Variable-scale multispectral remote sensing image segmentation method using spectral index

Abstract

The invention discloses a variable-scale multispectral remote sensing image segmentation method using a spectral index, which belongs to the technical field of remote sensing image processing, and comprises the following steps: firstly, determining the gradient of a multispectral remote sensing image by using a vector field model, generating a mark according to the gradient, selecting a proper spectral index, and carrying out image segmentation according to the spectral index; and carrying out image segmentation by using watershed transformation. According to the variable-scale multispectral remote sensing image segmentation method using the spectral index, optimization is carried out by using the spectral index, the capability of distinguishing different ground feature types is enhanced, and the segmentation precision is improved; according to the method, watershed transformation is adopted for segmentation, so that the image segmentation result is more stable and efficient, the method can adapt to remote sensing images of different scales, the quality and effect of remote sensing data processing are improved, and mathematical morphology-based variational scale watershed segmentation is optimized.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing image processing, and in particular to a variable-scale multispectral remote sensing image segmentation method using spectral indices. Background Art

[0002] Scale-variable segmentation of various geographic objects in high-resolution remote sensing imagery has been the focus of much research. However, selecting the most appropriate scale based on the geographic distribution of ground objects remains a major challenge. Optimal segmentation of remote sensing images based on different ground object scales has become a core issue.

[0003] The watershed transform is a widely used method in which the quality of the markers directly determines the segmentation scale. Therefore, the final segmentation quality depends on the marker generation. If the markers can be generated based on the object's own scale, the watershed transform can be used to obtain variable-scale segmentation results for the object. Spectral index-based markers can also help overcome over-segmentation, as they are sensitive to irrelevant local minima generated by the gradient.

[0004] In summary, compared to other imagery (such as medical images, finger images, eye images, and facial images), remote sensing images contain more complex scenes: multi-scale nested features of land cover types (e.g., from individual trees to forest patches), the superposition of spectral and spatial heterogeneity (e.g., mixed crops within agricultural plots), and topological constraints between different land objects (e.g., the spatial connection between road networks and buildings), all of which make watershed transform-based segmentation face more severe theoretical and technical challenges. Existing methods have not yet fully addressed the challenge of achieving accurate variable-scale segmentation through self-organizing label generation mechanisms. Summary of the Invention

[0005] The purpose of the present invention is to provide a variable-scale multispectral remote sensing image segmentation method using spectral indices, which uses spectral indices for optimization, enhances the ability to distinguish different types of land features, and improves segmentation accuracy; uses watershed transform for segmentation, making the image segmentation results more stable and efficient, and can adapt to remote sensing images of different scales, improve the quality and effect of remote sensing data processing, and optimize the variational-scale watershed segmentation based on mathematical morphology.

[0006] To achieve the above object, the present invention provides a scale-varying multispectral remote sensing image segmentation method using spectral index, comprising the following steps:

[0007] S1, generating labels from multispectral gradients using a vector field model;

[0008] S2, generating a spectral index mark based on the histogram;

[0009] S3, labeled image with merged gradient index and spectral index;

[0010] S4. Use watershed transform for image segmentation.

[0011] Preferably, the S1 is based on the first basic form of the vector field model, uses the Canny method to derive the gradient of the multispectral remote sensing image, and generates a marker from the multispectral gradient;

[0012] Let I(x,y) appear in the multispectral image in the form of a vector field, whose band is I i (x,y), i=1,…,n,I i The gradient of (x,y) in the i-th band is:

[0013]

[0014] Its first basic form is:

[0015]

[0016] Among them, the matrix The eigenvector of is used to obtain the direction of maximum and minimum change; the eigenvalue of the matrix G represents the gradient of the image, and the eigenvector of G determines the edge direction; the maximum and minimum eigenvalues ​​λ ± is defined as:

[0017]

[0018] The multispectral gradient is defined as:

[0019]

[0020] Preferably, in said S2, a spectral index is selected, and a spectral index marker image is generated using kernel density estimation based on the spectral index analysis;

[0021] First, given x1,x2,x3,…,x n ∈R is a random sample of the distribution F of f(x), then the sum density estimate of f(x) is:

[0022]

[0023] Among them, k(x,x i ) is the kernel function, and the following Gaussian function is used as the kernel function:

[0024]

[0025] Where σ is the variance

[0026] When x=T peak The peak point A when x=T trough The valley value at can be obtained by the following formula:

[0027]

[0028] Where Δp f (i) indicates that sign() indicates the sign function, Indicates the variable point i corresponding to the maximum value.

[0029] Preferably, in said S3, mathematical morphology is selected to combine gradient and spectral index to reconstruct the marked image;

[0030] First get the intersection of the gradient and spectral index marker images as follows:

[0031] INSECT = S ∩ G;

[0032] Where S is the spectral labeled image;

[0033] Second, the gradient marker minus the spectral index marker is defined as:

[0034] G'=(G-INSECT)ΘB;

[0035] Among them, Θ represents the erosion operation, and B is the structural element in mathematical morphology;

[0036] Then, we use the intersection INSECT as the mark and S as the mask to reconstruct a new mark. The S reconstructed from INSECT is denoted as RS and is defined as:

[0037] RS = THIN(S∪INSECT);

[0038] Among them, THIN() represents the thinning operation in mathematical morphology;

[0039] This reconstruction step is repeated while converting the spectral index marks into one-pixel-wide lines;

[0040] Finally, the final marker CM is defined as:

[0041]

[0042] Preferably, in the watershed transformation of S4, the number of labels generated by the spectral index is determined by the threshold value of the spectral index;

[0043] Based on the final labeled image, a watershed transform is performed using the minimum stacking technique, using local variance as a tool to explore the optimal ratio. Based on the image, the local variance is defined as:

[0044]

[0045] Among them, Var(I i,j) is the local mean of the image, e, I i,j 、 Represent neighborhood sum, image pixel value and image variance respectively.

[0046] Therefore, the present invention adopts the above-mentioned variable-scale multispectral remote sensing image segmentation method using spectral index, uses spectral index for optimization, enhances the ability to distinguish different types of land objects, and improves segmentation accuracy; uses watershed transform for segmentation, making the image segmentation results more stable and efficient, and can adapt to remote sensing images of different scales, thereby improving the quality and effect of remote sensing data processing, so as to optimize the variational-scale watershed segmentation based on mathematical morphology.

[0047] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 The present invention is a flowchart of an embodiment of a variable-scale multispectral remote sensing image segmentation method using spectral indices. DETAILED DESCRIPTION

[0049] The technical solution of the present invention is further described below with reference to the accompanying drawings and embodiments.

[0050] Unless otherwise defined, technical or scientific terms used in the present invention shall have the same meaning as commonly understood by one of ordinary skill in the art to which the present invention belongs.

[0051] Example 1

[0052] like Figure 1 As shown, the present invention provides a variable scale multispectral remote sensing image segmentation method using spectral index, comprising the following steps:

[0053] S1. Generate markers from multispectral gradients using a vector field model: Based on the first basic form of the vector field model, the Canny method is used to derive the gradient of the multispectral remote sensing image and generate markers from the multispectral gradient. The specific implementation method is as follows:

[0054] Let I(x,y) appear in the multispectral image in the form of a vector field, whose band is I i (x,y), i=1,…,n,I i The gradient of (x,y) in the i-th band can be written as:

[0055]

[0056] Its first basic form (i.e., the square criterion) is:

[0057]

[0058] Among them, the matrix The eigenvectors of can be used to obtain the directions of maximum and minimum changes. The eigenvalues ​​of the matrix G can represent the gradient of the image, and the eigenvectors of G can be used to determine the edge direction. The maximum and minimum eigenvalues ​​λ ± is defined as:

[0059]

[0060] The multispectral gradient can be defined as:

[0061]

[0062] S2. Generate spectral index markers based on histogram: Select appropriate spectral index and generate spectral index marker images using kernel density estimation based on spectral index analysis. The specific implementation method is as follows:

[0063] First, given x1,x2,x3,…,x n ∈R is a random sample of the distribution F of f(x), then the sum density estimate of f(x) is:

[0064]

[0065] Among them, k(x,x i ) is the kernel function. The following Gaussian function is used as the kernel function:

[0066]

[0067] Among them, σ is the variance, d is the feature space distance between samples;

[0068] Peak point A (x=T peak ) and valley value (x=T trough ) can be obtained by the following formula:

[0069]

[0070] Where Δp f (i) indicates that sign() indicates the sign function, Indicates the variable point i corresponding to the maximum value.

[0071] S3. Marker image combining gradient index and spectral index: Mathematical morphology is used to combine gradient and spectral index to reconstruct the marker image. The specific implementation method is as follows:

[0072] First get the intersection of the gradient and spectral index marker images as follows:

[0073] INSECT = S ∩ G;

[0074] Where S is the spectral labeled image;

[0075] Second, the gradient marker minus the spectral index marker is defined as:

[0076] G'=(G-INSECT)ΘC;

[0077] Among them, Θ represents the erosion operation and B is the structural element in mathematical morphology.

