A superpixel segmentation method, system, device and storage medium

Abstract

The present invention discloses a superpixel segmentation method, system, device and storage medium, comprising: obtaining a remote sensing image I to be segmented; inputting the remote sensing image I to be segmented into a non-similarity model between pixels and seed points with boundary contour constraints, and obtaining the non-similarity between pixels and seed points in the remote sensing image I to be segmented, wherein the non-similarity model between pixels and seed points with boundary contour constraints is constructed based on the diffusion flux, spatial distance and spectral distance of pixels with respect to seed points, and the diffusion flux of pixels with respect to seed points is calculated based on an anisotropic diffusion model; and performing superpixel segmentation on the remote sensing image I to be segmented according to the non-similarity between pixels and seed points in the remote sensing image I to be segmented. The method, system, device and storage medium have high superpixel segmentation accuracy.

Description

Technical Field

[0001] The present invention belongs to the field of remote sensing image processing and relates to a superpixel segmentation method, system, device and storage medium. Background Art

[0002] Remote sensing image interpretation is crucial for intelligent Earth observation. As the spatial resolution of remote sensing images continues to improve, detailed information about ground objects becomes richer. At the same time, the spatial distribution of objects becomes more complex, heterogeneity increases, and boundary uncertainty increases. Traditional pixel-based image processing methods are extremely sensitive to noise, often resulting in "salt and pepper" artifacts. To further unlock the information contained in images, the concept of "object" has been proposed, and excellent results have been achieved in many areas.

[0003] Superpixel segmentation is an effective tool for object detection. Its goal is to cluster adjacent pixels with the same features into subregions in the image domain space. Currently, superpixel segmentation methods can be categorized into three types: graph-based, gradient-based, and clustering-based. Graph-based methods represent images as undirected graphs. The edge weight connecting two nodes is used to measure the similarity between the two pixels. Image segmentation is then equivalent to undirected graph partitioning. Representative graph-based algorithms include Normalized Cumulative Cut (NCut) and Entropy Ratio Superpixel (ERS). NCut is an unbiased measure of the uncorrelation between graph subgraphs, which improves the noise sensitivity of traditional graph-based algorithms. ERS formulates the superpixel segmentation problem as a graph topology optimization problem based on the entropy rate of random walks on the graph. It encourages the generation of superpixels of similar size and can better preserve object boundaries. Graph-based methods strictly enforce connectivity within superpixels. However, most of them have high model complexity. Gradient-based methods evolve active contours based on image gradients. Representative algorithms include TurboPixel and Spatially Constrained Watershed (SCoW). TurboPixels is a geometric flow-based algorithm that evolves contours based on boundary velocity. TurboPixel respects local image boundaries and limits under-segmentation and over-segmentation through compactness constraints. SCoW introduces spatial constraints into the labeling control flow to obtain compact and uniformly distributed superpixels, achieving a balance between uniformity and compactness. Gradient-based methods rely more heavily on gradient information, making them less suitable for complex image processing. Clustering-based methods treat superpixel segmentation as a clustering problem coupled with spatial distances. Representative algorithms include Simple Linear Iterative Clustering (SLIC) and Linear Spectral Clustering (LSC). SLIC converts the image from RGB color space to LAB color space and models the distance between pixels and seed points using normalized spatial and spectral distances. LSC maps each pixel to a point in a high-dimensional space and achieves segmentation based on weighted K-means. Clustering-based methods can easily control compactness and have been widely studied due to their simple principles and strong scalability. In recent years, many improved methods have been proposed to improve superpixel segmentation accuracy from the perspectives of compactness and goodness of fit. However, unlike natural images, remote sensing images are characterized by large data volumes, numerous objects, complex spatial structures, rich feature information, and fuzzy boundaries. This makes it difficult to effectively segment a variety of features at a uniform scale, such as large, uniform convex structures and small, narrow structures. Many superpixels are prone to over-segmentation or crossing boundaries, resulting in low superpixel segmentation accuracy, which affects the accuracy of subsequent remote sensing image processing. Summary of the Invention

[0004] The purpose of the present invention is to overcome the shortcomings of the above-mentioned prior art and provide a superpixel segmentation method, system, device and storage medium, which have high superpixel segmentation accuracy.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] In one aspect, the present invention provides a superpixel segmentation method, comprising:

[0007] Obtain the remote sensing image I to be segmented;

[0008] Inputting the remote sensing image to be segmented I into a dissimilarity model between pixels and seed points with boundary contour constraints to obtain dissimilarity between pixels and seed points in the remote sensing image to be segmented I, wherein the dissimilarity model between pixels and seed points with boundary contour constraints is constructed based on a diffusion flux, a spatial distance, and a spectral distance of the pixel with respect to the seed point, and the diffusion flux of the pixel with respect to the seed point is calculated based on an anisotropic diffusion model;

[0009] Superpixel segmentation is performed on the remote sensing image I to be segmented according to the dissimilarity between pixels and seed points in the remote sensing image I to be segmented.

