CN113111834B - Global local spectrum weight band selection method based on attention - Google Patents
Global local spectrum weight band selection method based on attention Download PDFInfo
- Publication number
- CN113111834B CN113111834B CN202110443963.9A CN202110443963A CN113111834B CN 113111834 B CN113111834 B CN 113111834B CN 202110443963 A CN202110443963 A CN 202110443963A CN 113111834 B CN113111834 B CN 113111834B
- Authority
- CN
- China
- Prior art keywords
- band
- pixel sample
- pixel
- weight matrix
- wave
- 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
- 238000001228 spectrum Methods 0.000 title abstract description 20
- 238000010187 selection method Methods 0.000 title abstract description 6
- 239000011159 matrix material Substances 0.000 claims abstract description 35
- 238000013528 artificial neural network Methods 0.000 claims abstract description 27
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 14
- 230000003595 spectral effect Effects 0.000 claims abstract description 12
- 238000000034 method Methods 0.000 claims description 13
- 238000003064 k means clustering Methods 0.000 claims description 11
- 238000007781 pre-processing Methods 0.000 claims description 7
- 238000012549 training Methods 0.000 abstract description 3
- 230000006870 function Effects 0.000 description 4
- 238000011176 pooling Methods 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 230000003247 decreasing effect Effects 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 241001270131 Agaricus moelleri Species 0.000 description 1
- 238000000701 chemical imaging Methods 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- ZZUFCTLCJUWOSV-UHFFFAOYSA-N furosemide Chemical compound C1=C(Cl)C(S(=O)(=O)N)=CC(C(O)=O)=C1NCC1=CC=CO1 ZZUFCTLCJUWOSV-UHFFFAOYSA-N 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
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
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/194—Terrestrial scenes using hyperspectral data, i.e. more or other wavelengths than RGB
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A40/00—Adaptation technologies in agriculture, forestry, livestock or agroalimentary production
- Y02A40/10—Adaptation technologies in agriculture, forestry, livestock or agroalimentary production in agriculture
Abstract
The invention discloses a wave band selection method of global local spectrum weight based on attention, which is characterized in that a pixel sample in a preprocessed hyperspectral data set is input into a fully-connected neural network based on attention to obtain a weight matrix of the pixel sample. And reconstructing hyperspectral data through a convolutional neural network, obtaining the spectral weight of each pixel sample under the condition that the loss of the reconstructed spectrum and the original spectrum is not reduced, training the spectral weight of each pixel by using a K-means algorithm, and selecting a wave band with the maximum variance in each cluster to obtain a final wave band selection result. So that the selection of the optimal band subset can be made taking into account both global and local information of the spectrum.
Description
Technical Field
The invention relates to the technical field of hyperspectral imaging, in particular to a wave band selection method of global local spectral weight based on attention.
Background
The hyperspectral camera has higher spectral resolution and more abundant information than the common standard RGB camera, and is widely applied to the situations of poor imaging environment, complex background and the like which are difficult to distinguish by human eyes at the present stage, such as the fields of observing weather by using satellite hyperspectrum, carrying out target detection by using underwater hyperspectrum and the like. Because hyperspectral has the characteristics of large data volume and high redundancy, a large amount of time is consumed in the process of acquiring, transmitting and processing data, and meanwhile, the phenomenon of Hughes is easy to cause, and the classification precision is improved and then reduced along with the increase of the number of wave bands. Band selection has proven to be effective in avoiding such problems. Band selection is an effective dimension reduction method without damaging the original physical meaning of hyperspectral spectrum, and more researchers have been focused in recent years. The method of using a depth network for band selection is not absolute, but mostly only a fully connected network or a convolution network is used, if only the fully connected network is used, the relation between global spectrums can be captured, but the characteristic that each neuron in the fully connected network is connected with all neurons in the upper layer can cause information between continuous bands to be lost, and on the contrary, only the convolution network which is more focused on local information is used, and the loss of global information is caused.
Disclosure of Invention
The invention provides a wave band selection method of global local spectrum weight based on attention so as to overcome the technical problems.
The invention discloses a wave band selection method of global local spectrum weight based on attention, which comprises the following steps:
inputting a hyperspectral dataset, and carrying out data preprocessing on the hyperspectral dataset so that all pixel samples conform to the same distribution; the pixel sample comprises a plurality of band values;
inputting the preprocessed pixel samples into a fully-connected neural network to obtain a band weight matrix of each pixel sample;
the weight matrix and pixel point data of the pixel sample are subjected to Hadamard product, and a result is input into a convolutional neural network to obtain an optimized pixel sample;
solving the optimized pixel sample and the original pixel sample through a loss function, and obtaining a weight matrix of the optimized pixel sample when the fully connected neural network and the convolution neural network are converged;
clustering the weight matrix of the optimized pixel sample through a K-means clustering algorithm; and obtaining the band index with the largest variance in each band cluster according to the clustering result.
Further, the inputting hyperspectral dataset and the data preprocessing the hyperspectral dataset so that all pixel samples conform to the same distribution includes:
inputting a PavaU hyperspectral dataset, wherein the PavaU hyperspectral dataset is represented by D, D epsilon R N×L Wherein, N is the number of pixel samples, L is the number of wave bands;
preprocessing a pixel sample in a PaviaU hyperspectral dataset by formula (1), expressed as:
where a represents a sample of pixels having a dimension of n×l.
Further, the inputting the preprocessed pixel samples into a fully connected neural network to obtain a band weight matrix of each pixel sample includes:
obtaining a weight matrix of the pixel sample by the formula (2), wherein the weight matrix is expressed as follows:
w＝softmax(f(x)) (2)
where f represents the fully-connected neural network, x represents the pixel sample, and softmax represents the softmax layer of the fully-connected neural network.
Further, the performing hadamard product on the weight matrix and the pixel point data of the pixel sample, and inputting the result into a convolutional neural network to obtain an optimized pixel sample, including:
the optimized pixel samples are obtained by equation (3):
restrction＝f(x*w) (3)
where, let z=x×w, f consist of five subfunction nests, expressed as:
further, the clustering processing of the weight matrix of the optimized pixel sample by the K-means clustering algorithm comprises the following steps: dividing the wave bands into different wave band clusters by using a K-means clustering algorithm according to Euclidean distance between the wave bands; the K-means clustering algorithm comprises the following steps: randomly selecting K wave bands from L wave bands of the weight matrix of the optimized pixel sample as initial points; randomly selecting the positions of K wave bands as the mass centers of the initial wave band clusters, and forming wave band clusters of K wave band categories; calculating the distance between the wave bands except the K wave bands and the centroid of each initial wave band cluster, and marking the wave band class as a wave band cluster corresponding to the centroid closest to the centroid of the initial wave band cluster; if the band cluster to which the bands except the K bands belong is unchanged, returning to the previous step, updating the position of the mass center in the band cluster, and then recalculating the distance between the bands except the K bands and the mass center of each initial band cluster and classifying the bands into the corresponding band cluster again; the algorithm ends when the centroid of each of the band clusters no longer changes.
Further, the obtaining the band index with the largest variance in each band cluster according to the clustering result includes:
dividing all wave bands in the weight matrix of the optimized image sample into C 1 ，C 2 ，C 3 …C k Clusters, the selected band subset is obtained through clustering, and the cluster is expressed as:
wherein,the band index represented by the largest variance in the kth band cluster.
The method comprises the steps of inputting the pixel samples in the preprocessed hyperspectral data set into a fully-connected neural network to obtain a weight matrix of the pixel samples. And reconstructing hyperspectral data through a convolutional neural network, obtaining the spectral weight of each pixel sample under the condition that the loss of the reconstructed spectrum and the original spectrum is not reduced, training the spectral weight of each pixel by using a K-means algorithm, and selecting the spectrum with the maximum variance in each cluster to obtain a final band selection result. So that the selection of the optimal band subset can be made taking into account both global and local information of the spectrum.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings may be obtained according to the drawings without inventive effort to a person skilled in the art.
FIG. 1 is a flow chart of the method of the present invention;
FIG. 2 is a schematic diagram of obtaining band values of different weights of a pixel sample through a fully connected neural network in the present invention;
FIG. 3 is a schematic representation of spectral reconstruction by convolutional neural network in accordance with the present invention;
FIG. 4 is a graph comparing the effects of the present invention with those of UBS and ONR algorithms for band selection on a Pavia U dataset;
FIG. 5 is a flowchart of the K-means algorithm in the present invention.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments of the present invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
As shown in fig. 1, the present embodiment provides a method for selecting a band based on a global local spectral weight of attention, including:
101. inputting a hyperspectral dataset, and carrying out data preprocessing on the hyperspectral dataset so that all pixel samples conform to the same distribution; the pixel sample comprises a plurality of wave band values;
specifically, the PaviaU dataset is data obtained by photographing Pavia university with a ROSIS sensor in 2002, which contains 610×340 pixels, 9 categories of interest and 103 wavelength bands, the wavelength ranges from 0.43 to 0.86 μm, and here the dataset PaviaU is taken as input, so n=610×340, l=103.
Pavia U hyperspectral dataset is denoted by D, D ε R N×L Wherein, N is the number of pixel samples, L is the number of wave bands; 90% of the pixels are randomly decimated for training and 10% of the pixels are used for verification. The pixel samples in the PaviaU hyperspectral dataset are preprocessed in order to make all the pixel samples conform to the same distribution so that the algorithm can converge quickly, and meanwhile, the internal data relativity of each sample is not lost, which is expressed by the following formula (1):
where a represents a sample of pixels having a dimension of n×l.
102. Inputting the preprocessed pixel samples into a fully-connected neural network to obtain a band weight matrix of each pixel sample;
specifically, the preprocessed pixel samples are input into the fully-connected neural network for processing, and the relatively important wave bands are extracted while the less important wave bands are filtered, so that the wave bands of interest can be given a larger weight, and the less important wave bands are given a smaller weight. As shown in fig. 2, each pixel sample is input, and finally, the band values of different weights (different colors and different weights) in each pixel sample are obtained (only different gray scales can be used for distinguishing due to the limitation of the drawing in the specification). Because the importance of each wave band is consistent for each sample when the spectrum dimensions are consistent, jump connection is added to accelerate the convergence of the algorithm, and a weight matrix w is obtained through a fully connected neural network, and the whole process can be expressed as:
w＝softmax(f(x)) (2)
where f represents the fully-connected neural network, x represents the pixel sample, and softmax represents the softmax layer of the fully-connected neural network.
103. The method comprises the steps of (1) carrying out Hadamard product on a weight matrix and pixel point data of a pixel sample, and inputting a result into a convolutional neural network to obtain an optimized pixel sample;
specifically, because the CNN network (convolutional neural network) has robustness to the information processing of the context, and meanwhile, the hyperspectral image has the characteristic of high correlation of adjacent bands, the relationship between the bands and the upper and lower adjacent bands can be easily captured by adopting the CNN network, and the network structure of the CNN reconstructed spectrum is shown in fig. 3. The input of the CNN network is obtained by carrying out Hadamard product on a weight matrix w obtained through the fully-connected neural network and original input data. In this reconstruction module, a pooling operation and an inverse pooling operation are inserted, where the pooling operation is to reduce the dimension of the spectrum raw data of the hyperspectral image, and memorize the spectrum with the most characteristic (i.e. the band with the largest band value between every two bands) at the same time, and then reconstruct new spectrum data under the precondition of maintaining the spectrum with the most characteristic in the inverse pooling operation. The process is expressed as:
restrction＝f(x*w) (3)
where z=x×w, f is nested by a plurality of functions, since it contains jump junctions, we consist of 5 sub-functions, expressed as:
obtaining parameters of formula (4) f, so that a reconstructed pixel point can be obtained after the product of the input weight and the original pixel point, the reconstructed pixel point is enabled to be as close to the original pixel point as possible, and then a loss function is given: loss= (structure-x) 2 Strategies for bringing loss as close to 0 as possible, here using learning rate reduction and network early cessation, when validating set loThe learning rate is reduced by an order of magnitude of 0.1 when ss is no longer decreasing 5 consecutive times, and the network is stopped when ss is no longer decreasing 8 consecutive times.
