Image separation method, device, storage medium and electronic device

Abstract

The present application discloses an image separation method, device, storage medium and electronic device, and relates to the field of image processing technology, wherein the method comprises: performing noise estimation on an image to be separated to obtain an estimated noise matrix corresponding to the image to be separated; performing kernelization processing using a kernel minimum noise transformation method based on the estimated noise matrix and the estimated pixel matrix corresponding to the image to be separated to obtain a kernelized matrix; based on the kernelized matrix, performing image reconstruction using a preset machine learning method based on local linear modeling to obtain a target image matrix; performing background separation processing on the target image matrix to obtain a target background image and a target image. The method in the present application can be applied to spectral images of any noise level in any scene for image separation, thereby improving image separation efficiency.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to an image separation method, device, storage medium and electronic equipment. Background Art

[0002] Image segmentation refers to dividing the original image into several non-intersecting regions based on multiple features such as grayscale, color, geometric shape, spatial texture, etc., and making certain features in the same region show similarity or consistency, while different regions are obviously different. Simply put, it is to separate the target from the background in an image. For grayscale images, pixels inside the region generally have grayscale similarity, while there is generally grayscale discontinuity at the boundary of the region. Traditional segmentation methods include: threshold-based image segmentation methods, region-based image segmentation methods, and edge detection-based segmentation methods. Most traditional image segmentation algorithms use knowledge in mathematics, digital image processing, etc., but the disadvantage of threshold-based image segmentation methods is that they mainly rely on grayscale histograms, rarely consider the spatial position relationship of pixels, are very sensitive to noise, and are prone to lose some boundary information when the background is complex, resulting in incomplete segmentation. For images with small contrast between the target and the background (such as images with small distinction between cells and background), it is difficult to obtain accurate target boundaries. In addition, threshold segmentation only considers the grayscale information of the image without considering the spatial information of the image, and is very sensitive to noise and grayscale unevenness. The disadvantages of region-based image segmentation methods include: large storage space and computing time overhead, easy to be affected by overlapping interference between internal tissues of the target, difficult to determine the termination conditions of growth and division, and often irregular boundaries, discontinuous regions and holes. In addition, region growing technology may obtain inaccurate results in noisy images because edges may be very difficult to detect. Image segmentation methods based on edge detection have poor segmentation effects when the edges of the image are unclear or contain multiple edges. In addition, the method has poor noise resistance because edge detection requires the use of prior knowledge related to edge features, and the default parameters cannot be guaranteed to be applicable to all images. When the default parameter values ​​are inconsistent with the actual application, ideal edge detection results cannot be obtained, and therefore ideal segmentation results cannot be obtained. Summary of the invention

[0003] In view of this, the present invention provides an image separation method, device, storage medium and electronic device, the main purpose of which is to solve the problems of inaccurate segmentation, waste of resources and low separation efficiency in the current image separation process.

[0004] To solve the above problems, the present application provides an image separation method, comprising:

[0005] Perform noise estimation on the image to be separated to obtain an estimated noise matrix corresponding to the image to be separated;

[0006] Based on the estimated noise matrix and the estimated pixel matrix corresponding to the image to be separated, a kernel minimum noise transformation method is used to perform kernelization processing to obtain a kernelized matrix;

[0007] Based on the kernelization matrix, a preset machine learning method based on local linear modeling is used to perform image reconstruction to obtain a target image matrix;

[0008] The target image matrix is ​​subjected to background separation processing to obtain a target background image and a target image.

[0009] Optionally, performing noise estimation on the image to be separated to obtain an estimated noise matrix corresponding to the image to be separated specifically includes:

[0010] For each pixel point in the image to be separated, a preset window is used to collect the neighborhood pixels of each pixel point to obtain a number of neighborhood pixel values ​​corresponding to each pixel point;

[0011] Calculating and processing each of the neighborhood pixel values ​​corresponding to the same pixel point to obtain an estimated pixel value corresponding to the same pixel point, so as to obtain an estimated pixel matrix corresponding to the image to be separated;

[0012] Calculation processing is performed based on the initial pixel matrix of the image to be separated and the estimated pixel matrix to obtain an estimated noise matrix corresponding to the image to be separated.

[0013] Optionally, the performing kernelization processing based on the estimated noise matrix and the estimated pixel matrix corresponding to the image to be separated using a kernel minimum noise transform method to obtain a kernelized matrix specifically includes:

[0014] Performing average value calculation processing based on the initial pixel matrix of the image to be separated to obtain a mean value matrix corresponding to the image to be separated;

[0015] Based on the mean matrix, respectively performing calculation processing on the estimated noise matrix and the estimated pixel matrix to obtain a noise center matrix corresponding to the estimated noise matrix and a signal center matrix corresponding to the estimated pixel matrix;

[0016] Performing calculation processing based on the noise center matrix to obtain a noise covariance matrix corresponding to the noise center matrix;

[0017] Performing calculation processing based on the signal center matrix to obtain a signal covariance matrix corresponding to the signal center matrix;

[0018] Based on the noise covariance matrix and the signal covariance matrix, a preset kernel function is used to perform nonlinear mapping processing to obtain the kernelization matrix.

[0019] Optionally, based on the kernelization matrix, a preset machine learning method based on local linear modeling is used to perform image reconstruction to obtain a target image matrix, specifically including:

[0020] Performing feature extraction on the kernelization matrix to obtain a feature matrix corresponding to the image to be separated;

[0021] The image to be separated is reconstructed based on the feature matrix to obtain the target image matrix.

