CN113807438B - Anti-hybrid semi-supervised hyperspectral image classification method based on semantic preservation - Google Patents

Anti-hybrid semi-supervised hyperspectral image classification method based on semantic preservation Download PDF

Info

Publication number
CN113807438B
CN113807438B CN202111093418.8A CN202111093418A CN113807438B CN 113807438 B CN113807438 B CN 113807438B CN 202111093418 A CN202111093418 A CN 202111093418A CN 113807438 B CN113807438 B CN 113807438B
Authority
CN
China
Prior art keywords
network
sample
data
hyperspectral image
semantic
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
Application number
CN202111093418.8A
Other languages
Chinese (zh)
Other versions
CN113807438A (en
Inventor
王雷全
朱同川
赵欣
李忠伟
吴春雷
周家梁
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.)
China University of Petroleum East China
Original Assignee
China University of Petroleum East China
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 China University of Petroleum East China filed Critical China University of Petroleum East China
Priority to CN202111093418.8A priority Critical patent/CN113807438B/en
Publication of CN113807438A publication Critical patent/CN113807438A/en
Application granted granted Critical
Publication of CN113807438B publication Critical patent/CN113807438B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G06F18/2155—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the incorporation of unlabelled data, e.g. multiple instance learning [MIL], semi-supervised techniques using expectation-maximisation [EM] or naïve labelling
    • 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/047—Probabilistic or stochastic 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
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Probability & Statistics with Applications (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Image Analysis (AREA)

Abstract

The invention relates to a semantic reservation-based anti-hybrid semi-supervised hyperspectral image classification method, which comprises the following steps in a training stage: step 1, extracting characteristics of a labeled sample through a characteristic extraction network; step 2, extracting features of two non-tag samples randomly extracted from a sample set through the feature extraction network, mixing the extracted features of the two non-tag samples, and sending the mixed features into a reconstruction network; step 3, reconstructing a sample through the reconstruction network, and sending the reconstructed sample into an anti-semantic reduction network; and 4, optimizing parameters of the feature extraction network, the reconstruction network and the antagonism semantic restoration network through a joint loss function. The method comprises the following steps in a test stage: and 5, sending the unlabeled sample of the hyperspectral image to be classified into the feature extraction network, and obtaining a label serving as a classification result through a classification function after extracting the features.

