CN111461087A - Hyperspectral anomaly detection method based on spectrum preserving sparse self-encoder - Google Patents

Hyperspectral anomaly detection method based on spectrum preserving sparse self-encoder Download PDF

Info

Publication number
CN111461087A
CN111461087A CN202010545143.6A CN202010545143A CN111461087A CN 111461087 A CN111461087 A CN 111461087A CN 202010545143 A CN202010545143 A CN 202010545143A CN 111461087 A CN111461087 A CN 111461087A
Authority
CN
China
Prior art keywords
encoder
hyperspectral
self
network
data
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202010545143.6A
Other languages
Chinese (zh)
Inventor
李恒
吴泽彬
魏洁
颜斌
徐洋
韦志辉
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing University of Science and Technology
Original Assignee
Nanjing University of Science and Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Nanjing University of Science and Technology filed Critical Nanjing University of Science and Technology
Priority to CN202010545143.6A priority Critical patent/CN111461087A/en
Publication of CN111461087A publication Critical patent/CN111461087A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • G06V20/13—Satellite images
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • 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
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Astronomy & Astrophysics (AREA)
  • Remote Sensing (AREA)
  • Multimedia (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a hyperspectral image anomaly detection method based on a spectrum preserving sparse self-encoder, which comprises the following steps of: solving a wavelet coefficient of the hyperspectral data; constructing a linear unmixing constrained wavelet self-coding network, replacing the inner product of a coding layer with a spectral angular distance, selecting a Relu function as an activation function of the coding layer, introducing a normalization layer and a dropout layer, adding a penalty term and a regular term in a loss function, and constructing a network for keeping a sparse self-coder based on a spectrum; inputting the hyperspectral data to be measured into a sparse self-encoder network, setting network parameters, performing unmixing operation on the network parameters to obtain background end member data, and calculating a reconstruction error to obtain required abnormal target data. The method can be used for quickly and accurately detecting the abnormal target of the hyperspectral image.

Description

Hyperspectral anomaly detection method based on spectrum preserving sparse self-encoder
Technical Field
The invention belongs to the technical field of remote sensing image processing, and particularly relates to a hyperspectral anomaly detection method based on a spectrum preserving sparse self-encoder.
Background
The hyperspectral image target detection is an important content of hyperspectral data processing and plays an important role in various fields of human society. The hyperspectral target detection can be divided into a supervision method and an unsupervised method according to whether a target signal is known or not, and the unsupervised method, namely the hyperspectral image anomaly detection method can detect an anomalous target different from a background pixel under the condition of unknown prior information, so that the extensive research of domestic and foreign personnel is obtained. The hyperspectral image has the following characteristics:
(1) the spectrum has a plurality of wave bands, and the wave bands are basically continuous.
(2) The image not only contains abundant ground feature information, but also contains spectral information.
(3) The amount of hyperspectral data is large and the amount of data redundancy is large.
(4) Spatial resolution and signal-to-noise ratio are low.
(5) There are mixed picture elements.
Therefore, the traditional anomaly detection algorithm cannot well separate the background target from the abnormal target, the detection precision is not very high, and with the development of data driving, machine learning and manifold learning, the hyperspectral remote sensing image anomaly detection technology needs to be further researched.
The sparse representation model does not need to assume the distribution of the model, the self-adaptive capacity of the learning dictionary can be improved, and the algorithm has the main ideas that 1, adjacent pixels of a hyperspectral image have spectral similarity, 2, pixels in the hyperspectral image are located in a low-dimensional subspace, and 3, the minimization problem of solving L0 norm is defined as a standard linear programming problem to recover the sparse representation of a sample.
However, the above method still has the following problems: (1) the number of iterations is calculated; (2) the detection calculation amount is large; (3) the operation and convergence speed is slow.
Disclosure of Invention
The invention aims to provide a high-spectrum image abnormal target detection method based on a spectrum preserving sparse self-encoder, which is high in speed and precision.
The technical solution for realizing the purpose of the invention is as follows: a hyperspectral anomaly detection method based on a spectrum preserving sparse self-encoder comprises the following steps:
step 1, solving a wavelet coefficient of hyperspectral data;
step 2, constructing a wavelet self-coding network of linear unmixing constraint, replacing the inner product of a coding layer with a spectral angular distance, adding a penalty term and a regular term into a loss function, and constructing a spectrum preserving sparse self-coder network based on wavelet multi-scale transformation;
and 3, inputting the hyperspectral data to be measured into the spectrum maintaining sparse self-encoder network constructed in the step 2, performing unmixing operation to obtain background end member data, and calculating a reconstruction error to obtain required abnormal target data.
Compared with the prior art, the invention has the remarkable advantages that: (1) according to the self-encoder network based on deep learning, a spectrum unmixing technology is introduced into the problem of anomaly detection, so that the interference of background information can be effectively inhibited, the influence of noise and nonlinear correlation on spectrum information is overcome, and high-level spectrum and spatial features are extracted from hyperspectral image data, so that the accuracy of anomaly detection is improved; (2) the network structure of the traditional self-encoder is modified, and the limitations such as sparse penalty term and regular term are added in the loss function, so that the learning capability of the network is more excellent, and the accuracy of abnormal detection is improved; (3) after network parameters are set, the Adam optimizer based on random gradient is used for updating model parameters, so that the problem of manually calculating very complex derivatives is solved, and the operation speed is increased.
The hyperspectral anomaly detection method based on the spectrum preserving sparse self-encoder provided by the invention is described in detail below with reference to the attached drawings.
Drawings
FIG. 1 is a flow chart of a hyperspectral anomaly detection method based on a spectrum preserving sparse self-encoder according to the invention.
Detailed Description
With reference to fig. 1, the hyperspectral anomaly detection method based on the spectrum preserving sparse self-encoder of the invention comprises the following steps:
step 1, for hyperspectral data
Figure 347311DEST_PATH_IMAGE001
Performing a' trous wavelet transform to obtain a wavelet coefficient, and unmixing the wavelet coefficient;
Figure 955010DEST_PATH_IMAGE002
wherein
Figure 7280DEST_PATH_IMAGE003
In order to be a low-frequency coefficient,
Figure 473640DEST_PATH_IMAGE004
in order to be a high-frequency coefficient,
Figure 927755DEST_PATH_IMAGE005
an end-member matrix and an abundance matrix representing the low-frequency segments,
Figure 706355DEST_PATH_IMAGE006
an end-member matrix and an abundance matrix representing the high-frequency segments,
Figure 245921DEST_PATH_IMAGE007
the number of spectral bands of the hyperspectral image,
Figure 17437DEST_PATH_IMAGE008
is the number of pixels;
since the low and high frequencies have the same abundance coefficients, there are
Figure 326058DEST_PATH_IMAGE009
；
Figure 275560DEST_PATH_IMAGE010
Is a constructed wavelet coefficient matrix ofThe expression is as follows:
Figure 36842DEST_PATH_IMAGE011
wherein the content of the first and second substances,
Figure 113514DEST_PATH_IMAGE012
and after the unmixing formula of the wavelet domain is obtained, the wavelet coefficient is used as the input of the self-encoder.
And 2, optimally designing a network structure and a loss function of the self-encoder based on a linear unmixing model, replacing the inner product of an encoding layer with a spectral angular distance SAD, selecting a Relu function as an activation function of the encoding layer, introducing a normalization layer and a dropout layer, adding a penalty term and a regular term in the loss function, and constructing a spectrum preserving sparse self-encoder network based on wavelet multi-scale transformation.
(1) In order to extract the features of the spectral curves in the network, the spectral angular distance SAD is used to replace the inner product of the coding layer to obtain more recognizable and separable hidden features:
Figure 276642DEST_PATH_IMAGE013
Figure 397045DEST_PATH_IMAGE014
is a weight matrix between the coding layer and the hidden layer, whereinKThe number of end-members is indicated,Dthe number of spectral bands representing the hyperspectral image,
Figure 911203DEST_PATH_IMAGE015
is a mixed spectral image of the light source,nindicating the number of samples.
Figure 24521DEST_PATH_IMAGE016
Normalizing SAD values to interval [0,1 ]]，
Figure 42156DEST_PATH_IMAGE017
Represent the similarity of spectra andhhigher values of (d) mean higher similarity of spectral curves.
(2) Aiming at the problems that the traditional self-encoder network adopts the sigmoid function as the activation function, which causes large calculation amount and complex reverse derivation calculation, and easily causes gradient disappearance, the invention adopts the Relu function as the activation function, thereby saving the calculation amount. And the Relu function enables the output of the neuron input value smaller than 0 to be 0, so that the network has sparsity, and the overfitting problem is relieved to a certain extent.
(3) In order to solve the problem of uncertainty of the trainable parameters of the self-encoder network, a normalization layer is introduced before a Relu function:
Figure 599039DEST_PATH_IMAGE018
wherein the content of the first and second substances,
Figure 334914DEST_PATH_IMAGE019
is composed of
Figure 750458DEST_PATH_IMAGE020
The average value of (a) of (b),
Figure 622599DEST_PATH_IMAGE021
to prevent the denominator from being a very small value of 0,
Figure 350384DEST_PATH_IMAGE022
presentation pair
Figure 573555DEST_PATH_IMAGE023
The normalization is carried out in such a way that,
Figure 28676DEST_PATH_IMAGE024
is a parameter that can be trained in a particular way,
Figure 20903DEST_PATH_IMAGE025
。
Figure 919588DEST_PATH_IMAGE026
(4) in order to solve the problems of overfitting and gradient disappearance in the network training process, a dropout layer is introduced, the hidden layer output after the activation function processing is randomly cleared, and the sparsity of the network is further increased.
Figure 364476DEST_PATH_IMAGE027
(5) In a hidden layer of the sparse self-encoder, the minimum activation value of each column is set to be 0, namely K-1 maximum activation values are reserved in the hidden layer, so that the self-encoder obtains an accurate sparsity, and K is the number of end members of the hyperspectral image.
Figure 170758DEST_PATH_IMAGE028
(6) To satisfy the constraint of abundance sum of 1, use
Figure 33803DEST_PATH_IMAGE029
Normalization is carried out on the abundance matrix through norm, and the network structure of the processed self-encoder is as follows:
Figure 837811DEST_PATH_IMAGE030
Figure 238837DEST_PATH_IMAGE031
wherein
Figure 583230DEST_PATH_IMAGE032
Is the weight matrix between the hidden layer and the decoding layer,
Figure 799317DEST_PATH_IMAGE033
for the purpose of self-encoding the output of the network,Dthe number of spectral bands representing the hyperspectral image,Kthe number of end-members is indicated,
Figure 774226DEST_PATH_IMAGE034
is a mixed spectral image of the light source,nindicating the number of samples.
(7) The original loss function uses euclidean: (
Figure 459285DEST_PATH_IMAGE035
) Norm to calculate the reconstruction error, in order to minimize the error between the input and output data, first, the spectral preserving sparse self-encoder uses a weighted logarithm term to constrain the average activation value of the hidden neuron output to a given sparse value close to 0
Figure 545053DEST_PATH_IMAGE036
Close and add it as a penalty term to the loss function; secondly, in order to emphasize the sparsity of the hidden layer, introduce
Figure 379760DEST_PATH_IMAGE029
Regularization term
Figure 525571DEST_PATH_IMAGE037
(ii) a Then, the weight matrix is aligned
Figure 432347DEST_PATH_IMAGE038
Is added with
Figure 321806DEST_PATH_IMAGE039
And the regular constraint reduces the weight and avoids overfitting. The modified loss function is:
Figure 512484DEST_PATH_IMAGE040
wherein the content of the first and second substances,
Figure 829196DEST_PATH_IMAGE041
are the weight coefficients of the different components,
Figure 223269DEST_PATH_IMAGE042
in order to be the sparsity weighting factor,
Figure 916418DEST_PATH_IMAGE043
representing hidden neuronsjThe activity of (2).
Step 3, inputting the hyperspectral data to be measured into the sparse self-encoder network constructed in the step 2, setting network parameters, performing unmixing operation on the network parameters to obtain background end member data, and calculating a reconstruction error to obtain required abnormal target data;
parameter(s)
Figure 463068DEST_PATH_IMAGE044
Are respectively set as 10-2，10-50.0035, 0.04, sparse values will be given
Figure 216261DEST_PATH_IMAGE045
Set to 0.02 and learning rate to 10-3An Adam optimization algorithm based on random gradient is adopted, an automatic derivation mechanism is combined, and model parameters are optimized through a minimum loss function.
Inputting the hyperspectral data to be measured into a sparse self-encoder network with set parameters, performing unmixing operation on the data to be measured to obtain background end member data, and performing unmixing operation on the hyperspectral image to be measured
Figure 97629DEST_PATH_IMAGE046
Subtracting the obtained background end metadata to obtain the required abnormal target data, namely:
Figure 328890DEST_PATH_IMAGE047
the flow of the hyperspectral image abnormal target detection algorithm based on the spectrum preserving sparse self-encoder is as follows:
inputting: raw hyperspectral dataset
Figure 228582DEST_PATH_IMAGE046
Weight matrix { W1,W2}, regularization parameter, parameter
Figure 152676DEST_PATH_IMAGE048
Hidden layer end element number: hiddenSize, batch size of training set: n, number of epochs of training set: m, iteration of each epoch training
And (3) outputting: abnormal pixel data
Figure 521340DEST_PATH_IMAGE049
The algorithm process is as follows:
1: data setxDividing the training set into N batchs
2：for i∈{1,2，…，epoch：M} do
3：for j∈{1,2，…iteration} do
4: calculation of step 1 for each batch do of the training set
5: performing error calculation in the step 2, and using a random gradient optimization and automatic derivation mechanism to perform model parameters
Figure 556292DEST_PATH_IMAGE050
Perform the update
6：end for
7：end for
8：end for
9: outputting ROC curve and AUC value to the obtained abnormal detection result
In summary, the method is different from a traditional anomaly detection method, a traditional self-encoder network structure is optimally designed based on a self-encoder network of deep learning, the limitations such as sparse penalty terms and regular terms are added in a loss function, so that the learning capability of the network is more excellent, hyperspectral data to be detected is input into a modified network, a spectrum unmixing technology is introduced into the anomaly detection problem, the interference of background information is effectively inhibited, the influence of noise and nonlinear correlation on the spectrum information is overcome, high-level spectrum and space features are extracted from the hyperspectral image data, and the accuracy of anomaly detection is improved; after network parameters are set, model parameters are updated through a random gradient optimization method, and the abnormal detection calculation speed is increased.

Claims (6)

1. A hyperspectral anomaly detection method based on a spectrum preserving sparse self-encoder is characterized by comprising the following steps:
step 1, solving a wavelet coefficient of hyperspectral data;
step 2, constructing a wavelet self-coding network of linear unmixing constraint, replacing the inner product of a coding layer with a spectral angular distance, adding a penalty term and a regular term into a loss function, and constructing a spectrum preserving sparse self-coder network based on wavelet multi-scale transformation;
and 3, inputting the hyperspectral data to be measured into the spectrum maintaining sparse self-encoder network constructed in the step 2, performing unmixing operation to obtain background end member data, and calculating a reconstruction error to obtain required abnormal target data.
2. The hyperspectral anomaly detection method based on the spectrum preserving sparse self-encoder according to claim 1, wherein the method for solving the wavelet coefficient of the hyperspectral data in the step 1 is as follows:
for high spectral data
Figure 125848DEST_PATH_IMAGE001
Performing a' trous wavelet transform to obtain a wavelet coefficient, and unmixing the wavelet coefficient;
Figure 29082DEST_PATH_IMAGE002
wherein
Figure 568648DEST_PATH_IMAGE003
In order to be a low-frequency coefficient,
Figure 44891DEST_PATH_IMAGE004
in order to be a high-frequency coefficient,
Figure 478146DEST_PATH_IMAGE005
an end-member matrix and an abundance matrix representing the low-frequency segments,
Figure 693227DEST_PATH_IMAGE006
an end member matrix and an abundance matrix representing the high frequency segments;
since the low and high frequencies have the same abundance coefficients, there are
Figure 343258DEST_PATH_IMAGE007
；
Figure 324989DEST_PATH_IMAGE008
The expression of the constructed wavelet coefficient matrix is as follows:
Figure 114216DEST_PATH_IMAGE009
wherein the content of the first and second substances,
Figure 483886DEST_PATH_IMAGE010
and after the unmixing formula of the wavelet domain is obtained, the wavelet coefficient is used as the input of the self-encoder.
3. The hyperspectral anomaly detection method based on the spectrum preserving sparse self-encoder according to claim 1 is characterized in that the specific method in the step 2 is as follows:
the method comprises the steps of optimally designing a network structure and a loss function of the self-encoder based on a linear unmixing model, replacing an inner product of an encoding layer with a spectral angular distance, selecting a Relu function as an activation function of the encoding layer, introducing a normalization layer and a dropout layer, adding a punishment term and a regular term in the loss function, and constructing a spectrum preserving sparse self-encoder network based on wavelet multi-scale transformation.
4. The hyperspectral anomaly detection method based on the spectrum preserving sparse self-encoder according to claim 3 is characterized in that a spectrum preserving sparse self-encoder network based on wavelet multi-scale transformation is constructed, and the method specifically comprises the following steps:
using spectral angular distancesSADInstead of inner products of coding layersObtaining hidden features:
Figure 621213DEST_PATH_IMAGE011
Figure 406636DEST_PATH_IMAGE012
is a weight matrix between the coding layer and the hidden layer, whereinKThe number of end-members is indicated,Dthe number of spectral bands representing the hyperspectral image,
Figure 315948DEST_PATH_IMAGE013
is a mixed spectral image of the light source,nrepresents the number of samples;
Figure 59782DEST_PATH_IMAGE014
normalizing SAD values to interval [0,1 ]]，
Figure 949984DEST_PATH_IMAGE015
Representing the similarity of the spectra;
adopting a Relu function as an activation function, wherein the Relu function enables the output of the neuron input value smaller than 0 to be 0;
a normalization layer is introduced before the Relu function:
Figure 476780DEST_PATH_IMAGE016
wherein the content of the first and second substances,
Figure 614501DEST_PATH_IMAGE017
is composed of
Figure 233963DEST_PATH_IMAGE018