[0078] Then, we use the intersection INSECT as the mark and S as the mask to reconstruct a new mark. The S reconstructed from INSECT (denoted as RS) is defined as:

[0079] RS = THIN(S∪INSECT);

[0080] Among them, THIN() represents the thinning operation in mathematical morphology;

[0081] This reconstruction step is repeated while converting the spectral index marks into one-pixel-wide lines.

[0082] Finally, the final marker CM is defined as:

[0083]

[0084] S4. Image segmentation using watershed transform: In the watershed transform, the threshold of the spectral index determines the number of markers generated based on the spectral index and is the key to affecting the segmentation performance.

[0085] Based on the final labeled image, the minimum value superposition technique is used to perform watershed transformation. The specific implementation method is:

[0086] Use local variance as a tool to explore the optimal ratio. Depending on the image, local variance can be defined as:

[0087]

[0088] Among them, Var(I i,j ) is the local mean of the image, e, I i,j 、 Represent neighborhood sum, image pixel value and image variance respectively.

[0089] Therefore, the present invention adopts the above-mentioned variable-scale multispectral remote sensing image segmentation method using spectral index, uses spectral index for optimization, enhances the ability to distinguish different types of land objects, and improves segmentation accuracy; uses watershed transform for segmentation, making the image segmentation results more stable and efficient, and can adapt to remote sensing images of different scales, thereby improving the quality and effect of remote sensing data processing, so as to optimize the variational-scale watershed segmentation based on mathematical morphology.

[0090] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solutions of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A scale-varying multispectral remote sensing image segmentation method using spectral indices, characterized by: The following steps are involved: S1, generating labels from multispectral gradients using a vector field model; S2, generating a spectral index mark based on the histogram; S3, labeled image with merged gradient index and spectral index; S4. Use watershed transform for image segmentation.

2. The variable-scale multispectral remote sensing image segmentation method using spectral index according to claim 1, characterized in that: Said S1 is based on the first basic form of the vector field model, utilizes the Canny method to derive the gradient of the multispectral remote sensing image, and generates a marker from the multispectral gradient; Let I(x,y) appear in the multispectral image in the form of a vector field, whose band is I i (x,y), i=1,…,n,I i The gradient of (x,y) in the i-th band is: Its first basic form is: Among them, the matrix The eigenvector of is used to obtain the direction of maximum and minimum change; the eigenvalue of the matrix G represents the gradient of the image, and the eigenvector of G determines the edge direction; the maximum and minimum eigenvalues ​​λ ± is defined as: The multispectral gradient is defined as:

3. The variable-scale multispectral remote sensing image segmentation method using spectral index according to claim 1, characterized in that: In said S2, a spectral index is selected, and a spectral index marker image is generated using kernel density estimation according to the spectral index analysis; First, given x1,x2,x3,…,x n ∈R is a random sample of the distribution F of f(x), then the sum density estimate of f(x) is: Among them, k(x,x i ) is the kernel function, and the following Gaussian function is used as the kernel function: Among them, σ is the variance, d is the feature space distance between samples; When x=T peak The peak point A when x=T trough The valley value at can be obtained by the following formula: Where Δp f (i) indicates that sign() indicates the sign function, Indicates the variable point i corresponding to the maximum value.

4. The variable-scale multispectral remote sensing image segmentation method using spectral index according to claim 1, characterized in that: In S3, mathematical morphology is selected to combine gradient and spectral index to reconstruct the marked image; First get the intersection of the gradient and spectral index marker images as follows: INSECT = S ∩ G; Where S is the spectral labeled image; Second, the gradient marker minus the spectral index marker is defined as: G'=(G-INSECT)ΘC; Among them, Θ represents the corrosion operation, B is the structural element; Then, we use the intersection INSECT as the mark and S as the mask to reconstruct a new mark. The S reconstructed from INSECT is denoted as RS and is defined as: RS = THIN(S∪INSECT); Among them, THIN() represents the thinning operation; This reconstruction step is repeated while converting the spectral index marks into one-pixel-wide lines; Finally, the final marker CM is defined as:

5. The variable scale multispectral remote sensing image segmentation method using spectral index according to claim 1, characterized in that: In the watershed transformation of S4, the number of markers generated by the spectral index is determined by the threshold of the spectral index; In terms of images, the local variance is defined as:. Among them, Var(I i,j ) is the local mean of the image, e, I i,j 、 Represent neighborhood sum, image pixel value and image variance respectively.

Images