CN113065518A - Hyperpixel space spectrum multi-kernel hyperspectral image classification method based on LBP (local binary pattern) features - Google Patents
Hyperpixel space spectrum multi-kernel hyperspectral image classification method based on LBP (local binary pattern) features Download PDFInfo
- Publication number
- CN113065518A CN113065518A CN202110447325.4A CN202110447325A CN113065518A CN 113065518 A CN113065518 A CN 113065518A CN 202110447325 A CN202110447325 A CN 202110447325A CN 113065518 A CN113065518 A CN 113065518A
- Authority
- CN
- China
- Prior art keywords
- pixels
- pixel
- kernel
- super
- lbp
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 66
- 238000001228 spectrum Methods 0.000 title claims abstract description 56
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 41
- 230000003595 spectral effect Effects 0.000 claims abstract description 37
- 230000011218 segmentation Effects 0.000 claims abstract description 27
- 238000012847 principal component analysis method Methods 0.000 claims abstract description 9
- 238000001914 filtration Methods 0.000 claims abstract description 8
- 238000012706 support-vector machine Methods 0.000 claims abstract description 8
- 230000004927 fusion Effects 0.000 claims abstract description 7
- 230000006870 function Effects 0.000 claims description 25
- 239000000126 substance Substances 0.000 claims description 16
- 238000000605 extraction Methods 0.000 claims description 11
- 238000004458 analytical method Methods 0.000 claims description 9
- 239000002131 composite material Substances 0.000 claims description 7
- 238000012549 training Methods 0.000 claims description 7
- 238000004364 calculation method Methods 0.000 claims description 3
- 230000009467 reduction Effects 0.000 abstract description 2
- 238000001514 detection method Methods 0.000 description 4
- 244000025254 Cannabis sativa Species 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 235000008331 Pinus X rigitaeda Nutrition 0.000 description 2
- 235000011613 Pinus brutia Nutrition 0.000 description 2
- 241000018646 Pinus brutia Species 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 230000007123 defense Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 244000075850 Avena orientalis Species 0.000 description 1
- 235000007319 Avena orientalis Nutrition 0.000 description 1
- 240000004658 Medicago sativa Species 0.000 description 1
- 235000017587 Medicago sativa ssp. sativa Nutrition 0.000 description 1
- 229910000831 Steel Inorganic materials 0.000 description 1
- 241000209140 Triticum Species 0.000 description 1
- 235000021307 Triticum Nutrition 0.000 description 1
- 240000008042 Zea mays Species 0.000 description 1
- 235000005824 Zea mays ssp. parviglumis Nutrition 0.000 description 1
- 235000002017 Zea mays subsp mays Nutrition 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000007635 classification algorithm Methods 0.000 description 1
- 235000005822 corn Nutrition 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012805 post-processing Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000000513 principal component analysis Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 239000010959 steel Substances 0.000 description 1
Images
Classifications
-
- 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
-
- 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
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Astronomy & Astrophysics (AREA)
- Remote Sensing (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a super-pixel space spectrum multi-kernel hyperspectral image classification method based on LBP characteristics. Firstly, the image subjected to the dimensionality reduction by the principal component analysis method is subjected to superpixel segmentation to generate a hyperspectral image with a superpixel index. Then, spatial features between super-pixels and in the super-pixels are extracted by adopting a weighted average filtering and LBP algorithm to obtain spatial kernels between the super-pixels and spatial kernels in the super-pixels, and the extracted spectral kernels are combined for fusion. Finally, the combined kernel is input into a support vector machine classifier to generate a classification result graph. According to the method, the LBP algorithm is combined with the super-pixels, the LBP algorithm is used for extracting the edge characteristic information in the super-pixels, the problems that the pixel edge information is lost and the classification of the edge pixels is inaccurate due to the fact that the average value of all pixels in the super-pixels is used for replacing all pixel values in the super-pixels in the traditional multi-core method can be effectively solved, and the classification accuracy and efficiency are improved.
Description
Technical Field
The invention relates to the technical field of hyperspectral remote sensing image processing, in particular to a method for classifying a superpixel space-spectrum multi-kernel hyperspectral image based on Local Binary Pattern (LBP) characteristics.
Background
The appearance of hyperspectral remote sensing is a revolution in the remote sensing world, and substances which are not detectable in broadband remote sensing can be detected in hyperspectral remote sensing. Hundreds of identical narrow-band spectral channels can be obtained from the hyperspectral image, and richer spectral information can be provided to support the fine identification of various ground surface covering materials. Therefore, the hyperspectral image is receiving more and more attention, and is applied to classification, target detection, anomaly detection, spectrum unmixing and the like. The hyperspectral image classification task plays a substantial important role in the fields of crop detection in geological exploration, national defense, military and the like, and is worthy of further research.
However, while hyperspectral image classification is widely used, it also faces significant challenges, such as the well-known hough phenomenon. The Hughes phenomenon refers to the phenomenon that the classification precision is increased firstly and then decreased along with the increase of the number of the wave bands participating in the operation in the hyperspectral analysis process. To solve this problem, feature extraction is considered as a key step of hyperspectral image processing. However, due to the spatial variability of spectral features, hyperspectral image feature extraction is one of the most challenging tasks in hyperspectral image processing.