[0022] Optionally, the performing feature extraction on the kernelization matrix to obtain a feature matrix corresponding to the image to be separated specifically includes:

[0023] Decomposing the kernelized matrix to obtain sub-kernel matrices;

[0024] Performing calculation processing based on each of the sub-kernel matrices to obtain a similarity matrix corresponding to the kernelization matrix;

[0025] Performing diagonal processing on a target sub-core matrix in the sub-core matrix to obtain a diagonal matrix corresponding to the target sub-core matrix;

[0026] Performing calculations based on the diagonal matrix and the similarity matrix to obtain an initial eigenvector corresponding to the similarity matrix;

[0027] Orthogonalization is performed based on the initial eigenvector to obtain a feature matrix corresponding to the image to be separated.

[0028] Optionally, performing background separation processing on the target image matrix to obtain a target background image and a target image specifically includes:

[0029] Performing calculation processing based on each pixel point in the target image matrix to obtain an average vector corresponding to the target image matrix;

[0030] Performing calculation based on the average vector and the first vectors corresponding to the pixels to obtain an initial similarity between the average vector and the first vectors;

[0031] Calculate and process based on each of the initial similarities to obtain an initial similarity distance and an initial threshold for starting an iteration round;

[0032] Based on each of the initial similarities, background separation processing is performed on the target image matrix of the starting iteration round to obtain an initial background image and an initial target image;

[0033] When the initial similarity distance at the beginning of the iteration round satisfies a first preset condition, determining the initial target image as the target image and determining the initial background image as the target background image;

[0034] When the initial similarity distance at the beginning of the iteration round does not meet the first preset condition, the background separation processing process of the next round is performed on the initial target image, and the iteration is repeated until the current similarity distance of the current iteration round and the historical similarity distance of the previous iteration round meet the second preset condition, and the current target image of the current round is determined to be the target image and the current background image of the current round is determined to be the target background image.

[0035] Optionally, the performing calculation based on the average vector and the first vectors corresponding to the pixels to obtain the initial similarity between the average vector and the first vectors specifically includes:

[0036] Performing calculation based on the average vector and the first vector corresponding to each pixel point to obtain each initial Euclidean distance value;

[0037] Normalizing the average vector and each of the first vectors respectively to obtain a first normalized vector corresponding to the average vector and a second normalized vector corresponding to each of the first vectors;

[0038] Calculation processing is performed based on the first normalized vector and each of the second normalized vectors to obtain the initial similarity between the average vector and each of the first vectors.

[0039] In order to solve the above problems, the present application provides an image separation device, comprising:

[0040] A noise estimation module, used to perform noise estimation on the image to be separated to obtain an estimated noise matrix corresponding to the image to be separated;

[0041] A kernelization processing module, configured to perform kernelization processing using a kernel minimum noise transformation method based on the estimated noise matrix and an estimated pixel matrix corresponding to the image to be separated, so as to obtain a kernelization matrix;

[0042] An image reconstruction module, used to perform image reconstruction based on the kernelization matrix by using a preset machine learning method based on local linear modeling to obtain a target image matrix;

[0043] The background separation module is used to perform background separation processing on the target image matrix to obtain a target background image and a target image.

[0044] In order to solve the above problem, the present application provides a storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above image separation method are implemented.

[0045] In order to solve the above problem, the present application provides an electronic device, which at least includes a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps of the above-mentioned image separation method when executing the computer program in the memory.

[0046] The image separation method of the present application has good adaptability to hyperspectral images under different scenes, different noise levels, different spatial resolutions and different spectral resolutions. The method of the present application can achieve autonomous hyperspectral image target background separation without preprocessing or postprocessing operations. The traditional kernel minimum noise transform has the problems of high computational complexity and low execution efficiency. The introduction of a preset machine learning method based on local linear modeling can significantly reduce the computational complexity of obtaining matrix eigenvalues ​​and improve image separation efficiency.

[0047] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0049] Figure 1 A schematic diagram of a process of an image separation method provided in an embodiment of the present application is shown;

[0050] Figure 2 A schematic diagram of a process of an image separation method provided by another embodiment of the present application is shown;

[0051] Figure 3 A schematic diagram of background separation based on an outlier removal method and an iterative strategy provided in an embodiment of the present application is shown;

[0052] Figure 4 A structural block diagram of an image separation device provided in another embodiment of the present application is shown. DETAILED DESCRIPTION

[0053] Various aspects and features of the present application are described herein with reference to the accompanying drawings.

[0054] It should be understood that various modifications may be made to the embodiments of the present application. Therefore, the above description should not be considered as limiting, but only as an example of an embodiment. Other modifications within the scope and spirit of the present application will occur to those skilled in the art.

[0055] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments of the present application and, together with the general description of the present application given above and the detailed description of the embodiments given below, serve to explain the principles of the present application.

[0056] These and other characteristics of the present application will become apparent from the following description of a preferred form of embodiment given as a non-limiting example with reference to the accompanying drawings.

[0057] It should also be understood that although the present application has been described with reference to some specific examples, those skilled in the art will be able to surely implement many other equivalent forms of the present application.

[0058] The above and other aspects, features and advantages of the present application will become more apparent in view of the following detailed description when taken in conjunction with the accompanying drawings.

[0059] Specific embodiments of the present application are described hereinafter with reference to the accompanying drawings; however, it should be understood that the embodiments applied for are merely examples of the present application, which may be implemented in a variety of ways. Well-known and / or repeated functions and structures are not described in detail to avoid unnecessary or redundant details that obscure the present application. Therefore, the specific structural and functional details applied for herein are not intended to be limiting, but merely serve as a basis and representative basis for the claims to teach those skilled in the art to use the present application in a variety of ways with substantially any suitable detailed structure.