The average value of (a) of (b),
Figure 440822DEST_PATH_IMAGE019
to prevent the denominator from being a very small value of 0,
Figure 535424DEST_PATH_IMAGE020
presentation pair
Figure 527651DEST_PATH_IMAGE021
The normalization is carried out in such a way that,
Figure 19812DEST_PATH_IMAGE022
is a parameter that can be trained in a particular way,
Figure 746591DEST_PATH_IMAGE023
；
Figure 224977DEST_PATH_IMAGE024
introducing a dropout layer, and randomly resetting the output of the hidden layer after the activation function processing:
Figure 789819DEST_PATH_IMAGE025
in a hidden layer of a sparse self-encoder, setting the minimum activation value of each column as 0:
Figure 531510DEST_PATH_IMAGE026
i.e. remain in hidden layersK-1 of the maximum activation values,Kthe number of the end members of the hyperspectral image;
by using
Figure 945917DEST_PATH_IMAGE027
Normalization is carried out on the abundance matrix through norm, and the network structure of the processed self-encoder is as follows:
Figure 555890DEST_PATH_IMAGE028
Figure 991551DEST_PATH_IMAGE029
wherein
Figure 684569DEST_PATH_IMAGE030
Is the weight matrix between the hidden layer and the decoding layer,
Figure 838470DEST_PATH_IMAGE031
for the purpose of self-encoding the output of the network,
Figure 674970DEST_PATH_IMAGE032
is a mixed spectral image of the light source,nrepresents the number of samples;
the spectrum preserving sparse self-encoder utilizes a weighted logarithm term to restrain the average activation value output by the hidden layer neuron so as to enable the average activation value to be matched with a given sparse value close to 0
Figure 965137DEST_PATH_IMAGE033
Close and add it as a penalty term to the loss function; introduction of
Figure 32319DEST_PATH_IMAGE027
Regularization term
Figure 407937DEST_PATH_IMAGE034
(ii) a For the weight matrix
Figure 45198DEST_PATH_IMAGE035
Adding
Figure 189872DEST_PATH_IMAGE036
Canonical constraint, the modified loss function is:
Figure 568900DEST_PATH_IMAGE037
wherein the content of the first and second substances,
Figure 946661DEST_PATH_IMAGE038
are the weight coefficients of the different components,
Figure 108652DEST_PATH_IMAGE039
the sparsity-weighting factor is a function of,
Figure 124144DEST_PATH_IMAGE040
representing hidden neuronsjThe activity of (2).
5. The hyperspectral anomaly detection method based on the spectrum preserving sparse self-encoder according to claim 4 is characterized in that step 3 is to input hyperspectral data to be detected into the spectrum preserving sparse self-encoder network constructed in step 2, set and adjust network parameters, perform unmixing operation on the network parameters to obtain background end member data, calculate reconstruction errors to obtain required anomalous target data, and specifically comprise the following steps:
setting parameters separately
Figure 346178DEST_PATH_IMAGE041
Learning rate, and optimizing model parameters by adopting an Adam optimization algorithm based on a random gradient through a minimization loss function;
inputting the hyperspectral data to be measured into a sparse self-encoder network with set parameters, performing unmixing operation on the data to be measured to obtain background end member data, and performing unmixing operation on the hyperspectral image to be measured
Figure 414497DEST_PATH_IMAGE042
Subtracting the obtained background end metadata to obtain the required abnormal target data, namely:
Figure 645758DEST_PATH_IMAGE043
。
6. the hyperspectral anomaly detection method based on spectrum preserving sparse self-encoder according to claim 5, characterized in that parameters
Figure 512826DEST_PATH_IMAGE044
Are respectively set as 10-2，10-50.0035, 0.04, sparse value
Figure 109024DEST_PATH_IMAGE045
Set to 0.02 and learning rate to 10-3。
CN202010545143.6A 2020-06-16 2020-06-16 Hyperspectral anomaly detection method based on spectrum preserving sparse self-encoder Pending CN111461087A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010545143.6A CN111461087A (en) 2020-06-16 2020-06-16 Hyperspectral anomaly detection method based on spectrum preserving sparse self-encoder