[0010] The superpixel segmentation method of the present invention is further improved in that:

[0011] The dissimilarity D between pixel i and seed point j in the remote sensing image I to be segmented ij for:

[0012]

[0013] in, is the spectral distance between pixel i and seed point j, is the spatial distance between pixel i and seed point j, is the diffusion flux between pixel i and seed point j, N Spatial is the normalization parameter of spatial distance, N Spectral is the spectral distance normalization parameter, N Flux is the diffusion flux normalization parameter.

[0014] The spatial distance between pixel i and seed point j for:

[0015]

[0016] Among them, (a j ,b j ) is the grid coordinate of the seed point j in the image space domain, (x i ,y i ) is the grid position coordinate of pixel i in the image space domain.

[0017] The spectral distance between pixel i and seed point j for:

[0018]

[0019] Among them, S j is the spectral measurement vector of seed point j, I i is the spectral measure vector of pixel i.

[0020] Diffusion flux between pixel i and seed point j for:

[0021]

[0022] in, is the coordinate point (x i ,y i ) value.

[0023] The updating process of the diffusion flux map in the non-similarity model is:

[0024]

[0025] Among them, U J (x,y,t+1) is the diffusion flux diagram at time t+1, U J (x, y, t) is the diffusion flux diagram at time t, λ is the stability coefficient, R is the diffusion distance, c f (·) is the diffusion coefficient in the f direction, |·| represents the modulus value, represents the image gradient in the f direction, represents the flux gradient in the direction f.

[0026] The specific operation of performing superpixel segmentation on the remote sensing image to be segmented I according to the dissimilarity between the pixels and the seed points in the remote sensing image to be segmented I is:

[0027] According to the dissimilarity between pixels and seed points in the remote sensing image I to be segmented, the K-means clustering algorithm is used to perform superpixel segmentation on the remote sensing image I to be segmented.

[0028] In a second aspect of the present invention, the present invention provides a superpixel segmentation system, comprising:

[0029] An acquisition module is used to acquire the remote sensing image I to be segmented;

[0030] a dissimilarity calculation module, configured to input the remote sensing image to be segmented I into a dissimilarity model between pixels and seed points with boundary contour constraints, to obtain dissimilarity between pixels and seed points in the remote sensing image to be segmented I, wherein the dissimilarity model between pixels and seed points with boundary contour constraints is constructed based on a diffusion flux, spatial distance, and spectral distance of the pixels with respect to the seed points, and the diffusion flux of the pixels with respect to the seed points is calculated based on an anisotropic diffusion model;

[0031] The superpixel segmentation module is used to perform superpixel segmentation on the remote sensing image I to be segmented based on the dissimilarity between the pixels and the seed points in the remote sensing image I to be segmented.

[0032] In a third aspect of the present invention, the present invention provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the superpixel segmentation method when executing the computer program.

[0033] In a fourth aspect of the present invention, the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the superpixel segmentation method are implemented.

[0034] The present invention has the following beneficial effects:

[0035] During specific operation, the superpixel segmentation method, system, device and storage medium described in the present invention input the remote sensing image I to be segmented into a non-similarity model between pixels and seed points with boundary contour constraints, obtain the non-similarity between pixels and seed points in the remote sensing image I to be segmented, and then perform superpixel segmentation based on this. Specifically, the present invention introduces the diffusion flux of pixels about seed points on the basis of spectral distance and spatial distance, wherein the diffusion flux of pixels about seed points is calculated based on the anisotropic diffusion model, and constructs a non-similarity model between pixels and seed points with boundary contour constraints to improve the superpixel fitting ability of complex landform boundaries, improve the superpixel segmentation accuracy, and further improve the subsequent remote sensing image interpretation accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The accompanying drawings, which constitute part of the present invention, are intended to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are intended to explain the present invention and do not constitute an undue limitation of the present invention. In the accompanying drawings:

[0037] Figure 1a is a flow chart of the method of the present invention;

[0038] Figure 1b Flowchart for the calculation of diffusion flux in the present invention;

[0039] Figure 2a This is a schematic diagram of Landsat8 imagery;

[0040] Figure 2b This is the standard classification result map of Landsat8 image;

[0041] Figure 3a This is the superpixel segmentation result of NCut;

[0042] Figure 3b This is the superpixel segmentation result of ERS;

[0043] Figure 3c This is the superpixel segmentation result of Turbopixels;

[0044] Figure 3d This is the superpixel segmentation result of Scow;

[0045] Figure 3e This is the superpixel segmentation result of SLIC;

[0046] Figure 3f This is the superpixel segmentation result of LSC;

[0047] Figure 3g This is the superpixel segmentation result image based on the diffusion coefficient c1 of the present invention;

[0048] Figure 3h This is the superpixel segmentation result image based on the diffusion coefficient c2 of the present invention;

[0049] Figure 4a This is the remote sensing image classification result map of NCut;

[0050] Figure 4b This is the ERS remote sensing image classification result map;

[0051] Figure 4c This is the remote sensing image classification result map of Turbopixels;

[0052] Figure 4d This is Scow's remote sensing image classification result map;

[0053] Figure 4e This is the remote sensing image classification result map of SLIC;

[0054] Figure 4f This is the LSC remote sensing image classification result map;

[0055] Figure 4g This is a remote sensing image classification result diagram based on the diffusion coefficient c1 of the present invention;

[0056] Figure 4hThis is a remote sensing image classification result diagram based on the diffusion coefficient c2 of the present invention;

[0057] Figure 5a for Figure 4a The standard map of area 1 indicated by the middle arrow;

[0058] Figure 5b for Figure 4a The enlarged image of area 1 indicated by the middle arrow;

[0059] Figure 5c for Figure 4b Enlarged view of area 1 indicated by the middle arrow;

[0060] Figure 5d for Figure 4c The enlarged image of area 1 indicated by the middle arrow;

[0061] Figure 5e for Figure 4d Enlarged view of area 1 indicated by the middle arrow;

[0062] Figure 5f for Figure 4e The enlarged image of area 1 indicated by the middle arrow;

[0063] Figure 5g for Figure 4f Enlarged view of area 1 indicated by the middle arrow;

[0064] Figure 5h for Figure 4g Enlarged view of area 1 indicated by the middle arrow;

[0065] Figure 5i for Figure 4h The enlarged image of area 1 indicated by the middle arrow;

[0066] Figure 6a for Figure 4a The standard map of area 2 is indicated by the middle arrow;

[0067] Figure 6b for Figure 4a The enlarged image of area 2 indicated by the middle arrow;

[0068] Figure 6c for Figure 4b The enlarged image of area 2 indicated by the middle arrow;

[0069] Figure 6d for Figure 4c The enlarged image of area 2 indicated by the middle arrow;

[0070] Figure 6e for Figure 4d The enlarged image of area 2 indicated by the middle arrow;

[0071] Figure 6f for Figure 4eThe enlarged image of area 2 indicated by the middle arrow;

[0072] Figure 6g for Figure 4f The enlarged image of area 2 indicated by the middle arrow;

[0073] Figure 6h for Figure 4g The enlarged image of area 2 indicated by the middle arrow;

[0074] Figure 6i for Figure 4h The enlarged image of area 2 indicated by the middle arrow;

[0075] Figure 7 This is a system structure diagram of the present invention. DETAILED DESCRIPTION

[0076] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0077] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0078] The present invention is described in further detail below with reference to the accompanying drawings:

[0079] The principle of the present invention is: to introduce the concept of diffusion flux on the basis of spectral distance and spatial distance, to add a boundary constraint mechanism of land objects in the clustering process, to improve the fitting ability of superpixels to the boundaries of complex land objects, to improve the segmentation accuracy of superpixels, and then to further improve the interpretation accuracy of subsequent remote sensing images. Specifically, a number of seed points are randomly determined in the image domain. Assuming that the seed points are concentration emitters, the concentration is concentrated at the seed points at the initial moment, and diffuses to any point in the image domain over time according to the concentration gradient and the image gradient. The concentration gradient controls the diffusion direction, and the image gradient controls the diffusion speed. The diffusion speed is a monotonically decreasing function of the image gradient. The diffusion speed reaches a maximum value in the homogeneous area and reaches a minimum value at the boundary, thereby achieving the purpose of exploring the boundary in the clustering process. The cumulative value of the diffusion amount that diffuses continuously over a period of time is called the diffusion flux.