[0060] This specification may use the phrases "in one embodiment," "in another embodiment," "in yet another embodiment," or "in other embodiments," all of which may refer to one or more of the same or different embodiments according to the present application.

[0061] The present application embodiment provides an image separation method, such as Figure 1 As shown, including:

[0062] Step S101: performing noise estimation on the image to be separated to obtain an estimated noise matrix corresponding to the image to be separated;

[0063] In the specific implementation of this step, the image to be separated can be a hyperspectral image, and the neighborhood features of the hyperspectral image can be used for noise estimation. Specifically, for each pixel point in the image to be separated, a preset window is used to collect the neighborhood pixels of each pixel point to obtain a number of neighborhood pixel values ​​corresponding to each pixel point; wherein, the preset window can be a 3×3 window, and the present application does not limit the size of the preset window. Calculations are performed on each of the neighborhood pixel values ​​corresponding to the same pixel point to obtain an estimated pixel value corresponding to the same pixel point, so as to obtain an estimated pixel matrix corresponding to the image to be separated; calculations are performed based on the initial pixel matrix of the image to be separated and the estimated pixel matrix to obtain an estimated noise matrix corresponding to the image to be separated.

[0064] Step S102: performing a kernelization process using a kernel minimum noise transformation method based on the estimated noise matrix and the estimated pixel matrix corresponding to the image to be separated, to obtain a kernelized matrix;

[0065] In the specific implementation process of this step, average value calculation processing is performed based on the initial pixel matrix of the image to be separated to obtain a mean matrix corresponding to the image to be separated; based on the mean matrix, calculation processing is performed on the estimated noise matrix and the estimated pixel matrix respectively to obtain a noise center matrix corresponding to the estimated noise matrix and a signal center matrix corresponding to the estimated pixel matrix; calculation processing is performed based on the noise center matrix to obtain a noise covariance matrix corresponding to the noise center matrix; calculation processing is performed based on the signal center matrix to obtain a signal covariance matrix corresponding to the signal center matrix; based on the noise covariance matrix and the signal covariance matrix, a preset kernel function is used to perform nonlinear mapping processing to obtain the kernelization matrix.

[0066] Step S103: Based on the kernelization matrix, a preset machine learning method based on local linear modeling is used to perform image reconstruction to obtain a target image matrix;

[0067] In the specific implementation process of this step, the preset machine learning method based on local linear modeling can be the Nystrom method. Specifically, feature extraction is performed on the kernelization matrix to obtain a feature matrix corresponding to the image to be separated; the image to be separated is reconstructed based on the feature matrix to obtain the target image matrix.

[0068] Step S104: performing background separation processing on the target image matrix to obtain a target background image and a target image.

[0069] In the specific implementation process of this step, background separation processing is iteratively performed based on the target image matrix until the current threshold of the current iteration round meets the preset conditions, and the current target image of the current round is determined to be the target image and the current background image of the current round is determined to be the target background image.

[0070] The image separation method of the present application has good adaptability to hyperspectral images under different scenes, different noise levels, different spatial resolutions and different spectral resolutions. The method of the present application can achieve autonomous hyperspectral image target background separation without preprocessing or postprocessing operations. The traditional kernel minimum noise transform has the problems of high computational complexity and low execution efficiency. The introduction of a preset machine learning method based on local linear modeling can significantly reduce the computational complexity of obtaining matrix eigenvalues ​​and improve image separation efficiency.

[0071] Another embodiment of the present application provides another image separation method, such as Figure 2 As shown, including:

[0072] Step S201: performing noise estimation on the image to be separated to obtain an estimated noise matrix corresponding to the image to be separated;

[0073] In the specific implementation of this step, the image to be separated can be an image category such as a hyperspectral image. For each pixel point in the image to be separated, a preset window is used to collect the neighborhood pixels of each pixel point to obtain a number of neighborhood pixel values ​​corresponding to each pixel point; the preset window size can be 3×3, and this application does not limit the preset window size. The neighborhood pixel values ​​corresponding to the same pixel point are calculated and processed to obtain an estimated pixel value corresponding to the same pixel point. To obtain the estimated pixel matrix Z corresponding to the image to be separated N ; Calculate the estimated pixel value z i,j,k The mathematical formula is as shown in the following formula (1):

[0074]

[0075] Among them, z i,j,k is the pixel value of the i-th row and h-th column of the k-th spectral segment of the original hyperspectral image; is the estimated pixel value corresponding to the pixel value of the i-th row and h-th column of the k-th spectral band of the original hyperspectral image. Estimated pixel matrix Z N An estimated pixel matrix composed of estimated pixel values ​​corresponding to each pixel point of the image to be separated. Based on the initial pixel matrix of the image to be separated and the estimated pixel matrix, a calculation process is performed to obtain an estimated noise matrix corresponding to the image to be separated. Specifically, the data calculation formula for each estimated noise pixel in the estimated noise matrix is ​​as shown in the following formula (2):

[0076]

[0077] Among them, n i,j,k For z i,j,k The estimated noise value.

[0078] Step S202: performing average value calculation processing based on the initial pixel matrix of the image to be separated to obtain a mean value matrix corresponding to the image to be separated;

[0079] In the specific implementation of this step, due to the influence of atmospheric effects and instrument noise, the spectral image is often interfered by various noises, and the spectral image interfered by noise is prone to image degradation. For optical images, the signal and noise are independent of each other. For a hyperspectral image with n pixels and b spectral bands, Z(p) is an image matrix with n rows and b columns. Z(p) is the sum of the signal part and the noise part. The mathematical expression of Z(p) can be shown in the following formula (3):

[0080] Z(p)=Z S (p)+Z N (p) (3);

[0081] Where Z(p) is the pixel vector at position p, Z S (p) is the signal component contained in Z(p), Z N (p) is the noise component included in Z(p).

