CN111563544B - Maximum signal-to-noise ratio hyperspectral data dimension reduction method for multi-scale superpixel segmentation - Google Patents
Maximum signal-to-noise ratio hyperspectral data dimension reduction method for multi-scale superpixel segmentation Download PDFInfo
- Publication number
- CN111563544B CN111563544B CN202010343479.4A CN202010343479A CN111563544B CN 111563544 B CN111563544 B CN 111563544B CN 202010343479 A CN202010343479 A CN 202010343479A CN 111563544 B CN111563544 B CN 111563544B
- Authority
- CN
- China
- Prior art keywords
- pixel
- low
- super
- dimension
- dimensional
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/30—Noise filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A40/00—Adaptation technologies in agriculture, forestry, livestock or agroalimentary production
- Y02A40/10—Adaptation technologies in agriculture, forestry, livestock or agroalimentary production in agriculture
Abstract
The invention discloses a multi-scale super-pixel segmentation maximum signal-to-noise ratio hyperspectral data dimension reduction method, which comprises the following steps: preprocessing original hyperspectral image data by adopting principal component analysis, extracting principal components, and generating a false color map; based on the set super-pixel segmentation number, a simple iterative linear clustering method is adopted to segment the false color map, so that a plurality of super-pixel blocks are obtained; reducing the dimension of each super pixel by adopting a maximum signal-to-noise ratio method to obtain low-dimension characteristics of a low-dimension subspace; arranging the low-dimensional features according to the original pixel positions to obtain low-dimensional feature representation of the original hyperspectral image data; classifying the low-dimensional characteristic representation by adopting a support vector machine to obtain a prediction tag; setting super-pixel segmentation numbers of different scales, and repeating the steps to obtain a plurality of groups of prediction labels; and determining the final prediction label of each pixel from the multiple groups of prediction labels by adopting a maximum voting method.
Description
Technical Field
The invention relates to the field of remote sensing image processing, in particular to a multi-scale super-pixel segmentation maximum signal-to-noise ratio hyperspectral data dimension reduction method.
Background
Hyperspectral data plays an important role in a plurality of remote sensing applications such as agriculture, mineralogy, monitoring, astronomy, environmental science and the like.
The hyperspectral data size is large, the information of the contained ground objects is rich, but the correlation among spectrums is strong, the information is redundant, in the practical application, all data are directly utilized, the time consumption of data processing can be increased, and pixel misclassification is easily caused. Therefore, dimension reduction of hyperspectral data is an important link in hyperspectral image processing.
The maximum signal-to-noise ratio is a linear transformation that is weighted by the maximized signal-to-noise ratio. Noise is separated in the process of realizing data dimension reduction. Essentially two principal component transformations are performed. Firstly, filtering the whole image or image data with the same property by utilizing high-pass filtering to obtain a noise covariance matrix; then, performing principal component analysis on the covariance matrix of the original image; and finally, combining the noise covariance matrix obtained in the first step to complete the whole dimension reduction process.
However, the above-mentioned hyperspectral data dimension reduction method adopts unified projection transformation for different ground object areas, and ignores the spatial correlation of hyperspectral data.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides a multi-scale super-pixel segmentation maximum signal-to-noise ratio hyperspectral data dimension reduction method.
The invention provides a multi-scale super-pixel segmentation maximum signal-to-noise ratio hyperspectral data dimension reduction method, which comprises the following steps:
step 1) preprocessing original hyperspectral image data by adopting principal component analysis, extracting principal components, and generating a false color map;
step 2) based on the set super-pixel segmentation number, segmenting the false color map by adopting a simple iterative linear clustering method to obtain a plurality of super-pixel blocks;
step 3) reducing the dimension of each super pixel by adopting a maximum signal-to-noise ratio method to obtain low-dimension characteristics of a low-dimension subspace;
step 4) arranging the low-dimensional features according to the original pixel positions to obtain low-dimensional feature representation of the original hyperspectral image data;
step 5) classifying the low-dimensional feature representation by adopting a support vector machine to obtain a prediction tag;
step 6) setting super-pixel segmentation numbers of different scales, and repeating the steps 2) -5) to obtain a plurality of groups of prediction labels;
and 7) determining the final prediction label of each pixel from the multiple groups of prediction labels by adopting a maximum voting method.
