Ice fracture identification method based on multi-source remote sensing data

Abstract

The invention relates to the field of image data processing, in particular to an ice fissure recognition method based on multi-source remote sensing data, which comprises the following steps: performing edge detection on the multi-source remote sensing data to obtain an ice fissure bottom image; preprocessing the ice fissure bottom image, and distinguishing a glacier fissure development area and a non-glacier fissure area; carrying out edge detection, tubular flow field identification and dynamic threshold optimization on the glacier fissure development area, and extracting to obtain glacier fissure data; and performing precision verification on the glacier fracture data to obtain final glacier fracture distribution information. By improving the valley bottom edge detection algorithm and the dynamic threshold optimization strategy and combining the high-resolution image, accurate identification of the ice crack is realized, the advantages of remote sensing data are effectively exerted, low-cost identification is realized, and a more reliable data result is obtained.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of image data processing, and in particular to a method for identifying ice cracks based on multi-source remote sensing data. BACKGROUND

[0002] At present, there are few studies on ice crack extraction, which cannot meet the requirements of automatic extraction of ice cracks with strong adaptability. The mainstream method for extracting ice cracks is visual interpretation, which can realize the tracing of cracks 50 years ago by manually identifying cracks in remote sensing images. However, the sources of crack extraction include ground penetrating radar (GPR), optical remote sensing images, synthetic aperture radar images and laser exploration data. However, the above-mentioned crack extraction methods generally have the problems of high cost, low efficiency and unsuitability for large-scale glacier crack monitoring.

[0003] Using high-precision optical satellite images can significantly improve the efficiency of ice crack identification. By histogram equalization processing of 10-meter resolution panchromatic images, the image contrast is effectively improved, and the accurate identification of single ice crack is realized. However, the above-mentioned method cannot evaluate the dynamic evolution of large-scale disordered ice crack belts, and cannot accurately and efficiently determine the evolution of ice crack shape profile. SUMMARY

[0004] In view of the fact that the ice crack is difficult to identify due to the mixture of glacier shadow and ice rock boundary, and the fact that the existing method using only single-source remote sensing data cannot make up for the insufficient coverage of remote sensing images and too few image data in different regions, and cannot efficiently and accurately identify the dynamic evolution of ice cracks, the present application provides a method for identifying ice cracks based on multi-source remote sensing data, which comprises the following steps:

[0005] S100: obtaining multi-source remote sensing data; performing valley bottom edge detection based on the multi-source remote sensing data to obtain an ice crack bottom image;

[0006] S200: preprocessing the ice crack bottom image to obtain an ice crack development area and a non-ice crack area;

[0007] S300: performing edge detection and tubular flow field identification on the ice crack development area and optimizing the dynamic threshold to extract ice crack data;

[0008] S400: performing precision verification on the ice crack data to obtain final ice crack distribution information.

[0009] Preferably, in S100, the multi-source remote sensing data is obtained, specifically as follows:

[0010] The region is photographed by a drone and a satellite to obtain multi-source remote sensing data; wherein the multi-source remote sensing data includes drone images and satellite remote sensing images.

[0011] Preferably, in S100, according to the multi-source remote sensing data, a valley bottom edge is detected to obtain an ice crack bottom image, specifically:

[0012] The multi-source remote sensing data is subjected to gray level gradient direction recognition to obtain the lowest valley point of the region in the corresponding direction.

[0013] The positions of the lowest valley points of the region in all directions are subjected to smoothing processing to obtain the ice crack bottom image.

[0014] Preferably, in S200, before the ice crack bottom image is preprocessed, the following steps are included:

[0015] The ice crack bottom image is reduced in size according to a preset ratio to obtain a reduced image;

[0016] The pixel gray level of the reduced image is calculated according to the pixel gray level of the ice crack bottom image.

[0017] Preferably, in S200, the ice crack bottom image is preprocessed to distinguish the glacial crack development region and the non-glacial crack region, specifically:

[0018] The reduced image is subjected to crack segment recognition to obtain the crack position;

[0019] According to the crack position, the glacial crack development region and the non-glacial crack region are distinguished.