[0082] Similarly, the covariance matrix of the image can be written as the sum of the signal covariance matrix and the noise covariance matrix. The covariance matrix of the image is shown in the following formula (4):

[0083] S(Z)=S(Z S )+S(Z N ) (4);

[0084] Among them, S(Z) is the covariance matrix of the image, S(Z S ) and S(Z N ) are the signal covariance matrix and the noise covariance matrix respectively. Based on the initial pixel matrix of the image to be separated, the average value calculation process is performed to obtain the mean matrix Z corresponding to the image to be separated mean ; Specifically, the mean matrix Z with n rows and b columns mean It is shown in the following formula (5):

[0085]

[0086] in, is the average value of the i-th band.

[0087] Step S203: Based on the mean matrix, respectively perform calculation processing on the estimated noise matrix and the estimated pixel matrix to obtain a noise center matrix corresponding to the estimated noise matrix and a signal center matrix corresponding to the estimated pixel matrix;

[0088] In the specific implementation process of this step, based on the mean matrix, the estimated noise matrix is ​​calculated and processed to obtain a noise center matrix corresponding to the estimated noise matrix; the calculation mathematical formula of the noise center matrix is ​​shown in the following formula (6):

[0089] X N =Z N +Z mean (6);

[0090] Among them, X N is the noise center matrix, Z N is the estimated noise matrix, Z mean is a mean matrix. Based on the mean matrix, the estimated pixel matrix is ​​calculated and processed to obtain a signal center matrix X corresponding to the estimated pixel matrix; the calculation formula of the signal center matrix is ​​as shown in the following formula (7):

[0091] X=Z+Z mean (7);

[0092] Where X is the signal center matrix, Z is the estimated pixel matrix, and Z mean is the mean matrix.

[0093] Step S204: performing calculation processing based on the noise center matrix to obtain a noise covariance matrix corresponding to the noise center matrix;

[0094] In the specific implementation process of this step, calculation processing is performed based on the noise center matrix to obtain the noise covariance matrix S corresponding to the noise center matrix N ; Noise covariance matrix S N The calculation formula is as shown in the following formula (8):

[0095]

[0096] Among them, S N is the noise covariance matrix, X N is the noise center matrix.

[0097] Step S205: performing calculation processing based on the signal center matrix to obtain a signal covariance matrix corresponding to the signal center matrix;

[0098] In the specific implementation process of this step, calculation processing is performed based on the signal center matrix X to obtain a signal covariance matrix S corresponding to the signal center matrix X; the calculation formula of the signal covariance matrix S is shown in the following formula (9):

[0099] S=X T X / (n-1) (9);

[0100] Among them, S is the signal covariance matrix and X is the signal center matrix.

[0101] Step S206: Based on the noise covariance matrix and the signal covariance matrix, a preset kernel function is used to perform nonlinear mapping processing to obtain the kernelization matrix;

[0102] In the specific implementation process of this step, the principal component analysis method is used to perform calculation processing based on the noise covariance matrix and the signal covariance matrix to obtain a noise variance matrix corresponding to the noise covariance matrix and a signal variance matrix corresponding to the signal covariance matrix; the noise score matrix is ​​obtained by calculation processing based on the signal variance matrix and the noise variance matrix; the noise score NF is the ratio of the noise variance matrix to the signal variance matrix, and for a linear combination a T X(p), the calculation formula of the noise score matrix can be shown as follows (10):

[0103]

[0104] Where a is the characteristic matrix of NF. T is the transposed matrix of a, and kernelization is performed based on the noise score matrix to obtain the kernelized matrix 1 / NF. Specifically, the kernel minimum noise transform minimizes NF and uses it to sort the new components of image quality. Minimizing NF is equivalent to maximizing 1 / NF, and by setting a∝X T b The mathematical expression of 1 / NF can be shown as follows:

[0105]

[0106] Where X T is the transposed matrix of the signal center matrix X. In the kernel minimum noise transform, the nonlinear mapping used to transform the estimated noise matrix and the estimated pixel matrix into the feature space can be used to kernelize 1 / NF. The mathematical expression for kernelizing 1 / NF is shown in the following formula (12):

[0107] 1 / NF=b T Φ(X)Φ(X) T Φ(X)Φ(X) T b / b TΦ(X)Φ(X N ) T Φ(X N )Φ(X) T b (12);

[0108] The nonlinear mapping function expression is as shown in the following formula (13):

[0109] Φ:X→Φ(X) (13);

[0110] Where Φ(X) is the kernelized X matrix. The mapping function of formula 13 can be a logical function Sigmoid function, a hyperbolic tangent function Tanh function, etc. This application does not limit the mapping function. Let κ = Φ(X)Φ(X) T , κ y =Φ(X)Φ(X N ) T ; Then formula (12) can be expressed by the following formula (14):

[0111]

[0112] The maximum Rayleigh entropy method can be used to solve formula (14). When 1 / NF takes the maximum value, the following formula (15) can be obtained:

[0113]

[0114] Formula (15) can be further transformed into formula (16):

[0115]

[0116] Formula (17) is obtained from formula (16):

[0117]

[0118] in, and λ is The eigenvector and eigenvalue of can be calculated by formula (17) to get the matrix b. The result Y of kernel minimum noise transform feature extraction can be expressed as follows:

[0119] Y=Φ(X)a=Φ(X)Φ(X) T b = κb (18).