Description

Anti-hybrid semi-supervised hyperspectral image classification method based on semantic preservation
Technical Field
The invention relates to the technical field of image processing, in particular to the technical field of machine learning and hyperspectral image classification, and more particularly relates to a hyperspectral image classification method based on semantic preservation and anti-hybrid semi-supervision, which can be used for feature recognition of hyperspectral images.
Background
The hyperspectral remote sensing can acquire information of hundreds of continuous spectrum segments of the surface object, and provides rich spectral information to enhance the distinguishing capability of the surface object. The hyperspectral image classification plays a vital role and is a prerequisite for developing remote sensing applications such as forest checking, urban area monitoring and resource exploration.
In recent years, a hyperspectral classification method based on deep learning has great potential. The hyperspectral image classification is carried out through the neural network, so that the classification accuracy of the hyperspectral image is greatly improved.
Qizhe Xie et al in paper "Unsupervised Data Augmentation For Consistency Training" proposed using consistency training over a large amount of unlabeled data to constrain model predictions from varying over input noise, which is mostly in the form of gaussian noise or data enhancement. However, the data enhancement method applied to the hyperspectral image is less than the common RGB image method, and cannot well achieve the effect that model prediction remains unchanged after noise is added to an unlabeled image.
Zhi Zhang et al in paper Bagof Freebies for Training Object Detection Neural Networks propose mixup, an algorithm for image blending enhancement that is applied in computer vision, which can blend images between different classes to expand the training dataset. However, the mixup method explicitly fuses image tags by means of a loss function weighted addition, which has no practical physical meaning.
Disclosure of Invention
The invention aims at overcoming the defects of the prior art, and provides a hyperspectral classification method for anti-hybrid semi-supervision based on semantic preservation, which mainly comprises the following functional parts: 1. a supervised classification section; 2. a hidden space Mixup sample reconstruction portion; 3. the anti-regularization part of semantic preservation. The method based on semi-supervised consistency training has the problems that from three aspects of data amplification of label-free data, implicit Mixup strategy, and antagonism regularization strategy for realizing semantic preservation, interpolation and mixed hidden coding are adopted in hidden space, the result is reconstructed, an image after interpolation (mixing) is implicitly generated, a feature extraction network and a reconstruction network can generate semantic meaningful combinations of corresponding data points, and the structure of the data can be effectively learned under the condition that an explicit label is not used; and inputting the interpolated data points generated by the reconstruction network into an anti-semantic restoration network to restore the interpolated mixing coefficients. Thus, data enhancement is carried out on a large amount of unlabeled data, the training model extracts characteristics, and useful characterization is learned. And finally, testing hyperspectral sample data by using the trained model.
According to an embodiment of the present invention, there is provided a semantic preservation-based anti-hybrid semi-supervised hyperspectral image classification method, wherein a hyperspectral image sample set is used in a training phase, the sample set comprising labeled sample X L And unlabeled exemplar X U The method comprises the following steps in a training phase: step 1, extracting network f through characteristics θ () Extracting features of the labeled sample, and optimizing a parameter theta by using a loss function; step 2, extracting the network f through the characteristics θ () Extracting features of two unlabeled samples randomly extracted from a sample set, mixing the extracted features of the two unlabeled samples, and sending the mixed features into a reconstruction networkStep 3, by means of the reconstruction network +.>Reconstructing the sample, and sending the reconstructed sample to the anti-semantic restoration network d η () The method comprises the steps of carrying out a first treatment on the surface of the Step 4, extracting the network f through the characteristics θ () Said reconstruction network->And said antagonistic semantic reduction network d η () To jointly optimize the parameters θ, +.>And η, wherein the feature extraction network f in step 1 and step 2 θ () The parameter theta of (2) is shared, the said specialSign extraction network f θ () And said antagonistic semantic reduction network d η () Is a convolutional neural network, said reconstruction network +.>Is a deconvolution neural network. The method comprises the following steps in a test stage: step 5, sending the unlabeled sample of the hyperspectral image to be classified into the feature extraction network f θ () And obtaining the label serving as a classification result through a classification function after extracting the characteristics.
Therefore, the invention has the beneficial effects that:
1) The data amplification semi-supervised hyperspectral classification method based on the hidden space mixup is provided, the problem that the RGB image enhancement method is not suitable for hyperspectral images is avoided, the relationship between samples is considered through the data amplification mode of the hidden space mixup, and meanwhile the influence of homonymous foreign matters and homonymous foreign matters on semi-supervised classification can be relieved;