As an improvement of the above method, the step 1) specifically includes:
normalizing each spectrum channel characteristic X of all pixels of original hyperspectral image data to obtain normalized characteristic Y of each pixel;
calculating covariance matrix cov (Y, Y T )：
Wherein n is the number of samples;
covariance matrix cov (Y, Y T ) Is described as a feature value lambda and a feature vector v;
for the characteristic values lambda arranged from large to small, selecting the largest 3 characteristic values, namely lambda 1 ,λ 2 ,λ 3 The method comprises the steps of carrying out a first treatment on the surface of the Lambda is set to 1 ,λ 2 ,λ 3 Corresponding feature vector v 1 ,ν 2 ,ν 3 Respectively forming a feature vector matrix P as row vectors;
converting the normalized feature Y into a low-dimensional space through a feature vector matrix P to obtain a low-dimensional feature representation Z,
Z＝PY
thereby obtaining a false color map.
As an improvement of the above method, the step 2) specifically includes:
step 201) setting the super pixel segmentation number as K, uniformly taking K pixels as an initialized cluster center for the false color image, and then, the distance between adjacent cluster centers is as follows:
wherein N is the pixel value of the false color image;
step 202), re-optimizing the cluster center in a 3×3 neighborhood with the initial cluster center as the center, and limiting the pixel search range to 2s×2s for searching;
for each searched pixel point I i Respectively calculating the pixel point and the clustering center I 0 Distance d between colors c And a spatial distance d s ，
Wherein, (l) 0 ,a 0 ,b 0 ),(l i ,a i ,b i ) Respectively represent pixel points I 0 And I i Corresponding trichromatic values;
wherein, (x) 0 ,y 0 ),(x i ,y i ) Respectively represent pixel points I 0 And I i Corresponding spatial coordinates in the false color image;
the measurement distance D between the pixel point and the clustering center is calculated by the following formula:
wherein m is a fixed constant;
step 203), searching cluster centers in a range of 2S multiplied by 2S by taking each pixel as a center, calculating the measurement distance between the pixel and each cluster center, and taking the cluster center with the shortest distance as the cluster center of the pixel;
repeating the steps, and continuously iterating until the clustering center of each pixel is not changed any more, so that K super pixels are finally obtained;
step 204), a marking table is established, initial values are set to be-1, discontinuous superpixels are distributed to adjacent superpixels again according to the Z-shaped trend, superpixels with the size smaller than a certain threshold are distributed to corresponding labels after traversing, and a plurality of superpixel blocks are obtained after traversing all the superpixel points.
As an improvement of the above method, the step 3) specifically includes:
selecting superpixel data Z k ∈R L×N Where k=1, …, K represents all super-pixel numbers, L represents the pixel feature dimension, and N represents the pixel number;
solving the following formula according to the maximum noise ratio algorithm to obtain a projection matrix W k So that the signal S is uncorrelated k And noise N k The ratio between them is the largest and,
wherein Cov (S k )，Cov(N k ) Representing uncorrelated signal S k And noise N k Variance of (d), data Z k Variance Cov (Z) k ) The method comprises the following steps:
Cov(Z k )＝Cov(S k )+Cov(N k )；
projection matrix W k Is formed by Cov (N) k ) -1 Cov(Z k ) Is composed of feature vectors corresponding to the L largest feature values; data Z k In the low-dimensional subspace can be expressed as:
thereby obtaining the low-dimensional subspace low-dimensional characteristic H k ,k＝1,…,K。
As an improvement of the above method, the step 4) specifically includes: arranging low-dimensional features H according to the original hyperspectral image positions k K=1, …, K, resulting in a low dimensional feature h= { H of the original hyperspectral image 1 ,…,H K }。
The invention also provides a maximum signal-to-noise ratio hyperspectral data dimension reduction system for multi-scale super-pixel segmentation, which comprises the following components: the device comprises a preprocessing module, a dimension reduction processing module and a voting module; wherein, the liquid crystal display device comprises a liquid crystal display device,
the preprocessing module is used for preprocessing the original hyperspectral image data by adopting principal component analysis, extracting principal components and generating a false color map;
the dimension reduction processing module is used for dividing the false color map by adopting a simple iterative linear clustering method based on the set super-pixel division number to obtain a plurality of super-pixel blocks; reducing the dimension of each super pixel by adopting a maximum signal-to-noise ratio method to obtain low-dimension characteristics of a low-dimension subspace; arranging the low-dimensional features according to the original pixel positions to obtain low-dimensional feature representation of the original hyperspectral image data; classifying the low-dimensional characteristic representation by adopting a support vector machine to obtain a prediction tag; repeatedly processing according to the set super-pixel segmentation numbers with different scales to obtain a plurality of groups of prediction labels;
and the voting module is used for determining the final prediction label of each pixel from a plurality of groups of prediction labels by adopting a maximum voting method.