[0120] Step S207: extracting features from the kernelization matrix to obtain a feature matrix corresponding to the image to be separated;

[0121] In the specific implementation process of this step, the kernel matrix is decomposed to obtain each sub-kernel matrix; calculation and processing are performed based on each sub-kernel matrix to obtain a similarity matrix corresponding to the kernel matrix. Specifically, for a kernel matrix P with n rows and n columns, its similarity matrix can be expressed by the following formula (19):

[0122]

[0123] Among them, A represents the upper left matrix with m rows and m columns in P, B represents the upper right matrix with m rows and s columns, (m + s = n, m << s), and C represents a matrix with s rows and s columns. C can be expressed by the following formula (20):

[0124] C = B T A -1 B (20);

[0125] Then the similarity matrix Q of formula (19) can be expressed by the following formula (21):

[0126]

[0127] Diagonalization processing is performed on the target sub-kernel matrix in the sub-kernel matrix to obtain a diagonal matrix corresponding to the target sub-kernel matrix. Specifically, the matrix A in the similarity matrix Q is the target sub-kernel matrix, and the mathematical expression for diagonalizing the matrix A is shown by the following formula (22):

[0128] A = UAU T (22);

[0129] Calculation and processing are performed based on the diagonal matrix and the similarity matrix to obtain an initial eigenvector corresponding to the similarity matrix. The calculation mathematical formula of the initial eigenvector can be shown by the following formula (23):

[0130]

[0131] Orthogonalization processing is performed based on the initial eigenvector to obtain an eigenmatrix corresponding to the image to be separated. Specifically, the K matrix is defined as shown by the following formula (24):

[0132]

[0133] The K matrix is orthogonally processed to obtain a K orthogonal matrix. The K orthogonal matrix can be shown by the following formula (25):

[0134]

[0135] Based on the K orthogonal matrix and the initial eigenvector, a calculation process is performed to obtain a feature matrix V corresponding to the image to be separated. The calculation formula of the feature matrix is ​​as shown in the following formula (26):

[0136]

[0137] Step S208: reconstructing the image to be separated based on the feature matrix to obtain the target image matrix;

[0138] In the specific implementation process of this step, formula (18) is used to map the pixel values ​​corresponding to all feature matrices into the transformation result matrix Y and output the feature extraction result of the minimum noise transformation to obtain the target image matrix.

[0139] Step S209: performing calculation processing based on each pixel point in the target image matrix to obtain an average vector corresponding to the target image matrix;

[0140] In the specific implementation process of this step, calculation processing is performed based on each pixel point in the target image matrix to obtain the average vector m corresponding to the target image matrix.

[0141] Step S210: performing calculation processing based on the average vector and the first vectors corresponding to the pixels to obtain an initial similarity between the average vector and the first vectors;

[0142] In the specific implementation process of this step, calculation processing is performed based on the average vector m and the first vector n corresponding to each pixel point to obtain the initial similarity between the average vector and each first vector; first, calculation processing is performed based on the average vector m and the first vector n corresponding to each pixel point to obtain each initial Euclidean distance value e(m, n); the calculation mathematical formula of each initial Euclidean distance value is shown in the following formula (27):

[0143]

[0144] Among them, k is the dimension of the spectral image, m i is the i-th element constituting the average vector, n i To constitute the i-th element of the first vector, the average vector m and the first vector n corresponding to each pixel point are normalized respectively to obtain a first normalized vector corresponding to the average vector and a second normalized vector corresponding to each first vector; the mathematical expression for calculating the normalized vector can be expressed as follows:

[0145]

[0146] Based on the first normalized vector and each of the second normalized vectors, calculation processing is performed to obtain the initial similarity NED(m, n) between the average vector and each of the first vectors. The mathematical expression for calculating the initial similarity can be expressed as the following formula (29):

[0147] NED(m,n)=e(N m , N n ) (29);

[0148] where m and n are k-dimensional spectra, e(m, n) is the Euclidean distance between m and n, and N m and N n are the normalized vectors of m and n, and is the expected value of m and n.

[0149] Step S211: performing calculation based on each of the initial similarities to obtain an initial similarity distance and an initial threshold for starting an iteration round;

[0150] In the specific implementation process of this step, screening is performed based on the initial similarities to obtain the maximum similarity NED max , minimum similarity NED min And the median similarity NED Median ; Based on the maximum similarity and the minimum similarity, a calculation process is performed to obtain an initial similarity distance; the calculation expression of the initial similarity distance is shown in the following formula (30):

[0151] Ramge=NED max -NED min (30);

[0152] Based on the initial similarity distance and the median similarity, a calculation process is performed to obtain an initial threshold value; the calculation expression of the initial threshold value is shown in the following formula (31):

[0153] Threshold=NED Median +a*Ramge (31);

[0154] Among them, NED Median is the median of the NED set, a is an adjustable constant, and the value of a can be set according to actual needs.

[0155] Step S212: performing background separation processing on the target image matrix of the starting iteration round based on each of the initial similarities to obtain an initial background image and an initial target image;

[0156] In the specific implementation process of this step, a judgment is made based on each of the initial similarities and the initial threshold to obtain a judgment result; when the judgment result is that each first target similarity in each of the similarities is greater than the initial threshold, the pixel point corresponding to each of the first target similarities is determined to be the initial background pixel; when the judgment result is that each second target similarity in each of the similarities is less than or equal to the initial threshold, the pixel point corresponding to each of the second target similarities is determined to be the initial target pixel.

[0157] Step S213: when the initial similarity distance at the beginning of the iteration round satisfies the first preset condition, determining the initial target image as the target image and determining the initial background image as the target background image;

[0158] In the specific implementation of this step, the first preset condition may be that when the initial similarity distance is less than or equal to a first preset threshold, the initial target image is determined to be the target image and the initial background image is determined to be the target background image.