2) The anti-reconstruction method for the amplified sample based on semantic preservation is provided, a pseudo tag does not need to be generated for the amplified sample, the implicit semantic information of the amplified sample is obtained in a countermeasure mode, and the problem of classification accuracy reduction caused by false tag errors is avoided.
Drawings
Fig. 1 is a schematic architecture diagram of a convolutional neural network (AdvMix) implementing a semantic retention based anti-hybrid semi-supervised hyperspectral classification method according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a semantic preservation-based anti-hybrid semi-supervised hyperspectral classification method according to an embodiment of the present invention;
fig. 3 is a flow diagram of a semantic retention based anti-hybrid semi-supervised hyperspectral classification method according to an embodiment of the present invention.
Detailed Description
The implementation of the technical scheme is described in further detail below with reference to the accompanying drawings.
Those skilled in the art will appreciate that while the following description refers to numerous technical details regarding embodiments of the present invention, this is meant as an example only, and not meant to be limiting, of the principles of the present invention. The present invention can be applied to other than the technical details exemplified below without departing from the principle and spirit of the present invention.
In addition, in order to avoid limiting the description of the present specification to redundancy, in the description in the present specification, some technical details that can be obtained in the prior art material may be omitted, simplified, changed, etc., as will be understood by those skilled in the art, and this does not affect the disclosure sufficiency of the present specification.
The following detailed description of specific embodiments refers to the accompanying drawings.
FIG. 1 is a schematic architecture diagram of a hyperspectral classification method implementing semantic preservation-based challenge-hybrid semi-supervision, in which a feature extraction network is parameter-shared, according to an embodiment of the present invention; fig. 2 is a schematic diagram of a semantic retention based anti-hybrid semi-supervised hyperspectral classification method according to an embodiment of the present invention.
As shown in fig. 1 and 2, the semantic retention-based anti-hybrid semi-supervised hyperspectral classification method mainly involves four functional parts: the device comprises a data preprocessing part, a data segmentation part, a model training part and a prediction classification part. The model training and prediction classification part comprises a supervised classification part, a hidden space Mixup sample reconstruction part and an anti-regularization part with reserved semantics.
The data preprocessing part is used for preprocessing hyperspectral image data (comprising a training set and a testing set), including cleaning and normalizing of images, and the data segmentation part is used for segmenting the preprocessed data so as to meet the input requirement of the feature extraction network.
In the training stage, a training sample set is constructed from the segmented data, the labeled data is input to a supervised part of an upper branch of the model of fig. 1, features are extracted through a feature extraction network, and then a classification result of hyperspectral pixels is obtained through a classification part by using a loss function. Simultaneously, two pieces of data without labels are input to an unsupervised part of a lower branch of the model in fig. 1, the two pieces of data are respectively input to two feature extraction networks to obtain hidden codes, the hidden codes are subjected to sample interpolation recombination through a mixed coefficient, then the reconstruction is performed through a reconstruction network, and the obtained reconstruction result is input to an anti-semantic reduction network to recover the mixed coefficient.
In the test stage, the segmented data is used as a test sample set to be input into a feature extraction network for processing, and then hyperspectral features are obtained. And inputting the hyperspectral features into the classification part to obtain a classification result of the hyperspectral pixels.
Specifically, as shown in fig. 3, the semantic retention-based anti-hybrid semi-supervised hyperspectral classification method according to the embodiment of the present invention includes the following steps:
step S100, carrying out data preprocessing on hyperspectral image data, including data cleaning and data normalization,
the data normalization is used for uniformly mapping sample image data to a [0,1] interval and enabling the data to accord with normal distribution by utilizing a curvilinearity method;
step S200, dividing the image by using a fixed window, dividing all pixels into a sample set with window size w×w (for example, taking w as 8, the sample set is a set of 8×8 image blocks), forming a label data set according to whether the data has labels or notAnd no tag dataset->Wherein->For labeled hyperspectral samples, y i Is->Corresponding tag of N l For the number of labeled samples, +.>Is a label-free hyperspectral sample, the quantity of which is HW-N l B is the number of spectral bands of the hyperspectral image. As an example, w may be 8 and b 32.
