LU503093B1 - A Hyperspectral Image Classification Method and System Based on Local Correntropy Matrix - Google Patents
A Hyperspectral Image Classification Method and System Based on Local Correntropy Matrix Download PDFInfo
- Publication number
- LU503093B1 LU503093B1 LU503093A LU503093A LU503093B1 LU 503093 B1 LU503093 B1 LU 503093B1 LU 503093 A LU503093 A LU 503093A LU 503093 A LU503093 A LU 503093A LU 503093 B1 LU503093 B1 LU 503093B1
- Authority
- LU
- Luxembourg
- Prior art keywords
- correntropy
- local
- matrix
- spectral
- pixels
- Prior art date
Links
- 239000011159 matrix material Substances 0.000 title claims abstract description 41
- 238000000034 method Methods 0.000 title claims abstract description 28
- 238000012549 training Methods 0.000 claims abstract description 16
- 230000009467 reduction Effects 0.000 claims abstract description 9
- 230000003595 spectral effect Effects 0.000 claims description 41
- 238000012360 testing method Methods 0.000 claims description 12
- 238000012545 processing Methods 0.000 claims description 7
- 238000001228 spectrum Methods 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 4
- 238000012706 support-vector machine Methods 0.000 abstract description 12
- 238000013461 design Methods 0.000 abstract description 3
- 238000004088 simulation Methods 0.000 description 7
- 230000000694 effects Effects 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 3
- 238000000701 chemical imaging Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000003384 imaging method Methods 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 244000075850 Avena orientalis Species 0.000 description 1
- 235000007319 Avena orientalis Nutrition 0.000 description 1
- PEDCQBHIVMGVHV-UHFFFAOYSA-N Glycerine Chemical compound OCC(O)CO PEDCQBHIVMGVHV-UHFFFAOYSA-N 0.000 description 1
- 240000004658 Medicago sativa Species 0.000 description 1
- 235000017587 Medicago sativa ssp. sativa Nutrition 0.000 description 1
- 235000008331 Pinus X rigitaeda Nutrition 0.000 description 1
- 235000011613 Pinus brutia Nutrition 0.000 description 1
- 241000018646 Pinus brutia Species 0.000 description 1
- 241000209140 Triticum Species 0.000 description 1
- 235000021307 Triticum Nutrition 0.000 description 1
- 241001532014 Xanthorrhoea Species 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
- 238000004458 analytical method Methods 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 235000005822 corn Nutrition 0.000 description 1
- 230000007123 defense Effects 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000010921 in-depth analysis Methods 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 230000000116 mitigating effect Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000008092 positive effect Effects 0.000 description 1
- 230000002265 prevention Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/10—Image acquisition
- G06V10/12—Details of acquisition arrangements; Constructional details thereof
- G06V10/14—Optical characteristics of the device performing the acquisition or on the illumination arrangements
- G06V10/143—Sensing or illuminating at different wavelengths
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/58—Extraction of image or video features relating to hyperspectral data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/776—Validation; Performance evaluation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Multimedia (AREA)
- General Physics & Mathematics (AREA)
- Computing Systems (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Databases & Information Systems (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Spectroscopy & Molecular Physics (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a hyperspectral image classification method and system based on local correntropy matrix. Firstly, dimensionality reduction is performed on hyperspectral image data to remove noise and reduce computational complexity, and then the boundary filling by boundary copying is carried out on the dimensionality-reduced image. Secondly, for each pixel in the image, a local correntropy matrix is constructed to extract the local features. Finally, the obtained local correntropy matrix is input into the support vector machine as a feature for classification, and the class label of each pixel is obtained, which can obtain nonlinear spectral-spatial features. Because the method belongs to the manual feature design method, it does not need a large number of training samples to learn features, has small sample complexity, and is more suitable for practical application.
Description
A Hyperspectral Image Classification Method and System Based on LU503093
Local Correntropy Matrix
The invention belongs to the technical field of remote sensing image processing, in particular to a hyperspectral image classification method and system, which can be used in the fields of precision agriculture, environmental monitoring, urban planning, disaster prevention and mitigation, and the like.
