environment, it is necessary to analyze its characteristics from both the sea surface and land perspectives. Regarding the sea surface, depending on the different incident angles and wind strengths during the two image acquisition periods, the sea surface may exhibit three scenarios: one image is brighter than the other, both are brighter, or both are darker. Simultaneously, due to radiation differences, the backscattering coefficients of the land in the two images may also differ significantly.

[0050] Historical waterline and historical SAR images are acquired, and a priori information is generated using a change detection method. In a preferred embodiment of this invention, the historical SAR image is represented as... The current SAR image is represented as ,right and The nonlinear mapping for change detection is as follows The change detection result image is represented as The change detection result image value is between 0 and 1, representing the degree of change of a pixel in two real-phase SAR images. The larger the value, the greater the change; otherwise, there is a small change or no change. For waterline detection, the focus is on the area near the waterline. Therefore, it is only necessary to consider the change detection of similar strip areas extending from the waterline in historical images to the sea surface and land.

[0051] Assume the size of the expansion to both sides of the waterline in the historical SAR image is... The region of change becomes An adaptive threshold was obtained using Otsu's method. Perform segmentation to obtain the segmentation results. A value of 1 indicates a significant change, while a value of 0 indicates a small or no change. Subsequent methods require the information generated by change detection to... and .

[0052] A waterline detection model is constructed, and the posterior probability category label is obtained by maximizing the likelihood probability. Specifically, in a preferred embodiment of this invention, the waterline detection model includes:

[0053] For the current SAR image Represented by image pixels , Indicates the first There are 10 pixels, and the label for each pixel is 1. , Indicates the sea surface. Representing land; the problem of detecting water edges can be transformed into a label assignment problem for each pixel belonging to either sea or land. A common method for label assignment is posterior probability. If the maximum posterior probability of a pixel belonging to either sea or land is known, then it should belong to the category with the higher maximum posterior probability. The posterior probability can be expressed as:

[0054]

[0055] in, Let be the likelihood probability. Let be the prior probability of the label, with a constant denominator. Maximizing the posterior probability is equivalent to maximizing the product of the likelihood probability and the prior probability. Assuming the prior probability is uniformly distributed, maximizing the posterior probability is equivalent to maximizing the likelihood probability. have If there are seed points, the likelihood probability is expressed as:

[0056]

[0057] in, It is a normalized constant. The pixel distribution describes the degree of association with the label; Let the label probability be . If the distribution is uniform, then the likelihood probability is expressed as:

[0058]

[0059] in, Indicates that the label is The m-th seed point.

[0060] The category labels are combined with a random walk algorithm to infer the current category attributes.

[0061] In a specific implementation, as a preferred embodiment of the present invention, the inference of the current category attribute specifically includes:

[0062] Assume that VV polarimetric SAR image of 1 pixel It can generate with A weighted graph with nodes, represented as: ,in It is a set of nodes. It is a set of edges. The weight matrix has edge connections; in the node set any node Representing an image The Middle 1 pixel, in the edge set any edge in Indicates at node In the four-neighbor or eight-neighbor domains of the node The connection; in middle The definition of is:

[0063]

[0064] in, and They are nodes and nodes The mean of the block it belongs to. These are control parameters;

[0065] Using assumptions similar to Kim's, the summation term of the likelihood probability is expressed as:

[0066]

[0067] in, Let the steady-state probability be... Represented as a vector The steady-state probability of a random walk based on change detection is then defined as:

[0068]

[0069] Where c is a weighting factor between 0 and 1. It is the category label for the area outside the waterline on both sides of the historical SAR image data, and U is the current node to and The transition probability of the nodes. The row-normalized transition matrix is ​​expressed as:

[0070]

[0071] in, It is a diagonal matrix. , , , Including two types of markers, used in addition to expanding on both sides of the waterline in historical SAR image data. , Each element in the text has a corresponding tag of 1, otherwise it is 0; for Mapping of regional change detection information Defined as:

[0072]

[0073] in, For normalization operations, An indicator related to historical SAR imagery, changes, and change segmentation is defined as:

[0074]

[0075] Strip area based on historical waterfront extension Historical land and sea labels and the degree of change detection within the area are used to infer the current category attributes. The steady-state probability of the corresponding label is increased based on the category attributes. Since historical data and its waterline are used, the labeling does not require manual setting, greatly reducing manual intervention. In addition, besides... Outside of the designated area, the land and sea labels for the remaining areas are fixed; the parts requiring RW (Range Weighing) are mainly concentrated in... region, and This method, which only occupies a small portion of the image to be detected, can effectively alleviate the problem of arrival probability decay caused by long-distance travel.

[0076] Using the inferred current category attributes as a reference, the current waterline is solved to achieve random walk waterline detection.