104. Solving the optimized pixel sample and the original pixel sample through a loss function, and obtaining a weight matrix of the optimized pixel sample when the fully connected neural network and the convolution neural network are converged;
in particular, the weight matrix may be regarded as a hyperspectral image with more pronounced spectral characteristics than the original hyperspectral image. The purpose of continuously iterating the optimized pixel sample and the original pixel sample for loss is to make the whole neural network converged, wherein the whole neural network comprises a fully-connected neural network and a convolutional neural network, and after the whole network is converged, the output of the fully-connected neural network, namely a w weight matrix, is taken out when the whole network is converged.
105. Clustering the weight matrix of the optimized pixel sample by a K-means clustering algorithm; and obtaining the band index with the largest variance in each band cluster according to the clustering result.
Specifically, the bands are divided into different clusters using a K-means clustering algorithm according to the euclidean distance between the bands, which is used here as shown in fig. 5.
K-means clustering algorithm:
randomly selecting K wave bands from L wave bands of the weight matrix of the optimized pixel sample as initial points;
randomly selecting the positions of K wave bands as the mass centers of the initial wave band clusters, and forming wave band clusters of K wave band categories;
calculating the distance between the wave bands except K wave bands and the centroid of each initial wave band cluster, and marking the wave band class as the wave band cluster corresponding to the centroid closest to the centroid of the initial wave band cluster;
if the band cluster to which the bands except the K bands belong is unchanged, returning to the previous step, updating the position of the mass center in the band cluster, and then recalculating the distance between the bands except the K bands and the mass center of each initial band cluster and classifying the distances into the corresponding band cluster again;
the algorithm ends when the centroid of each band cluster no longer changes.
Searching a wave band with the largest variance value in each cluster in the result of each cluster to obtain a final result. Since a larger band variance indicates a larger data range and a larger information amount. The specific process is as follows:
dividing all wave bands in the weight matrix of the optimized image sample into C 1 ，C 2 ，C 3 …C k Clusters, the selected band subset is obtained through clustering, and the cluster is expressed as:
wherein,the band index represented by the largest variance in the kth band cluster.
As shown in fig. 4, the algorithm effect of the present invention is superior to the effect of the UBS, ONR algorithm when the number of selected bands is less than 15.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and not for limiting the same; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some or all of the technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit of the invention.