[0159] Step S214: When the initial similarity distance at the beginning of the iteration round does not meet the first preset condition, the background separation processing process of the next round is performed for the initial target image, and the process is iterated in a loop until the current similarity distance of the current iteration round and the historical similarity distance of the previous iteration round meet the second preset condition, and the current target image of the current round is determined to be the target image and the current background image of the current round is determined to be the target background image.

[0160] In the specific implementation of this step, the second preset condition may be that the difference between the current similarity distance of the current iteration round and the historical similarity distance of the previous iteration round is less than or equal to the second preset threshold, and the second preset threshold may be set according to actual needs. Figure 3 As shown, it is a schematic diagram of background separation based on the outlier removal method and iterative strategy of the present application. When the initial similarity distance at the beginning of the iteration round does not meet the first preset condition, the background separation process of the next round is performed for the initial target image, and the iteration is repeated until the current similarity distance of the current iteration round and the historical similarity distance of the previous iteration round meet the second preset condition, and the current target image of the current round is determined to be the target image and the current background image of the current round is determined to be the target background image.

[0161] The image separation method of the present application has good adaptability to hyperspectral images under different scenes, different noise levels, different spatial resolutions and different spectral resolutions. The method of the present application can achieve autonomous hyperspectral image target background separation without preprocessing or postprocessing operations. The traditional kernel minimum noise transform has the problems of high computational complexity and low execution efficiency. The introduction of a preset machine learning method based on local linear modeling can significantly reduce the computational complexity of obtaining matrix eigenvalues ​​and improve image separation efficiency.

[0162] Another embodiment of the present application provides an image separation device, such as Figure 4 As shown, including:

[0163] Noise estimation module 1, used to perform noise estimation on the image to be separated to obtain an estimated noise matrix corresponding to the image to be separated;

[0164] A kernelization processing module 2 is used to perform kernelization processing based on the estimated noise matrix and the estimated pixel matrix corresponding to the image to be separated by using a kernel minimum noise transformation method to obtain a kernelization matrix;

[0165] An image reconstruction module 3 is used to reconstruct an image based on the kernelization matrix by using a preset machine learning method based on local linear modeling to obtain a target image matrix;

[0166] The background separation module 4 is used to perform background separation processing on the target image matrix to obtain a target background image and a target image.

[0167] In the specific implementation process, the noise estimation module 1 is specifically used to: for each pixel point in the image to be separated, use a preset window to collect the neighborhood pixels of each pixel point to obtain a number of neighborhood pixel values ​​corresponding to each pixel point; perform calculations and processing on each of the neighborhood pixel values ​​corresponding to the same pixel point to obtain the estimated pixel value corresponding to the same pixel point, so as to obtain an estimated pixel matrix corresponding to the image to be separated; perform calculations and processing based on the initial pixel matrix of the image to be separated and the estimated pixel matrix to obtain an estimated noise matrix corresponding to the image to be separated.

[0168] In the specific implementation process, the kernelization processing module 2 is specifically used for: performing average value calculation processing based on the initial pixel matrix of the image to be separated to obtain a mean matrix corresponding to the image to be separated; based on the mean matrix, performing calculation processing on the estimated noise matrix and the estimated pixel matrix respectively to obtain a noise center matrix corresponding to the estimated noise matrix and a signal center matrix corresponding to the estimated pixel matrix; performing calculation processing based on the noise center matrix to obtain a noise covariance matrix corresponding to the noise center matrix; performing calculation processing based on the signal center matrix to obtain a signal covariance matrix corresponding to the signal center matrix; based on the noise covariance matrix and the signal covariance matrix, a preset kernel function is used to perform nonlinear mapping processing to obtain the kernelization matrix.

[0169] In the specific implementation process, the image reconstruction module 3 is specifically used for: extracting features from the kernelization matrix to obtain a feature matrix corresponding to the image to be separated; and reconstructing the image to be separated based on the feature matrix to obtain the target image matrix.

[0170] In the specific implementation process, the image reconstruction module 3 is also used to: decompose the kernelization matrix to obtain each sub-kernel matrix; perform calculation processing based on each sub-kernel matrix to obtain a similarity matrix corresponding to the kernelization matrix; diagonalize the target sub-kernel matrix in the sub-kernel matrix to obtain a diagonal matrix corresponding to the target sub-kernel matrix; perform calculation processing based on the diagonal matrix and the similarity matrix to obtain an initial eigenvector corresponding to the similarity matrix; perform orthogonalization processing based on the initial eigenvector to obtain a feature matrix corresponding to the image to be separated.

[0171] In the specific implementation process, the background separation module 4 is specifically used to: perform calculation processing based on each pixel point in the target image matrix to obtain an average vector corresponding to the target image matrix; perform calculation processing based on the average vector and the first vectors corresponding to each pixel point to obtain the initial similarity between the average vector and each first vector; perform calculation processing based on each initial similarity to obtain an initial similarity distance and an initial threshold for starting an iteration round; perform background separation processing on the target image matrix at the beginning of the iteration round based on each initial similarity to obtain an initial background image and an initial target image; when the initial similarity distance at the beginning of the iteration round meets the first preset condition, determine the initial target image as the target image and determine the initial background image as the target background image; when the initial similarity distance at the beginning of the iteration round does not meet the first preset condition, perform the background separation processing process of the next round on the initial target image, and iterate cyclically until the current similarity distance of the current iteration round and the historical similarity distance of the previous iteration round meet the second preset condition, determine the current target image of the current round as the target image and determine the current background image of the current round as the target background image.