Wherein, as an example, the training set is composed of 10 samples randomly extracted from each class in the sample set, and the test set is the rest samples after the training set is extracted from the sample set, and 100 samples are randomly extracted from each class.
And step S300, extracting the characteristics of the labeled data, sending the characteristic result to the full-connection layer, and sending the characteristic result to a softmax function to generate a classification result. The cross entropy loss function is used during the training phase, defined as follows:
wherein x is l Is a labeled sample, y is its label, X L Representing a labeled sample dataset, p () is the model output distribution.
Step S400, respectively extracting features of two unlabeled samples through a feature extraction network, then carrying out interpolation mixing on the obtained two hidden codes (features), and reconstructing the mixed data through a reconstruction network to obtain mixed sample data. The loss function of the feature extraction network and the reconstruction network training in the training phase is defined as follows:
where λ is a scalar super parameter used to control the weights between the terms, f θ ()、And d η () Respectively representing a feature extraction network, a reconstruction network and an antagonism semantic restoration network,
is sample->And->The mixing coefficient of the mixed reconstructed sample in the hidden space is alpha.
Step S500, the mixed sample data output by the reconstruction network is input to an anti-semantic reduction network, and the anti-semantic reduction network restores the mixed coefficients. The penalty function of the anti-semantic reduction network in the training phase is defined as follows:
as an example, the feature extraction network employs a Wide-res net with a widening factor of 2 and a number of layers of 28 layers.
In the reconstruction network, the deconvolution output is first made 4×4×16 by one deconvolution layer. A 2D convolution block is then connected, consisting of two consecutive 3 x 3 2D convolution layers and an up-sampling layer. The 2D convolution block is performed twice in total, the output structure is 16×16 x 16. The two 2D convolutional layers are then concatenated to give an output channel number of the target channel number 32, i.e. an output structure of 16 x 32. The 3 x 3 2D convolutional layer is then concatenated until the target resolution 8 x 32 is obtained.
The structure of the anti-semantic reduction network is the same as that of the feature extraction network, and the average value of the last layer of activation functions is calculated in order to enable the anti-semantic reduction network to output a single scalar value.
To verify the method, classification and testing was performed using the Pavia University hyperspectral dataset as an example. The Pavia University data is a portion of the hyperspectral data of an image made by an on-board reflectance optical spectroscopy imager (ROSIS-03) in germany in the year 2003 for parkia city in italy. The spectrum imager continuously images 115 wave bands within the wavelength range of 0.43-0.86 mu m, and the spatial resolution of the formed image is 1.3m. The data is 610 x 340 in size and therefore contains 2207400 pixels in total, but contains a large number of background pixels, and contains only 42776 pixels in total, which contain 9 types of features in total. The number of training samples, the number of test samples, and the total number of samples for each category are shown in Table 1.
TABLE 1 Pavia University dataset training, test sample number
Under the above sample conditions, the methods of the present invention (Advmix) were subjected to comparative tests with the four methods of SVM, SSSR, SSGAT, ASSRF, and the overall classification accuracy (OA), the average classification accuracy (AA), and the Kappa coefficient were recorded. The test results are shown in Table 2.
TABLE 2 comparison of Classification Properties
In table 2, SVM is a hyperspectral classification method using a conventional classifier SVM, SSSR is a method of semi-supervised hyperspectral classification based on spatial regularization, SSGAT is a method of semi-supervised hyperspectral image classification based on a graph attention network, and ASSRF is a method of semi-supervised hyperspectral classification based on an active learning random forest.
As can be seen from Table 2, the method provided by the invention has a good classification result, the classification effect is improved to different degrees, and the overall precision and the average precision are superior to those of the comparison method.
In summary, according to the semantic reservation-based anti-hybrid semi-supervised hyperspectral image classification method, hyperspectral data characteristic information is extracted from tagged data through a characteristic extraction network; and carrying out feature extraction on the unlabeled data through a feature extraction network, carrying out interpolation mixing through a mixing coefficient, reconstructing the mixed data through a reconstruction network, and finally sending the mixed data after feature extraction into an anti-semantic network to restore the interpolation coefficient. The problem that hyperspectral images are less than the RGB image data enhancement method is avoided, and the accuracy of hyperspectral ground object classification is improved.
Finally, those skilled in the art will appreciate that various modifications, adaptations, and alternatives of the above-described embodiments of the present invention can be made without departing from the scope of the present invention as defined by the appended claims.