Claims (3)
1. A method for selecting a band of attention-based global local spectral weights, comprising:
inputting a hyperspectral dataset, and carrying out data preprocessing on the hyperspectral dataset so that all pixel samples conform to the same distribution; the pixel sample comprises a plurality of band values;
inputting the preprocessed pixel samples into a fully-connected neural network to obtain a band weight matrix of each pixel sample;
the weight matrix and pixel point data of the pixel sample are subjected to Hadamard product, and a result is input into a convolutional neural network to obtain an optimized pixel sample;
solving the optimized pixel sample and the original pixel sample through a loss function, and obtaining a weight matrix of the optimized pixel sample when the fully connected neural network and the convolution neural network are converged;
clustering the weight matrix of the optimized pixel sample through a K-means clustering algorithm; obtaining a band index with the largest square difference in each band cluster according to the clustering result;
the inputting hyperspectral dataset and performing data preprocessing on the hyperspectral dataset so that all pixel samples conform to the same distribution, comprising:
inputting a PavaU hyperspectral dataset, wherein the PavaU hyperspectral dataset is represented by D, D epsilon R N×L Wherein, N is the number of pixel samples, L is the number of wave bands;
preprocessing a pixel sample in a PaviaU hyperspectral dataset by formula (1), expressed as:
wherein, A represents a pixel sample with dimension of N x L;
inputting the preprocessed pixel samples into a fully connected neural network to obtain a band weight matrix of each pixel sample, wherein the band weight matrix comprises the following steps:
obtaining a weight matrix of the pixel sample by the formula (2), wherein the weight matrix is expressed as follows:
w＝softmax(f(x)) (2)
wherein f represents the fully-connected neural network, x represents the pixel sample, and softmax represents the softmax layer of the fully-connected neural network;
the weight matrix and pixel point data of the pixel sample are subjected to Hadamard product, and a result is input into a convolutional neural network to obtain an optimized pixel sample, wherein the Hadamard product comprises:
the optimized pixel samples are obtained by equation (3):
restruction＝f(x*w) (3)
where, let z=x×w, f consist of five subfunction nests, expressed as:
2. the method for selecting a band of attention-based global local spectral weights according to claim 1, wherein clustering the weight matrix of the optimized pixel samples by a K-means clustering algorithm comprises:
dividing the wave bands into different wave band clusters by using a K-means clustering algorithm according to Euclidean distance between the wave bands; the K-means clustering algorithm comprises the following steps:
randomly selecting K wave bands from L wave bands of the weight matrix of the optimized pixel sample as initial points;
randomly selecting the positions of K wave bands as the mass centers of the initial wave band clusters, and forming wave band clusters of K wave band categories;
calculating the distance between the wave bands except the K wave bands and the centroid of each initial wave band cluster, and marking the wave band class as a wave band cluster corresponding to the centroid closest to the centroid of the initial wave band cluster;
if the band cluster to which the bands except the K bands belong is unchanged, returning to the previous step, updating the position of the mass center in the band cluster, and then recalculating the distance between the bands except the K bands and the mass center of each initial band cluster and classifying the bands into the corresponding band cluster again;
the algorithm ends when the centroid of each of the band clusters no longer changes.
3. The method for selecting a band based on global local spectral weights according to claim 2, wherein the obtaining the band index with the largest variance in each band cluster according to the clustering result comprises:
dividing all wave bands in the weight matrix of the optimized image sample into C 1 ,C 2 ,C 3 …C k And clustering to obtain a selected band subset, wherein the selected band subset is expressed as:
wherein,the band index represented by the largest variance in the kth band cluster.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110443963.9A CN113111834B (en) | 2021-04-23 | 2021-04-23 | Global local spectrum weight band selection method based on attention |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110443963.9A CN113111834B (en) | 2021-04-23 | 2021-04-23 | Global local spectrum weight band selection method based on attention |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113111834A CN113111834A (en) | 2021-07-13 |
| CN113111834B true CN113111834B (en) | 2024-04-05 |
Family
ID=76719696