A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor executing the computer program to perform the method described above.
A computer readable storage medium storing a computer program which, when executed by a processor, causes the processor to perform the method described above.
Compared with the prior art, the invention has the advantages that:
the invention considers the diversity of different ground object areas, adopts different projection transformation, and reduces the dimension of hyperspectral data.
Drawings
FIG. 1 is a flow chart of a hyperspectral data dimension reduction process based on the maximum signal to noise ratio of multi-scale superpixel segmentation;
FIG. 2 is a hyperspectral image of example 1 of the present invention;
FIG. 3 is a pseudo-color plot of hyperspectral data of example 1 of the present invention;
FIG. 4 is a graph showing the effect of super-pixel segmentation in accordance with embodiment 1 of the present invention;
fig. 5 is a low-dimensional characteristic diagram of hyperspectral data of example 1 of the present invention.
Detailed Description
The invention aims at solving the problem that the dimension reduction effect is poor due to the fact that the unified projection transformation is adopted for the maximum signal-to-noise ratio method and is directly applied to dimension reduction of hyperspectral data, and provides a maximum signal-to-noise ratio hyperspectral data dimension reduction method based on multi-scale super-pixel segmentation.
A maximum signal-to-noise ratio hyperspectral data dimension reduction method based on multi-scale super-pixel segmentation comprises the following steps:
step 1: preprocessing original hyperspectral data by adopting principal component analysis, extracting 3 principal components, and generating a false color map;
normalizing each spectrum channel characteristic X of all pixels of the hyperspectral data to obtain a normalized characteristic Y of each pixel;
calculating covariance matrix cov (Y, Y T ) The method comprises the following steps:
wherein n is the number of samples;
covariance matrix cov (Y, Y T ) Is described as a feature value lambda and a feature vector v;
for the characteristic values from large to small, selecting the maximum 3 lambda 1 ,λ 2 ,λ 3 Then corresponding 3 eigenvectors v 1 ,ν 2 ,ν 3 Respectively forming a feature vector matrix P as row vectors;
converting the normalized feature Y into a low-dimensional space through a feature vector matrix P to obtain a low-dimensional feature representation Z:
Z＝PY。
step 2: setting the number of super pixels, and dividing the false color map by adopting a simple iterative linear clustering method to obtain a plurality of super pixel blocks.
Super-pixel segmentation is performed on the false color image:
initializing a clustering center: assuming that the number of super-pixel divisions is K and the number of pseudo-color image pixels to be divided is N, each super-pixel contains N/K pixels, and the adjacent cluster center distance is about
Optimizing a clustering center: re-optimizing the clustering center in a 3×3 neighborhood with the clustering center as the center, limiting the pixel searching range to 2s×2s, and converging an acceleration algorithm;
calculating the distance: for each searched pixel point I i Respectively calculate it and cluster center I 0 The distance between them; the distance comprises a color distance and a space distance, and the distance calculation method comprises the following steps:
wherein d c Represents the color distance, d s Represents the spatial distance, (l) 0 ,a 0 ,b 0 ),(l i ,a i ,b i ) Respectively represent pixel points I 0 And I i Corresponding trichromatic values, (x) 0 ,y 0 ),(x i ,y i ) Respectively represent pixel points I 0 And I i Corresponding spatial coordinates in the whole image, N s Is the maximum spatial distance within a class, defined asIs applicable to each cluster. Maximum color distance N c Depending on the image and the cluster, a fixed one is usually takenConstant m (value range [1, 40)]Typically 10) is taken instead. The mixing metric D' is as follows:
the final distance measure D is as follows:
searching a clustering center in a range by taking each pixel as a center, calculating the measurement distance between the pixel and each clustering center, and taking the clustering center with the shortest distance as the clustering center of the pixel;
iterative optimization: the cluster center of each pixel is not changed any more, and K super pixels are finally obtained; error convergence, thereby determining a plurality of cluster centers and obtaining a plurality of super pixels;
enhancing connectivity: and (3) creating a marking table, wherein elements in the table are-1, discontinuous super pixels are reassigned to adjacent super pixels according to the Z-shaped trend (from left to right and from top to bottom), and traversed pixel points are assigned to corresponding labels until all points are traversed.