Claims (10)

1. A semantic preservation-based anti-hybrid semi-supervised hyperspectral image classification method,
wherein a hyperspectral image sample set is used in the training stage, and the sample set comprises a labeled sample X L And unlabeled exemplar X U ，
The method comprises the following steps in a training stage:
step 1, extracting network f through characteristics θ () Extracting features of the labeled sample, and optimizing a parameter theta by using a loss function;
step 2, extracting the network f through the characteristics θ () Extracting features of two unlabeled samples randomly extracted from a sample set, mixing the extracted features of the two unlabeled samples, and sending the mixed features into a reconstruction network
Step 3, through the reconfiguration networkReconstructing the sample, and sending the reconstructed sample to the anti-semantic restoration network d η ()；
Step 4, extracting the network f through the characteristics θ () The reconstruction networkAnd said antagonistic semantic reduction network d η () To jointly optimize the parameters θ, +.>And eta, and the sum eta,
wherein the feature extraction network f in step 1 and step 2 θ () Is shared by the parameters θ of the feature extraction network f θ () And said antagonistic semantic reduction network d η () Is a convolutional neural network, the reconstruction networkIs a deconvolution neural network,
the method comprises the following steps in a test stage:
step 5, sending the unlabeled sample of the hyperspectral image to be classified into the feature extraction network f θ () And obtaining the label serving as a classification result through a classification function after extracting the characteristics.
2. The method of claim 1, wherein a labeled sample X L And unlabeled exemplar X U W×w image blocks obtained by dividing hyperspectral image, the labeled sample is expressed asThe unlabeled exemplar is denoted +.>Wherein (1)>For sample data of labeled samples, y i Is->Corresponding tag of N l For the number of labeled samples, +.>Sample data, which is a sample without labels, in an amount of HW-N l Wherein, the method comprises the steps of, wherein,h and W are the number of samples in the height and width directions of the hyperspectral image, respectively, B is the number of bands of the hyperspectral image,
wherein in step 2, the feature extraction network f θ () The corresponding characteristics of the two unlabeled samples are obtained after the characteristic extraction of the two unlabeled samplesWherein (1)>Sample data for two unlabeled exemplars,
in step 3, by reconstructing the networkObtaining a reconstituted sample->Wherein alpha is a mixing coefficient, alpha is 0,1]。
3. Method according to claim 2, wherein in step 4 the network d is restored by antagonizing semantics η () Obtaining the reduced mixing coefficient
4. The method of claim 1, wherein prior to step 1, the hyperspectral image data is subjected to data cleaning and data normalization,
the data normalization is used for uniformly mapping the hyperspectral image data to the [0,1] interval and enabling the hyperspectral image data to conform to normal distribution by utilizing a curvilinearity method.
5. The method of claim 1, wherein the feature extraction network f θ () Is a Wide-ResNet network, for high slaveAnd carrying out feature extraction on the labeled sample data extracted by the spectral image, wherein the Wide-ResNet feature extraction network is trained by using a cross entropy loss function.
6. The method of claim 5, wherein the cross entropy loss function is defined as follows:
wherein E represents a mathematical expectation, x l Is labeled sample X L Y is its label, and p () is the model output distribution.
7. The method of claim 6, the antagonistic semantic reduction network d η () The loss function of (2) is defined as follows:
8. the method according to any one of claims 5 to 7, wherein in step 4 the joint loss function is defined as follows:
where λ is a scalar super parameter.
9. The method according to any one of claims 1 to 7, wherein the parameter θ is optimized by a random gradient descent method,And eta.
10. The method according to claim 8A method in which a random gradient descent method is used to optimize the parameter θ,And eta.
CN202111093418.8A 2021-09-17 2021-09-17 Anti-hybrid semi-supervised hyperspectral image classification method based on semantic preservation Active CN113807438B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111093418.8A CN113807438B (en) 2021-09-17 2021-09-17 Anti-hybrid semi-supervised hyperspectral image classification method based on semantic preservation

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111093418.8A CN113807438B (en) 2021-09-17 2021-09-17 Anti-hybrid semi-supervised hyperspectral image classification method based on semantic preservation