The most common hyperspectral image feature extraction method is a kernel method, and the joint learning and utilization of spatial information and spectral information can be realized by adopting a simple linear weighting mode. The kernel method is to fully mine and utilize spectral information and spatial information contained in hyperspectral data, construct a new kernel function, realize fusion of different heterogeneous characteristics under the meaning of sample similarity measurement, and then complete classification under an SVM frame so as to effectively improve classification performance. From the research progress in recent years, the design aiming at the hyperspectral kernel function goes through three stages of spectrum weighting kernel, empty spectrum mixing kernel and multi-kernel learning. Compared with a classification method only considering spectral information, the classification method combining spatial information and spectral information can achieve a better classification effect. However, since the size and shape of the spatial region used by the spatial kernel are fixed, the spatial texture of the hyperspectral image is often not fully utilized.
The super-pixel segmentation can change the shape of an area in a self-adaptive mode according to the space structure of a hyperspectral image, pixels are grouped by utilizing the similarity of features among the pixels, a small number of super-pixels are used for replacing a large number of pixels to express picture features, and the complexity of image post-processing can be reduced to a great extent. The superpixel-based composite kernel method treats a superpixel as a local neighborhood, and spatial information can be obtained by using the superpixel, so that the selection of an optimal spatial neighborhood is avoided. Compared with a single-core method based on the super-pixels, the multi-core method based on the super-pixels not only utilizes the spatial information in the super-pixels, but also utilizes the spatial information among the super-pixels, and has higher classification precision. However, the super-pixel based approach also has drawbacks: spatial information within a super-pixel is represented by the mean of the pixels, and edge information of the pixels within the super-pixel is lost.
Disclosure of Invention
The invention provides a super-pixel space-spectrum multi-kernel hyperspectral image classification method based on LBP (local binary pattern) characteristics, which aims to solve the problems that the space information in super-pixels is represented by the mean value of the pixels, the edge information in the super-pixels is lost, the image classification efficiency is low, the precision is poor and the like in the conventional super-pixel based method.
The invention provides a super-pixel space-spectrum multi-kernel hyperspectral image classification method based on LBP characteristics, which comprises the following steps of:
step 2: obtaining three kernel functions by utilizing the hyperspectral image with the superpixel index, wherein the three kernel functions comprise a spectrum kernel, a space kernel based in the superpixel and a space kernel based between the superpixels;
and step 3: and fusing the obtained three kernel functions, inputting the obtained fusion result into a support vector machine classifier, and classifying after training to obtain a classification result graph.
Further, in one implementation, the step 1 includes:
1-1, performing wave band selection on the hyperspectral image by using a principal component analysis method to obtain three unrelated principal component images;
step 1-2, applying an entropy rate superpixel segmentation algorithm to the principal component image to generate a superpixel segmentation image;
and 1-3, combining the super-pixel segmentation image with an original hyperspectral image to generate the hyperspectral image with the super-pixel index.
Further, in one implementation, the step 1-1 includes:
and linearly projecting the hyperspectral images into a group of new coordinate spaces by using a principal component analysis method, selecting the first three principal components of the hyperspectral images to form principal component images, and using the principal component images for superpixel segmentation in the step 1-2.
Further, in one implementation, the step 1-2 includes:
step 1-2-1, selecting the number of super pixels according to the following formula and the complexity of the texture of the main component image:
L＝Lbase×Rtexture
Rtexture＝n/N
wherein L is the number of super pixels, LbaseIs the number of basis super pixels, RtextureThe texture ratio is obtained, N represents the number of nonzero elements in the main component image after texture analysis and filtration, and N represents the number of nonzero elements in the main component image before texture analysis and filtration;
step 1-2-2, firstly constructing a graph G (V, E) on a main component image by superpixel segmentation;
v is a vertex set corresponding to a basic image pixel, and E is an edge set of paired similar points between adjacent pixels;
step 1-2-3, by selecting edges
Dividing the graph into a plurality of connected sub-graphs, each sub-graph corresponding to a super-pixel;
step 1-2-4, adding an entropy rate term H (A) and a balance term B (-) to an objective function of superpixel segmentation according to the following formula:
where λ is the weight introduced to control H (A) and B (-), λ ≧ 0.
Further, in one implementation, the step 2 includes:
step 2-1, collecting all spectral pixels in the hyperspectral image; the single super pixel is composed of adjacent spectral pixels with similar structural features, the spectral pixels are collected to obtain spectral features of the hyperspectral image, and a spectral kernel is obtained through calculation according to the spectral features;
2-2, obtaining a difference value between a central pixel and a neighborhood pixel of the central pixel by using a local binary pattern algorithm and a threshold marking method, and analyzing a local texture structure of the main component image to realize extraction of a spatial kernel in a super pixel;
and 2-3, extracting spatial features among the superpixels by adopting a weighted average algorithm, executing weight value replacement operation in each superpixel, and combining all the superpixels obtained by replacement to obtain a spatial kernel among the superpixels.
Further, in one implementation, the step 2-1 includes:
step 2-1-1, collecting all spectrum pixels in the hyperspectral image, wherein all spectrum pixels in the hyperspectral image form spectrum characteristics of the hyperspectral image;
step 2-1-2, calculating a spectrum kernel by using the collected spectrum pixels:
wherein the content of the first and second substances,
is a nuclear of the light spectrum,
for the input n spectral pixels, σ is a scale parameter for controlling the smoothness of the kernel measure, and σ belongs to a positive real number.
