CN114332534B - Hyperspectral image small sample classification method - Google Patents
Hyperspectral image small sample classification method Download PDFInfo
- Publication number
- CN114332534B CN114332534B CN202111641407.9A CN202111641407A CN114332534B CN 114332534 B CN114332534 B CN 114332534B CN 202111641407 A CN202111641407 A CN 202111641407A CN 114332534 B CN114332534 B CN 114332534B
- Authority
- CN
- China
- Prior art keywords
- pixel
- sample
- hyperspectral image
- super
- category
- 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
- 238000000034 method Methods 0.000 title claims abstract description 34
- 238000012549 training Methods 0.000 claims abstract description 54
- 238000001228 spectrum Methods 0.000 claims abstract description 37
- 230000003595 spectral effect Effects 0.000 claims description 27
- 238000013145 classification model Methods 0.000 claims description 17
- 238000012360 testing method Methods 0.000 claims description 9
- 238000005070 sampling Methods 0.000 claims description 4
- 238000012706 support-vector machine Methods 0.000 claims description 4
- 238000007635 classification algorithm Methods 0.000 claims description 3
- 238000013135 deep learning Methods 0.000 claims description 3
- 230000004927 fusion Effects 0.000 claims description 3
- 238000003384 imaging method 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
- 238000004422 calculation algorithm Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000010521 absorption reaction Methods 0.000 description 1
- 238000007792 addition Methods 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000000701 chemical imaging Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 238000003331 infrared imaging Methods 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000008092 positive effect Effects 0.000 description 1
- 238000004451 qualitative analysis Methods 0.000 description 1
- 238000004445 quantitative analysis Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000013526 transfer learning Methods 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a hyperspectral image small sample classification method, which adopts a method for searching superpixels, marks all pixels in the superpixels to which a training sample belongs as the same kind, and uses the same kind as an expansion sample to expand the small sample for the first time, and adopts a pairwise pairing method to expand the sample number for the second time on the basis of the first expansion, further expands the sample number, and increases the spectrum information of a single sample by using a pixel pair method, thereby solving the technical problems of low quantity, difficult acquisition and low classification precision of the existing hyperspectral image training sample, improving the sample number, further improving the classification accuracy, and improving the spectrum information contained by the single sample in a pixel pair mode, and further improving the classification accuracy.
Description
Technical Field
The invention belongs to the technical field of hyperspectral image classification, and particularly relates to a hyperspectral image small sample classification method.
Background
The hyperspectral remote sensing technology integrates spectrum and imaging, and can obtain image information and spectrum information at the same time. The spectrum information has a plurality of spectrum bands, and each pixel has a spectrum with continuous and higher resolution. The characteristic of hyperspectral map integration improves the accuracy of the hyperspectral map integrated method applied to the characterization of the morphology of the ground object and the qualitative and quantitative analysis of the ground object and the target, so that the hyperspectral imaging technology has wide application in the aspects of resource exploration, ecological environment monitoring, land coverage classification, target identification and the like.
In hyperspectral image practical applications, there are usually only a small number of available training samples, and the acquisition of the training samples requires collection according to experience or field investigation, which is time-consuming, laborious and expensive. The limited training samples available for hyperspectral images can lead to an inability to improve classification accuracy. The problem of model discomfort caused by the limited training sample of the hyperspectral image is solved through an algorithm and a strategy, so that the classification accuracy is improved, and the method is a hot spot for researches of researchers at present.
Aiming at the problem of limited hyperspectral image training samples, various small sample strategies are proposed in the prior art, including data expansion, transfer learning, active learning and the like, wherein the data expansion is most widely applied.
The data expansion is an effective method for solving the problem of small samples, and the sample expansion is realized by creating new training samples from given known samples, so that the accuracy of classifying the small samples of the hyperspectral image is improved to a certain extent, but the problem of classifying the small samples is still the focus of the hyperspectral image research, especially the problem of classifying the very small samples.