[0080] Combining diffusion flux, spatial distance and spectral distance, a non-similarity model between pixels and seed points with boundary contour constraints is constructed, and then superpixel segmentation is achieved based on the K-means clustering algorithm. Applying this invention to remote sensing image classification can effectively improve the accuracy of remote sensing image classification.

[0081] Example 1

[0082] Specifically, refer to Figure 1a and Figure 1b , the superpixel segmentation method of the present invention comprises the following steps:

[0083] 1) Input remote sensing image I;

[0084] Given a high-resolution remote sensing image I = {I i (x i ,y i ):(x i ,y i )∈Ω,i=1,...,n}, where i is the pixel index, (x i ,y i ) is the grid point coordinate of pixel i in the image space domain, I i is the spectral measurement vector of pixel i, Ω is the image space domain, Ω={(x i ,y i ):i=1,...,n}.

[0085] 2) Model parameter initialization;

[0086] 21) Initialize the seed point set S;

[0087] Assume that the segmentation scale is C. To avoid the seed points falling on the boundary, The position with the lowest gradient value in the 3×3 neighborhood window at the center of the uniform grid is used as the seed point, and the initial seed point set S = {S j(a j ,b j ):j=1,...,m}, where j is the seed point (superpixel) index, m is the number of seed points (superpixels), (a j ,b j ) is the grid coordinate of seed point j in the image space domain, S j is the spectral measurement vector of seed point (superpixel) j;

[0088] 22) Initialize the spatial distance normalization parameter N Spatial , spectral distance normalization parameter N spectral , diffusion flux normalization parameter N Flux And the image gradient histogram threshold η.

[0089] 3) Calculate the spatial distance between the pixel and the seed point

[0090] 4) Calculate the spectral distance between the pixel and the seed point

[0091] 5) Calculate the diffusion flux of pixels in the image about the seed point;

[0092] 51) Based on the seed point set, construct the initial diffusion flux map U J (x, y, t=0), where (x i ,y i )=(a j ,b j ), then U j (x i ,y i ,t=0)=1;

[0093] 52) Construct spectral gradient map Where f is the gradient direction, |f| = 4 is the number of directions, f = {(s, p): (s, p) ∈ {(0, +1), (-1, 0), (0, -1), (+1, 0)}}, and the gradient images in each direction are:

[0094]

[0095]

[0096]

[0097]

[0098] 53) Adaptively determining a diffusion gradient threshold according to a gradient image histogram threshold η, wherein the gradient value corresponding to when the histogram cumulative value is greater than η is used as the diffusion gradient threshold;

[0099] 54) Calculate the diffusion coefficient c. There are two ways to calculate the diffusion coefficient c:

[0100]

[0101] or

[0102] Where δ is the diffusion gradient threshold, |·| represents the modulus value;

[0103] 55) Update the diffusion flux map;

[0104]

[0105] Among them, λ is the coefficient that controls the stability of the model, and R is the diffusion distance;

[0106] 56) Repeat step 55) until the iteration stop condition is reached, wherein the iteration stop condition is that the diffusion radius of the pixel about the seed point is greater than 2N Spatial or the maximum number of iterations is reached.

[0107] 6) Calculate the dissimilarity D between the pixel and the seed point with boundary contour constraints ij ;

[0108]

[0109] in,

[0110]

[0111] 7) Get the superpixel label L to which the pixel belongs i =arg min(D ij );

[0112] 8) Eliminate isolated pixels and update the seed point set S, which is updated with the mean of the spatial and spectral features of all pixels in the superpixel;

[0113] 9) Repeat steps 3 to 8) until the iteration stops when the iteration stopping condition is reached, wherein the iteration stopping condition is that the seed point set converges or the iteration reaches the maximum number of times.

[0114] It should be noted that the present invention introduces the concept of diffusion flux on the basis of spectral distance and spatial distance to construct the non-similarity between pixels and seed points. This metric adds a boundary constraint mechanism of land objects in the clustering process based on the anisotropic diffusion model, improves the superpixel fitting ability of complex land object boundaries, and solves the problem that traditional superpixel segmentation methods are difficult to effectively segment land objects with large differences in shape, scale, texture and other features at the same scale, thereby improving the accuracy of superpixel segmentation and further improving the interpretation accuracy of subsequent remote sensing images.