[0020] Preferably, in S200, the reduced image is subjected to crack segment recognition to obtain the crack position, specifically:

[0021] The pixel gray level of the reduced image is obtained, and according to the pixel gray level, the pixels belonging to the crack segment and the non-crack segment in the reduced image are distinguished, thereby obtaining the crack position.

[0022] Preferably, in S300, the glacial crack development region is subjected to edge detection and tubular flow field recognition and dynamic threshold optimization to extract the glacial crack data, specifically:

[0023] The Sobel operator is used to perform edge detection on the glacial crack development region to obtain an edge gradient field of the glacial crack development region; and the glacial crack development region is subjected to image sharpening processing according to the edge gradient field.

[0024] The glacial crack development region is subjected to tubular flow field analysis to obtain the glacial crack profile of the glacial crack development region.

[0025] The glacier crack development region is subjected to dynamic threshold optimization processing on the glacier crack profile to obtain glacier crack data of the glacier crack development region.

[0026] Preferably, in S300, an edge of the glacier crack development region is detected by using a Sobel operator to obtain an edge gradient field of the glacier crack development region; and the glacier crack development region is subjected to image sharpening processing according to the edge gradient field, specifically:

[0027] The edge of the glacier crack development region is detected by using a Sobel operator to obtain an edge gradient field of the glacier crack development region in a horizontal direction and a vertical direction;

[0028] The gradient amplitude of the edge gradient field is subjected to standardization processing, and the glacier crack development region is subjected to image sharpening processing by using a Laplace operator.

[0029] Preferably, in S300, the glacier crack development region is subjected to tubular flow field analysis to obtain a glacier crack profile of the glacier crack development region, specifically:

[0030] The glacier crack development region is subjected to tubular flow field analysis to obtain a glacier crack propagation geometry activity profile evolution process, so as to obtain a profile in a main extension direction and a crack width direction of the glacier crack;

[0031] In S300, the glacier crack profile is subjected to dynamic threshold optimization processing to obtain glacier crack data of the glacier crack development region, specifically:

[0032] The profile in the main extension direction and the crack width direction of the glacier crack is subjected to dynamic threshold optimization processing to obtain glacier crack topological structure data of the glacier crack development region.

[0033] Preferably, in S400, the glacier crack data is subjected to precision verification to obtain final glacier crack distribution information, specifically:

[0034] A confusion matrix is constructed by comparing expert interpreted images, and overall precision of the glacier crack data is determined according to the confusion matrix;

[0035] A Kappa coefficient of the glacier crack data is determined according to the overall precision, producer precision and user precision;

[0036] Whether the glacier crack data is reliable is determined according to the Kappa coefficient, so that reliable glacier crack data is taken as the final glacier crack distribution information.

[0037] Compared with the prior art, the present application has the following beneficial effects:

[0038] Firstly, the application realizes accurate identification of ice cracks by improving the valley edge detection algorithm and dynamic threshold optimization strategy, combined with high-resolution images, with an overall accuracy of 90.7% and a Kappa coefficient of 0.91, significantly better than the ice crack identification accuracy of the traditional threshold method.

[0039] Secondly, the application is suitable for large-scale ice crack identification and can quickly and accurately identify the position of ice cracks. Considering the wide coverage and large amount of information of high-resolution images, the image size is too large to quickly identify whether there is ice crack in the high-resolution image. By reducing the image to the minimum size, the ice crack information is retained and part of the image noise is removed.

[0040] Thirdly, the application uses Sobel operator to perform edge detection on the ice crack image, uses Laplace operator as second-order differential operator to perform image sharpening processing, and introduces an effective method for segmenting line-shaped structures in digital images based on Sobel operator. Based on TuFF, the segmentation is performed by propagating geometric active contours. In order to improve the geometric smoothness of the segmentation result, a contour length regularization term is introduced in the numerical solution. After defining the crack indicator function, the geometric driven segmentation of the crack is realized based on the level set theory framework. The evolving contour is represented as the zero level set of the function by defining the implicit function, which has the advantage of implicit representation, making the evolving contour adaptively change the topological structure of ice crack bifurcation and closure.