Disclosure of Invention
The invention aims to provide a hyperspectral image small sample classification method, which adopts a superpixel method to perform primary expansion on small samples, adopts a pairwise pairing mode to perform secondary expansion on the basis of primary expansion, integrates the advantages of superpixel and pixel pairs, remarkably improves the number of training samples, solves the technical problem that the existing hyperspectral image training samples are limited, and improves the classification accuracy.
The invention is realized by adopting the following technical scheme:
the method for classifying the hyperspectral image small samples comprises the following steps:
1, randomly selecting N training samples from a hyperspectral image;
2, performing a first expansion on the N training samples by adopting the following steps:
1) Selecting a center pixel in an equidistant sampling mode according to the number of super pixels;
2) For each neighborhood of center pixels, D is employed all_sam ＝D spec_sam +rD spat Calculating first spatial spectrum distances between each pixel and the central pixel, selecting a first S-S pixel set corresponding to the first spatial spectrum distances as a super pixel, and if the super pixel contains a training sample, marking the pixel category in the super pixel as a category consistent with the contained training sample;
to calculate the spectral difference using the spectral angle, p i Spectral information for the ith pixel in the neighborhood, p c Spectral information for the center pixel c;
the neighborhood is 2s x 2s pixels surrounding one central pixel; d (D) spat R is the weight for balancing the spectrum information and the space information;
3) For each neighborhood of center pixels, D is employed all_pca ＝D spec_pca +rD spat Calculating a second spatial spectrum distance between each pixel in a neighborhood and the central pixel, selecting a pixel set corresponding to the first S x S second spatial spectrum distances as a super pixel, and if the super pixel contains a training sample, marking the pixel category in the super pixel as a category consistent with the contained training sample;
to calculate the spectral difference using the principal components, q i1 ，q i2 ，q i3 The first three principal components, q, respectively, of the ith pixel in a neighborhood c1 ，q c2 ，q c3 The first three principal components of the center pixel c, respectively;
4) Fusing the super pixels obtained by calculating the first spatial spectrum distance with the super pixels obtained by calculating the second spatial spectrum distance to obtain a first expansion sample;
3, performing pairwise pairing on the samples subjected to the first expansion to perform second expansion;
4, establishing a hyperspectral image classification model by adopting a training sample after two expansion;
and 5, predicting the category of the unknown hyperspectral image sample by adopting the established hyperspectral image classification model.
Further, in step 2, the first extended sample is obtained by fusing the superpixel obtained by calculating the first spatial spectrum distance and the superpixel obtained by calculating the second spatial spectrum distance, including:
c is the attribution category, j is each sample of the hyperspectral image, lable1 represents the attribution category of the extended sample obtained by calculating the first spectral distance, and Lable2 represents the attribution category of the extended sample obtained by calculating the second spectral distance.
Further, the step 3 specifically includes: lable of the expanded sample fusion(j) And performing pairwise pairing, marking as a category if two training samples belong to the same category, otherwise discarding.
In step 4, a hyperspectral image classification model is established by adopting a deep learning, support vector machine, extreme learning machine or AdaBoost classification algorithm; wherein, the discrimination standard adopts total classification precision, average classification precision or Kappa coefficient.
In step 5, the pixels in the test sample and the n-n neighborhood of the test sample are formed into pixel pairs in a pairwise pairing mode, the pixel pairs are input into the hyperspectral image classification model, and the classification with the largest number is selected as the final classification by the prediction result through a voting strategy.
Compared with the prior art, the invention has the advantages and positive effects that: in the hyperspectral image small sample classification method provided by the invention, a method for searching superpixels is adopted, all pixels in the superpixels to which one training sample belongs are marked as the same type, the first expansion is carried out on the small samples as the expansion samples, the second expansion is carried out by adopting a pairwise pairing method on the basis of the first expansion, the number of the samples is further expanded, the spectral information of a single sample is increased according to the pixel pair mode, the technical problem that the number of the existing hyperspectral image training samples is low is solved, the number of the samples is increased, the classification accuracy is improved, the spectral information contained in the single sample is also increased in a pixel pair mode, and the classification accuracy is further improved.
Other features and advantages of the present invention will become more apparent from the following detailed description of embodiments of the present invention, which is to be read in connection with the accompanying drawings.
Drawings
FIG. 1 is a flow chart of a method for classifying small samples of hyperspectral images according to the present invention;
fig. 2 is a diagram illustrating an embodiment of a classification method for small samples of hyperspectral images according to the present invention.
Detailed Description
The following describes the embodiments of the present invention in further detail with reference to the drawings.
The super-pixel is a method based on image segmentation, and the region blocks formed by adjacent pixels with similar color, brightness, texture and other characteristics are formed, and in the hyperspectral ground object remote sensing image, the probability that the adjacent pixels are of the same kind is very high, so that the image can be segmented into sub-regions with the same quality as much as possible by adopting the super-pixel method.
The invention aims to solve the problem of low classification accuracy caused by small sample number in the existing hyperspectral image classification, combines a super-pixel thought to expand samples, further combines a pixel pair mode to expand the samples and simultaneously improves the spectrum information of a single sample, so that the accuracy of hyperspectral image classification is remarkably improved in the modes of improving the sample number and the spectrum information content of the single sample.
As shown in FIG. 1, the method for classifying the hyperspectral image small samples provided by the invention comprises the following steps:
step S1: n training samples are randomly selected from the hyperspectral image.
As shown in fig. 2, an example of a hyperspectral image in which N training samples are randomly selected, as indicated by the white dots in the figure.
Step S2: and selecting the center pixel by adopting an equidistant sampling mode according to the number of the super pixels.
Taking the number of super pixels as M as an example, selecting M center pixels in an equidistant sampling mode.
Step S3: for each neighborhood of center pixels, a spatial distance of each pixel from the center pixel is calculated.
The neighborhood of each center pixel is 2s pixels surrounding the center pixel, that is, a search for similar pixels is performed in the region 2s around the center of the super pixel.
According toCalculating a spatial distance from an ith pixel to a center pixel c, where (x i ，y i ) Is the spatial coordinates of the ith pixel, (x) c ，y c ) The spatial coordinates of the center pixel c, d is the diagonal length of the neighborhood.
Step S41: the spectral difference is calculated at a spectral angle for each neighborhood of center pixels.
For each neighborhood of center pixels, use is made ofCalculating a spectral difference, wherein p i Spectral information for the ith pixel in the neighborhood, p c Is the spectral information of the center pixel c.
Step S42: for a neighborhood of each center pixel, a first spatial spectrum distance of each pixel from the center pixel is calculated.
By D all_sam ＝D spec_sam +rD spat And calculating a first spatial spectrum distance between each pixel and the central pixel, wherein r is a weight for balancing the spectrum information and the spatial information.
Step S43: and selecting the pixel set corresponding to the first spatial spectrum distances of the previous S as a super pixel, and if the super pixel contains a training sample, marking the pixel types in the super pixel as the types consistent with the contained training sample.
And selecting the pixel set corresponding to the first S first spatial spectrum distances as a superpixel aiming at the neighborhood of each central pixel to obtain a first superpixel set containing M superpixels.
If a training sample is included in one super-pixel in the M super-pixels, and the class of the training sample is assumed to be X, all pixel classes in the super-pixel are marked as class X consistent with the included training sample.
Step S51: the spectral difference is calculated in principal components for each neighborhood of the center pixel.
For each neighborhood of center pixels, use is made of
Calculating a spectral difference, wherein q i1 ，q i2 ，q i3 The first three principal components, q, respectively, of the ith pixel in a neighborhood c1 ，q c2 ，q c3 The first three principal components of the center pixel c, respectively.
Step S52: for each neighborhood of the center pixel, a second spatial spectrum distance of each pixel from the center pixel is calculated.
For each neighborhood of center pixels, D is employed all_pca ＝D spec_pca +rD spat A second spatial spectral distance is calculated for each pixel from the center pixel.
Step S53: and selecting the pixel set corresponding to the first S-S second spatial spectrum distances as a super pixel, and if the super pixel contains training samples, marking the pixel types in the super pixel as the types consistent with the contained training samples.
And selecting the pixel set corresponding to the first S times S second spatial spectrum distances as a superpixel aiming at the neighborhood of each central pixel, and obtaining a second superpixel set containing M superpixels.
If a training sample is included in one super-pixel in the M super-pixels, and the class of the training sample is assumed to be Y, all pixel classes in the super-pixel are marked as a class Y consistent with the included training sample.
Step S6: and fusing the super pixels obtained by calculating the first spatial spectrum distance with the super pixels obtained by calculating the second spatial spectrum distance to obtain a first expansion sample.
The extended sample obtained according to the first superpixel set obtained in step S43 and the extended sample obtained according to the second superpixel set obtained in step S53 are fused in the following manner:
wherein C is the attribution category, j is each sample of the hyperspectral image, lable1 represents the attribution category of the extended sample obtained by calculating the first spectral distance, that is, the attribution category of the extended sample obtained in step S43, and Lable2 represents the attribution category of the extended sample obtained by calculating the second spectral distance, that is, the attribution category of the extended sample obtained in step S53.
Through the method, the pixels with the consistent categories in the super pixels obtained by two methods are reserved, and the pixels with the inconsistent categories are discarded, so that a sample set after the first expansion is implemented is obtained.
Step S7: and carrying out pairwise pairing on the samples subjected to the first expansion for the second expansion.
Lable of the expanded sample fusion(j) Pairing every two samples, and carrying out class assignment on the paired samples according to the class of the paired samples, namely marking the paired samples as a class if the paired training samples belong to the same class, otherwise discarding the paired training samples; specifically, the paired pixel pairs are assigned categories according to the following rules:
wherein S is ij ＝[s i ,s j ]，s i Sum s j Is two paired training samples.
The method of pairwise pairing is adopted to expand the once expanded sample for the second time, and the single training sample is formed in the form of pixel pairs, so that the spectral information quantity of the single sample is improved, and the classification accuracy is improved.
Step S8: and (5) establishing a hyperspectral image classification model by adopting the training samples after the two expansion.
Establishing a hyperspectral image classification model by adopting a deep learning, support vector machine, extreme learning machine or AdaBoost classification algorithm; wherein, the discrimination standard adopts total classification precision, average classification precision or Kappa coefficient; the total classification accuracy reflects the frequency of the classification result of the random sample consistent with the classification of the real mark, the average classification accuracy represents the average value of the percentage of each type to be correctly classified, and the Kappa coefficient reflects the similarity degree of the classified image and the real mark image and also represents the overall classification error of the image.
Step S9: and predicting the category of the unknown hyperspectral image sample by using the established hyperspectral image classification model.
The method comprises the steps of building a hyperspectral image classification model by using a training sample after super-pixel expansion and pixel pair expansion, during testing, using a super-pixel pair mode by a test sample, forming pixel pairs by the test sample and pixels in an n-n adjacent area of the test sample in a pairwise pairing mode, namely forming n-1 pixel pairs, inputting the n-1 prediction results into the hyperspectral image classification model, and selecting the class with the largest voting number as the final class of the test sample through a voting strategy.
According to the invention, the expansion of the original small sample is realized by the super-pixel pair method, an accurate and stable hyperspectral image classification model is established under the condition of the small sample, the problems of complex acquisition, time and labor waste and high cost of the existing small sample are solved, a good classification effect can be achieved under the condition of the small sample and the extremely small sample, the cost is effectively reduced, the manpower and material resources are saved, the accuracy of the classification model is greatly improved, the classification and identification of hyperspectral images can be rapidly, nondestructively and accurately realized, and the effect realized by the invention is described by using data of a specific embodiment.
The method comprises the following steps:
1. and acquiring a hyperspectral image and selecting a training sample.
Hyperspectral image data are exemplified by the Indian pins dataset. The Indian Pines dataset was collected by an onboard visible infrared imaging spectrometer AVIRIS sensor, imaging an Indian pine tree in Indiana in 1992, and intercepting 145X 145 as hyperspectral image classification test data for 16 classes. The wavelength range is 0.4-2.5um, which is formed by imaging in 220 continuous wave bands, removing the 104 th-108 th, 150 th-163 th and 220 th wave bands of the covering water absorption area, and reducing the band number to 200.
N training samples are randomly selected from hyperspectral image data, and the category and statistical table of Indian pins are shown in table 1, and the category distribution is shown in figure 2.
List one
2. Performing two expansion on the original training sample by adopting a super-pixel and pixel pair method
The number of super pixels is 400, and the accuracy of the fused expanded error training samples, the expanded total training samples and the expanded training samples is shown as a second table.
Watch II
Taking 5-sample Indian pins as an example, after pairwise pairing to form super-pixel pairs, the new training samples are 1961 (1961-1) numbers which are 48044 times of the original training samples.
3. And establishing a hyperspectral image classification model, and predicting the class of the unknown hyperspectral image sample.
Taking a support vector machine algorithm as an example, establishing a hyperspectral image classification model, and taking a small sample hyperspectral classification result as a table III:
watch III
As can be seen from the third table, the classification result of the hyperspectral image under the small sample is obviously improved by the super-pixel pair method.
It should be noted that the above description is not intended to limit the invention, but rather the invention is not limited to the above examples, and that variations, modifications, additions or substitutions within the spirit and scope of the invention will be within the scope of the invention.
Claims (5)
1. A method for classifying a small sample of a hyperspectral image, comprising:
1, randomly selecting N training samples from a hyperspectral image;
2, performing a first expansion on the N training samples by adopting the following steps:
1) Selecting a center pixel in an equidistant sampling mode according to the number of super pixels;
2) For each neighborhood of center pixels, D is employed all_sam ＝D spec_sam +rD spat Calculating first spatial spectrum distances between each pixel and the central pixel, selecting a first S-S pixel set corresponding to the first spatial spectrum distances as a super pixel, and if the super pixel contains a training sample, marking the pixel category in the super pixel as a category consistent with the contained training sample;
to calculate the spectral difference using the spectral angle, p i Spectral information for the ith pixel in the neighborhood, p c Spectral information for the center pixel c;
the neighborhood is 2s x 2s pixels surrounding one central pixel; d (D) spat R is the weight for balancing the spectrum information and the space information;
3) For each neighborhood of center pixels, D is employed all_pca ＝D spec_pca +rD spat Computing each of a neighborhoodThe second spatial spectrum distance between the pixel and the central pixel, selecting the pixel set corresponding to the first S times S second spatial spectrum distances as a super pixel, and if the super pixel contains a training sample, marking the pixel category in the super pixel as a category consistent with the contained training sample;
to calculate the spectral difference using the principal components, q i1 ，q i2 ，q i3 The first three principal components, q, respectively, of the ith pixel in a neighborhood c1 ，q c2 ，q c3 The first three principal components of the center pixel c, respectively;
4) Fusing the super pixels obtained by calculating the first spatial spectrum distance with the super pixels obtained by calculating the second spatial spectrum distance to obtain a first expansion sample;
3, performing pairwise pairing on the samples subjected to the first expansion to perform second expansion;
4, establishing a hyperspectral image classification model by adopting a training sample after two expansion;
and 5, predicting the category of the unknown hyperspectral image sample by adopting the established hyperspectral image classification model.
2. The method of classifying small samples of hyperspectral images according to claim 1, wherein in step 2, the first time of the expanded samples is obtained by fusing the superpixels obtained by calculating the first spatial spectrum distance with the superpixels obtained by calculating the second spatial spectrum distance, comprising:
for the attribution category, j is each sample of the hyperspectral image, lable1 represents the attribution category of the extended sample obtained by calculating the first spectral distance, lable2 represents the attribution category of the extended sample obtained by calculating the second spectral distance.
3. The method for classifying hyperspectral image small samples as claimed in claim 2, wherein the step 3 specifically comprises:
lable of the expanded sample fusion(j) And performing pairwise pairing, marking as a category if two training samples belong to the same category, otherwise discarding.
4. The method for classifying small samples of hyperspectral images according to claim 1, wherein in step 4, a hyperspectral image classification model is built by using a deep learning, support vector machine, extreme learning machine, or AdaBoost classification algorithm; wherein, the discrimination standard adopts total classification precision, average classification precision or Kappa coefficient.
5. The method for classifying small samples of hyperspectral images according to claim 1, wherein in step 5, the pixels in the neighborhood of the test sample and the pixel pairs are paired in pairs, the hyperspectral image classification model is input, and the most number of categories of the prediction result are selected as the final categories through voting strategies.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111641407.9A CN114332534B (en) | 2021-12-29 | 2021-12-29 | Hyperspectral image small sample classification method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111641407.9A CN114332534B (en) | 2021-12-29 | 2021-12-29 | Hyperspectral image small sample classification method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114332534A CN114332534A (en) | 2022-04-12 |
| CN114332534B true CN114332534B (en) | 2024-03-29 |
Family
ID=81016688
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111641407.9A Active CN114332534B (en) | 2021-12-29 | 2021-12-29 | Hyperspectral image small sample classification method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114332534B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115861720B (en) * | 2023-02-28 | 2023-06-30 | 人工智能与数字经济广东省实验室(广州) | Small sample subclass image classification and identification method |
| CN117237743B (en) * | 2023-11-09 | 2024-02-27 | 深圳爱莫科技有限公司 | Small sample quick-elimination product identification method, storage medium and processing equipment |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2016091017A1 (en) * | 2014-12-09 | 2016-06-16 | 山东大学 | Extraction method for spectral feature cross-correlation vector in hyperspectral image classification |
| WO2018045626A1 (en) * | 2016-09-07 | 2018-03-15 | 深圳大学 | Super-pixel level information fusion-based hyperspectral image classification method and system |
| CN107909120A (en) * | 2017-12-28 | 2018-04-13 | 南京理工大学 | Based on alternative label K SVD and multiple dimensioned sparse hyperspectral image classification method |
| CN108009559A (en) * | 2016-11-02 | 2018-05-08 | 哈尔滨工业大学 | A kind of Hyperspectral data classification method based on empty spectrum united information |
| CN108985360A (en) * | 2018-06-29 | 2018-12-11 | 西安电子科技大学 | Hyperspectral classification method based on expanding morphology and Active Learning |
| CN111695467A (en) * | 2020-06-01 | 2020-09-22 | 西安电子科技大学 | Spatial spectrum full convolution hyperspectral image classification method based on superpixel sample expansion |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110717354B (en) * | 2018-07-11 | 2023-05-12 | 哈尔滨工业大学 | Super-pixel classification method based on semi-supervised K-SVD and multi-scale sparse representation |
-
2021
- 2021-12-29 CN CN202111641407.9A patent/CN114332534B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2016091017A1 (en) * | 2014-12-09 | 2016-06-16 | 山东大学 | Extraction method for spectral feature cross-correlation vector in hyperspectral image classification |
| WO2018045626A1 (en) * | 2016-09-07 | 2018-03-15 | 深圳大学 | Super-pixel level information fusion-based hyperspectral image classification method and system |
| CN108009559A (en) * | 2016-11-02 | 2018-05-08 | 哈尔滨工业大学 | A kind of Hyperspectral data classification method based on empty spectrum united information |
| CN107909120A (en) * | 2017-12-28 | 2018-04-13 | 南京理工大学 | Based on alternative label K SVD and multiple dimensioned sparse hyperspectral image classification method |
| CN108985360A (en) * | 2018-06-29 | 2018-12-11 | 西安电子科技大学 | Hyperspectral classification method based on expanding morphology and Active Learning |
| CN111695467A (en) * | 2020-06-01 | 2020-09-22 | 西安电子科技大学 | Spatial spectrum full convolution hyperspectral image classification method based on superpixel sample expansion |
Non-Patent Citations (4)
| Title |
|---|
| 基于小样本学习的高光谱遥感图像分类算法;张婧;袁细国;;聊城大学学报(自然科学版);20200804(06);4-14 * |
| 基于空谱联合协同表征的高光谱图像分类算法;刘颖;刘蕊;李大湘;杨凡超;;计算机工程与设计;20200316(03);33-34 * |
| 结合主动学习与标签传递算法的高光谱图像分类;王立国;商卉;石瑶;;哈尔滨工程大学学报;20201231(05);117-123 * |
| 融合空谱特征和集成超限学习机的高光谱图像分类;谷雨;徐英;郭宝峰;;测绘学报;20180915(09);82-93 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114332534A (en) | 2022-04-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107292339B (en) | Unmanned aerial vehicle low-altitude remote sensing image high-resolution landform classification method based on feature fusion | |
| CN106503739A (en) | The target in hyperspectral remotely sensed image svm classifier method and system of combined spectral and textural characteristics | |
| CN110263717B (en) | Method for determining land utilization category of street view image | |
| CN106339674B (en) | The Hyperspectral Image Classification method that model is cut with figure is kept based on edge | |
| CN114332534B (en) | Hyperspectral image small sample classification method | |
| CN111639587B (en) | Hyperspectral image classification method based on multi-scale spectrum space convolution neural network | |
| CN114897816B (en) | Mask R-CNN mineral particle identification and particle size detection method based on improved Mask | |
| CN106295124A (en) | Utilize the method that multiple image detecting technique comprehensively analyzes gene polyadenylation signal figure likelihood probability amount | |
| CN112699756B (en) | Hyperspectral image-based tea origin identification method and system | |
| CN110956207B (en) | Method for detecting full-element change of optical remote sensing image | |
| CN102855485B (en) | The automatic testing method of one grow wheat heading | |
| CN104820841B (en) | Hyperspectral classification method based on low order mutual information and spectrum context waveband selection | |
| CN108229551A (en) | A kind of Classification of hyperspectral remote sensing image method based on compact dictionary rarefaction representation | |
| CN111144250A (en) | Land coverage classification method fusing radar and optical remote sensing data | |
| CN116030352B (en) | Long-time-sequence land utilization classification method integrating multi-scale segmentation and super-pixel segmentation | |
| CN113033315A (en) | Rare earth mining high-resolution image identification and positioning method | |
| CN112446298A (en) | Hyperspectral nondestructive testing method for wheat scab | |
| CN107145831A (en) | Based on vector probabilistic diffusion and markov random file Hyperspectral Image Classification method | |
| CN113177548A (en) | Key area identification method for immune fixed electrophoresis | |
| CN105930863A (en) | Determination method for spectral band setting of satellite camera | |
| CN110929739B (en) | Automatic impervious surface range remote sensing iterative extraction method | |
| Wang et al. | Fast high-order sparse subspace clustering with cumulative MRF for hyperspectral images | |
| CN111882573A (en) | Cultivated land plot extraction method and system based on high-resolution image data | |
| CN111751295A (en) | Modeling method and application of wheat powdery mildew severity detection model based on imaging hyperspectral data | |
| Dhanuja et al. | Areca nut disease detection using image processing technology |
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 |