[0172] In the specific implementation process, the background separation module 4 is also used to: perform calculations based on the average vector and the first vectors corresponding to each of the pixel points to obtain each initial Euclidean distance value; normalize the average vector and each of the first vectors respectively to obtain a first normalized vector corresponding to the average vector and a second normalized vector corresponding to each of the first vectors; perform calculations based on the first normalized vector and each of the second normalized vectors to obtain the initial similarity between the average vector and each of the first vectors.

[0173] The image separation method of the present application has good adaptability to hyperspectral images under different scenes, different noise levels, different spatial resolutions and different spectral resolutions. The method of the present application can achieve autonomous hyperspectral image target background separation without preprocessing or postprocessing operations. The traditional kernel minimum noise transform has the problems of high computational complexity and low execution efficiency. The introduction of a preset machine learning method based on local linear modeling can significantly reduce the computational complexity of obtaining matrix eigenvalues ​​and improve image separation efficiency.

[0174] Another embodiment of the present application provides a storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the following method steps are implemented:

[0175] Step 1: performing noise estimation on the image to be separated to obtain an estimated noise matrix corresponding to the image to be separated;

[0176] Step 2: Based on the estimated noise matrix and the estimated pixel matrix corresponding to the image to be separated, a kernel minimum noise transformation method is used to perform kernelization processing to obtain a kernelized matrix;

[0177] Step 3: Based on the kernelization matrix, a preset machine learning method based on local linear modeling is used to perform image reconstruction to obtain a target image matrix;

[0178] Step 4: performing background separation processing on the target image matrix to obtain a target background image and a target image.

[0179] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0180] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0181] The specific implementation process of the above method steps can refer to the embodiments of any of the above image separation methods, and this embodiment will not be repeated here.

[0182] The image separation method of the present application has good adaptability to hyperspectral images under different scenes, different noise levels, different spatial resolutions and different spectral resolutions. The method of the present application can achieve autonomous hyperspectral image target background separation without preprocessing or postprocessing operations. The traditional kernel minimum noise transform has the problems of high computational complexity and low execution efficiency. The introduction of a preset machine learning method based on local linear modeling can significantly reduce the computational complexity of obtaining matrix eigenvalues ​​and improve image separation efficiency.

[0183] Another embodiment of the present application provides an electronic device, which may be a server, and includes a processor, a memory, a network interface, and a database connected via a system bus. The processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes a non-volatile and / or volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the electronic device is used to communicate with an external client via a network connection. When the electronic device program is executed by the processor, it implements functions or steps on the server side of an image separation method.

[0184] In one embodiment, an electronic device is provided, which may be a client. The electronic device includes a processor, a memory, a network interface, a display screen, and an input device connected via a system bus. The processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the electronic device is used to communicate with an external server via a network connection. When the electronic device program is executed by the processor, the functions or steps on the client side of an image separation method are implemented.

[0185] Another embodiment of the present application provides an electronic device, comprising at least a memory and a processor, wherein the memory stores a computer program, and the processor implements the following method steps when executing the computer program in the memory:

[0186] Step 1: performing noise estimation on the image to be separated to obtain an estimated noise matrix corresponding to the image to be separated;

[0187] Step 2: Based on the estimated noise matrix and the estimated pixel matrix corresponding to the image to be separated, a kernel minimum noise transformation method is used to perform kernelization processing to obtain a kernelized matrix;

[0188] Step 3: Based on the kernelization matrix, a preset machine learning method based on local linear modeling is used to perform image reconstruction to obtain a target image matrix;

[0189] Step 4: performing background separation processing on the target image matrix to obtain a target background image and a target image.

[0190] The specific implementation process of the above method steps can be found in the above-mentioned embodiment of the image separation method, and this embodiment will not be repeated here.

[0191] The image separation method of the present application has good adaptability to hyperspectral images under different scenes, different noise levels, different spatial resolutions and different spectral resolutions. The method of the present application can achieve autonomous hyperspectral image target background separation without preprocessing or postprocessing operations. The traditional kernel minimum noise transform has the problems of high computational complexity and low execution efficiency. The introduction of a preset machine learning method based on local linear modeling can significantly reduce the computational complexity of obtaining matrix eigenvalues ​​and improve image separation efficiency.

[0192] The above embodiments are only exemplary embodiments of the present application and are not intended to limit the present application. The protection scope of the present application is defined by the claims. Those skilled in the art may make various modifications or equivalent substitutions to the present application within the essence and protection scope of the present application, and such modifications or equivalent substitutions shall also be deemed to fall within the protection scope of the present application.

Claims

1. An image separation method, characterized in that: include: Perform noise estimation on the image to be separated to obtain an estimated noise matrix corresponding to the image to be separated; Based on the estimated noise matrix and the estimated pixel matrix corresponding to the image to be separated, a kernel minimum noise transformation method is used to perform kernelization processing to obtain a kernelized matrix; Based on the kernelization matrix, a preset machine learning method based on local linear modeling is used to perform image reconstruction to obtain a target image matrix; Performing background separation processing on the target image matrix to obtain a target background image and a target image; Based on the kernelization matrix, a preset machine learning method based on local linear modeling is used to perform image reconstruction to obtain a target image matrix, which specifically includes: Performing feature extraction on the kernelization matrix to obtain a feature matrix corresponding to the image to be separated; Reconstructing the image to be separated based on the feature matrix to obtain the target image matrix; The performing background separation processing on the target image matrix to obtain the target background image and the target image specifically includes: Performing calculation processing based on each pixel point in the target image matrix to obtain an average vector corresponding to the target image matrix; Performing calculation based on the average vector and the first vectors corresponding to the pixels to obtain an initial similarity between the average vector and the first vectors; Calculate and process based on each of the initial similarities to obtain an initial similarity distance and an initial threshold for starting an iteration round; Based on each of the initial similarities, background separation processing is performed on the target image matrix of the starting iteration round to obtain an initial background image and an initial target image; When the initial similarity distance at the beginning of the iteration round satisfies a first preset condition, determining the initial target image as the target image and determining the initial background image as the target background image; When the initial similarity distance at the beginning of the iteration round does not meet the first preset condition, the background separation processing process of the next round is performed on the initial target image, and the iteration is repeated until the current similarity distance of the current iteration round and the historical similarity distance of the previous iteration round meet the second preset condition, and the current target image of the current round is determined to be the target image and the current background image of the current round is determined to be the target background image.

2. The method according to claim 1, characterized in that The performing noise estimation on the image to be separated to obtain an estimated noise matrix corresponding to the image to be separated specifically includes: For each pixel point in the image to be separated, a preset window is used to collect the neighborhood pixels of each pixel point to obtain a number of neighborhood pixel values ​​corresponding to each pixel point; Calculating and processing each of the neighborhood pixel values ​​corresponding to the same pixel point to obtain an estimated pixel value corresponding to the same pixel point, so as to obtain an estimated pixel matrix corresponding to the image to be separated; Calculation processing is performed based on the initial pixel matrix of the image to be separated and the estimated pixel matrix to obtain an estimated noise matrix corresponding to the image to be separated.

3. The method according to claim 1, characterized in that The step of performing kernelization processing based on the estimated noise matrix and the estimated pixel matrix corresponding to the image to be separated by using a kernel minimum noise transformation method to obtain a kernelized matrix specifically includes: Performing average value calculation processing based on the initial pixel matrix of the image to be separated to obtain a mean value matrix corresponding to the image to be separated; Based on the mean matrix, respectively performing calculation processing on the estimated noise matrix and the estimated pixel matrix to obtain a noise center matrix corresponding to the estimated noise matrix and a signal center matrix corresponding to the estimated pixel matrix; Performing calculation processing based on the noise center matrix to obtain a noise covariance matrix corresponding to the noise center matrix; Performing calculation processing based on the signal center matrix to obtain a signal covariance matrix corresponding to the signal center matrix; Based on the noise covariance matrix and the signal covariance matrix, a preset kernel function is used to perform nonlinear mapping processing to obtain the kernelization matrix.

4. The method according to claim 1, characterized in that The extracting features from the kernelization matrix to obtain a feature matrix corresponding to the image to be separated specifically includes: Decomposing the kernelized matrix to obtain sub-kernel matrices; Performing calculation processing based on each of the sub-kernel matrices to obtain a similarity matrix corresponding to the kernelization matrix; Performing diagonal processing on a target sub-core matrix in the sub-core matrix to obtain a diagonal matrix corresponding to the target sub-core matrix; Performing calculations based on the diagonal matrix and the similarity matrix to obtain an initial eigenvector corresponding to the similarity matrix; Orthogonalization is performed based on the initial eigenvector to obtain a feature matrix corresponding to the image to be separated.

5. The method according to claim 4, characterized in that The performing calculation based on the average vector and the first vectors corresponding to the pixels to obtain the initial similarity between the average vector and the first vectors specifically includes: Performing calculation based on the average vector and the first vector corresponding to each pixel point to obtain each initial Euclidean distance value; Normalizing the average vector and each of the first vectors respectively to obtain a first normalized vector corresponding to the average vector and a second normalized vector corresponding to each of the first vectors; Calculation processing is performed based on the first normalized vector and each of the second normalized vectors to obtain the initial similarity between the average vector and each of the first vectors.

6. An image separation device, characterized in that: include: A noise estimation module, used to perform noise estimation on the image to be separated to obtain an estimated noise matrix corresponding to the image to be separated; A kernelization processing module, configured to perform kernelization processing using a kernel minimum noise transformation method based on the estimated noise matrix and an estimated pixel matrix corresponding to the image to be separated, so as to obtain a kernelization matrix; An image reconstruction module is used to reconstruct an image based on the kernelization matrix using a preset machine learning method based on local linear modeling to obtain a target image matrix; specifically, the image reconstruction based on the kernelization matrix using a preset machine learning method based on local linear modeling to obtain a target image matrix includes: extracting features from the kernelization matrix to obtain a feature matrix corresponding to the image to be separated; reconstructing the image to be separated based on the feature matrix to obtain the target image matrix; The background separation module is used to perform background separation processing on the target image matrix to obtain a target background image and a target image; specifically, it is used to perform calculation processing based on each pixel point in the target image matrix to obtain an average vector corresponding to the target image matrix; perform calculation processing based on the average vector and the first vectors corresponding to each pixel point to obtain an initial similarity between the average vector and each first vector; perform calculation processing based on each initial similarity to obtain an initial similarity distance and an initial threshold value for starting an iteration round; perform background separation processing on the target image matrix for starting an iteration round based on each initial similarity to obtain an initial an initial background image and an initial target image; when the initial similarity distance at the start of an iteration round satisfies a first preset condition, determining the initial target image as the target image and determining the initial background image as the target background image; when the initial similarity distance at the start of an iteration round does not satisfy the first preset condition, performing the background separation process of the next round on the initial target image, iterating in a loop until the current similarity distance of the current iteration round and the historical similarity distance of the previous iteration round satisfy a second preset condition, determining the current target image of the current round as the target image and determining the current background image of the current round as the target background image.

7. A storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the image separation method according to any one of claims 1 to 5 are implemented.

8. An electronic device, characterized in that: The method comprises at least a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the image separation method according to any one of claims 1 to 5 when executing the computer program on the memory.

Images