In recent years, with the rapid development of hyperspectral imaging technology, hyperspectral images have been widely used in military surveying and reconnaissance, urban planning analysis, medical information diagnosis, precision agriculture, disaster warning, and other national defense and civil fields. However, because hyperspectral images have the characteristics of high data dimension, few training samples and large variation of spectral feature space, traditional image classification methods are often ineffective in processing hyperspectral images. How to extract effective features from redundant nonlinear hyperspectral data and realize high-precision classification is an urgent problem to be solved.
With the rapid development of remote sensing technology, the band range of hyperspectral data is becoming wider and wider, and the spectral resolution is also increasing. With the help of computer technology, there are more and more application scenarios of hyperspectral data, but the classification of hyperspectral data is still facing great difficulties and challenges.
Firstly, hyperspectral data is large and complex. The bands of hyperspectral images are numerous and continuous, which leads to high feature dimensions. In hyperspectral classification, with the increase of feature dimensions, the classification effect will first increase and then decrease, which is called Huge phenomenon or curse of dimensionality.
And not all bands have a positive effect on classification, and there are data redundancies in a large number of data. Therefore, it is significant to reduce the dimension of hyperspectral data band reasonably to improve the classification effect.
Secondly, there is a lot of imaging interference. Hyperspectral image is a multi-dimensional image containing spatial and spectral information. Due to the weather, illumination and imaging angle during spectral imaging, the data acquisition process is easily interfered by various uncertainties, which will lead to different surface materials showing the same spectral 1 characteristics or the same materials showing different spectral reflection phenomena, that is, LYS03093 "the same spectrum of foreign bodies" and "different spectra of the same objects". Therefore, it is necessary to consider both spatial features and spectral features when designing classification methods.
Finally, there are not enough labeled samples in practical applications. On the one hand, the data sets obtained by the spaceborne hyperspectral imager are few and generally not public.
On the other hand, in remote sensing, labeling hyperspectral data requires human experts to spend a lot of time collecting information on the spot, and it is a very time-consuming and laborious task to obtain label samples. Achieving superior classification results under the condition that the available samples are very limited is still a problem to be solved in the current hyperspectral image field.
To sum up, on the basis of an in-depth analysis of hyperspectral image characteristics, the invention designs a spectral-spatial feature learning method based on correntropy matrix, which has broad application prospects and important theoretical value.
The purpose of the present invention is to propose a hyperspectral image classification method based on correntropy matrix to solve the difficulties in extracting spectral-spatial features of hyperspectral images in the prior art and improve the performance of spectral image classification.
The technical solution for realizing the purpose of the invention is as follows: Firstly, the dimension reduction is performed on the hyperspectral image to remove noise and reduce the amount of calculation of the subsequent steps; Secondly, for each pixel in the image, a local correntropy matrix is constructed to extract the local features of the hyperspectral image;
Finally, the obtained local correntropy matrix is input into the Support Vector Machine (SVM) as features for classification. The specific steps are as follows:
Step 1, input a hyperspectral image, and reduce the dimension of the input image;
Step 2, performing boundary filling processing on the dimension-reduced image by boundary copying;
Step 3, building a local correntropy matrix pixel by pixel, and vectorizing the local correntropy matrix as the spectral-spatial characteristics of the pixels;
Step 4, selecting p% pixels from the hyperspectral image as training samples, and the remaining (100- 7 )% labeled pixels as test samples; 2
Step 5, training the SVM classifier by using the spectral-spatial features of the training LYS03093 samples;
Step 6, input the spectral-spatial characteristics of the test sample into the SVM classifier to obtain the class label of the test sample, and complete the classification.
Further, the specific implementation of step 3 is as follows, 3a) Set the sliding window size to T x T; 3b) Let the center pixel of the window bep1, and its neighboring pixels bepi, i =2, 3, ... ,
T? The cosine distance between this pixel and its surrounding pixels can be obtained by this formula: cos(p,, P,) _ (PP) i=23 17 [Pia (Pi,
Where (-)and |-I2 represent vector inner product and Frobenius norm, respectively. 3c) Selecting the most similar first K-1 pixels to obtain adjacent pixels which are similar in space and spectrum; 3d) Using K pixels to construct correntropy matrix representation. If two different spectral bands are expressed as b, and b, respectively, according to the calculation formula of correntropy, the correntropy between the two spectral bands can be obtained as follows: 1 &
Corr(b,,b)=—> k,(b,.b,)
Ka by means the "th spectral value in the b, spectral band, and there are k spectral values in each spectral band, that is, the spectral values of the selected k pixels in this spectral band.