Step 3: and reducing the dimension by adopting a maximum signal-to-noise ratio method one by one to obtain the low-dimensional subspace low-dimensional characteristics.
The maximum signal-to-noise ratio is adopted to reduce the dimension by super pixel:
selecting superpixel data Z k ∈R L×N Where k=1, …, K represents all super-pixel numbers, L represents the pixel feature dimension, and N represents the number of pixels. Hypothesis data Z k Can be derived from uncorrelated signal S k And noise N k Superposition, namely:
Z k ＝S k +N k
maximum noise ratio algorithm for finding a projection matrix W k So that there is no correlation signal S k And noise N k The ratio between them is the largest. According to the nature of the varianceData Z k Variance Cov (Z) k ) The method comprises the following steps:
Cov(Z k )＝Cov(S k )+Cov(N k )
wherein Cov (S k )，Cov(N k ) Representing uncorrelated signal S k And noise N k Is a variance of (c).
Projection matrix W k The method can be obtained by solving the following problems:
projection matrix W k Is formed by Cov (N) k ) -1 Cov(Z k ) The feature vectors corresponding to the L largest feature values. Data Z k In the low-dimensional subspace can be expressed as:
step 4: according to the original pixel position, arranging the low-dimensional characteristics to obtain the low-dimensional characteristic representation of the whole hyperspectral image; arranging the low-dimensional features H according to the original image position k K=1, …, K, resulting in a low dimensional characteristic h= { H of the entire hyperspectral image 1 ,…,H K }。
Step 5: and classifying the low-dimensional features by adopting a support vector machine to obtain the predictive label.
Step 6: setting the number of super pixels with different scales, repeating the step 2-5 to obtain a plurality of groups of prediction labels, setting the number of super pixels with different scales of 2n+1 groups, and repeating the step 2-5 to obtain the 2n+1 groups of prediction labels of the pixels.
Step 7: and determining the final prediction label of each pixel by using a maximum voting method.
The technical scheme of the invention is described in detail below with reference to the accompanying drawings and examples.
Example 1
FIG. 1 is a flow chart of a method for dimension reduction of maximum signal-to-noise ratio hyperspectral data based on multi-scale superpixel segmentation.
The programming software used was Matalab R2015a, and a piece of hyperspectral image data with a size of 610×340×103 shown in fig. 2 was selected as the original image data. The method for reducing the dimension of the hyperspectral data based on the maximum signal to noise ratio of multi-scale super-pixel segmentation is adopted, and the specific process for reducing the dimension of the original hyperspectral data is as follows:
and step 1, preprocessing original hyperspectral data by adopting a principal component analysis method, extracting the first 3 principal components, generating a false color image, normalizing each spectral channel characteristic X of all pixels of the hyperspectral data, and obtaining normalized characteristics Y of each pixel.
Calculating covariance matrix cov (Y, Y T ) The method comprises the following steps:
covariance matrix cov (Y, Y T ) Is described as a feature value lambda and a feature vector v;
for the characteristic values from large to small, selecting the maximum 3 lambda 1 ,λ 2 ,λ 3 Then corresponding 3 eigenvectors v 1 ,ν 2 ,ν 3 Respectively forming a feature vector matrix P as row vectors;
the data is transformed into the projection space of 3 eigenvectors resulting in a low dimensional eigenvalue Z, i.e. z=py, as shown in fig. 3.
And 2, setting the super-pixel segmentation number as 100, and performing super-pixel segmentation on the false color map by adopting a simple linear clustering algorithm.