[0041] Fourthly, the application introduces a dynamic threshold optimization algorithm system to solve the adaptive defect of the traditional threshold method. By constructing a numerical optimization model with good convergence, an optimization strategy based on gradient descent is used to realize intelligent solution of the segmentation threshold. Specifically, the candidate threshold is initialized based on the maximum inter-class variance criterion, and then the objective function is established in the feature space to quantitatively evaluate the segmentation quality. The threshold parameter is updated constantly by Jacobi iteration matrix until the convergence condition is met, which improves the segmentation robustness in complex ice surface environment. By improving the weight distribution mechanism of the classical mean iteration algorithm, the regional contrast factor is introduced to dynamically weight the gray mean value in each iteration process, and the candidate threshold is regularized and constrained by combining the neighborhood space correlation, ensuring stable segmentation performance in snow-covered areas and shadow interference areas.

[0042] Fifthly, the application provides a more efficient and convenient ice crack extraction method, which is particularly suitable for mixed scenes of shadows and ice-rock boundaries on steep slopes of mountains, and can automatically and accurately identify the ice cracks through remote sensing data, taking advantage of remote sensing data, with low identification cost and more reliable data results. BRIEF DESCRIPTION OF DRAWINGS

[0043] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description only represent some embodiments of the present application, and all other drawings obtained by those of ordinary skill in the art without creative effort based on these drawings also belong to the protection scope of the present application.

[0044] Figure 1 is a flow chart of an ice crack identification method based on multi-source remote sensing data provided by the present application.

[0045] Figure 2 is a valley bottom edge detection process.

[0046] Figure 3 is a Sobel operator calculation instance.

[0047] Figure 4 is a crack extraction graph obtained by edge detection and tubular flow field identification.

[0048] Figure 5 is a crack extraction graph after dynamic threshold optimization.

[0049] Figure 6 is an ice crack distribution graph of a certain field mountain area.

[0050] Figure 7 is a graph of ice crack length and direction distribution law of a certain field mountain area. DETAILED DESCRIPTION

[0051] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings. It can be understood that the specific embodiments described here are only used to explain the present application, but not to limit the present application. In addition, it should be noted that, in order to facilitate the description, only the parts related to the present application are shown in the drawings, and not all the structures. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort belong to the protection scope of the present application.

[0052] The terms "include" and "have" and any variations thereof in the present application are intended to cover non-exclusive inclusion. For example, a process, method, method, product or device including a series of steps or units is not limited to the listed steps or units, but optionally also includes steps or units not listed, or optionally also includes other steps or units inherent to these processes, methods, products or devices.

[0053] Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the application. The appearances of the phrase in various places in the specification are not necessarily all referring to the same embodiment, nor are they necessarily mutually exclusive of one another. As will be apparent to those of ordinary skill in the art, embodiments described herein can be combinable with other embodiments.

[0054] Reference is made to Figure 1 As shown in the drawings, the application provides a method for identifying ice cracks based on multi-source remote sensing data, which comprises the following steps:

[0055] S100: acquiring multi-source remote sensing data; performing valley bottom edge detection according to the multi-source remote sensing data to obtain an ice crack bottom image;

[0056] S200: pre-processing the ice crack bottom image to distinguish an ice crack development area and a non-ice crack area;

[0057] S300: performing edge detection and tubular flow field identification on the ice crack development area and optimizing a dynamic threshold to extract ice crack data;

[0058] S400: verifying the accuracy of the ice crack data to obtain final ice crack distribution information.

[0059] Further, in S100, the multi-source remote sensing data is acquired, specifically:

[0060] The region is photographed by a drone and a satellite to obtain multi-source remote sensing data; wherein the multi-source remote sensing data includes drone images and satellite remote sensing images.