The local correntropy matrix can be expressed as:
M con = {Corr(b,,6,)} i,j=l
B represents the number of spectral bands of hyperspectral images (after dimension reduction).
Each off-diagonal element in the local correntropy matrix represents the correntropy of different spectral bands, that is, the relationship between different spectral bands. 3e) Vectorize local correntropy matrix; 3f) Move the window, and circularly execute steps 3b)-3e) to sequentially obtain each pixel's local correntropy matrix representation and vectorize it.
Furthermore, in step 2, the method of boundary filling 1s to replace the new filling values with the nearest original boundary values, so as to ensure that the center of the sliding window in step 3 can fall on the original image boundary. 3
Furthermore, in step 5, the SVM classifier uses logarithm-Euclidean kernel function. LU503093
Compared with the prior art, the invention has obvious advantages: Firstly, compared with the hyperspectral feature extraction algorithm based on spectral information, the local correntropy matrix constructed by the invention fully considers the space-spectrum integration characteristics of hyperspectral data. Secondly, the nonlinear spectral-spatial features can be obtained by using the invention's local correntropy matrix, so the invention's method can obtain more accurate classification results. Thirdly, because the method belongs to the manual feature design method, it does not need a large number of training samples to learn features, and has small sample complexity, which makes it more suitable for practical application.
Fig. 1 1s a flowchart of hyperspectral image classification method based on local correntropy matrix of the present invention.
In order to make the purpose, technical scheme and advantages of the present invention clearer, the technical scheme and effects of the present invention will be further described in detail below with reference to the attached drawings.
Referring to fig. 1, the implementation steps of the present invention are as follows:
Step 1, input a hyperspectral image, and reduce the dimension of the image, assuming that the dimension after dimension reduction is d,
Step 2, performing boundary filling processing on the dimension-reduced image by boundary copying;
Step 3, building a local correntropy matrix pixel by pixel, and vectorizing the local correntropy matrix as the spectral-spatial characteristics of the pixels;
Step 4, select p% pixels from the hyperspectral image as training samples (p=2 in this embodiment), and the remaining (100- 7 )% labeled pixels as test samples;
Step 5, training SVM classifier by using the spectral-spatial characteristics of training samples, and the support vector machine uses logarithmic-Euclidean kernel function;
Step 6, input the spectral-spatial characteristics of the test sample into the SVM classifier to obtain the class label of the test sample, and complete the classification.
Further, the specific implementation of step 3 1s as follows, 4
3a) Set the sliding window size toT x T, in this embodiment, 7= 15; LU503093 3b) Let the center pixel of the window be Pr and its neighboring pixels be Pi, i=2,3, T° , The cosine distance between this pixel and its surrounding pixels can be obtained by this formula: cos(p,, P,) _ (Pop) =2,3,.--,1° le, heil
Where (-)and I-12 represent vector inner product and Frobenius norm, respectively. 3c) Select the most similar first K-1 pixels (in this embodiment, K takes 125) to obtain adjacent pixels that are similar in space and spectrum; 3d) K pixels are used to construct local correntropy matrix representation. If two different spectral bands are expressed as b and b, respectively, according to the calculation formula of correntropy, the correntropy between the two spectral bands can be obtained as follows:
Corr (b, 6)=+$4 (b,,b,,) 1200) = ato Ons On by means the th spectral value in the b, spectral band, and there are K spectral values in each spectral band, that is, the spectral values of the selected K pixels in this spectral band.
The local correntropy matrix can be expressed as:
B
Moon = {Corr(b, bl},
B represents the number of spectral bands of hyperspectral images (after dimension reduction).
Each off-diagonal element in the local correntropy matrix represents the correntropy of different spectral bands, that is, the relationship between different spectral bands. 3e) Vectorize local correntropy matrix; 3f) Move the window, and circularly execute steps 3b)-3e) to sequentially obtain the local correntropy matrix representation of each pixel and vectorize it.
Furthermore, in step 2, the method of boundary filling is to replace the new filling values with the nearest original boundary values, so as to ensure that the center of the sliding window in step 3 can fall on the original image boundary.