[0077] In a specific implementation, as a preferred embodiment of the present invention, the step of solving the current waterline specifically includes:

[0078] steady-state probability After mathematical processing, we get:

[0079]

[0080] in, ,get ,

[0081]

[0082] in, Represents image pixel nodes Belongs to the label Likelihood probability A set of.

[0083] The current solution for the waterline is as follows:

[0084]

[0085] By label The category is used to obtain the current image. The pixel category is used to complete the waterline detection.

[0086] Example

[0087] like Figure 2 As shown, this embodiment compares the results of the algorithm of the present invention with those of the RW algorithm, RWR algorithm, SMRW algorithm, and HIDRWR algorithm. Figure 2 In the image, (a) represents a historical image; (b) represents a current image; (c) shows the result of the RW algorithm; (d) shows the result of the RWR algorithm; (e) shows the result of the SMRW algorithm; and (f) shows the result of the HIDRWR algorithm. Figure 2 (a) and Figure 2 (b) It can be seen that there is a significant difference in backscattering on land between the two images. The land in the historical image is generally darker, while the land in the current image is generally brighter, with some weak boundary areas near the coastline. Due to the different tide levels at the time of imaging, coupled with the influence of human activities, the waterline in the two images differs considerably. The experimental results show that the RW and RWR methods produce significant extraction biases in this area. Both the SMRW and HIDRWR methods can extract the approximate waterline, but HIDRWR has the best fit.

[0088] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1.A waterline detection method based on historical waterline and image information driven random walk, characterized in that, The application relates to a water edge line detection method and device. The method comprises the following steps: The step of generating the prior by using the change detection method comprises the following steps: The historical SAR image is represented as The current SAR image is represented as The non-linear mapping for change detection on and is The change detection result image is represented as ; Assuming the waterline of historical SAR image is extended to both sides with size , the changed strip area becomes , the adaptive threshold is obtained by using Otsu method, and the segmentation result is obtained by segmenting ; The water edge line detection model is constructed, and a category label of posterior probability is obtained by maximizing likelihood probability; the water edge line detection model comprises the following steps: For the current SAR image , the image pixels are represented as , , where the first pixel corresponds to the label , , where the sea surface is represented as , and the land is represented as . If a pixel is known to belong to the sea surface and land label with the maximum posterior probability, it should belong to the class label with the greater corresponding maximum posterior probability. The posterior probability is represented as wherein, is the likelihood probability, is the prior probability of the label, and the denominator is a constant. Assuming that the prior probability is uniformly distributed, maximizing the posterior probability is equivalent to maximizing the likelihood probability, assuming that the label There are seed points, the likelihood probability is represented as: wherein, is a normalization constant, is a pixel distribution, describing the degree of association with a label; is a label probability, assuming a uniform distribution of the likelihood probability is expressed as: wherein, denotes the m-th seed point of the label denotes the m-th seed point of the label The category label is combined with a random walk algorithm to infer a current stage category attribute; the step comprises the following steps: Assume a VV polarimetric SAR image with pixels , a weighted graph with nodes can be generated, the weighted graph is represented as , where is a set of nodes, is a set of edges, is a weight matrix with edges connected; in the set of nodes , any one node represents the th pixel in the image , in the set of edges , any one edge represents the connection of the node in the four-neighborhood or eight-neighborhood of the node ; in , the definition of is: wherein, and are the mean of the blocks where the nodes and the nodes are located, is a control parameter; If a Kim assumption is adopted, the summation term of the likelihood probability is expressed as: where, For the steady state probability, the steady state probability In vector representation, we have The steady state probability of the random walk based on change detection is then defined as where c is a weight factor between 0 and 1, is the class label of the outside region of the waterline of the historical SAR image data, U is the transition probability from the current node to and the nodes in the middle; is the row-normalized transition matrix, which is represented as: wherein, is a diagonal matrix, , , , includes two types of markers, in addition to the expansion of the waterline on both sides of the historical SAR image data , is change detection information mapping of the region, which is defined as: wherein, is a normalization operation, is an indicator related to historical SAR images, changes and change segmentation, defined as: Strip region based on historical waterline expansion Historical shoreline label and change detection degree inference present stage class attribute within The inferred current stage category attribute is taken as a reference to solve a current water edge line, and random walk water edge line detection is realized. 2.The waterline detection method based on historical waterline and image information driven random walk of claim 1, wherein, The step of solving the current water edge line comprises the following steps: On the steady-state probability Mathematical manipulation gives: wherein , to obtain , wherein, representing image pixel nodes belonging to a label likelihood probabilities a set of; The solving result of the current water edge line is as follows: By the class of the label of the current image pixels, the waterline detection is completed.

Images