Initializing a clustering center: assuming that the number of super pixels is K and the number of the pseudo-color image pixels to be segmented is N, each super pixel comprises N/K pixels, and the distance between adjacent cluster centers is about
Optimizing a clustering center: the cluster centers are re-optimized within a 3 x 3 neighborhood centered around the cluster center. The pixel searching range is limited to 2S multiplied by 2S, and the acceleration algorithm converges;
calculating the distance: for each searched pixel point I i Respectively calculate it and cluster center I 0 Distance between them. The distance includes a color distance and a spatial distance. The distance calculation method comprises the following steps:
wherein d c Represents the color distance, d s Represents the spatial distance, (l) 0 ,a 0 ,b 0 ),(l i ,a i ,b i ) Respectively represent pixel points I 0 And I i Corresponding trichromatic values, (x) 0 ,y 0 ),(x i ,y i ) Respectively represent pixel points I 0 And I i Corresponding spatial coordinates in the whole image, N s Is the maximum spatial distance within a class, defined asIs applicable to each cluster. Maximum color distance N c Depending on the image and the cluster, a constant m (value range 1,40]Typically 10) is taken instead. The final distance measure D' is as follows:
searching cluster centers in a 2S multiplied by 2S range by taking each pixel as a center, calculating the measurement distance between the pixel and each cluster center, and taking the cluster center with the shortest distance as the cluster center of the pixel;
iterative optimization: the steps are iterated continuously until the error converges;
enhancing connectivity: and (3) creating a marking table, wherein elements in the table are-1, discontinuous super pixels are reassigned to adjacent super pixels according to the Z-shaped trend (from left to right and from top to bottom), and traversed pixel points are assigned to corresponding labels until all points are traversed. The segmentation effect is shown in fig. 4.
And 3, setting a dimension reduction parameter as 30 by taking the super-pixel segmentation diagram as a guide, and adopting a maximum signal-to-noise ratio method to reduce the dimension of the hyperspectral data by super-pixel.
And adopting the maximum signal-to-noise ratio to reduce the dimension by super-pixel.
Selecting superpixel data Z k ∈R L×N Where k=1, …, K represents all super-pixel numbers, L represents the pixel feature dimension, and N represents the number of pixels. Hypothesis data Z k Can be derived from uncorrelated signal S k And noise N k Superposition, namely:
Z k ＝S k +N k
maximum noise ratio algorithm for finding a projection matrix W k So that there is no correlation signal S k And noise N k The ratio between them is the largest. Data Z according to the nature of the variance k Is the variance of:
Cov(Z k )＝Cov(S k )+Cov(N k )
wherein Cov (S k )，Cov(N k ) Representing uncorrelated signal S k And noise N k Is a variance of (c).
Projection matrix W k The method can be obtained by solving the following problems:
projection matrix W k Is formed by Cov (N) k ) -1 Cov(Z k ) The feature vectors corresponding to the L largest feature values. Data Z k In the low-dimensional subspace can be expressed as:
step 4, arranging the low-dimensional features H according to the original image positions k K=1, …, K, resulting in a low dimensional characteristic h= { H of the entire hyperspectral image 1 ,…,H K }. As shown in fig. 5.
And 5, selecting 30% of pixel low-dimensional characteristics as training samples in the experiment, using the rest pixels as test samples, and classifying by using a support vector machine.
Step 6, setting the multi-scale super-pixel segmentation number asThe above steps 2-5 were repeated for nine scales to the power of { -4, -3, …,4} to obtain classification results, and experimental data are shown in Table 1.
Setting the number of the 2n+1 groups of super pixels with different scales, and repeating the steps 2-5 to obtain 2n+1 groups of predictive labels of the pixels.
And 7, determining the final prediction label of each pixel by using a maximum voting method for nine groups of prediction results with different scales.
The label experimental data are shown in the following table:
table 1: comparison of experimental results
Example 2
Embodiment 2 of the present invention provides a system for reducing the maximum signal-to-noise ratio hyperspectral data of multi-scale superpixel segmentation, comprising: the device comprises a preprocessing module, a dimension reduction processing module and a voting module; wherein, the liquid crystal display device comprises a liquid crystal display device,
the preprocessing module is used for preprocessing the original hyperspectral image data by adopting principal component analysis, extracting principal components and generating a false color map;
the dimension reduction processing module is used for dividing the false color map by adopting a simple iterative linear clustering method based on the set super-pixel division number to obtain a plurality of super-pixel blocks; reducing the dimension of each super pixel by adopting a maximum signal-to-noise ratio method to obtain low-dimension characteristics of a low-dimension subspace; arranging the low-dimensional features according to the original pixel positions to obtain low-dimensional feature representation of the original hyperspectral image data; classifying the low-dimensional characteristic representation by adopting a support vector machine to obtain a prediction tag; repeatedly processing according to the set super-pixel segmentation numbers with different scales to obtain a plurality of groups of prediction labels;
and the voting module is used for determining the final prediction label of each pixel from the multiple groups of prediction labels by adopting a maximum voting method.
Example 3
A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the method of embodiment 1 when executing the computer program.
Example 4
A computer readable storage medium storing a computer program which, when executed by a processor, causes the processor to perform the method of embodiment 1.