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110443963.9A Active CN113111834B (en) | 2021-04-23 | 2021-04-23 | Global local spectrum weight band selection method based on attention |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113111834B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2017066825A1 (en) * | 2015-10-19 | 2017-04-27 | National Ict Australia Limited | Spectral reconstruction |
| CN111160273A (en) * | 2019-12-31 | 2020-05-15 | 北京云智空间科技有限公司 | Hyperspectral image space spectrum combined classification method and device |
| CN111191514A (en) * | 2019-12-04 | 2020-05-22 | 中国地质大学（武汉） | Hyperspectral image band selection method based on deep learning |
| WO2021003951A1 (en) * | 2019-07-08 | 2021-01-14 | 南京信息工程大学 | Hyperspectral image classification method based on label-constrained elastic network graph model |
| CN112580480A (en) * | 2020-12-14 | 2021-03-30 | 河海大学 | Hyperspectral remote sensing image classification method and device |
-
2021
- 2021-04-23 CN CN202110443963.9A patent/CN113111834B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2017066825A1 (en) * | 2015-10-19 | 2017-04-27 | National Ict Australia Limited | Spectral reconstruction |
| WO2021003951A1 (en) * | 2019-07-08 | 2021-01-14 | 南京信息工程大学 | Hyperspectral image classification method based on label-constrained elastic network graph model |
| CN111191514A (en) * | 2019-12-04 | 2020-05-22 | 中国地质大学（武汉） | Hyperspectral image band selection method based on deep learning |
| CN111160273A (en) * | 2019-12-31 | 2020-05-15 | 北京云智空间科技有限公司 | Hyperspectral image space spectrum combined classification method and device |
| CN112580480A (en) * | 2020-12-14 | 2021-03-30 | 河海大学 | Hyperspectral remote sensing image classification method and device |
Non-Patent Citations (1)
| Title |
|---|
| 遥感影像地物分类多注意力融和U型网络法;李道纪;郭海涛;卢俊;赵传;林雨准;余东行;;测绘学报(第08期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113111834A (en) | 2021-07-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wang et al. | Adaptive spectral–spatial multiscale contextual feature extraction for hyperspectral image classification | |
| CN112232280B (en) | Hyperspectral image classification method based on self-encoder and 3D depth residual error network | |
| CN110210313B (en) | Hyperspectral remote sensing image classification method based on multi-scale PCA-3D-CNN (principal component analysis-three dimensional-CNN) space spectrum combination | |
| Farmonov et al. | Crop type classification by DESIS hyperspectral imagery and machine learning algorithms | |
| Abdalla et al. | Nutrient status diagnosis of infield oilseed rape via deep learning-enabled dynamic model | |
| CN107145836B (en) | Hyperspectral image classification method based on stacked boundary identification self-encoder | |
| Dou et al. | Band selection of hyperspectral images using attention-based autoencoders | |
| Zhou et al. | Deep feature alignment neural networks for domain adaptation of hyperspectral data | |
| CN111144214B (en) | Hyperspectral image unmixing method based on multilayer stack type automatic encoder | |
| CN110309811A (en) | A kind of hyperspectral image classification method based on capsule network | |
| Arun et al. | Capsulenet-based spatial–spectral classifier for hyperspectral images | |
| CN109190511B (en) | Hyperspectral classification method based on local and structural constraint low-rank representation | |
| CN111160392A (en) | Hyperspectral classification method based on wavelet width learning system | |
| CN115512162A (en) | Terrain classification method based on attention twin network and multi-mode fusion features | |
| CN113052130B (en) | Hyperspectral image classification method based on depth residual error network and edge protection filtering | |
| CN110781832B (en) | Hyperspectral image abnormal target detection method based on combined depth confidence network | |
| CN112633301A (en) | Traditional Chinese medicine tongue image greasy feature classification method based on depth metric learning | |
| CN113111834B (en) | Global local spectrum weight band selection method based on attention | |
| CN112784907A (en) | Hyperspectral image classification method based on spatial spectral feature and BP neural network | |
| CN111062888A (en) | Hyperspectral image denoising method based on multi-target low-rank sparsity and spatial-spectral total variation | |
| CN114299398B (en) | Small sample remote sensing image classification method based on self-supervision contrast learning | |
| CN116563606A (en) | Hyperspectral image classification method based on dual-branch spatial spectrum global feature extraction network | |
| Shao et al. | Dual-channel networks with optimal-band selection strategy for arbitrary cropped hyperspectral images classification | |
| CN115439431A (en) | Stack type noise reduction self-encoder based on spectrum loss function and hyperspectral anomaly detection method based on collaborative representation | |
| CN114418027A (en) | Hyperspectral image characteristic waveband selection method based on waveband attention mechanism |
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 |