[0061] Considering that different remote sensing data sources have different image shooting accuracy and range coverage for a region, a single remote sensing data source cannot obtain accurate images for the whole region. In order to make up for the defects of a single remote sensing data source, such as insufficient remote sensing image coverage or a small number of remote sensing images for different regions, the region where the glacier is located is photographed by a drone and a satellite in multiple ways to obtain multi-source remote sensing data. Specifically, the region where the glacier is located can be photographed by a drone to obtain image data, and / or visible light image data can be obtained from a remote sensing satellite, thereby forming multi-source remote sensing data. The multi-source remote sensing data includes but is not limited to drone images, satellite remote sensing images, and SAR images. Preferably, a drone can be used to take aerial photographs of the region where the glacier is located to obtain glacier inventory data images. GF-2 satellite, ASTER GDEM satellite, and Google maps can also be used to collect GF-2 images, ASTER GDEM, and Google images of the region where the glacier is located, thereby enriching the multi-source remote sensing data and providing sufficient and reliable image data sources for identifying ice cracks in the region where the glacier is located.

[0062] Further, in S100, according to the multi-source remote sensing data, a valley bottom edge is detected to obtain an ice crack bottom image, specifically:

[0063] The multi-source remote sensing data is subjected to gray level gradient direction recognition to obtain the lowest valley point of the region in the corresponding direction;

[0064] The position of the lowest valley point of the region in all directions is subjected to smoothing processing to obtain the ice crack bottom image.

[0065] The actual ice crack recognition and extraction will be affected by the gradient gray scale change and noise. Therefore, the application designs an edge detection algorithm specially used for ice crack detection, please refer to Figure 2 In an image of a mountainous glacier, the overall illumination of the image is uneven, and the gray level of the valley generally increases from the bottom to the top, but in the local area, the gray level is low due to the existence of cracks and shadows. In order to recognize the gray level gradient direction of the valley, the lowest valley point in a certain direction of the region should be detected, and the direction of the position is marked. For example Figure 2 As shown in the figure, three lines of "1", "2", "3" are marked on the left and right sides of the trapezoidal region; if the detected center pixel "0" is a valley point candidate, the gray value thereof should satisfy the following conditions: the gray value of the center pixel "0" should be lower than that of the line "1", the gray value of the line "1" should be lower than that of the line "2", and the gray value of the line "2" should be lower than that of the line "3". Then, the weighted average value of each line is calculated, and in the weighting process, the weight of the center pixel should be larger, and the weight of the other pixels should be smaller. Since fractional differentiation can effectively smooth the image, especially in the case of processing fine edges, fractional differentiation is used here to calculate the weight coefficient.

[0066] The valley point has a unique directional feature in the four directions of front, back, left and right, so the four directions should be detected respectively. In addition, the multi-source remote sensing image contains a large amount of noise, in order to reduce the noise interference, a Gaussian smoothing function is used for filtering processing, wherein the Gaussian smoothing function has a width parameter sigma (usually referred to as a scale space parameter), and the selection of the parameter sigma depends on the size distribution of the white spots in the image, and those skilled in the art can set the size of the parameter sigma according to the actual size distribution of the white spots, which will not be described in detail here.

[0067] Further, in S200, before the ice crack bottom image is preprocessed, including:

[0068] The ice crack bottom image is reduced according to a preset ratio to obtain a reduced image;

[0069] According to the pixel gray scale of the ice crack bottom image, the pixel gray scale of the reduced image is calculated.

[0070] The image resolution of multi-source remote sensing data is high, and the coverage is wide, so that the amount of multi-source remote sensing data information is huge, and the image size is too large to quickly identify whether the high-resolution image exists ice crack. In order to efficiently distinguish the ice crack development area and the non-ice crack area, the image needs to be reduced to the minimum size, so as to retain the ice crack information and remove part of the image noise. Specifically, the gray scale of the corresponding result pixel in the reduced image can be calculated by using the gray scales of four adjacent pixels in the original ice crack bottom image, that is, the gray scale of the result pixel is calculated according to the gray scales of the above four adjacent pixels. For example, set f(x, y) as the original ice crack bottom image, where x and y respectively represent the rows and columns of the original ice crack bottom image, and x=1, 2, 3...n, y=1, 2, 3...m, and set the reduced image as f(x k , y k ), where x k =1,...,n / 2 k , y k =1,...,m / 2 k , k is a positive integer, and k≤K, n≥2 K , m≥2 K . Through the above manner, the size of the original ice crack bottom image can be reduced while the ice crack information of the original ice crack bottom image is maximally retained, and the signal-to-noise ratio of the reduced image is improved.

[0071] Further, in S200, the ice crack bottom image is preprocessed to distinguish the ice crack development area and the non-ice crack area, specifically:

[0072] The reduced image is subjected to crack segment identification to obtain the crack position.

[0073] According to the crack position, the ice crack development area and the non-ice crack area are distinguished.

[0074] Further, in S200, the reduced image is subjected to crack segment identification to obtain the crack position, specifically:

[0075] The pixel gray scale level is obtained from the reduced image, and according to the pixel gray scale level, the pixels belonging to the crack segment and the non-crack segment in the reduced image are distinguished, so as to obtain the crack position.

[0076] Further, in S300, the ice crack development area is subjected to edge detection and tubular flow field identification and dynamic threshold optimization, and the ice crack data is extracted, specifically:

[0077] The Sobel operator is used to detect the edge of the glacier crack development area and obtain the edge gradient field of the glacier crack development area; based on the edge gradient field, the image of the glacier crack development area is sharpened;

[0078] Conduct tubular flow field analysis on the glacier crack development area to obtain the glacier crack profile in the glacier crack development area;

[0079] The dynamic threshold optimization processing is performed on the glacier crack contour to obtain the glacier crack data in the glacier crack development area; wherein, the glacier crack data includes the glacier crack contour evolution data.

[0080] After the original ice crack bottom image is reduced and pre-processed to obtain a reduced image, a post-processing function based on fracture mechanics is used to identify whether a segment can represent a part of the crack, thereby identifying and calibrating the crack position, which facilitates the subsequent further processing and classification of the crack. The present invention uses an edge detector to detect the reduced image, wherein the edge detector algorithm assumes that the crack grayscale must be lower than the non-crack area on the glacier surface. If the grayscale is greater than the non-crack area, the reduced image needs to be inversely operated first. If the reduced image contains both dark cracks and bright cracks, the original reduced image and the image obtained by the inverse operation need to be crack detected separately to obtain the crack position. Then, based on the crack position, the glacier crack development area and the non-glacier crack area are distinguished. Generally speaking, the area covered by the cracks is determined as the glacier crack development area, and the area not covered by the cracks is determined as the non-glacier crack area.

[0081] Furthermore, in S300, the Sobel operator is used to perform edge detection on the glacier crack development area to obtain the edge gradient field of the glacier crack development area; based on the edge gradient field, the image of the glacier crack development area is sharpened, specifically as follows:

[0082] The Sobel operator is used to detect the edge of the glacier crack development area, and the edge gradient field of the glacier crack development area in the horizontal and vertical directions is obtained;

[0083] The gradient amplitude of the edge gradient field is normalized, and the Laplace operator is used to sharpen the image in the area where glacier cracks are developed.

[0084] See also Figure 3 The present invention uses the Sobel operator to detect the edge of the glacier crack development area. The Sobel operator uses two convolution kernels and Horizontal and vertical edge detection are performed separately, and Figure 3 Rendering using convolution kernel Perform edge detection in the horizontal direction. Specifically, the convolution kernel and As follows:

[0085] .

[0086] Specifically, the Sobel operator is first used to calculate the edge gradient field as follows:

[0087] Horizontal direction gradient ; vertical direction gradient ; wherein, represents the image of the glacier crack development area;

[0088] The horizontal direction gradient and the vertical direction gradient are then used to calculate the gradient amplitude and the gradient direction of the edge:

[0089] Gradient amplitude ;

[0090] Gradient direction ;

[0091] The gradient amplitude of the edge gradient field is normalized, and a Laplacian operator is used as a second-order differential operator to perform image sharpening processing on the glacier crack development area.

[0092] Further, in S300, the tubular flow field analysis is performed on the glacier crack development area to obtain the glacier crack profile of the glacier crack development area, specifically:

[0093] The tubular flow field analysis is performed on the glacier crack development area to obtain the geometric active contour evolution process of the glacier crack propagation, so as to obtain the profile in the main extension direction and the crack width direction of the glacier crack.

[0094] The tubular flow field (TuFF) method is introduced, which is an effective method for segmenting line structures in digital images. The geometric active contour is implemented using a level set. The geometric active contour propagates under the influence of the tubular flow field, and the control equation is as follows:

[0095]

[0096] In the above control equation, , is a direction coefficient for controlling the influence on the curve evolution, and , is a crack development direction indicator function; for the position where the long strip structure exists, , the value of is relatively high (generally around 1), and for the position where the long strip structure does not exist, , The value of is low (usually 0); is a geometric active contour The unit normal vector field at each point on the surface; , denote the axial component and the normal component respectively, that is, Indicates the main extension direction of the crack, represents the direction orthogonal to the main extension direction of the crack (such as the crack width direction). In the above control equation, , Control the propagation rate of the profile along the main crack extension direction and the crack width direction respectively. Figure 4 , (a) is the original image of the glacier crack development area, and (b) is the crack extraction image corresponding to the original image of the glacier crack development area. In addition, in order to improve the geometric smoothness of the crack segmentation result, the contour length regularization term is introduced in the numerical solution: ,in is the smoothing coefficient, Geometric active contour Divergence, Geometric active contour In this way, edge smoothing is achieved by feeding back the change in contour curvature, effectively suppressing contour distortion caused by noise.

[0097] In S300, dynamic threshold optimization processing is performed on the glacier crack profile to obtain glacier crack data in the glacier crack development area, specifically:

[0098] Dynamic threshold optimization processing is performed on the contours of the glacier cracks in the main extension direction and the crack width direction to obtain the glacier crack topology data in the glacier crack development area.

[0099] In order to solve the adaptive defects of the traditional threshold method, a dynamic threshold optimization algorithm system is introduced. After defining the crack development direction indicator function, the geometry-driven segmentation of the crack is realized based on the level set theory framework. The level set theory framework defines the implicit function A Geometry Active Contour is expressed as the zero level set of the implicit function, that is: The advantage of the above method lies in its implicit representation characteristics, which enables the geometric active contour to adaptively change the topological structure such as crack bifurcation and closure. The reconstructed level set evolution equation is as follows:

[0100] .

[0101] The above reconstructed level set evolution equation is numerically discretized by introducing an entropy-stabilized format and solved by the finite difference method, which can effectively deal with the numerical dissipation problem in the evolution of the level set function. In particular, the reinitialization technique is used to periodically maintain Distance function, ensuring the numerical stability of the evolution process. Figure 5 (a) is the crack image before dynamic threshold optimization, and (b) is the crack extraction image after dynamic threshold optimization. Furthermore, to enhance segmentation robustness on complex glacier surfaces, the weight allocation mechanism of the classic mean iteration algorithm has been improved. A regional contrast factor is introduced during each iteration to dynamically weight the grayscale mean. At the same time, the candidate thresholds are regularized by combining neighborhood spatial correlation. This ensures that the algorithm maintains stable segmentation performance in both snow-covered and shadow-interfered areas.

[0102] Furthermore, in S400, the accuracy of the glacier crack data is verified to obtain the final glacier crack distribution information, specifically:

[0103] A confusion matrix was constructed by comparing expert interpretation images, and the overall accuracy of the glacier crevasse data was determined based on the confusion matrix.

[0104] Determine the Kappa coefficient of glacier crevasse data based on overall accuracy, producer accuracy, and user accuracy;

[0105] The reliability of the glacier crack data is determined based on the Kappa coefficient, and the reliable glacier crack data is used as the final glacier crack distribution information.

[0106] To quantitatively evaluate the performance of the glacier crevasse data acquisition process, a spatial overlay analysis method was used for accuracy verification. The validation dataset consisted of 18 high-resolution remote sensing images of typical GF-2 glacier areas, encompassing diverse areas such as bare ice, snow-covered areas, and moraine-covered areas. Fifteen of these images were selected for crevasse distribution, varying in illumination, orientation, and size. A baseline ground truth map was established through expert visual interpretation, ensuring pixel-level crevasse annotation accuracy.

[0107] Specifically, an overall accuracy (OA) of the glacier crevasse data is determined according to a confusion matrix constructed by comparing the expert interpretation of the image; a Kappa coefficient of the glacier crevasse data is determined according to the overall accuracy, a producer's accuracy (PA) and a user's accuracy (UA); these accuracies reflect the accuracy of image classification from different aspects, and quantify the missed detection rate and the false detection rate of the algorithm in crevasse pixel recognition. The consistency degree of the extraction result and the artificial interpretation result is measured by calculating the Kappa coefficient (KC) through a corresponding formula, and the calculation formula is: In the above formula, is the proportion of observation consistency, is the proportion of expected consistency. When the kappa is greater than a preset value (such as the preset value is 0.81), it is determined whether the glacier crevasse data is reliable, so that the reliable glacier crevasse data is used as the final glacier crevasse distribution information.

[0108] Specifically, the accuracy verification result of the expert visual interpretation is shown in Table 1 as follows:

[0109] Table 1

[0110] Method OA / % PA / % UA / % Kappa coefficient (K) Conventional threshold segmentation method 76.2 68.5 72.1 0.62 U-Net benchmark model 85.4 81.3 79.6 0.78 The method of the present invention 0.98 0.95 0.87 0.91

[0111] As shown in Table 1, the glacier crevasse extraction method provided by the present application is more efficient and convenient, and is particularly suitable for the mixed scene of the shadow and the ice-rock boundary of the glacier on the steep slope of the mountain, can automatically and accurately identify the glacier crevasse through remote sensing data, has the advantage of remote sensing data, has low identification cost and more reliable data results.

[0112] Please refer to Figures 6-7 , which are the crevasse distribution maps of the Jiepu Glacier in the Chongduipu River Basin in the middle section of the Himalayas, and the length and direction distribution law of the glacier crevasse, and the contour size distribution of the glacier crevasse can be comprehensively and accurately identified and extracted by the method of the present application, which will not be described in detail here.

[0113] From the above description of the embodiments, those skilled in the art can clearly understand that the embodiments can be implemented by means of a general hardware platform as necessary, and of course can also be implemented by means of a combination of hardware and software. Based on such understanding, the above technical solutions can be embodied in the form of a computer program product, and the present application can be embodied in the form of a computer program product implemented on one or more computer usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer usable program code.

[0114] Finally, it should be noted that the above examples are only used to illustrate the technical solutions of the present application, and are not intended to limit the same, and other examples can also be used; although the present application has been described in detail with reference to the foregoing examples, those skilled in the art should understand that the technical solutions recorded in the foregoing examples can still be modified, or some technical features therein can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for identifying ice cracks based on multi-source remote sensing data, characterized in that: The method comprises the following steps: S100: Acquire multi-source remote sensing data; perform valley bottom edge detection based on the multi-source remote sensing data to obtain an image of the bottom of the ice crevasse; S200: Preprocessing the bottom image of the ice crack to distinguish between glacier crack development areas and non-glacier crack areas; S300: performing edge detection and tubular flow field identification and dynamic threshold optimization on the glacier crack development area to extract glacier crack data; S400: Verify the accuracy of the glacier crack data to obtain final glacier crack distribution information.

2. The method according to claim 1, characterized in that In S100, multi-source remote sensing data is obtained, specifically: The region is photographed by drones and satellites to obtain multi-source remote sensing data; wherein the multi-source remote sensing data includes drone images and satellite remote sensing images.

3. The method according to claim 2, characterized in that In S100, valley bottom edge detection is performed based on the multi-source remote sensing data to obtain an image of the bottom of the ice crevasse, specifically: Identifying the grayscale gradient direction of the multi-source remote sensing data to obtain the lowest valley point of the region in the corresponding direction; The position of the lowest valley point in all directions of the region is smoothed to obtain the bottom image of the ice crack.

4. The method according to claim 1, wherein In S200, before pre-processing the ice crevasse bottom image, the following steps are included: reducing the image of the bottom of the ice crevasse according to a preset ratio to obtain a reduced image; The pixel grayscale of the reduced image is calculated according to the pixel grayscale of the ice crevasse bottom image.

5. The method according to claim 4, characterized in that In S200, the bottom image of the ice crack is pre-processed to distinguish between the glacier crack development area and the non-glacier crack area, specifically: performing crack segment identification on the reduced image to obtain crack positions; According to the crack positions, glacier crack development areas and non-glacier crack areas are distinguished.

6. The method according to claim 5, characterized in that In S200, crack segments are identified on the reduced image to obtain crack locations, specifically: Pixel grayscales are acquired from the reduced image, and pixels belonging to crack segments and non-crack segments in the reduced image are distinguished according to the pixel grayscales, thereby obtaining crack positions.

7. The method according to claim 6, characterized in that In S300, edge detection, tubular flow field identification, and dynamic threshold optimization are performed on the glacier crack development area to extract glacier crack data, specifically: Using the Sobel operator to perform edge detection on the glacier crack development area to obtain an edge gradient field of the glacier crack development area; performing image sharpening processing on the glacier crack development area according to the edge gradient field; Performing tubular flow field analysis on the glacier crack development area to obtain a glacier crack profile in the glacier crack development area; Dynamic threshold optimization processing is performed on the glacier crack profile to obtain glacier crack data in the glacier crack development area; wherein the glacier crack data includes glacier crack profile evolution data.

8. The method according to claim 7, characterized in that In S300, edge detection is performed on the glacier crack development area using the Sobel operator to obtain an edge gradient field of the glacier crack development area; and image sharpening processing is performed on the glacier crack development area based on the edge gradient field, specifically: Using the Sobel operator to perform edge detection on the glacier crack development area, and obtaining edge gradient fields of the glacier crack development area in the horizontal and vertical directions; The gradient amplitude of the edge gradient field is normalized, and the Laplace operator is used to perform image sharpening processing on the glacier crack development area.

9. The method according to claim 8, characterized in that In S300, a tubular flow field analysis is performed on the glacier crack development area to obtain the glacier crack profile of the glacier crack development area, specifically: Performing tubular flow field analysis on the glacier crack development area to obtain the evolution process of the geometric active contour of the glacier crack propagation, thereby obtaining the contour of the glacier crack in the main extension direction and the crack width direction; In S300, dynamic threshold optimization processing is performed on the glacier crack profile to obtain glacier crack data in the glacier crack development area, specifically: Dynamic threshold optimization processing is performed on the contours of the glacier cracks in the main extension direction and the crack width direction to obtain the glacier crack topology data of the glacier crack development area.

10. The method according to claim 1, characterized in that In S400, the accuracy of the glacier crack data is verified to obtain the final glacier crack distribution information, specifically: Constructing a confusion matrix by comparing expert interpretation images, and determining the overall accuracy of the glacier crevasse data based on the confusion matrix; Determining a Kappa coefficient of the glacier crevasse data based on the overall accuracy, producer accuracy, and user accuracy; Whether the glacier crack data is reliable is determined based on the Kappa coefficient, so that the reliable glacier crack data is used as the final glacier crack distribution information.

Images