Finally, it should be noted that the above embodiments are only for illustrating the technical solution of the present invention and are not limiting. Although the present invention has been described in detail with reference to the embodiments, it should be understood by those skilled in the art that modifications and equivalents may be made thereto without departing from the spirit and scope of the present invention, which is intended to be covered by the appended claims.
Claims (6)
1. A method for dimension reduction of maximum signal-to-noise ratio hyperspectral data for multi-scale superpixel segmentation, the method comprising:
step 1) preprocessing original hyperspectral image data by adopting principal component analysis, extracting principal components, and generating a false color map;
step 2) based on the set super-pixel segmentation number, segmenting the false color map by adopting a simple iterative linear clustering method to obtain a plurality of super-pixel blocks;
step 3) reducing the dimension of each super pixel by adopting a maximum signal-to-noise ratio method to obtain low-dimension characteristics of a low-dimension subspace;
step 4) arranging the low-dimensional features according to the original pixel positions to obtain low-dimensional feature representation of the original hyperspectral image data;
step 5) classifying the low-dimensional feature representation by adopting a support vector machine to obtain a prediction tag;
step 6) setting super-pixel segmentation numbers of different scales, and repeating the steps 2) -5) to obtain a plurality of groups of prediction labels;
step 7) determining a final prediction label of each pixel from a plurality of groups of prediction labels by adopting a maximum voting method;
the step 1) specifically comprises the following steps:
normalizing each spectrum channel characteristic X of all pixels of original hyperspectral image data to obtain normalized characteristic Y of each pixel;
calculating covariance matrix cov (Y, Y T )：
Wherein n is the number of samples;
covariance matrix cov (Y, Y T ) Is described as a feature value lambda and a feature vector v;
for the characteristic values lambda arranged from large to small, selecting the largest 3 characteristic values, namely lambda 1 ,λ 2 ,λ 3 The method comprises the steps of carrying out a first treatment on the surface of the Lambda is set to 1 ,λ 2 ,λ 3 Corresponding feature vector v 1 ,ν 2 ,ν 3 Respectively forming a feature vector matrix P as row vectors;
converting the normalized feature Y into a low-dimensional space through a feature vector matrix P to obtain a low-dimensional feature representation Z,
Z＝PY
thereby obtaining a false color map;
the step 4) is specifically as follows: arranging low-dimensional features H according to the original hyperspectral image positions k K=1, …, K, resulting in a low dimensional feature h= { H of the original hyperspectral image 1 ,…,H K Where K represents all the super-pixel numbers.
2. The method for dimension reduction of maximum signal-to-noise ratio hyperspectral data for multi-scale superpixel segmentation according to claim 1, wherein said step 2) specifically comprises:
step 201) setting the super pixel segmentation number as K, uniformly taking K pixels as an initialized cluster center for the false color image, and then, the distance between adjacent cluster centers is as follows:
wherein N is the pixel value of the false color image;
step 202), re-optimizing the cluster center in a 3×3 neighborhood with the initial cluster center as the center, and limiting the pixel search range to 2s×2s for searching;
for each searched pixel point I i Respectively calculating the pixel point and the clustering center I 0 Distance d between colors c And a spatial distance d s ，
Wherein, (l) 0 ,a 0 ,b 0 ),(l i ,a i ,b i ) Dividing intoRespectively represent pixel point I 0 And I i Corresponding trichromatic values;
wherein, (x) 0 ,y 0 ),(x i ,y i ) Respectively represent pixel points I 0 And I i Corresponding spatial coordinates in the false color image;
the measurement distance D between the pixel point and the clustering center is calculated by the following formula:
wherein m is a fixed constant;
step 203), searching cluster centers in a range of 2S multiplied by 2S by taking each pixel as a center, calculating the measurement distance between the pixel and each cluster center, and taking the cluster center with the shortest distance as the cluster center of the pixel;
repeating the steps, and continuously iterating until the clustering center of each pixel is not changed any more, so that K super pixels are finally obtained;
step 204), a marking table is established, initial values are set to be-1, discontinuous superpixels are distributed to adjacent superpixels again according to the Z-shaped trend, superpixels with the size smaller than a certain threshold are distributed to corresponding labels after traversing, and a plurality of superpixel blocks are obtained after traversing all the superpixel points.