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010545143.6A CN111461087A (en) 2020-06-16 2020-06-16 Hyperspectral anomaly detection method based on spectrum preserving sparse self-encoder

Publications (1)

Publication Number Publication Date
CN111461087A true CN111461087A (en) 2020-07-28

Family

ID=71682040

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010545143.6A Pending CN111461087A (en) 2020-06-16 2020-06-16 Hyperspectral anomaly detection method based on spectrum preserving sparse self-encoder

Country Status (1)

Country Link
CN (1) CN111461087A (en)

Cited By (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112183561A (en) * 2020-11-09 2021-01-05 山东中医药大学 Joint fusion subtraction automatic encoder algorithm for image feature extraction
CN112215267A (en) * 2020-09-25 2021-01-12 天津大学 Hyperspectral image-oriented depth space spectrum subspace clustering method
CN112261018A (en) * 2020-10-13 2021-01-22 中国光大银行股份有限公司 Abnormal object detection method and device, storage medium and electronic device
CN112488102A (en) * 2020-11-30 2021-03-12 中南大学 Medical image classification method and device based on multi-view learning and depth supervision self-encoder
CN112699838A (en) * 2021-01-13 2021-04-23 武汉大学 Hyperspectral mixed pixel nonlinear blind decomposition method based on spectral diagnosis characteristic weighting
CN112990107A (en) * 2021-04-19 2021-06-18 中国人民解放军国防科技大学 Hyperspectral remote sensing image underwater target detection method and device and computer equipment
CN113379859A (en) * 2021-06-03 2021-09-10 西安理工大学 Hyperspectral anomaly detection method based on full convolution cascade self-encoder
CN113435366A (en) * 2021-06-30 2021-09-24 南京理工大学 Multi-time hyperspectral image Bayesian unmixing method for wavelet domain
CN113591816A (en) * 2021-09-29 2021-11-02 广东省科学院智能制造研究所 Hyperspectral anomaly detection method and system based on self-supervision guide coding network
CN113762629A (en) * 2021-09-10 2021-12-07 南京机电职业技术学院 Athlete optimal load strength prediction method based on improved AE
CN115223047A (en) * 2022-08-08 2022-10-21 西南交通大学 Spectrum unmixing method based on self-supervision learning
CN116091832A (en) * 2023-02-16 2023-05-09 哈尔滨工业大学 Tumor cell slice hyperspectral image classification method based on self-encoder network

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103871087A (en) * 2014-03-20 2014-06-18 西北工业大学 Hyperspectral unmixing compressive sensing method based on three-dimensional total variation sparse prior
CN108389188A (en) * 2018-01-30 2018-08-10 大庆师范学院 A kind of sparse EO-1 hyperion Anomaly target detection method

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103871087A (en) * 2014-03-20 2014-06-18 西北工业大学 Hyperspectral unmixing compressive sensing method based on three-dimensional total variation sparse prior
CN108389188A (en) * 2018-01-30 2018-08-10 大庆师范学院 A kind of sparse EO-1 hyperion Anomaly target detection method

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
AHMAD W. BITAR等: "Sparsity-based Cholesky Factorization and Its Application to Hyperspectral Anomaly Detection", 《ARXIV.ORG》 *
张晓慧: "基于改进稀疏表达的高光谱异常目标检测", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 *

Cited By (20)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112215267A (en) * 2020-09-25 2021-01-12 天津大学 Hyperspectral image-oriented depth space spectrum subspace clustering method
CN112215267B (en) * 2020-09-25 2022-11-01 天津大学 Hyperspectral image-oriented depth space spectrum subspace clustering method
CN112261018A (en) * 2020-10-13 2021-01-22 中国光大银行股份有限公司 Abnormal object detection method and device, storage medium and electronic device
CN112261018B (en) * 2020-10-13 2023-01-31 中国光大银行股份有限公司 Abnormal object detection method and device, storage medium and electronic device
CN112183561A (en) * 2020-11-09 2021-01-05 山东中医药大学 Joint fusion subtraction automatic encoder algorithm for image feature extraction
CN112183561B (en) * 2020-11-09 2024-04-30 山东中医药大学 Combined fusion-subtraction automatic encoder algorithm for image feature extraction
CN112488102A (en) * 2020-11-30 2021-03-12 中南大学 Medical image classification method and device based on multi-view learning and depth supervision self-encoder
CN112699838B (en) * 2021-01-13 2022-06-07 武汉大学 Hyperspectral mixed pixel nonlinear blind decomposition method based on spectral diagnosis characteristic weighting
CN112699838A (en) * 2021-01-13 2021-04-23 武汉大学 Hyperspectral mixed pixel nonlinear blind decomposition method based on spectral diagnosis characteristic weighting
CN112990107A (en) * 2021-04-19 2021-06-18 中国人民解放军国防科技大学 Hyperspectral remote sensing image underwater target detection method and device and computer equipment
CN113379859B (en) * 2021-06-03 2022-11-18 西安理工大学 Hyperspectral anomaly detection method based on full convolution cascade self-encoder
CN113379859A (en) * 2021-06-03 2021-09-10 西安理工大学 Hyperspectral anomaly detection method based on full convolution cascade self-encoder
CN113435366A (en) * 2021-06-30 2021-09-24 南京理工大学 Multi-time hyperspectral image Bayesian unmixing method for wavelet domain
CN113762629A (en) * 2021-09-10 2021-12-07 南京机电职业技术学院 Athlete optimal load strength prediction method based on improved AE
CN113591816B (en) * 2021-09-29 2022-02-22 广东省科学院智能制造研究所 Hyperspectral anomaly detection method and system based on self-supervision guide coding network
CN113591816A (en) * 2021-09-29 2021-11-02 广东省科学院智能制造研究所 Hyperspectral anomaly detection method and system based on self-supervision guide coding network
CN115223047A (en) * 2022-08-08 2022-10-21 西南交通大学 Spectrum unmixing method based on self-supervision learning
CN115223047B (en) * 2022-08-08 2023-06-09 西南交通大学 Spectrum unmixing method based on self-supervision learning
CN116091832A (en) * 2023-02-16 2023-05-09 哈尔滨工业大学 Tumor cell slice hyperspectral image classification method based on self-encoder network
CN116091832B (en) * 2023-02-16 2023-10-20 哈尔滨工业大学 Tumor cell slice hyperspectral image classification method based on self-encoder network

Similar Documents

Publication Publication Date Title
CN111461087A (en) Hyperspectral anomaly detection method based on spectrum preserving sparse self-encoder
Huyan et al. Hyperspectral anomaly detection via background and potential anomaly dictionaries construction
Chen et al. Identification of tomato leaf diseases based on combination of ABCK-BWTR and B-ARNet
Zhang et al. Understanding the learning mechanism of convolutional neural networks in spectral analysis
Zhang et al. Noise reduction in the spectral domain of hyperspectral images using denoising autoencoder methods
WO2020073737A1 (en) Quantitative spectroscopic data analysis and processing method based on deep learning
Zhang et al. Anomaly detection using improved deep SVDD model with data structure preservation
Zhang et al. Classification modeling method for near-infrared spectroscopy of tobacco based on multimodal convolution neural networks
Liu et al. NIRS feature extraction based on deep auto-encoder neural network
CN110008948B (en) Hyperspectral image target detection method based on variational self-coding network
CN113951830B (en) Brain disease classification method based on 3D attention convolution and self-supervision learning
Yuan et al. Moldy peanuts identification based on hyperspectral images and Point-centered convolutional neural network combined with embedded feature selection
CN115861887A (en) Fungus detection method for black garlic
CN115564996A (en) Hyperspectral remote sensing image classification method based on attention union network
Liu et al. Spectral group attention networks for hyperspectral image classification with spectral separability analysis
Cheng et al. Hyperspectral imaging classification based on LBP feature extraction and multimodel ensemble learning
Yu et al. Deep spectral–spatial feature fusion-based multiscale adaptable attention network for hyperspectral feature extraction
Jiang et al. Quantitative analysis of aflatoxin B1 in moldy peanuts based on near-infrared spectra with two-dimensional convolutional neural network
Guan et al. A recognition method of mushroom mycelium varieties based on near-infrared spectroscopy and deep learning model
CN112560949B (en) Hyperspectral classification method based on multilevel statistical feature extraction
Gull et al. A deep learning approach for multi‐stage classification of brain tumor through magnetic resonance images
CN116312860B (en) Agricultural product soluble solid matter prediction method based on supervised transfer learning
CN115330901B (en) Image reconstruction method and device based on compressed sensing network
CN117011595A (en) Hyperspectral image feature extraction method based on approximate NMR model
CN108492283B (en) Hyperspectral image anomaly detection method based on band-constrained sparse representation

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication

Application publication date: 20200728

RJ01 Rejection of invention patent application after publication