Publications (2)

Publication Number Publication Date
CN113807438A CN113807438A (en) 2021-12-17
CN113807438B true CN113807438B (en) 2023-08-29

Family

ID=78939690

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111093418.8A Active CN113807438B (en) 2021-09-17 2021-09-17 Anti-hybrid semi-supervised hyperspectral image classification method based on semantic preservation

Country Status (1)

Country Link
CN (1) CN113807438B (en)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111340064A (en) * 2020-02-10 2020-06-26 中国石油大学(华东) Hyperspectral image classification method based on high-low order information fusion
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
CN113378795A (en) * 2021-07-12 2021-09-10 浙江科技学院 Road scene semantic segmentation method based on GAN and cross-modal feature fusion
CN113392931A (en) * 2021-07-02 2021-09-14 西安电子科技大学 Hyperspectral open set classification method based on self-supervision learning and multitask learning

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
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
CN111340064A (en) * 2020-02-10 2020-06-26 中国石油大学(华东) Hyperspectral image classification method based on high-low order information fusion
CN113392931A (en) * 2021-07-02 2021-09-14 西安电子科技大学 Hyperspectral open set classification method based on self-supervision learning and multitask learning
CN113378795A (en) * 2021-07-12 2021-09-10 浙江科技学院 Road scene semantic segmentation method based on GAN and cross-modal feature fusion

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
基于多损失约束与注意力块的图像修复方法;曹真;杨云;齐勇;李程辉;;陕西科技大学学报(第03期);全文 *

Also Published As

Publication number Publication date
CN113807438A (en) 2021-12-17

Similar Documents

Publication Publication Date Title
CN110648334A (en) Multi-feature cyclic convolution saliency target detection method based on attention mechanism
Fu et al. DSAGAN: A generative adversarial network based on dual-stream attention mechanism for anatomical and functional image fusion
CN113962893A (en) Face image restoration method based on multi-scale local self-attention generation countermeasure network
CN110473142B (en) Single image super-resolution reconstruction method based on deep learning
CN113392931B (en) Hyperspectral open set classification method based on self-supervision learning and multitask learning
Tursun et al. MTRNet++: One-stage mask-based scene text eraser
CN113865859B (en) Gear box state fault diagnosis method for multi-scale multi-source heterogeneous information fusion
CN110309811A (en) A kind of hyperspectral image classification method based on capsule network
Zhao et al. ADRN: Attention-based deep residual network for hyperspectral image denoising
CN113066065B (en) No-reference image quality detection method, system, terminal and medium
CN115909006B (en) Mammary tissue image classification method and system based on convolution transducer
CN111914909A (en) Hyperspectral change detection method based on space-spectrum combined three-direction convolution network
CN111563577B (en) Unet-based intrinsic image decomposition method for skip layer frequency division and multi-scale identification
Nalbalwar et al. Detection of brain tumor by using ANN
Manu et al. GANID: a novel generative adversarial network for image dehazing
CN114882368A (en) Non-equilibrium hyperspectral image classification method
Zhou et al. MSAR‐DefogNet: Lightweight cloud removal network for high resolution remote sensing images based on multi scale convolution
Hussain et al. Image denoising to enhance character recognition using deep learning
CN113807438B (en) Anti-hybrid semi-supervised hyperspectral image classification method based on semantic preservation
CN117115675A (en) Cross-time-phase light-weight spatial spectrum feature fusion hyperspectral change detection method, system, equipment and medium
CN117391981A (en) Infrared and visible light image fusion method based on low-light illumination and self-adaptive constraint
CN116563606A (en) Hyperspectral image classification method based on dual-branch spatial spectrum global feature extraction network
Abdul Hamid et al. Texture image classification using improved image enhancement and adaptive SVM
CN115984585A (en) Attention mechanism-introduced multi-scale hyperspectral image feature extraction method
CN112862655B (en) JPEG image steganalysis method based on channel space 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