Further, in one implementation, the step 2-2 includes:
step 2-2-1, marking neighborhood pixels of the central point pixel according to a difference value between the central point pixel and the neighborhood pixels of the central point pixel by a threshold marking method;
step 2-2-2, after marking is finished, LBP codes of all pixels are obtained by using a local binary pattern algorithm, analysis of local texture structures of the images is achieved, and spatial kernels in the super pixels are extracted
The mathematical expression of the local binary pattern algorithm is as follows:
wherein, LBP(c、P)Representing LBP coding, c representing the center pixel, P representing the number of pixels around the center pixel, tiRepresenting adjacent pixel grey values, tcThe gray value of the pixel at the center point is represented, and s (-) is a sign function.
Further, in one implementation, the step 2-2-1 includes:
in a circular area, comparing a plurality of adjacent pixels with a central point pixel, if the gray value of any one of the adjacent pixels is larger than that of the central point pixel, marking the adjacent pixel as 1, and if the gray value of the adjacent pixel is smaller than that of the central point pixel, marking the adjacent pixel as 0.
Further, in one implementation, the step 2-3 includes:
step 2-3-1, extracting spatial features among superpixels by adopting a weighted average algorithm, namely for superpixel XiThe weighted average pixel is calculated according to the following formula:
wherein the content of the first and second substances,
in order to weight-average the pixels,
is a super pixel XijAverage of all pixels of (a);
wherein the content of the first and second substances,
weight value omega ofi,jThe following equation is used:
wherein the content of the first and second substances,
is a super pixel xiH is a kernel scale parameter;
step 2-3-1, the same weight value replacement operation is executed in each super pixel; combining all the super-pixels obtained by replacement to obtain a spatial kernel based on the super-pixels
Further, in one implementation, the step 3 includes:
step 3-1, fusing the three kernel functions obtained in the step 2 according to the following formula to form a composite kernel:
wherein the content of the first and second substances,
is a nuclear of the light spectrum,
to be based on the spatial kernel within the superpixel,
is based on the spatial kernel between the super pixels;
and 3-2, inputting the composite kernel into a support vector machine classifier, and classifying after finishing training to obtain a classification result graph.
The invention provides a super-pixel space spectrum multi-kernel hyperspectral image classification method based on LBP characteristics. Firstly, the image subjected to the dimensionality reduction by the principal component analysis method is subjected to superpixel segmentation to generate a hyperspectral image with a superpixel index. Then, spatial features between super-pixels and in the super-pixels are extracted by adopting a weighted average filtering and LBP algorithm to obtain spatial kernels between the super-pixels and spatial kernels in the super-pixels, and the extracted spectral kernels are combined for fusion. Finally, the combined kernel is input into a support vector machine classifier to generate a classification result graph. According to the method, the LBP algorithm is combined with the super-pixels, the LBP algorithm is used for extracting the edge characteristic information in the super-pixels, the problems that the pixel edge information is lost and the classification of the edge pixels is inaccurate due to the fact that the average value of all pixels in the super-pixels is used for replacing all pixel values in the super-pixels in the traditional multi-core method can be effectively solved, and the classification accuracy and efficiency are improved.
Compared with the prior art, the invention has the remarkable advantages that: (1) after the selection of the main component wave band of the hyperspectral image is completed, the two-dimensional superpixel image is generated by utilizing the entropy rate superpixel algorithm, and the efficiency and the precision of kernel extraction can be greatly improved. (2) On the basis of extracting the spectral kernel by adopting superpixel segmentation, the invention adopts an LBP algorithm to extract the spatial kernel in the superpixel and adopts a weighted average algorithm to extract the spatial kernel among the superpixels. The three kernel functions respectively play their roles, so that the classification accuracy can be effectively improved, and the classification efficiency can be improved. (3) The classification method fusing the three kernel functions is quick and effective, ingenious in method, novel in concept and good in application prospect.
Drawings
In order to more clearly illustrate the technical solution of the present invention, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious to those skilled in the art that other drawings can be obtained based on these drawings without creative efforts.
FIG. 1 is a schematic workflow diagram of a super-pixel space-spectrum multi-kernel hyperspectral image classification method based on LBP features according to an embodiment of the present invention;
fig. 2 is a schematic workflow diagram of a hyperspectral image with a superpixel index generated by combining an obtained superpixel segmentation graph and an original hyperspectral image in the method for classifying a superpixel spatio-spectral multi-kernel hyperspectral image based on LBP features according to the embodiment of the present invention.
Detailed Description
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
The embodiment of the invention discloses a method for classifying a superpixel space-spectrum multi-kernel hyperspectral image based on LBP characteristics, which is applied to the accurate classification of hyperspectral remote sensing images, is a key step of hyperspectral remote sensing image preprocessing, and plays a substantial key role in the application fields of geological exploration, crop detection, national defense and military and the like.