In addition, the invention also provides a hyperspectral image classification system based on local correntropy matrix, which comprises the following modules:
The dimension reduction module is used for inputting a hyperspectral image and LU503093 carrying out dimension reduction ;
The boundary processing module is used for performing boundary filling processing on the image after dimension reduction by boundary copying;
The local correntropy matrix construction module 1s used for constructing the local correntropy matrix pixel by pixel, and vectorizing the local correntropy matrix as the spectral- spatial characteristics of the pixels;
Sample selection module, which 1s used to select p% pixels from hyperspectral images as training samples, and the remaining (100- ? )% labeled pixels as test samples;
The training module is used for training the SVM classifier by using the spectral- spatial features of the training samples;
SVM classification module, which 1s used to input the spectral-spatial features of test samples into SVM classifier to obtain the class labels of test samples and complete the classification.
The specific implementation mode of each module corresponds to each step, and the present invention will not describe it.
The effect of the invention can be further illustrated by the following simulation experiments: (1) Simulation condition
The hardware conditions of the simulation of the invention are as follows: Windows 10,
Intel 17-6500U 2.50 GHz, 16 GB memory; The software platform is: MatlabR2020b;
The selected image source for simulation is the hyperspectral image of Indian Pines.
There are 16 types of ground objects in this image, and the number of labeled samples of each type is shown in Table 1.
Table 1 Number of labeled samples of different classes
Class Number
Alfalfa 46
Corn-notill 1428
Corn-mintill 830
Corn 237
Grass-pasture 483
Grass-trees 730
Grass-pasture-mowed 28
Hay-windrowed 478
Oats 20
Soybean-notill 972 6
Soybean-mintill 2455 LU503093
Soybean-clean 593
Wheat 205
Woods 1265
Buildings-Grass-Trees-Drives 386
Stone-Steel-Towers 93
Total 10249
The method of the invention and the existing KELM, EPF, MPM-LBP, SADL and MH-
ELM (ELM based on multiple hypotheses) methods are used for simulation. (2) Simulation content and results
Table 2 Classification results of different methods
Method SCMK HybridSN deepNRD RPNet LCMR The invention
OA 89.69 83.53 73.23 8132 91.80 94.54
The classification simulation of Figure 2(a) 1s carried out by the present invention and the existing five methods, in which:
OA is Overall Accuracy;
It can be seen from the classification result diagram that the classification method of the invention has better accuracy and classification effect. Compared with the prior art, the invention has obvious advantages in solving the problem of self-adaptive learning features in hyperspectral image classification, and the sample complexity is low. 7
Claims (4)
1. A hyperspectral image classification method based on the local correntropy matrix is LYS03093 characterized by comprising the following steps: Step 1, input a hyperspectral image, and reduce the dimension of the image; Step 2, performing boundary filling processing on the dimension-reduced image by boundary copying; Step 3, building a local correntropy matrix pixel by pixel and vectorizing the local correntropy matrix as the spectral-spatial features of the pixels; Step 4, selecting p% pixels from the hyperspectral image as training samples and the remaining (100- p )% labeled pixels as test samples; Step 5, training the SVM classifier by using the spectral-spatial features of the training samples; Step 6, input the spectral-spatial features of the test sample into the SVM classifier to obtain the class label of the test sample, and complete the classification.
2. The hyperspectral image classification method based on local correntropy matrix according to claim 1 is characterized in that: The specific implementation of step 3 is as follows, 3a) Set the sliding window size to T x T; 3b) Let the center pixel of the window bep1,and its neighboring pixels bepi, i =2, 3, ... ,
T>. The following formula can obtain the cosine distance between this pixel and its surrounding pixels: cos(pr.p) = APP) =2,3,.-, 7? [Pia (Pi, Where (-)and |-l2 represent vector inner product and Frobenius norm, respectively; 3c) Selecting the most similar first K-1 pixels to obtain adjacent pixels which are similar in space and spectrum; 3d) K pixels are used to construct a correntropy matrix. If two different spectral bands are expressed as b, and b, respectively, according to the calculation formula of correntropy, the correntropy between the two spectral bands can be obtained as follows: 1 £ Corr(b,,b,) =x 2k (5,59) Where, b,, represents the nth spectral value in the b, spectral band, and there are K spectral values in each spectral band, that is, the spectral values of the selected K pixels in this spectral band, then the local correntropy matrix can be expressed as: 8
M. = (Corrib) LU503093 i, j=l Where, B represents the number of spectral bands of hyperspectral image after dimension reduction, and each off-diagonal element in the local correntropy matrix represents the correntropy of different spectral bands, that is, the relationship between different spectral bands; 3e) Vectorize local correntropy matrix; 3f) Move the window, and circularly execute steps 3b)-3e) to sequentially obtain each pixel's local correntropy matrix representation and vectorize it.
3. The hyperspectral image classification method based on local correntropy matrix according to claim 2 is characterized in that: In step 2, the boundary filling method is to replace the new filling values with the nearest original boundary values, so as to ensure that the center of the sliding window in step 3 can fall on the original image boundary.
4. The hyperspectral image classification method based on local correntropy matrix according to claim 1 is characterized in that: In step 5, SVM classifier uses logarithm- Euclidean kernel function. 9
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
LU503093A LU503093B1 (en) | 2022-11-21 | 2022-11-21 | A Hyperspectral Image Classification Method and System Based on Local Correntropy Matrix |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
LU503093A LU503093B1 (en) | 2022-11-21 | 2022-11-21 | A Hyperspectral Image Classification Method and System Based on Local Correntropy Matrix |
Publications (1)
Publication Number | Publication Date |
---|---|
LU503093B1 true LU503093B1 (en) | 2024-05-21 |
Family
ID=91129039
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
LU503093A LU503093B1 (en) | 2022-11-21 | 2022-11-21 | A Hyperspectral Image Classification Method and System Based on Local Correntropy Matrix |
Country Status (1)
Country | Link |
---|---|
LU (1) | LU503093B1 (en) |
-
2022
- 2022-11-21 LU LU503093A patent/LU503093B1/en active IP Right Grant
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Chen et al. | Multi-scale spatial and channel-wise attention for improving object detection in remote sensing imagery | |
Song et al. | Hyperspectral image classification based on KNN sparse representation | |
CN106503739A (en) | The target in hyperspectral remotely sensed image svm classifier method and system of combined spectral and textural characteristics | |
CN108229551B (en) | Hyperspectral remote sensing image classification method based on compact dictionary sparse representation | |
Xiang et al. | Hyperspectral anomaly detection by local joint subspace process and support vector machine | |
Xu et al. | Spectral–spatial classification of hyperspectral image based on low-rank decomposition | |
CN109961096B (en) | Multimode hyperspectral image migration classification method | |
Boggavarapu et al. | A new framework for hyperspectral image classification using Gabor embedded patch based convolution neural network | |
CN110728197B (en) | Single-tree-level tree species identification method based on deep learning | |
Jia et al. | Spatial-spectral-combined sparse representation-based classification for hyperspectral imagery | |
CN113095409A (en) | Hyperspectral image classification method based on attention mechanism and weight sharing | |
CN114972885B (en) | Multi-mode remote sensing image classification method based on model compression | |
CN109034213B (en) | Hyperspectral image classification method and system based on correlation entropy principle | |
CN112818920B (en) | Double-temporal hyperspectral image space spectrum joint change detection method | |
Wang et al. | Urban building extraction from high-resolution remote sensing imagery based on multi-scale recurrent conditional generative adversarial network | |
Tamilarasi et al. | Automated building and road classifications from hyperspectral imagery through a fully convolutional network and support vector machine | |
Christophe et al. | Open source remote sensing: Increasing the usability of cutting-edge algorithms | |
Hong et al. | Object-oriented multiscale deep features for hyperspectral image classification | |
CN115457311A (en) | Hyperspectral remote sensing image band selection method based on self-expression transfer learning | |
Gao et al. | SSC-SFN: spectral-spatial non-local segment federated network for hyperspectral image classification with limited labeled samples | |
CN111079807A (en) | Ground object classification method and device | |
CN105513079A (en) | Detection method for large-scale time sequence remote sensing image change area | |
LU503093B1 (en) | A Hyperspectral Image Classification Method and System Based on Local Correntropy Matrix | |
CN109460788B (en) | Hyperspectral image classification method based on low-rank-sparse information combination network | |
Zhang et al. | Spatio-temporal subpixel mapping with cloudy images |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
FG | Patent granted |
Effective date: 20240521 |