CN108537742B - Remote sensing image panchromatic sharpening method based on generation countermeasure network - Google Patents
Remote sensing image panchromatic sharpening method based on generation countermeasure network Download PDFInfo
- Publication number
- CN108537742B CN108537742B CN201810193748.6A CN201810193748A CN108537742B CN 108537742 B CN108537742 B CN 108537742B CN 201810193748 A CN201810193748 A CN 201810193748A CN 108537742 B CN108537742 B CN 108537742B
- Authority
- CN
- China
- Prior art keywords
- network
- image
- network model
- generated
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 37
- 230000003595 spectral effect Effects 0.000 claims abstract description 28
- 238000012549 training Methods 0.000 claims abstract description 26
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 12
- 239000013598 vector Substances 0.000 claims abstract description 6
- 238000013507 mapping Methods 0.000 claims abstract description 5
- 238000002372 labelling Methods 0.000 claims abstract description 4
- 230000004927 fusion Effects 0.000 claims description 14
- 230000008569 process Effects 0.000 claims description 10
- 238000012545 processing Methods 0.000 claims description 9
- 238000012360 testing method Methods 0.000 claims description 7
- 238000010586 diagram Methods 0.000 claims description 6
- 230000004044 response Effects 0.000 claims description 3
- 238000001228 spectrum Methods 0.000 claims description 2
- 230000006870 function Effects 0.000 description 14
- 238000013135 deep learning Methods 0.000 description 8
- 230000009466 transformation Effects 0.000 description 5
- 238000004422 calculation algorithm Methods 0.000 description 4
- 238000001514 detection method Methods 0.000 description 4
- 238000005070 sampling Methods 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000007500 overflow downdraw method Methods 0.000 description 3
- 238000013459 approach Methods 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000000513 principal component analysis Methods 0.000 description 2
- 101000933252 Homo sapiens Protein BEX3 Proteins 0.000 description 1
- 102100025955 Protein BEX3 Human genes 0.000 description 1
- 230000003042 antagnostic effect Effects 0.000 description 1
- 238000000701 chemical imaging Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000007499 fusion processing Methods 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000008054 signal transmission Effects 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
- 238000012800 visualization Methods 0.000 description 1
Images
Classifications
-
- G06T5/73—
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T9/00—Image coding
- G06T9/002—Image coding using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10032—Satellite or aerial image; Remote sensing
- G06T2207/10036—Multispectral image; Hyperspectral image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10032—Satellite or aerial image; Remote sensing
- G06T2207/10041—Panchromatic image
Abstract
The invention relates to a remote sensing image panchromatic sharpening method based on a generation countermeasure network, which comprises the following steps: 1) constructing a data set: carrying out target classification and labeling on the remote sensing image data, and constructing an original image data set containing a full-color image and a spectral image; 2) designing and generating a network and judging a network model: the generated network model G of the GAN learns the mapping from random noise vectors and images in the data set to the generated sample image y, the generated sample image cannot be judged to be false by the judgment network model after the G is trained, the judgment network model D uses a convolutional neural network, the true and false of the generated data can be judged after the G is trained, and the classification problem of judging the true and false of the generated sample image can be completed as well as possible; 3) training generates a confrontation network model.
Description
Technical Field
The invention relates to the fields of remote sensing image processing, image fusion, deep learning and the like, in particular to a fusion method of a full-color image and a multispectral image in the field of remote sensing, and belongs to the technical field of remote sensing image fusion.
Background
The remote sensing image with higher spatial resolution and spectral resolution has obvious practical application significance in the aspects of object detection, geographic mapping, environmental monitoring and the like. However, due to the limitations of signal transmission bands and imaging sensor storage, most remote sensing satellites only provide low spatial/high spectral resolution Multispectral (MSI) images and high spatial/low spectral resolution Panchromatic (PAN) images, which are strongly complementary. By utilizing the difference advantages and complementarity of the two images, the fused images with clear spatial details and rich spectral information are fused, and the fusion technology is also called panchromatic sharpening and is a key preprocessing step in many remote sensing applications.
In the present invention, a Multispectral (MSI) image and a Hyperspectral (HIS) image may be collectively referred to as a spectral image. Typically, spectral imaging involves multiple narrow spectral bands obtained by the sensor, i.e., it contains multiple components.
In the panchromatic sharpening process, the key and requirement of the panchromatic and spectral image fusion method are as follows: the spectral resolution is fidelity, namely the fused image is consistent with the spectral information of the spectral image; the spatial resolution fidelity is that the fused image is consistent with the spatial information of the full-color image; the time and the calculation redundancy are low, namely, the large-scale multi-size remote sensing image fusion process can be quickly completed.
Currently, mainstream panchromatic sharpening methods can be classified into several categories, namely component replacement methods, multi-resolution analysis methods, and relative spectral contribution methods. The component replacement method performs transformation on a color space domain of a multispectral image mainly by Principal Component Analysis (PCA), Schmidt Orthogonalization (Schmidt Orthogonalization), intensity, hue, saturation (IHS) transformation, and the like, replaces a spatial information channel of the multispectral image with a panchromatic image, and obtains a fused image by inverse transformation. The multiresolution analysis method is mainly based on wavelet transform, Laplacian Pyramid (Laplacian Pyramid) and other tools, converts an image from a spatial domain to a transform domain, sets a responsive fusion rule according to the characteristics of transform coefficients, and finally inversely transforms to obtain a fused image. Relative spectral contribution methods, including the Brovey transform fusion method, the panchromatic plus multi-spectral method, replace the image component and apply linear combinations of spectral bands for processing.
In recent years, based on the appearance of large-scale data and the development of deep neural networks, deep learning methods have become an important research direction in the field of machine learning. The Convolutional Neural Network (CNN) has strong feature learning capability, feature data obtained by deep Network model learning has more essential representativeness to original data, and rich internal information can be extracted through a deep learning architecture model trained by large-scale data, so that visualization and classification problem processing are facilitated.
The generation countermeasure network (GAN) is a novel network in a deep learning algorithm, and the generation network and the discrimination network constructed by the convolutional neural network are used for carrying out the training of an antagonistic mode, and the modeling of a generation model is completed by utilizing the principle of binary zero sum game, so that the Nash equilibrium is finally achieved. Aiming at the scene of data loss, the generation model can help to generate related data and improve the data quantity, so that the learning efficiency is improved by using semi-supervised learning. The judgment model can judge the truth of the sample, the generation model is continuously strengthened, and the generated sample is closer to the real sample more and more through continuous iteration.
Disclosure of Invention
The invention aims to provide a remote sensing image panchromatic sharpening method capable of improving spatial resolution and spectral resolution while simplifying calculation redundancy. The invention trains an end-to-end fusion model of a full-color chart and a spectrogram through the existing remote sensing image data based on a deep learning method and a generation confrontation network model. The technical scheme adopted by the invention is as follows:
a remote sensing image panchromatic sharpening method based on a generation countermeasure network comprises the following steps:
1) constructing a data set: and carrying out target classification and labeling on the remote sensing image data, constructing an original image data set containing full-color images and spectral images, and dividing a training set and a testing set.
2) Designing and generating a network and judging a network model: the GAN's generated network model G learns the mapping to the generated sample image y from random noise vectors and images in the dataset, i.e. G: the generated sample image cannot be judged to be false by the judgment network model after being trained, the judgment network model D uses a convolution neural network, the true and false of the generated data can be judged after being trained, the classification problem of judging the true and false of the generated sample image can be completed as good as possible, and in the generation process of the final output sample image, the input full-color image and the spectral image have the same underlying structure and share the position of the protruded edge; generating a network model to increase skip connection, adopting an integral structure of U-Net, dividing the integral structure into a coding layer and a decoding layer, halving the length and the width of a characteristic diagram for each coding layer, halving the number of the characteristic layers, doubling the length and the width of the characteristic diagram for each decoding layer, doubling the number of the characteristic layers, connecting the coding layers in series through a channel, and then performing deconvolution processing; designing and judging a network model based on a convolutional neural network for classification, wherein the network is designed to comprise a concatenation layer and four convolutional layers, only the authenticity of each block in the generated fusion image is classified, the network is operated on the image in a convolutional manner, and the average value of all responses is taken to provide the final output of D;
3) training generates a confrontation network model: based on a designed network and a discrimination network model, a network G is generated for the network, samples are generated from random noise or latent variables, a network D is discriminated, a loss function applied to a full-color image and a fusion image is provided, the two are trained simultaneously until Nash equilibrium is reached, the discrimination network cannot correctly distinguish generated data and real data, a network structure is trained by utilizing a data set, a fully optimized network is obtained, and weight parameters in the network converge to global optimum.
Compared with the prior art, the invention has the advantages that:
the invention provides an image fusion technology based on a deep learning method innovatively, which is different from all the conventional thought of panchromatic sharpening methods.
Compared with the traditional method, the GAN can extract high-dimensional deep features implicit in large-scale data by utilizing a deep convolutional neural network, and the structure of the GAN can also reduce the information loss in the convolution process to the maximum extent. The GAN does not need to carry out forward and inverse transformation on the color space or other transformation domains of the original image, and can keep the stability and the continuity of data information; meanwhile, the images are directly fused through the trained model, so that the method is high in speed and efficiency.
And compared with the traditional generation model, the generation of the countermeasure network process is more efficient and the output is more real. GAN does not require different data to be generated in the sampling sequence, and the time redundancy of generating samples is low compared with the completely obvious belief networks such as NADE, Pixel RNN, WaveNet and the like. The GAN optimizes the likelihood itself, so the generation instance is more realistic than the variational auto-encoder (VAE) introduces a deterministic bias to optimize the lower bound of the log-likelihood. GAN does not require any particular dimensionality or reversibility of the underlying variables that generate the model input, as compared to non-linear independent component methods such as NICE, Real NVE, etc. In contrast to the boltzmann machine and the generation of random networks, the process of GAN generation instances only requires the model to run once, rather than multiple iterations in the form of markov chains.
Drawings
FIG. 1 is a flow chart of the experiment required by the present invention.
FIG. 2 is a schematic diagram of a configuration of a generative countermeasure network used in the present invention.
Fig. 3(a) panchromatic remote sensing image data (b) spectral remote sensing image data (c) remote sensing image data fused by a conventional method (d) remote sensing image data fused by the present invention.
FIG. 4 is a diagram illustrating the effect of the present invention.
Detailed Description
In order to make the technical solution of the present invention clearer, the following further describes a specific embodiment of the present invention. As shown in fig. 1, the present invention is specifically implemented by the following steps:
1. large-scale remote sensing image dataset construction
The method mainly selects remote sensing image sets such as SpaceNet on AWS and the like disclosed by a network to construct a data set, takes an RIO data set in SpaceNet as an example, and comprises 50cm image data and coordinates, wherein an image source is a WorldView-2 satellite of DigitaGlobe, and the method can be used for various application occasions such as image segmentation and detection.
SpaceNet is a large-scale remote sensing image data set hosted in an AWS cloud service platform of Amazon company, is completed by DigitalGlobe, CosmiQ Works and NVIDIA together, comprises an online storage bank of satellite images and marked training data, and is a publicly released satellite image data platform which has high resolution and is specially used for training a machine learning algorithm. Besides, the method also combines the NWPU VHR-10, a Chinese academy geospatial data cloud platform, the United States Geological Survey (USGS) and related remote sensing data of Google company to build a data set required by training and testing.
Taking 3band _ AOI _1_ RIO data in the SpaceNet data set as an example, the spectral image comprises three channels of red, green and blue. The ratio of spatial resolution of the panchromatic image to the spectral image is 4: 1, the resolution of the input full-color image is 256 × 256, and the network can sharpen the full-color image based on the spectral image. Each full-color image has a unique target spectral image corresponding to the full-color image, and after the full-color image is subjected to target classification and labeling, an original image data set containing the full-color image and the spectral image is constructed, and the data set is divided into 4: the scale of 1 is divided into a training set and a test set.
2. Panchromatic sharpening model design based on generation countermeasure network
As shown in the structure of fig. 2, the goal of generating the network model is to minimize the probability of discriminating the network model from discriminating the generated data as false as possible. And the generated network can 'cheat' the judgment of the judgment network as much as possible in the training process, and the judgment network can make a classification task of correct judgment on the image as much as possible.
The generating network G of GAN is a network that can learn the mapping from random noise vector z and image x in the dataset in 1 to the generated sample image y, i.e. G: { x, z } → y, G is trained, and the generated sample image cannot be distinguished as false by the distinguishing network model. And the judgment network model D uses a convolution neural network, and after the D is trained, the truth of the generated data can be judged, so that the classification problem of judging the truth of the generated sample image can be completed as well as possible.
The training target for GAN can be expressed as follows:
LcGAN(G,D)＝Ex,y[logD(x,y)]+Ex,z[log(1-D(x,G(x,z)))]
the above formula is the objective function to be optimized in the game problem of the generation model and the discrimination model. Wherein E isx,y[logD(x,y)]Represents the expectation of finding the discriminant loss function logD (x, y), x, z represents obeying the prior distribution, z is [ -1,1]Uniformly distributed acquired vectors within, E (x, z) representing the expectation derived from the random vectors. The G function and the D function respectively representGenerating an output image of the network and judging an output result of the network.
The generated model and the discriminant model both adopt a structure of Convolution (Convolution) -Batch Normalization-nonlinear operation Units (Rectified Linear Units, ReLu), and in the training process, the generated model minimizes the target, and the discriminant model maximizes the target, that is:
G*＝argminGmaxDLcGAN(G,D)
during the generation of the final output sample image, the input panchromatic image and the spectral image have the same underlying structure, sharing the location of the protruding edges. In order to enable the generative model to acquire the information, skip connection is added to the generative model, and the overall structure of U-Net is adopted. The network is divided into two parts of an encoding layer (eight layers in total) and a decoding layer (eight layers in total), wherein each encoding layer has the length and width of a feature map reduced by half, the number of feature layers is increased by half, each decoding layer has the length and width of the feature map doubled, the number of feature layers is increased and doubled, the encoding layers are connected in series through channels, and then deconvolution processing is carried out.
The image operation is carried out on the periphery of the input image, and the number of the convolution layers is designed to be 20, 4 times of down sampling and 4 times of up sampling. Specifically, for an n-th layer network, a jump connection is added between each ith layer and the nth-ith layer, and all channels in the ith layer and the nth-ith layer are connected. A ReLu active layer is connected behind the convolution layer except the last layer in the whole network to add a nonlinear unit for the network and reduce the risk of overfitting the network; and normalizing the final convolution layer.
The redundancy of the operation can be reduced through the downsampling process; and finally, realizing the up-sampling of the image by using two layers of deconvolution layers, and performing superposition processing on the characteristics output by the convolution layer in the previous stage and the characteristics output by the convolution layer in the later stage, so that the characteristics can be reused.
A discriminative network model is designed based on a Convolutional Neural Network (CNN) for classification. The network is designed to contain one concatenation layer, and four convolutional layers. The CNN of the parameter design is reduced, only the authenticity of each block in the generated fusion image is classified, the network is rolled on the image, and the average value of all responses is made to provide the final output of D.
3. Panchromatic sharpening model construction based on generation of countermeasure network
For pairs of images in the dataset, their orientation can be defined in the training as full color to spectral images. The input and output pairs of the image are preset, and the image data with different length-width ratios can be used by setting relevant parameters. Generating a network G, generating a vivid sample from random noise or potential variables, judging a network D, providing a loss function applied to a full-color image and a fusion image, training the two at the same time until Nash equilibrium is reached, and at the moment, the judgment network cannot correctly distinguish generated data from real data. And training the network structure by using a large-scale image data set to obtain a fully optimized network, wherein the weight parameters in the network are converged to be globally optimal. And then, detailed implementation steps of the network generation and network discrimination training processes are respectively explained.
(1) Generating a network model
Based on the programmable deep learning framework Tensorflow, firstly, a memory layer for storing networks of U-Net layers is defined, and a first coding layer encoder _1 is constructed. Inputting a batch of panchromatic remote sensing image data into the coding layer, processing by a convolution function, outputting an output which is a characteristic map (feature map) with reduced resolution and increased channel number, and storing the output into layers. The parameters of the conv function are a set of feature maps, the number of output channels, and the convolution kernel step.
And defining the number of output channels of the next second to eighth coding layers, and circularly generating the second to eighth coding layers.
And thirdly, defining the ratio of the output channel number of the next seven decoding layers to dropout. And then entering jump connection loop processing. The first layer cycle is to deconvolve the output of the layer-1 of the eighth layer coding first. And the second layer of circulation connects the deconvolution output and the seventh layer of output of the corresponding encoder in series through a channel, and then performs deconvolution output.
And fourthly, analogizing according to the third step, then connecting the output of the current seventh decoding layer and the output of the first layer of the encoder in series through a channel, and then performing deconvolution output. And completing generation of the model U-Net through the steps.
(2) Discriminating network model
The training purpose of the discrimination model D is to maximize the discrimination accuracy of the discrimination model D as much as possible. The discriminative network model is designed to contain one concatenation layer, and four convolution layers. Similarly, a memory is defined, and the layers are configured to output pairs of full-color images and spectral images in series through channels. Each convolution layer outputs the result of the previous layer after convolution operation.
It is also necessary to construct true discriminant models, false discriminant models, and discriminant objective functions and generate objective functions. The discrimination objective function (discriminantor _ loss) needs to make the true discrimination rate (predict _ real) approach to 1 and the false discrimination rate (predict _ fake) approach to 0. Generating the objective function (generator _ loss) requires that the predicate _ false be close to 0 and the value of target-outputs should be close to 0.
And (3) constructing a discriminant training function (discrima _ train) and a generating training function (gen _ train), namely performing multiple iterations and training tasks of the model according to the database constructed in the step 1, and selecting a batch gradient descent method according to a training strategy of the network.
4. Application effect verification of panchromatic sharpening model
The invention firstly compares the method with typical methods such as a PCI sharpening method and an ENVI-GS conversion method based on the data set 1. As shown in fig. 3, in order to make the subjective display more obvious, we take a three-channel image as an example to perform a fusion experiment, we perform result analysis of a comparison algorithm, and in terms of subjective visual effect, the method of the present invention can maintain good spectral information and spatial details for experimental data used for testing.
In order to verify the application effect of the network model, on the test data set constructed in the invention content 1, the existing target detection algorithms such as fast R-CNN and YOLO9000 based on deep learning are used, the data set based on the PASCAL VOC competition format is constructed according to the spectrum images before and after fusion, the military and civil targets such as airplanes, ships, oil storage tanks, bridges and ports in the image are labeled, and the target detection model is trained and tested.
And calling a built-in script to convert the picture format before training, and writing the category and other information in the label text into an LMDB format file. And calculating a mean value file of the images in the training set, removing a large number of useless background pixels in the images and hiding redundant information irrelevant to the target. And (4) cascading the classifier and the detector, and combining the output of the classifier as the input of the detector into a final recognition model. Finally, the test data are analyzed, relevant indexes such as accuracy and rapidity are counted, and the recognition degree of the system is evaluated.
Claims (1)
1. A remote sensing image panchromatic sharpening method based on a generation countermeasure network comprises the following steps:
1) constructing a data set: carrying out target classification and labeling on remote sensing image data, constructing an original image data set containing full-color images and spectral images, and dividing a training set and a testing set;
2) designing and generating a network and judging a network model: the GAN generation network model G is a mapping from random noise vectors and images in the data set to generated sample images through training, the G is trained, the generated sample images cannot be judged to be false by the judgment network model, the judgment network model D uses a convolutional neural network, after training, the true and false of the generated data can be judged, in the generation process of the final output sample image, the input full-color image and the spectrum image have the same bottom layer structure and share the position of a protruding edge; generating a network model to increase skip connection, adopting an integral structure of U-Net, dividing the integral structure into a coding layer and a decoding layer, halving the length and the width of a characteristic diagram for each coding layer, halving the number of the characteristic layers, doubling the length and the width of the characteristic diagram for each decoding layer, doubling the number of the characteristic layers, connecting the coding layers in series through a channel, and then performing deconvolution processing; designing and judging a network model based on a convolutional neural network for classification, wherein the network is designed to comprise a concatenation layer and four convolutional layers, only the authenticity of each block in the generated fusion image is classified, the network is operated on the image in a convolutional manner, and the average value of all responses is taken to provide the final output of D;
3) training generates a confrontation network model: based on a designed generation network and a discrimination network model, a generation network model G is generated, samples are generated from random noise or latent variables, a discrimination network D is provided, a loss function applied to a full-color image and a fusion image is provided, the two are trained simultaneously until Nash equilibrium is reached, the discrimination network cannot correctly distinguish generated data and real data, a network structure is trained by utilizing a data set, a fully optimized network is obtained, and weight parameters in the network converge to global optimum.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810193748.6A CN108537742B (en) | 2018-03-09 | 2018-03-09 | Remote sensing image panchromatic sharpening method based on generation countermeasure network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810193748.6A CN108537742B (en) | 2018-03-09 | 2018-03-09 | Remote sensing image panchromatic sharpening method based on generation countermeasure network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108537742A CN108537742A (en) | 2018-09-14 |
| CN108537742B true CN108537742B (en) | 2021-07-09 |
Family
ID=63486694
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810193748.6A Active CN108537742B (en) | 2018-03-09 | 2018-03-09 | Remote sensing image panchromatic sharpening method based on generation countermeasure network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108537742B (en) |
Families Citing this family (38)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108460391B (en) * | 2018-03-09 | 2022-03-22 | 西安电子科技大学 | Hyperspectral image unsupervised feature extraction method based on generation countermeasure network |
| CN109102029B (en) * | 2018-08-23 | 2023-04-07 | 重庆科技学院 | Method for evaluating quality of synthesized face sample by using information maximization generation confrontation network model |
| CN109389080B (en) * | 2018-09-30 | 2022-04-19 | 西安电子科技大学 | Hyperspectral image classification method based on semi-supervised WGAN-GP |
| CN109344778A (en) * | 2018-10-10 | 2019-02-15 | 成都信息工程大学 | Based on the unmanned plane road extraction method for generating confrontation network |
| CN109508647A (en) * | 2018-10-22 | 2019-03-22 | 北京理工大学 | A kind of spectra database extended method based on generation confrontation network |
| CN109495744B (en) * | 2018-10-29 | 2019-12-24 | 西安电子科技大学 | Large-magnification remote sensing image compression method based on joint generation countermeasure network |
| CN109409508B (en) * | 2018-11-06 | 2022-03-15 | 成都信息工程大学 | Method for solving model collapse based on generation of confrontation network use perception loss |
| US11520077B2 (en) | 2018-12-11 | 2022-12-06 | Exxonmobil Upstream Research Company | Automated reservoir modeling using deep generative networks |
| CN109671125B (en) * | 2018-12-17 | 2023-04-07 | 电子科技大学 | Highly-integrated GAN network device and method for realizing text image generation |
| US11017499B2 (en) * | 2018-12-21 | 2021-05-25 | Here Global B.V. | Method, apparatus, and computer program product for generating an overhead view of an environment from a perspective image |
| CN109785253B (en) * | 2018-12-25 | 2021-01-19 | 西安交通大学 | Panchromatic sharpening post-processing method based on enhanced back projection |
| CN109765462A (en) * | 2019-03-05 | 2019-05-17 | 国家电网有限公司 | Fault detection method, device and the terminal device of transmission line of electricity |
| CN109948796B (en) * | 2019-03-13 | 2023-07-04 | 腾讯科技（深圳）有限公司 | Self-encoder learning method, self-encoder learning device, computer equipment and storage medium |
| CN109948117B (en) * | 2019-03-13 | 2023-04-07 | 南京航空航天大学 | Satellite anomaly detection method for network self-encoder |
| CN110009013B (en) * | 2019-03-21 | 2021-04-27 | 腾讯科技（深圳）有限公司 | Encoder training and representation information extraction method and device |
| CN109885091B (en) * | 2019-03-21 | 2021-08-24 | 华北电力大学（保定） | Unmanned aerial vehicle autonomous flight control method and system |
| CN110163815B (en) * | 2019-04-22 | 2022-06-24 | 桂林电子科技大学 | Low-illumination reduction method based on multi-stage variational self-encoder |
| CN110322423B (en) * | 2019-04-29 | 2023-03-31 | 天津大学 | Multi-modal image target detection method based on image fusion |
| CN110189336B (en) * | 2019-05-30 | 2020-05-01 | 上海极链网络科技有限公司 | Image generation method, system, server and storage medium |
| CN110211046B (en) * | 2019-06-03 | 2023-07-14 | 重庆邮电大学 | Remote sensing image fusion method, system and terminal based on generation countermeasure network |
| CN110533578A (en) * | 2019-06-05 | 2019-12-03 | 广东世纪晟科技有限公司 | A kind of image interpretation method based on condition confrontation neural network |
| CN110751699B (en) * | 2019-10-15 | 2023-03-10 | 西安电子科技大学 | Color reconstruction method of optical remote sensing image based on convolutional neural network |
| CN111027603B (en) * | 2019-11-27 | 2022-07-05 | 湖北工业大学 | Image generation method for improving GAN model |
| CN111260594B (en) * | 2019-12-22 | 2023-10-31 | 天津大学 | Unsupervised multi-mode image fusion method |
| CN111274602B (en) * | 2020-01-15 | 2022-11-18 | 腾讯科技（深圳）有限公司 | Image characteristic information replacement method, device, equipment and medium |
| CN111340743B (en) * | 2020-02-18 | 2023-06-06 | 云南大学 | Semi-supervised multispectral and panchromatic remote sensing image fusion method and system |
| CN111325660B (en) * | 2020-02-20 | 2021-01-29 | 中国地质大学（武汉） | Remote sensing image style conversion method based on text data |
| CN111476269B (en) * | 2020-03-04 | 2024-04-16 | 中国平安人寿保险股份有限公司 | Balanced sample set construction and image reproduction identification method, device, equipment and medium |
| CN111428761B (en) * | 2020-03-11 | 2023-03-28 | 深圳先进技术研究院 | Image feature visualization method, image feature visualization device and electronic equipment |
| CN111368935B (en) * | 2020-03-17 | 2023-06-09 | 北京航天自动控制研究所 | SAR time-sensitive target sample amplification method based on generation countermeasure network |
| CN111105417B (en) * | 2020-03-17 | 2023-07-07 | 珠海欧比特宇航科技股份有限公司 | Image noise positioning method and system |
| CN111914935B (en) * | 2020-08-03 | 2022-07-15 | 哈尔滨工程大学 | Ship image target detection method based on deep learning |
| CN112287998A (en) * | 2020-10-27 | 2021-01-29 | 佛山市南海区广工大数控装备协同创新研究院 | Method for detecting target under low-light condition |
| CN112437451B (en) * | 2020-11-10 | 2022-08-02 | 南京大学 | Wireless network flow prediction method and device based on generation countermeasure network |
| CN112561782B (en) * | 2020-12-15 | 2023-01-03 | 哈尔滨工程大学 | Method for improving reality degree of simulation picture of offshore scene |
| CN113012069B (en) * | 2021-03-17 | 2023-09-05 | 中国科学院西安光学精密机械研究所 | Optical remote sensing image quality improving method combining deep learning under wavelet transform domain |
| CN113344846B (en) * | 2021-04-20 | 2023-02-21 | 山东师范大学 | Remote sensing image fusion method and system based on generation countermeasure network and compressed sensing |
| CN114036829B (en) * | 2021-11-02 | 2023-04-25 | 中国地质大学（武汉） | Geological profile generation method, system, equipment and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104112263A (en) * | 2014-06-28 | 2014-10-22 | 南京理工大学 | Method for fusing full-color image and multispectral image based on deep neural network |
| CN106295714A (en) * | 2016-08-22 | 2017-01-04 | 中国科学院电子学研究所 | A kind of multi-source Remote-sensing Image Fusion based on degree of depth study |
| CN107103590A (en) * | 2017-03-22 | 2017-08-29 | 华南理工大学 | A kind of image for resisting generation network based on depth convolution reflects minimizing technology |
| CN107180248A (en) * | 2017-06-12 | 2017-09-19 | 桂林电子科技大学 | Strengthen the hyperspectral image classification method of network based on associated losses |
| CN107239766A (en) * | 2017-06-08 | 2017-10-10 | 深圳市唯特视科技有限公司 | A kind of utilization resists network and the significantly face of three-dimensional configuration model ajusts method |
-
2018
- 2018-03-09 CN CN201810193748.6A patent/CN108537742B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104112263A (en) * | 2014-06-28 | 2014-10-22 | 南京理工大学 | Method for fusing full-color image and multispectral image based on deep neural network |
| CN106295714A (en) * | 2016-08-22 | 2017-01-04 | 中国科学院电子学研究所 | A kind of multi-source Remote-sensing Image Fusion based on degree of depth study |
| CN107103590A (en) * | 2017-03-22 | 2017-08-29 | 华南理工大学 | A kind of image for resisting generation network based on depth convolution reflects minimizing technology |
| CN107239766A (en) * | 2017-06-08 | 2017-10-10 | 深圳市唯特视科技有限公司 | A kind of utilization resists network and the significantly face of three-dimensional configuration model ajusts method |
| CN107180248A (en) * | 2017-06-12 | 2017-09-19 | 桂林电子科技大学 | Strengthen the hyperspectral image classification method of network based on associated losses |
Non-Patent Citations (3)
| Title |
|---|
| DAGAN: Deep De-Aliasing Generative Adversarial Networks for Fast Compressed Sensing MRI Reconstruction;Guang Yang等;《IEEE》;20171221;全文 * |
| IMG2DSM: Height Simulation From Single Imagery Using Conditional Generative Adversarial Net;Pedram Ghamisi等;《IEEE GEOSCIENCE AND REMOTE SENSING LETTERS》;20180305;全文 * |
| PanNet: A deep network architecture for pan-sharpening;Junfeng Yang等;《2017 IEEE International Conference on Computer Vision》;20171231;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108537742A (en) | 2018-09-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108537742B (en) | Remote sensing image panchromatic sharpening method based on generation countermeasure network | |
| Zheng et al. | FPGA: Fast patch-free global learning framework for fully end-to-end hyperspectral image classification | |
| CN108564109B (en) | Remote sensing image target detection method based on deep learning | |
| Chen et al. | Symmetrical dense-shortcut deep fully convolutional networks for semantic segmentation of very-high-resolution remote sensing images | |
| CN110363215B (en) | Method for converting SAR image into optical image based on generating type countermeasure network | |
| CN105740894B (en) | Semantic annotation method for hyperspectral remote sensing image | |
| CN111274869B (en) | Method for classifying hyperspectral images based on parallel attention mechanism residual error network | |
| CN109993072B (en) | Low-resolution pedestrian re-identification system and method based on super-resolution image generation | |
| CN108648197B (en) | Target candidate region extraction method based on image background mask | |
| Fu et al. | Joint camera spectral sensitivity selection and hyperspectral image recovery | |
| CN115690479A (en) | Remote sensing image classification method and system based on convolution Transformer | |
| CN110929736A (en) | Multi-feature cascade RGB-D significance target detection method | |
| CN106295613A (en) | A kind of unmanned plane target localization method and system | |
| CN112347888A (en) | Remote sensing image scene classification method based on bidirectional feature iterative fusion | |
| CN114187450A (en) | Remote sensing image semantic segmentation method based on deep learning | |
| CN107423747A (en) | A kind of conspicuousness object detection method based on depth convolutional network | |
| CN115205590A (en) | Hyperspectral image classification method based on complementary integration Transformer network | |
| CN112257741B (en) | Method for detecting generative anti-false picture based on complex neural network | |
| Wang et al. | Convolutional LSTM-based hierarchical feature fusion for multispectral pan-sharpening | |
| CN113807356B (en) | End-to-end low-visibility image semantic segmentation method | |
| CN115861756A (en) | Earth background small target identification method based on cascade combination network | |
| Shang et al. | Spatiotemporal reflectance fusion using a generative adversarial network | |
| Ma et al. | Multi-task interaction learning for spatiospectral image super-resolution | |
| CN113256733B (en) | Camera spectral sensitivity reconstruction method based on confidence voting convolutional neural network | |
| Jaffe et al. | Remote sensor design for visual recognition with convolutional neural networks |
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 |