[0115] It should be noted that the distance function corresponding to spatial distance and spectral distance in the present invention is only taken as an example by Euclidean distance, and can also be Mahalanobis distance, Manhattan distance, Chebyshev distance, Minkowski distance, correlation coefficient, probability density function and information entropy. In addition, the present invention is realized by taking K-means clustering as an example, and other machine learning methods, such as fuzzy clustering, Bayesian classification, etc. are not excluded. Simultaneously, diffusion coefficient model, the present invention is modeled as an example by taking c1 and c2, and other diffusion coefficient modeling modes are not excluded.

[0116] Comparative experiments;

[0117] The experimental environment is: simulation based on software programming on a CPU Core(TM) i5-7300U 2.60GHz, 8GB memory, and Windows 10 Ultimate system.

[0118] The present invention and currently more widely used superpixel segmentation methods, such as NCut, ERS, Turbopixels, SCow, SLIC and LSC, were applied to perform superpixel segmentation and subsequent classification experiments on Landsat8 images, and the accuracy of the segmentation and classification results was evaluated.

[0119] Figure 2a and Figure 2b They are Landsat8 images and standard classification results, which include five land feature types: forest, artificial surface, water body, farmland and bare soil.

[0120] Figures 3a to 3h They are NCut, ERS, Turbopixels, SCow, SLIC, LSC and the superpixel segmentation result diagram based on diffusion coefficients c1 and c2 of the present invention. It can be seen that the present invention can effectively segment large-area homogeneous objects and objects with narrow and fine features at the same time.

[0121] Figures 4a to 4h They are NCut, ERS, Turbopixels, SCow, SLIC, LSC and the remote sensing image classification results based on diffusion coefficients c1 and c2 of the present invention, and the arrows point to the enlarged areas as shown in Figures 5a to 6i As shown, the classification results of the present invention are closer to those of the standard classification.

[0122] In order to verify the effectiveness of the present invention, Figure 2bThe superpixel segmentation accuracy and classification accuracy of the present invention and the comparison algorithm are calculated with the true value. The results are shown in Table 1, including four indicators: Recall, UE, OA, and Kappa. Recall and UE are used to evaluate superpixel accuracy, while OA and Kappa are used to evaluate classification accuracy. The higher the indicators shown in Table 1, the better. As can be seen from Table 1, the accuracy of the present invention is higher than that of the comparison algorithm, and the accuracy fluctuation is small, which effectively verifies the effectiveness of the present invention.

[0123] Table 1

[0124]

[0125]

[0126] Example 2

[0127] refer to Figure 7 The superpixel segmentation system of the present invention comprises:

[0128] Acquisition module 1, used to acquire the remote sensing image I to be segmented;

[0129] a dissimilarity calculation module 2, configured to input the remote sensing image to be segmented I into a dissimilarity model between pixels and seed points with boundary contour constraints, and obtain dissimilarity between pixels and seed points in the remote sensing image to be segmented I, wherein the dissimilarity model between pixels and seed points with boundary contour constraints is constructed based on diffusion flux, spatial distance, and spectral distance of pixels with respect to seed points, and the diffusion flux of pixels with respect to seed points is calculated based on an anisotropic diffusion model;

[0130] The superpixel segmentation module 3 is used to perform superpixel segmentation on the remote sensing image I to be segmented according to the dissimilarity between the pixels and the seed points in the remote sensing image I to be segmented.

[0131] Example 3

[0132] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the superpixel segmentation method are implemented. The memory may include internal memory, such as a high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device. The processor, network interface, and memory are interconnected via an internal bus. The internal bus may be an industrial standard architecture bus, a peripheral component interconnect standard bus, an extended industrial standard architecture bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. The memory is used to store programs. Specifically, the program may include program code, and the program code includes computer operating instructions. The memory may include internal memory and non-volatile memory, and provides instructions and data to the processor.

[0133] Example 4

[0134] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the superpixel segmentation method. Specifically, the computer-readable storage medium includes, but is not limited to, volatile memory and / or non-volatile memory. The volatile memory may include random access memory (RAM) and / or cache memory, etc. The non-volatile memory may include read-only memory (ROM), hard disk, flash memory, optical disk, magnetic disk, etc.