3. The method for dimension reduction of maximum signal-to-noise ratio hyperspectral data for multi-scale super-pixel segmentation as set forth in claim 2, wherein said step 3) specifically includes:
selecting superpixel data Z k ∈R L×N Where k=1, …, K represents all super-pixel numbers, L represents the pixel feature dimension, and N represents the pixel number;
according to the maximum noise ratio algorithmSolving to obtain a projection matrix W k So that the signal S is uncorrelated k And noise N k The ratio between them is the largest and,
wherein Cov (S k )，Cov(N k ) Representing uncorrelated signal S k And noise N k Variance of (d), data Z k Variance Cov (Z) k ) The method comprises the following steps:
Cov(Z k )＝Cov(S k )+Cov(N k )；
projection matrix W k Is formed by Cov (N) k ) -1 Cov(Z k ) Is composed of feature vectors corresponding to the L largest feature values; data Z k In the low-dimensional subspace can be expressed as:
thereby obtaining the low-dimensional subspace low-dimensional characteristic H k ,k＝1,…,K。
4. A multi-scale super-pixel segmented maximum signal-to-noise ratio hyperspectral data dimension reduction system, the system comprising: the device comprises a preprocessing module, a dimension reduction processing module and a voting module; wherein, the liquid crystal display device comprises a liquid crystal display device,
the preprocessing module is used for preprocessing the original hyperspectral image data by adopting principal component analysis, extracting principal components and generating a false color map;
the dimension reduction processing module is used for dividing the false color map by adopting a simple iterative linear clustering method based on the set super-pixel division number to obtain a plurality of super-pixel blocks; reducing the dimension of each super pixel by adopting a maximum signal-to-noise ratio method to obtain low-dimension characteristics of a low-dimension subspace; arranging the low-dimensional features according to the original pixel positions to obtain low-dimensional feature representation of the original hyperspectral image data; classifying the low-dimensional characteristic representation by adopting a support vector machine to obtain a prediction tag; repeatedly processing according to the set super-pixel segmentation numbers with different scales to obtain a plurality of groups of prediction labels;
the voting module is used for determining a final prediction label of each pixel from a plurality of groups of prediction labels by adopting a maximum voting method;
the pretreatment module specifically comprises the following treatment processes:
normalizing each spectrum channel characteristic X of all pixels of original hyperspectral image data to obtain normalized characteristic Y of each pixel;
calculating covariance matrix cov (Y, Y T )：
Wherein n is the number of samples;
covariance matrix cov (Y, Y T ) Is described as a feature value lambda and a feature vector v;
for the characteristic values lambda arranged from large to small, selecting the largest 3 characteristic values, namely lambda 1 ,λ 2 ,λ 3 The method comprises the steps of carrying out a first treatment on the surface of the Lambda is set to 1 ,λ 2 ,λ 3 Corresponding feature vector v 1 ,ν 2 ,ν 3 Respectively forming a feature vector matrix P as row vectors;
converting the normalized feature Y into a low-dimensional space through a feature vector matrix P to obtain a low-dimensional feature representation Z,
Z＝PY
thereby obtaining a false color map;
the dimension reduction processing module is used for arranging the low-dimension features according to the original pixel positions to obtain low-dimension feature representation of the original hyperspectral image data; the method comprises the following steps:
arranging low-dimensional features H according to the original hyperspectral image positions k K=1, …, K, resulting in a low dimensional feature h= { H of the original hyperspectral image 1 ,…,H K Where K represents all the super-pixel numbers.
5. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the method of any of claims 1-3 when executing the computer program.