As shown in fig. 1, the method for classifying a hyper-pixel spatial spectrum multi-kernel hyper-spectral image based on LBP features provided in this embodiment includes the following steps:
Step 2: obtaining three kernel functions by utilizing the hyperspectral image with the superpixel index, wherein the three kernel functions comprise a spectrum kernel, a space kernel based in the superpixel and a space kernel based between the superpixels; specifically, in this step, all the spectral pixels in the hyperspectral image are collected according to the principle that each superpixel is composed of a group of adjacent spectral pixels, and the spectral information can be directly composed of the collected spectral pixels, so as to obtain the spectral characteristics of the hyperspectral image. And secondly, marking the difference between the central pixel and the neighborhood pixels by using an LBP algorithm through a threshold value, so as to analyze the local texture structure and realize the extraction of the space kernel in the super-pixel. And finally, extracting spatial features among the super pixels by adopting a weighted average algorithm, executing weight value replacement operation in each super pixel, and combining all the super pixels obtained by replacement to further obtain a spatial kernel among the super pixels.
And step 3: and fusing the obtained three kernel functions, inputting the obtained fusion result into a support vector machine classifier, and classifying after training to obtain a classification result graph.
According to the method, the LBP algorithm is combined with the super-pixels, the LBP algorithm is used for extracting the edge characteristic information in the super-pixels, the problems that the pixel edge information is lost and the classification of the edge pixels is inaccurate due to the fact that the average value of all pixels in the super-pixels is used for replacing all pixel values in the super-pixels in the traditional multi-core method can be effectively solved, and the classification accuracy and efficiency are improved.
In the method for classifying a superpixel space-spectrum multi-kernel hyperspectral image based on LBP features according to this embodiment, the step 1 includes:
1-1, performing wave band selection on the hyperspectral image by using a principal component analysis method to obtain three unrelated principal component images;
step 1-2, applying an entropy rate superpixel segmentation algorithm to the principal component image to generate a superpixel segmentation image;
and 1-3, combining the super-pixel segmentation image with an original hyperspectral image to generate the hyperspectral image with the super-pixel index.
In the method for classifying a superpixel space-spectrum multi-kernel hyperspectral image based on LBP features in this embodiment, the step 1-1 includes:
and linearly projecting the hyperspectral images into a group of new coordinate spaces by using a principal component analysis method, selecting the first three principal components of the hyperspectral images to form principal component images, and using the principal component images for superpixel segmentation in the step 1-2.
In the method for classifying a superpixel space-spectrum multi-kernel hyperspectral image based on LBP features in this embodiment, the step 1-2 includes:
step 1-2-1, selecting the number of super pixels according to the following formula and the complexity of the texture of the main component image:
L＝Lbase×Rtexture
Rtexture＝n/N
wherein L is the number of super pixels, LbaseIs the number of basis super pixels, RtextureThe texture ratio is obtained, N represents the number of nonzero elements in the main component image after texture analysis and filtration, and N represents the number of nonzero elements in the main component image before texture analysis and filtration;
step 1-2-2, firstly constructing a graph G (V, E) on a main component image by superpixel segmentation;
v is a vertex set corresponding to a basic image pixel, and E is an edge set of paired similar points between adjacent pixels;
step 1-2-3, by selecting edges
Dividing the graph into a plurality of connected sub-graphs, each sub-graph corresponding to a super-pixel;
step 1-2-4, adding an entropy rate term H (A) and a balance term B (-) to an objective function of superpixel segmentation according to the following formula:
where λ is the weight introduced to control H (A) and B (-), λ ≧ 0.
As shown in fig. 2, step 2 and step 3 described in this embodiment include kernel extraction and fusion of the hyperspectral images with superpixel indexes. The step 2 comprises the steps of obtaining three kernel functions by the hyperspectral image with the superpixel index to realize the extraction of three different types of features. In the method for classifying a superpixel space-spectrum multi-kernel hyperspectral image based on LBP features according to this embodiment, the step 2 includes:
step 2-1, collecting all spectral pixels in the hyperspectral image; the single super pixel is composed of adjacent spectral pixels with similar structural features, the spectral pixels are collected to obtain spectral features of the hyperspectral image, and a spectral kernel is obtained through calculation according to the spectral features;
2-2, obtaining a difference value between a central pixel and a neighborhood pixel of the central pixel by using a local binary pattern algorithm and a threshold marking method, and analyzing a local texture structure of the main component image to realize extraction of a spatial kernel in a super pixel;
and 2-3, extracting spatial features among the superpixels by adopting a weighted average algorithm, executing weight value replacement operation in each superpixel, and combining all the superpixels obtained by replacement to obtain a spatial kernel among the superpixels.
In the method for classifying a superpixel space-spectrum multi-kernel hyperspectral image based on LBP features in this embodiment, the step 2-1 includes:
step 2-1-1, collecting all spectrum pixels in the hyperspectral image, wherein all spectrum pixels in the hyperspectral image form spectrum characteristics of the hyperspectral image;
step 2-1-2, calculating a spectrum kernel by using the collected spectrum pixels:
wherein the content of the first and second substances,
is a nuclear of the light spectrum,
for the input n spectral pixels, σ is a scale parameter for controlling the smoothness of the kernel measure, and σ belongs to a positive real number.
In the method for classifying a superpixel space-spectrum multi-kernel hyperspectral image based on LBP features in this embodiment, the step 2-2 includes:
step 2-2-1, marking neighborhood pixels of the central point pixel according to a difference value between the central point pixel and the neighborhood pixels of the central point pixel by a threshold marking method;
step 2-2-2, after the marking is finished, the LBP codes of all pixels are obtained by using a local binary pattern algorithm to realize the pairAnalysis of local texture of image to extract spatial kernel in superpixel
The mathematical expression of the local binary pattern algorithm is as follows:
wherein, LBP(c、P)Representing LBP coding, c representing the center pixel, P representing the number of pixels around the center pixel, tiRepresenting adjacent pixel grey values, tcThe gray value of the pixel at the center point is represented, and s (-) is a sign function.
In the method for classifying a superpixel space-spectrum multi-kernel hyperspectral image based on LBP features according to this embodiment, the step 2-2-1 includes:
in a circular area, comparing a plurality of adjacent pixels with a central point pixel, if the gray value of any one of the adjacent pixels is larger than that of the central point pixel, marking the adjacent pixel as 1, and if the gray value of the adjacent pixel is smaller than that of the central point pixel, marking the adjacent pixel as 0.
In step 2-2 described in this embodiment, in the LBP algorithm, the difference between the central pixel and the neighboring pixels is marked by the threshold, so as to analyze the local texture structure of the hyperspectral image, and the method is gradual and is not affected by the change of the illumination condition. The original hyperspectral image retains the first 3 principal components after principal component analysis, the LBP algorithm is applied to each principal component to extract features, and all wave bands of the obtained local binary histogram are connected in series to form a spatial feature vector.
In the method for classifying a superpixel space-spectrum multi-kernel hyperspectral image based on LBP features in this embodiment, the steps 2 to 3 include:
step 2-3-1, extracting spatial features among superpixels by adopting a weighted average algorithm, namely for superpixel XiThe weighted average pixel is calculated according to the following formula:
wherein the content of the first and second substances,
in order to weight-average the pixels,
is a super pixel XijAverage of all pixels of (a);
wherein the content of the first and second substances,
weight value omega ofi,jThe following equation is used:
wherein the content of the first and second substances,
is a super pixel xiH is a kernel scale parameter;
step 2-3-1, the same weight value replacement operation is executed in each super pixel; combining all the super-pixels obtained by replacement to obtain a spatial kernel based on the super-pixels
In the method for classifying a superpixel space-spectrum multi-kernel hyperspectral image based on LBP features according to this embodiment, the step 3 includes:
step 3-1, fusing the three kernel functions obtained in the step 2 according to the following formula to form a composite kernel:
wherein the content of the first and second substances,
is a nuclear of the light spectrum,
to be based on the spatial kernel within the superpixel,
is based on the spatial kernel between the super pixels;
and 3-2, inputting the composite kernel into a support vector machine classifier, and classifying after finishing training to obtain a classification result graph.
The method is implemented on a public Indian Pines hyperspectral image data set, quantitative comparison is carried out on the method and a recently known hyperspectral image classification method, and the classification result is shown in table 1.
TABLE 1 Classification results obtained by carrying out the present invention on Indian Pines data
| Class name | SVMCK | SpATV | LBPELM | SC_MK | RMK | ASMGSSK | The method of the invention |
| Alfalfa | 0.5286 | 0.8510 | 0.9854 | 0.9610 | 0.9551 | 0.9878 | 0.9756 |
| Corn-no till | 0.9352 | 0.9709 | 0.9703 | 0.9416 | 0.9660 | 0.9755 | 0.9865 |
| Corn-min till | 0.9405 | 0.9839 | 0.9656 | 0.9734 | 0.9731 | 0.9908 | 0.9969 |
| Corn | 0.8433 | 0.9948 | 0.9689 | 0.9704 | 0.9710 | 0.9624 | 0.9948 |
| Grass/pasture | 0.9508 | 0.9581 | 0.9833 | 0.9667 | 0.9644 | 0.9844 | 0.9837 |
| Grass/tree | 0.9812 | 0.9849 | 0.9807 | 0.9982 | 0.9898 | 0.9911 | 0.9998 |
| Grass/pasture-mowed | 0.1696 | 0.5000 | 0.9364 | 0.9640 | 0.9652 | 0.9652 | 0.9640 |
| Hay-windrowed | 0.9907 | 1.0000 | 0.9950 | 0.9833 | 0.9937 | 0.9977 | 1.0000 |
| Oats | 0.0667 | 0.0556 | 0.9280 | 0.9111 | 0.9833 | 0.9056 | 1.0000 |
| Soybeans-no till | 0.8939 | 0.9437 | 0.9227 | 0.9410 | 0.9662 | 0.9777 | 0.9859 |
| Soybeans-min till | 0.9612 | 0.9904 | 0.9891 | 0.9698 | 0.9958 | 0.9914 | 0.9999 |
| Soybeans-clean till | 0.9218 | 0.9724 | 0.9831 | 0.9453 | 0.9668 | 0.9814 | 0.9919 |
| Wheat | 0.9901 | 0.9953 | 0.9901 | 0.9946 | 0.9963 | 0.9901 | 0.9946 |
| Woods | 0.9744 | 0.9944 | 0.9940 | 0.9938 | 0.9968 | 0.9938 | 0.9999 |
| Bldg-grass-tree-drives | 0.8185 | 0.9845 | 0.9952 | 0.9467 | 0.9912 | 0.9909 | 1.0000 |
| Stone-steel towers | 0.8930 | 0.7826 | 0.9253 | 0.9714 | 0.9860 | 0.9419 | 0.9774 |
| mean_OA | 0.9375 | 0.9751 | 0.9816 | 0.9670 | 0.9823 | 0.9861 | 0.9946 |
| STD_OA | 0.0026 | 0.0069 | 0.0045 | 0.0052 | 0.0030 | 0.0018 | 0.0017 |
| mean_kappa | 0.9286 | 0.8726 | 0.9789 | 0.9623 | 0.9799 | 0.9842 | 0.9939 |
| STD_kappa | 0.0030 | 0.0135 | 0.0032 | 0.0060 | 0.0034 | 0.0020 | 0.0019 |
In the present embodiment, the overall classification accuracy (OA), the kappa coefficient, and the standard deviation thereof are used as evaluation indexes. The optimal results in the table are shown in bold. Therefore, compared with other classification methods, the classification method provided by the invention has the advantages that the precision is improved by 1% -6%, and the standard deviation of the classification precision of the classification method is the minimum compared with other classification algorithms. This shows that under the condition of limited training samples, the algorithm designed by the invention not only has better classification performance, but also has more stable classification performance.
Compared with the prior art, the invention has the remarkable advantages that: (1) after the selection of the main component wave band of the hyperspectral image is completed, the two-dimensional superpixel image is generated by utilizing the entropy rate superpixel algorithm, and the efficiency and the precision of kernel extraction can be greatly improved. (2) On the basis of extracting the spectral kernel by adopting superpixel segmentation, the invention adopts an LBP algorithm to extract the spatial kernel in the superpixel and adopts a weighted average algorithm to extract the spatial kernel among the superpixels. The three kernel functions respectively play their roles, so that the classification accuracy can be effectively improved, and the classification efficiency can be improved. (3) The classification method fusing the three kernel functions is quick and effective, ingenious in method, novel in concept and good in application prospect.
In a specific implementation, the present invention further provides a computer storage medium, where the computer storage medium may store a program, and when the program is executed, the program may include some or all of the steps in each embodiment of the method for classifying a hyper-pixel spatial-spectral multi-kernel hyper-spectral image based on LBP features provided by the present invention. The storage medium may be a magnetic disk, an optical disk, a read-only memory (ROM), a Random Access Memory (RAM), or the like.
Those skilled in the art will readily appreciate that the techniques of the embodiments of the present invention may be implemented as software plus a required general purpose hardware platform. Based on such understanding, the technical solutions in the embodiments of the present invention may be essentially or partially implemented in the form of a software product, which may be stored in a storage medium, such as ROM/RAM, magnetic disk, optical disk, etc., and includes several instructions for enabling a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method according to the embodiments or some parts of the embodiments.
The same and similar parts in the various embodiments in this specification may be referred to each other. The above-described embodiments of the present invention should not be construed as limiting the scope of the present invention.
Claims (10)
1. The method for classifying the super-pixel space-spectrum multi-kernel hyperspectral images based on the LBP features is characterized by comprising the following steps of:
step 1, indexing a hyperspectral image by adopting an entropy rate superpixel segmentation algorithm to generate a hyperspectral image with a superpixel index;
step 2: obtaining three kernel functions by utilizing the hyperspectral image with the superpixel index, wherein the three kernel functions comprise a spectrum kernel, a space kernel based in the superpixel and a space kernel based between the superpixels;
and step 3: and fusing the obtained three kernel functions, inputting the obtained fusion result into a support vector machine classifier, and classifying after training to obtain a classification result graph.
2. The method for classifying the hyper-pixel spatial-spectrum multi-kernel hyper-spectral images based on the LBP (local binary pattern) characteristics according to claim 1, wherein the step 1 comprises the following steps:
1-1, performing wave band selection on the hyperspectral image by using a principal component analysis method to obtain three unrelated principal component images;
step 1-2, applying an entropy rate superpixel segmentation algorithm to the principal component image to generate a superpixel segmentation image;
and 1-3, combining the super-pixel segmentation image with an original hyperspectral image to generate the hyperspectral image with the super-pixel index.
3. The LBP feature-based superpixel space-spectrum multi-kernel hyperspectral image classification method according to claim 2, wherein the step 1-1 comprises:
and linearly projecting the hyperspectral images into a group of new coordinate spaces by using a principal component analysis method, selecting the first three principal components of the hyperspectral images to form principal component images, and using the principal component images for superpixel segmentation in the step 1-2.
4. The method for classifying the hyper-pixel spatial-spectrum multi-kernel hyper-spectral images based on the LBP (local binary pattern) characteristics according to claim 3, wherein the steps 1-2 comprise:
step 1-2-1, selecting the number of super pixels according to the following formula and the complexity of the texture of the main component image:
L＝Lbase×Rtexture
Rtexture＝n/N
wherein L is the number of super pixels, LbaseIs the number of basis super pixels, RtextureThe texture ratio is obtained, N represents the number of nonzero elements in the main component image after texture analysis and filtration, and N represents the number of nonzero elements in the main component image before texture analysis and filtration;
step 1-2-2, firstly constructing a graph G (V, E) on a main component image by superpixel segmentation;
v is a vertex set corresponding to a basic image pixel, and E is an edge set of paired similar points between adjacent pixels;
step 1-2-3, by selecting edges
Dividing the graph into a plurality of connected sub-graphs, each sub-graph corresponding to a super-pixel;
step 1-2-4, adding an entropy rate term H (A) and a balance term B (-) to an objective function of superpixel segmentation according to the following formula:
where λ is the weight introduced to control H (A) and B (-), λ ≧ 0.
5. The method for classifying the hyper-pixel spatial-spectrum multi-kernel hyper-spectral images based on the LBP (local binary pattern) characteristics according to claim 1, wherein the step 2 comprises the following steps:
step 2-1, collecting all spectral pixels in the hyperspectral image; the single super pixel is composed of adjacent spectral pixels with similar structural features, the spectral pixels are collected to obtain spectral features of the hyperspectral image, and a spectral kernel is obtained through calculation according to the spectral features;
2-2, obtaining a difference value between a central pixel and a neighborhood pixel of the central pixel by using a local binary pattern algorithm and a threshold marking method, and analyzing a local texture structure of the main component image to realize extraction of a spatial kernel in a super pixel;
and 2-3, extracting spatial features among the superpixels by adopting a weighted average algorithm, executing weight value replacement operation in each superpixel, and combining all the superpixels obtained by replacement to obtain a spatial kernel among the superpixels.
6. The method for classifying the hyper-pixel spatial-spectrum multi-kernel hyper-spectral images based on the LBP (local binary pattern) characteristics according to claim 5, wherein the step 2-1 comprises the following steps:
step 2-1-1, collecting all spectrum pixels in the hyperspectral image, wherein all spectrum pixels in the hyperspectral image form spectrum characteristics of the hyperspectral image;
step 2-1-2, calculating a spectrum kernel by using the collected spectrum pixels:
wherein the content of the first and second substances,
is a nuclear of the light spectrum,
for the input n spectral pixels, σ is a scale parameter for controlling the smoothness of the kernel measure, and σ belongs to a positive real number.
7. The method for classifying the hyper-pixel spatial-spectrum multi-kernel hyper-spectral images based on the LBP (local binary pattern) characteristics according to claim 5, wherein the step 2-2 comprises the following steps:
step 2-2-1, marking neighborhood pixels of the central point pixel according to a difference value between the central point pixel and the neighborhood pixels of the central point pixel by a threshold marking method;
step 2-2-2, after marking is finished, LBP codes of all pixels are obtained by using a local binary pattern algorithm, analysis of local texture structures of the images is achieved, and spatial kernels in the super pixels are extracted
The mathematical expression of the local binary pattern algorithm is as follows:
wherein, LBP(c、P)Representing LBP coding, c representing the center pixel, P representing the number of pixels around the center pixel, tiRepresenting adjacent pixel grey values, tcThe gray value of the pixel at the center point is represented, and s (-) is a sign function.
8. The method for classifying the hyper-pixel spatial-spectrum multi-kernel hyper-spectral images based on the LBP (local binary pattern) characteristics according to claim 7, wherein the step 2-2-1 comprises the following steps:
in a circular area, comparing a plurality of adjacent pixels with a central point pixel, if the gray value of any one of the adjacent pixels is larger than that of the central point pixel, marking the adjacent pixel as 1, and if the gray value of the adjacent pixel is smaller than that of the central point pixel, marking the adjacent pixel as 0.
9. The method for classifying the hyper-pixel spatial-spectrum multi-kernel hyper-spectral images based on the LBP (local binary pattern) characteristics according to claim 5, wherein the steps 2 to 3 comprise:
step 2-3-1, extracting spatial features among superpixels by adopting a weighted average algorithm, namely for superpixel XiThe weighted average pixel is calculated according to the following formula:
wherein the content of the first and second substances,
in order to weight-average the pixels,
is a super pixel XijAverage of all pixels of (a);
wherein the content of the first and second substances,
weight value omega ofi,jThe following equation is used:
wherein the content of the first and second substances,
is a super pixel xiH is a kernel scale parameter;
step 2-3-1, the same weight value replacement operation is executed in each super pixel; combining all the super-pixels obtained by replacement to obtain a spatial kernel based on the super-pixels
10. The method for classifying the hyper-pixel spatial-spectrum multi-kernel hyper-spectral images based on the LBP (local binary pattern) characteristics according to claim 1, wherein the step 3 comprises the following steps:
step 3-1, fusing the three kernel functions obtained in the step 2 according to the following formula to form a composite kernel:
wherein the content of the first and second substances,
is a nuclear of the light spectrum,
to be based on the spatial kernel within the superpixel,
is based on the spatial kernel between the super pixels;
and 3-2, inputting the composite kernel into a support vector machine classifier, and classifying after finishing training to obtain a classification result graph.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110447325.4A CN113065518A (en) | 2021-04-25 | 2021-04-25 | Hyperpixel space spectrum multi-kernel hyperspectral image classification method based on LBP (local binary pattern) features |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110447325.4A CN113065518A (en) | 2021-04-25 | 2021-04-25 | Hyperpixel space spectrum multi-kernel hyperspectral image classification method based on LBP (local binary pattern) features |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113065518A true CN113065518A (en) | 2021-07-02 |
Family
ID=76567559
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110447325.4A Pending CN113065518A (en) | 2021-04-25 | 2021-04-25 | Hyperpixel space spectrum multi-kernel hyperspectral image classification method based on LBP (local binary pattern) features |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113065518A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114842264A (en) * | 2022-05-17 | 2022-08-02 | 北京理工大学 | Hyperspectral image classification method based on multi-scale spatial spectral feature joint learning |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106503739A (en) * | 2016-10-31 | 2017-03-15 | 中国地质大学（武汉） | The target in hyperspectral remotely sensed image svm classifier method and system of combined spectral and textural characteristics |
| WO2018045626A1 (en) * | 2016-09-07 | 2018-03-15 | 深圳大学 | Super-pixel level information fusion-based hyperspectral image classification method and system |
| CN110796038A (en) * | 2019-10-15 | 2020-02-14 | 南京理工大学 | Hyperspectral remote sensing image classification method combined with rapid region growing superpixel segmentation |
| CN110866439A (en) * | 2019-09-25 | 2020-03-06 | 南京航空航天大学 | Hyperspectral image joint classification method based on multi-feature learning and superpixel kernel sparse representation |
-
2021
- 2021-04-25 CN CN202110447325.4A patent/CN113065518A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018045626A1 (en) * | 2016-09-07 | 2018-03-15 | 深圳大学 | Super-pixel level information fusion-based hyperspectral image classification method and system |
| CN106503739A (en) * | 2016-10-31 | 2017-03-15 | 中国地质大学（武汉） | The target in hyperspectral remotely sensed image svm classifier method and system of combined spectral and textural characteristics |
| CN110866439A (en) * | 2019-09-25 | 2020-03-06 | 南京航空航天大学 | Hyperspectral image joint classification method based on multi-feature learning and superpixel kernel sparse representation |
| CN110796038A (en) * | 2019-10-15 | 2020-02-14 | 南京理工大学 | Hyperspectral remote sensing image classification method combined with rapid region growing superpixel segmentation |
Non-Patent Citations (1)
| Title |
|---|
| 谷延锋: "高光谱遥感图像解译", 哈尔滨工业大学出版社, pages: 144 - 145 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114842264A (en) * | 2022-05-17 | 2022-08-02 | 北京理工大学 | Hyperspectral image classification method based on multi-scale spatial spectral feature joint learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109948693B (en) | Hyperspectral image classification method based on superpixel sample expansion and generation countermeasure network | |
| Adeel et al. | Diagnosis and recognition of grape leaf diseases: An automated system based on a novel saliency approach and canonical correlation analysis based multiple features fusion | |
| Pan et al. | R-VCANet: A new deep-learning-based hyperspectral image classification method | |
| Sağ et al. | Color image segmentation based on multiobjective artificial bee colony optimization | |
| Li et al. | Gaussian pyramid based multiscale feature fusion for hyperspectral image classification | |
| CN106503739A (en) | The target in hyperspectral remotely sensed image svm classifier method and system of combined spectral and textural characteristics | |
| Ortac et al. | Comparative study of hyperspectral image classification by multidimensional Convolutional Neural Network approaches to improve accuracy | |
| CN110458192B (en) | Hyperspectral remote sensing image classification method and system based on visual saliency | |
| Jia et al. | Spatial-spectral-combined sparse representation-based classification for hyperspectral imagery | |
| Huang et al. | Hyperspectral image classification via discriminant Gabor ensemble filter | |
| Jia et al. | A multiscale superpixel-level group clustering framework for hyperspectral band selection | |
| Tong et al. | Spectral–spatial and cascaded multilayer random forests for tree species classification in airborne hyperspectral images | |
| Lin et al. | Identification of pumpkin powdery mildew based on image processing PCA and machine learning | |
| CN113065518A (en) | Hyperpixel space spectrum multi-kernel hyperspectral image classification method based on LBP (local binary pattern) features | |
| Al-Badri et al. | Adaptive Non-Maximum Suppression for improving performance of Rumex detection | |
| CN113343900A (en) | Combined nuclear remote sensing image target detection method based on combination of CNN and superpixel | |
| Song et al. | Using dual-channel CNN to classify hyperspectral image based on spatial-spectral information | |
| Liu et al. | A band influence algorithm for hyperspectral band selection to classify moldy peanuts | |
| Li et al. | Group convolutional neural networks for hyperspectral image classification | |
| Wei et al. | The hyperspectral image clustering based on spatial information and spectral clustering | |
| Cai et al. | Multi-dimension CNN for hyperspectral image classificaton | |
| Parameshachari et al. | Plant Disease Detection and Classification Using Transfer Learning Inception Technique | |
| Huo et al. | Local graph regularized sparse reconstruction for salient object detection | |
| Bai et al. | Kernel low-rank entropic component analysis for hyperspectral image classification | |
| Li et al. | A novel denoising autoencoder assisted segmentation algorithm for cotton field |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210702 |
|
| RJ01 | Rejection of invention patent application after publication |