[0135] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0136] The present application is described with reference to the flow chart and / or block diagram of the method, device (system), and computer program product according to the embodiment of the present application. It should be understood that each flow process and / or box in the flow chart and / or block diagram and the combination of the flow process and / or box in the flow chart and / or block diagram can be realized by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processing machine or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device for realizing the function specified in one flow chart flow or multiple flows and / or one box or multiple boxes of the block diagram.

[0137] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a product including an instruction device that implements the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0138] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0139] 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 it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.

Claims

1. A superpixel segmentation method, characterized in that: include: Obtain the remote sensing image I to be segmented; Inputting the remote sensing image to be segmented I into a dissimilarity model between pixels and seed points with boundary contour constraints to obtain dissimilarity between pixels and seed points in the remote sensing image to be segmented I, wherein the dissimilarity model between pixels and seed points with boundary contour constraints is constructed based on a diffusion flux, a spatial distance, and a spectral distance of the pixel with respect to the seed point, and the diffusion flux of the pixel with respect to the seed point is calculated based on an anisotropic diffusion model; Perform superpixel segmentation on the remote sensing image I to be segmented according to the dissimilarity between the pixels and the seed points in the remote sensing image I to be segmented; The dissimilarity D between pixel i and seed point j in the remote sensing image I to be segmented ij for: in, is the spectral distance between pixel i and seed point j, is the spatial distance between pixel i and seed point j, is the diffusion flux of pixel i about seed point j, N Spatial is the normalization parameter of spatial distance, N Spectral is the spectral distance normalization parameter, N Flux is the diffusion flux normalization parameter.

2. The superpixel segmentation method according to claim 1, wherein: The spatial distance between pixel i and seed point j for: Among them, (a j ,b j ) is the grid coordinate of the seed point j in the image space domain, (x i ,y i ) is the grid position coordinate of pixel i in the image space domain.

3. The superpixel segmentation method according to claim 1, wherein: The spectral distance between pixel i and seed point j for: Among them, S j is the spectral measurement vector of seed point j, I i is the spectral measure vector of pixel i.

4. The superpixel segmentation method according to claim 1, wherein: The diffusion flux between the pixel i and the seed point j for: Among them, U j i (x i ,y i ,T) is the coordinate point (x i ,y i ) value.

5. The superpixel segmentation method according to claim 4, wherein: The updating process of the diffusion flux map in the non-similarity model is: Among them, U J (x,y,t+1) is the diffusion flux diagram at time t+1, U J (x, y, t) is the diffusion flux diagram at time t, λ is the stability coefficient, R is the diffusion distance, c f (·) is the diffusion coefficient in the f direction, |·| represents the modulus value, represents the image gradient in the f direction, represents the flux gradient in the direction f.

6. The superpixel segmentation method according to claim 1, wherein: The specific operation of performing superpixel segmentation on the remote sensing image to be segmented I according to the dissimilarity between the pixels and the seed points in the remote sensing image to be segmented I is: According to the dissimilarity between pixels and seed points in the remote sensing image I to be segmented, the K-means clustering algorithm is used to perform superpixel segmentation on the remote sensing image I to be segmented.

7. A superpixel segmentation system, characterized in that: include: An acquisition module (1) is used to acquire a remote sensing image I to be segmented; a dissimilarity calculation module (2) for inputting the remote sensing image to be segmented I into a dissimilarity model between pixels and seed points with boundary contour constraints, and obtaining dissimilarity between pixels and seed points in the remote sensing image to be segmented I, wherein the dissimilarity model between pixels and seed points with boundary contour constraints is constructed based on diffusion flux, spatial distance and spectral distance of pixels with respect to seed points, and the diffusion flux of pixels with respect to seed points is calculated based on an anisotropic diffusion model; A superpixel segmentation module (3) is used to perform superpixel segmentation on the remote sensing image I to be segmented based on the dissimilarity between pixels and seed points in the remote sensing image I to be segmented; The dissimilarity D between pixel i and seed point j in the remote sensing image I to be segmented ij for: in, is the spectral distance between pixel i and seed point j, is the spatial distance between pixel i and seed point j, is the diffusion flux of pixel i about seed point j, N Spatial is the normalization parameter of spatial distance, N Spectral is the spectral distance normalization parameter, N Flux is the diffusion flux normalization parameter.

8. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the superpixel segmentation method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the superpixel segmentation method according to any one of claims 1 to 6 are implemented.

Images