6. A computer readable storage medium, characterized in that the computer readable storage medium stores a computer program which, when executed by a processor, causes the processor to perform the method of any of claims 1-3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010343479.4A CN111563544B (en) | 2020-04-27 | 2020-04-27 | Maximum signal-to-noise ratio hyperspectral data dimension reduction method for multi-scale superpixel segmentation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010343479.4A CN111563544B (en) | 2020-04-27 | 2020-04-27 | Maximum signal-to-noise ratio hyperspectral data dimension reduction method for multi-scale superpixel segmentation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111563544A CN111563544A (en) | 2020-08-21 |
| CN111563544B true CN111563544B (en) | 2023-07-25 |
Family
ID=72071965
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010343479.4A Active CN111563544B (en) | 2020-04-27 | 2020-04-27 | Maximum signal-to-noise ratio hyperspectral data dimension reduction method for multi-scale superpixel segmentation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111563544B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112199990B (en) * | 2020-08-27 | 2023-06-23 | 重庆工商大学 | Hyperspectral image segmentation system and method |
| CN112633202B (en) * | 2020-12-29 | 2022-09-16 | 河南大学 | Hyperspectral image classification algorithm based on dual denoising combined multi-scale superpixel dimension reduction |
| CN112785583B (en) * | 2021-02-01 | 2022-03-04 | 哈尔滨工业大学 | Hyperspectral remote sensing image reflectivity recovery method based on superpixel segmentation |
| CN112800138B (en) * | 2021-02-04 | 2021-10-15 | 广东云曌医疗科技有限公司 | Big data classification method and system |
| CN112990368B (en) * | 2021-04-26 | 2021-07-30 | 湖南大学 | Polygonal structure guided hyperspectral image single sample identification method and system |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108447065A (en) * | 2018-03-12 | 2018-08-24 | 哈尔滨工业大学 | The super pixel dividing method of EO-1 hyperion driven based on factor weighted method pseudo color composing and color histogram |
| CN108876797A (en) * | 2018-06-08 | 2018-11-23 | 长安大学 | A kind of image segmentation system and method based on Spiking-SOM neural network clustering |
-
2020
- 2020-04-27 CN CN202010343479.4A patent/CN111563544B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108447065A (en) * | 2018-03-12 | 2018-08-24 | 哈尔滨工业大学 | The super pixel dividing method of EO-1 hyperion driven based on factor weighted method pseudo color composing and color histogram |
| CN108876797A (en) * | 2018-06-08 | 2018-11-23 | 长安大学 | A kind of image segmentation system and method based on Spiking-SOM neural network clustering |
Non-Patent Citations (1)
| Title |
|---|
| 李冬青.基于空-谱联合图嵌入的高光谱数据降维.《中国博士学位论文全文数据库 工程科技且辑(月刊)》.2019,第2-5章. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111563544A (en) | 2020-08-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111563544B (en) | Maximum signal-to-noise ratio hyperspectral data dimension reduction method for multi-scale superpixel segmentation | |
| Sun et al. | Graph-regularized fast and robust principal component analysis for hyperspectral band selection | |
| Uddin et al. | PCA-based feature reduction for hyperspectral remote sensing image classification | |
| Montazer et al. | An improved radial basis function neural network for object image retrieval | |
| Zhao et al. | Band-subset-based clustering and fusion for hyperspectral imagery classification | |
| CN114332621B (en) | Disease and pest identification method and system based on multi-model feature fusion | |
| CN110889865B (en) | Video target tracking method based on local weighted sparse feature selection | |
| CN111310598B (en) | Hyperspectral remote sensing image classification method based on 3-dimensional and 2-dimensional mixed convolution | |
| US20160078634A1 (en) | Methods and systems for image matting and foreground estimation based on hierarchical graphs | |
| Huo et al. | A batch-mode active learning algorithm using region-partitioning diversity for SVM classifier | |
| CN113723492A (en) | Hyperspectral image semi-supervised classification method and device for improving active deep learning | |
| CN109034213B (en) | Hyperspectral image classification method and system based on correlation entropy principle | |
| Zhang et al. | Polygon structure-guided hyperspectral image classification with single sample for strong geometric characteristics scenes | |
| CN111027636B (en) | Unsupervised feature selection method and system based on multi-label learning | |
| Hou et al. | Spatial–spectral weighted and regularized tensor sparse correlation filter for object tracking in hyperspectral videos | |
| Fu et al. | Robust multi-kernelized correlators for UAV tracking with adaptive context analysis and dynamic weighted filters | |
| CN110020674B (en) | Cross-domain self-adaptive image classification method for improving local category discrimination | |
| Ni et al. | Category-level assignment for cross-domain semantic segmentation in remote sensing images | |
| CN111582142B (en) | Image matching method and device | |
| CN111460966B (en) | Hyperspectral remote sensing image classification method based on metric learning and neighbor enhancement | |
| Artola et al. | Glad: A global-to-local anomaly detector | |
| CN115496933A (en) | Hyperspectral classification method and system based on space-spectrum prototype feature learning | |
| Sotiropoulos | Handling variable shaped & high resolution images for multi-class classification problem | |
| Varma et al. | HSIS-Net: Hyperspectral Image Segmentation Using Multi-view Active Learning Based FCSN. | |
| CN114202659A (en) | Fine-grained image classification method based on spatial symmetry irregular local region feature extraction |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |