CN113034361A - Remote sensing image super-resolution reconstruction method based on improved ESRGAN - Google Patents
Remote sensing image super-resolution reconstruction method based on improved ESRGAN Download PDFInfo
- Publication number
- CN113034361A CN113034361A CN202110236027.0A CN202110236027A CN113034361A CN 113034361 A CN113034361 A CN 113034361A CN 202110236027 A CN202110236027 A CN 202110236027A CN 113034361 A CN113034361 A CN 113034361A
- Authority
- CN
- China
- Prior art keywords
- network
- image
- convolution
- remote sensing
- layer
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 23
- 230000004913 activation Effects 0.000 claims abstract description 12
- 230000003321 amplification Effects 0.000 claims abstract description 10
- 238000003199 nucleic acid amplification method Methods 0.000 claims abstract description 10
- 238000010276 construction Methods 0.000 claims abstract description 5
- 238000012549 training Methods 0.000 claims abstract description 4
- 238000004364 calculation method Methods 0.000 claims description 9
- 238000012545 processing Methods 0.000 claims description 9
- 239000011159 matrix material Substances 0.000 claims description 8
- 230000004927 fusion Effects 0.000 claims description 7
- 230000008569 process Effects 0.000 claims description 6
- 230000003213 activating effect Effects 0.000 claims description 3
- 238000000605 extraction Methods 0.000 claims description 3
- 230000009286 beneficial effect Effects 0.000 abstract description 3
- 230000006870 function Effects 0.000 description 14
- 238000005516 engineering process Methods 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 2
- 230000000295 complement effect Effects 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000002349 favourable effect Effects 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformation in the plane of the image
- G06T3/40—Scaling the whole image or part thereof
- G06T3/4053—Super resolution, i.e. output image resolution higher than sensor resolution
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformation in the plane of the image
- G06T3/40—Scaling the whole image or part thereof
- G06T3/4007—Interpolation-based scaling, e.g. bilinear interpolation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformation in the plane of the image
- G06T3/40—Scaling the whole image or part thereof
- G06T3/4046—Scaling the whole image or part thereof using neural networks
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a remote sensing image super-resolution reconstruction method based on improved ESRGAN, which comprises the following steps: constructing an improved remote sensing image super-resolution reconstruction network model, which comprises a generation network and a discrimination network; generating a network composed of: the system comprises 64 convolution kernels with the size of 3x3, a residual error network formed by 23 RRDB modules and a LeakyReLU activation function; the judgment network comprises 6 layers, a full convolution network with even-number convolution kernel is adopted, and a BN layer and a LeakyReLU activation layer are added for construction; the first layer of inputs to the discrimination network are: carrying out channel merging on an image obtained by carrying out bicubic interpolation amplification on the original low-resolution remote sensing image real _ A and a fake _ B image generated by a generation network; and alternately training the generating network and the judging network, updating parameters of the generating network and the judging network, and finally obtaining an improved remote sensing image hyper-resolution reconstruction network model. The invention has the beneficial effects that: the high-definition image with the definition and the texture characteristics closer to the real high-resolution remote sensing image can be generated.
Description
Technical Field
The invention relates to the field of remote sensing image processing, in particular to a remote sensing image super-resolution reconstruction method based on improved ESRGAN.
Background
The satellite remote sensing image can rapidly provide information of the earth surface, but the satellite remote sensing images with medium and low resolutions have certain limitations for extracting high-precision ground features, updating maps, identifying targets and the like. The development of the satellite remote sensing image with high resolution makes the deep application of the remote sensing image possible, thereby providing favorable conditions for updating GIS data and applying GIS. The method is also significant for map updating, image matching, target detection and the like.
In the field of remote sensing, due to the influence of imaging technology and shooting equipment, a high-resolution remote sensing image is difficult to obtain, an unmanned aerial vehicle is often required to carry out aerial shooting, and both manpower and material resources are consumed, so that the technology for realizing image super-resolution reconstruction from the perspective of an algorithm becomes a hot research topic in multiple fields such as image processing and computer vision.
In recent years, more and more researchers carry out super-resolution reconstruction by using a deep learning method, and good progress is made, wherein the most impressive most deep-learning method is an ESRGAN super-resolution reconstruction network, the network adopts a multi-level RRDB module to replace a basic residual error network, and a BN module in a generated network is removed, so that the image super-resolution reconstruction technology is greatly improved. However, in the remote sensing field, due to the fact that processing such as compression and fusion exists in images shot by satellites, texture detail loss of the obtained low-resolution remote sensing images is serious, and the original ESRGAN network is prone to have the problems of artifacts, texture detail distortion and the like in the low-resolution remote sensing images shot by the reconstructed satellites.
Disclosure of Invention
In view of the above, the technical problems to be solved by the present invention are: how to generate a high-resolution remote sensing image with more vivid detail texture. Aiming at the situation, the invention provides an improved ESRGAN remote sensing image super-resolution reconstruction network, which reserves a strong generation network architecture of an original ESRGAN, and in the aspect of network discrimination, a VGG network adopted by the original ESRGAN is abandoned, a full convolution network with even-number-size convolution kernels is adopted, a BN layer and a LeakyReLU activation layer are added for construction, and finally, an 8x8 prediction matrix is output, and the matrix is averaged to finish discrimination.
The invention provides a remote sensing image super-resolution reconstruction method based on improved ESRGAN, which comprises the following steps:
s101: constructing an improved remote sensing image super-resolution reconstruction network model; the improved remote sensing image super-resolution reconstruction network model is based on an ESRGAN architecture and comprises a generation network net _ G and a judgment network net _ D; generating a false image fake _ B for deceiving a discrimination network by a network;
generating a network composed of: the system comprises 64 convolution kernels with the size of 3x3, a residual error network formed by 23 RRDB modules and a LeakyReLU activation function; the judgment network comprises 6 layers, a full convolution network with even-number convolution kernel is adopted, a BN layer and a LeakyReLU activation layer are added for construction, finally, a prediction matrix of 8x8 is output, and the matrix is averaged to finish the judgment;
the first layer of inputs to the discrimination network are: carrying out channel merging on an image obtained by carrying out bicubic interpolation amplification on the original low-resolution remote sensing image real _ A and a fake _ B image generated by a generation network;
s102: alternately training the generating network and the discriminating network: and generating a false image fake _ B for deceiving the discrimination network by using the generated network, updating parameters of the generated network and the discrimination network, and finally obtaining an improved remote sensing image hyper-resolution reconstruction network model.
Further, the specific structure of the discrimination network is as follows:
a first layer: 64 convolution networks of 4 × 4 convolution kernels with a convolution step size of 2;
a second layer: 128 convolution layers with convolution kernel of 4x4 and convolution step size of 2, and subsequently connected with BN layer and LeakyReLU layer;
and a third layer: 256 convolution layers with convolution kernel of 4x4 and convolution step size of 2, and subsequently connecting a BN layer and a LeakyReLU layer;
a fourth layer: 512 convolution layers with convolution kernel of 4x4 and convolution step size of 2, and subsequently connecting a BN layer and a LeakyReLU layer;
the fifth layer is 512 convolution layers with convolution kernels of 4x4 and convolution step size of 1, and is subsequently connected with a BN layer and a LeakyReLU layer;
a sixth layer: 1 convolution layer of 4 × 4 convolution kernels with a convolution step size of 1.
Further, the process of generating the network-generated false image fake _ B specifically includes:
s201: acquiring an original low-resolution remote sensing image real _ A and an original high-resolution remote sensing image real _ B;
s202: performing convolution operation on the low-resolution remote sensing image real _ A by using 64 convolution kernels with the size of 3 multiplied by 3 to obtain an original characteristic image;
s203: inputting the original characteristic image into a residual error network constructed by 23 RRDB modules in an ESRGAN network for characteristic extraction and characteristic fusion to obtain a processed characteristic image;
s204: performing convolution operation again on the feature images with the sizes of 64 and 3 multiplied by 3 after the convolution check processing to obtain the operated feature images;
s205: fusing the calculated characteristic image with the original characteristic image, performing nearest neighbor interpolation processing, and activating through a LeakyReLU function to generate a super-resolution image, namely a fake image fake _ B; the super-resolution image has the same size as the original high-resolution video real _ B.
Further, in step S102, the loss calculation formula for the generated network is as follows:
loss_G＝0.2*loss_G_pix+loss_G_feature+loss_G_gan (1)
in the formula (1), Loss _ G _ pix is L1Loss of pixel values of both the fake _ B image and the original high-resolution real _ B image generated by the generation network; the Loss _ G _ feature is L1Loss of pixel values of two characteristic images after a fake _ B image and a real _ B image are respectively input into a VGG16 network before activation; loss _ G _ gan is BCEWithLogitsLoss calculated according to the output value of the discrimination network net _ D;
when network loss calculation and parameter updating are generated, judging network parameters are kept unchanged; and finally, the loss _ G after fusion is the objective function of the generated network part.
Further, in step S102, the loss calculation formula of the discrimination network is as follows:
loss_D＝0.5*(loss_D_fake+loss_G_real) (2)
in the formula (2), loss _ D _ fake is BCEWithLoctitsLoss which inputs the output value of the discrimination network net _ D after the original low-resolution remote sensing image real _ A is subjected to bicubic interpolation amplification and fake _ B image generated by a generation network are subjected to channel merging; and the loss _ G _ real is BCEWithLogitsLoss which inputs the output value of the judgment network net _ D after the channel merging of the image obtained by the bicubic interpolation amplification of the original low-resolution remote sensing image real _ A and the original high-resolution remote sensing image real _ B.
The technical scheme provided by the invention has the beneficial effects that: the network provided by the invention can more accurately extract the characteristics of an input image and recover smoother and vivid textural characteristics, simultaneously changes the input of a discriminator, amplifies a low-definition remote sensing image and adds the amplified low-definition remote sensing image and an image channel generated by a generation network to input the amplified low-definition remote sensing image and the image channel into the discrimination network for discrimination.
Drawings
Fig. 1 is a schematic flow chart of a remote sensing image super-resolution reconstruction method based on improved ESRGAN according to the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, embodiments of the present invention will be further described with reference to the accompanying drawings.
Referring to fig. 1, the present invention provides a remote sensing image super-resolution reconstruction method based on improved ESRGAN, which specifically includes the following steps:
s101: constructing an improved remote sensing image super-resolution reconstruction network model; the improved remote sensing image super-resolution reconstruction network model is based on an ESRGAN architecture and comprises a generation network net _ G and a judgment network net _ D; generating a false image fake _ B for deceiving a discrimination network by a network;
generating a network composed of: the system comprises 64 convolution kernels with the size of 3x3, a residual error network formed by 23 RRDB modules and a LeakyReLU activation function; the discrimination network comprises 6 layers, and the specific structure is as follows:
a first layer: 64 convolution networks of 4 × 4 convolution kernels with a convolution step size of 2;
the first layer of inputs to the discrimination network are: carrying out channel merging on an image obtained by carrying out bicubic interpolation amplification on the original low-resolution remote sensing image real _ A and a fake _ B image generated by a generation network;
preferably, the input of the first layer is implemented using a torch.nnsequential () function; the process is as follows:
torch.nn.Sequential(torch.nn.Conv2d(in_channels＝6,out_channels＝64，kernel_size＝4,stride＝2,padding＝1),torch.nn.LeakyReLU())
where in _ channels represents the number of channels of the input fused image; out _ channels represents the number of channels of the output image, and also represents the number of convolution kernels; kernel _ size represents the size of the convolution kernel; stride represents the step size of the convolution kernel move; padding indicates the size of the input image boundary complement 0, and torch.
A second layer: 128 convolution layers with convolution kernel of 4x4 and convolution step size of 2, and subsequently connected with BN layer and LeakyReLU layer;
preferably, the implementation of the second layer calls the function as follows:
torch.nn.Sequential(torch.nn.Conv2d(in_channels＝64,out_channels＝128，kernel_size＝4,stride＝2,padding＝1),torch.nn.BatchNorm2d(128)，torch.nn.LeakyReLU(0.2,True))
wherein, the torch.nn.batchnorm2d represents the batch normalization process performed on the convolved images.
And a third layer: 256 convolution layers with convolution kernel of 4x4 and convolution step size of 2, and subsequently connecting a BN layer and a LeakyReLU layer;
preferably, the implementation of the third layer calls the function as follows:
torch.nn.Sequential(torch.nn.Conv2d(in_channels＝128,out_channels＝256，kernel_size＝4,stride＝2,padding＝1),torch.nn.BatchNorm2d(256)，torch.nn.LeakyReLU(0.2,True))；
a fourth layer: 512 convolution layers with convolution kernel of 4x4 and convolution step size of 2, and subsequently connecting a BN layer and a LeakyReLU layer;
preferably, the implementation of the fourth layer calls the function as follows:
torch.nn.Sequential(torch.nn.Conv2d(in_channels＝256,out_channels＝512，kernel_size＝4,stride＝2,padding＝1),torch.nn.BatchNorm2d(512)，torch.nn.LeakyReLU(0.2,True))；
the fifth layer is 512 convolution layers with convolution kernels of 4x4 and convolution step size of 1, and is subsequently connected with a BN layer and a LeakyReLU layer;
preferably, the implementation of the fifth layer calls the function as follows:
torch.nn.Sequential(torch.nn.Conv2d(in_channels＝512,out_channels＝512，kernel_size＝4,stride＝1,padding＝1),torch.nn.BatchNorm2d(512)，torch.nn.LeakyReLU(0.2,True))；
a sixth layer: 1 convolution layer of 4 × 4 convolution kernels with a convolution step size of 1.
Preferably, the implementation of the sixth layer calls the function as follows:
torch.nn.Conv2d(in_channels＝512,out_channels＝1，kernel_size＝4,stride＝1,padding＝1)；
s102: alternately training the generating network and the discriminating network: and generating a false image fake _ B for deceiving the discrimination network by using the generated network, updating parameters of the generated network and the discrimination network, and finally obtaining an improved remote sensing image hyper-resolution reconstruction network model.
The process of generating the network-generated false image fake _ B specifically includes:
s201: acquiring an original low-resolution remote sensing image real _ A and an original high-resolution remote sensing image real _ B;
s202: performing convolution operation on the low-resolution remote sensing image real _ A by using 64 convolution kernels with the size of 3 multiplied by 3 to obtain an original characteristic image;
the part is realized by a torch.nn.Conv2d () function, and the calling process specifically comprises the following steps:
torch.nn.Conv2d(in_channels＝3,out_channels＝64，kernel_size＝3,stride＝1,padding＝1)；
s203: inputting the original characteristic image into a residual error network constructed by 23 RRDB modules in an ESRGAN network for characteristic extraction and characteristic fusion to obtain a processed characteristic image;
the part is realized by constructing 23 RRDB modules by adopting a make _ layer method of models.
S204: performing convolution operation again on the feature images with the sizes of 64 and 3 multiplied by 3 after the convolution check processing to obtain the operated feature images;
this step is still implemented using the torch.nn.conv2d () function;
s205: fusing the calculated characteristic image with the original characteristic image, performing nearest neighbor interpolation processing, and activating through a LeakyReLU function to generate a super-resolution image, namely a fake image fake _ B; the super-resolution image is the same size as the original high-resolution video real _ B and is four times the original low-resolution image.
In step S102, a network loss calculation formula is generated as shown in formula (1):
loss_G＝0.2*loss_G_pix+loss_G_feature+loss_G_gan (1)
in the formula (1), Loss _ G _ pix is L1Loss of pixel values of both the fake _ B image and the original high-resolution real _ B image generated by the generation network; the Loss _ G _ feature is L1Loss of pixel values of two characteristic images after a fake _ B image and a real _ B image are respectively input into a VGG16 network before activation; loss _ G _ gan is BCEWithLogitsLoss calculated according to the output value of the discrimination network net _ D;
when network loss calculation and parameter updating are generated, judging network parameters are kept unchanged; and finally, the loss _ G after fusion is the objective function of the generated network part.
In step S102, the loss calculation formula of the network is determined as shown in formula (2):
loss_D＝0.5*(loss_D_fake+loss_G_real) (2)
in the formula (2), loss _ D _ fake is BCEWithLoctitsLoss which inputs the output value of the discrimination network net _ D after the original low-resolution remote sensing image real _ A is subjected to bicubic interpolation amplification and fake _ B image generated by a generation network are subjected to channel merging; loss _ G _ real is BCEWithLogitsLoss which is obtained by carrying out channel merging on an image obtained by carrying out bicubic interpolation amplification on original low-resolution remote sensing image real _ A and an original high-resolution remote sensing image real _ B and then inputting an output value of a judgment network net _ D
In general, compared with the prior art, the above technical solution contemplated by the present invention can achieve the following beneficial effects:
the VGG network adopted by the original ESRGAN is abandoned, the full convolution network of convolution kernel with even number size is adopted, the BN layer and the LeakyReLU activation layer are added for construction, finally, a prediction matrix of 8x8 is output, and the matrix is averaged to complete the judgment. The network can extract more accurate characteristics of an input image and recover more smooth and vivid textural characteristics, simultaneously changes the input of a discriminator, amplifies a low-definition remote sensing image and adds an image channel generated by a generation network to input the amplified low-definition remote sensing image and the image channel into the discrimination network for discrimination.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.
Claims (5)
1. A remote sensing image super-resolution reconstruction method based on improved ESRGAN is characterized in that: the method specifically comprises the following steps:
s101: constructing an improved remote sensing image super-resolution reconstruction network model; the improved remote sensing image super-resolution reconstruction network model is based on an ESRGAN architecture and comprises a generation network net _ G and a judgment network net _ D; generating a false image fake _ B for deceiving a discrimination network by a network;
generating a network composed of: the system comprises 64 convolution kernels with the size of 3x3, a residual error network formed by 23 RRDB modules and a LeakyReLU activation function; the judgment network comprises 6 layers, a full convolution network with even-number convolution kernel is adopted, a BN layer and a LeakyReLU activation layer are added for construction, finally, a prediction matrix of 8x8 is output, and the matrix is averaged to finish the judgment; the first layer of inputs to the discrimination network are: carrying out channel merging on an image obtained by carrying out bicubic interpolation amplification on the original low-resolution remote sensing image real _ A and a fake _ B image generated by a generation network;
s102: alternately training the generating network and the discriminating network: and generating a false image fake _ B for deceiving the discrimination network by using the generated network, updating parameters of the generated network and the discrimination network, and finally obtaining an improved remote sensing image hyper-resolution reconstruction network model.
2. The remote sensing image super-resolution reconstruction method based on the improved ESRGAN as claimed in claim 1, wherein: the specific structure of the discrimination network is as follows:
a first layer: 64 convolution networks of 4 × 4 convolution kernels with a convolution step size of 2;
a second layer: 128 convolution layers with convolution kernel of 4x4 and convolution step size of 2, and subsequently connected with BN layer and LeakyReLU layer;
and a third layer: 256 convolution layers with convolution kernel of 4x4 and convolution step size of 2, and subsequently connecting a BN layer and a LeakyReLU layer;
a fourth layer: 512 convolution layers with convolution kernel of 4x4 and convolution step size of 2, and subsequently connecting a BN layer and a LeakyReLU layer;
the fifth layer is 512 convolution layers with convolution kernels of 4x4 and convolution step size of 1, and is subsequently connected with a BN layer and a LeakyReLU layer;
a sixth layer: 1 convolution layer of 4 × 4 convolution kernels with a convolution step size of 1.
3. The remote sensing image super-resolution reconstruction method based on the improved ESRGAN as claimed in claim 1, wherein: the process of generating the network-generated false image fake _ B specifically includes:
s201: acquiring an original low-resolution remote sensing image real _ A and an original high-resolution remote sensing image real _ B;
s202: performing convolution operation on the low-resolution remote sensing image real _ A by using 64 convolution kernels with the size of 3 multiplied by 3 to obtain an original characteristic image;
s203: inputting the original characteristic image into a residual error network constructed by 23 RRDB modules in an ESRGAN network for characteristic extraction and characteristic fusion to obtain a processed characteristic image;
s204: performing convolution operation again on the feature images with the sizes of 64 and 3 multiplied by 3 after the convolution check processing to obtain the operated feature images;
s205: fusing the calculated characteristic image with the original characteristic image, performing nearest neighbor interpolation processing, and activating through a LeakyReLU function to generate a super-resolution image, namely a fake image fake _ B; the super-resolution image has the same size as the original high-resolution video real _ B.
4. The remote sensing image super-resolution reconstruction method based on the improved ESRGAN as claimed in claim 1, wherein: in step S102, a network loss calculation formula is generated as shown in formula (1):
loss_G＝0.2*loss_G_pix+loss_G_feature+loss_G_gan (1)
in the formula (1), Loss _ G _ pix is L1Loss of pixel values of both the fake _ B image and the original high-resolution real _ B image generated by the generation network; the Loss _ G _ feature is L1Loss of pixel values of two characteristic images after a fake _ B image and a real _ B image are respectively input into a VGG16 network before activation; loss _ G _ gan is BCEWithLogitsLoss calculated according to the output value of the discrimination network net _ D;
when network loss calculation and parameter updating are generated, judging network parameters are kept unchanged; and finally, the loss _ G after fusion is the objective function of the generated network part.
5. The remote sensing image super-resolution reconstruction method based on the improved ESRGAN as claimed in claim 1, wherein: in step S102, the loss calculation formula of the network is determined as shown in formula (2):
loss_D＝0.5*(loss_D_fake+loss_G_real) (2)
in the formula (2), loss _ D _ fake is BCEWithLoctitsLoss which inputs the output value of the discrimination network net _ D after the original low-resolution remote sensing image real _ A is subjected to bicubic interpolation amplification and fake _ B image generated by a generation network are subjected to channel merging; and the loss _ G _ real is BCEWithLogitsLoss which inputs the output value of the judgment network net _ D after the channel merging of the image obtained by the bicubic interpolation amplification of the original low-resolution remote sensing image real _ A and the original high-resolution remote sensing image real _ B.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110236027.0A CN113034361B (en) | 2021-03-03 | 2021-03-03 | Remote sensing image super-resolution reconstruction method based on improved ESRGAN |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110236027.0A CN113034361B (en) | 2021-03-03 | 2021-03-03 | Remote sensing image super-resolution reconstruction method based on improved ESRGAN |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113034361A true CN113034361A (en) | 2021-06-25 |
| CN113034361B CN113034361B (en) | 2022-10-14 |
Family
ID=76466139
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110236027.0A Expired - Fee Related CN113034361B (en) | 2021-03-03 | 2021-03-03 | Remote sensing image super-resolution reconstruction method based on improved ESRGAN |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113034361B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113516591A (en) * | 2021-07-20 | 2021-10-19 | 海南长光卫星信息技术有限公司 | Remote sensing image super-resolution reconstruction method, device, equipment and storage medium |
| WO2023000158A1 (en) * | 2021-07-20 | 2023-01-26 | 海南长光卫星信息技术有限公司 | Super-resolution reconstruction method, apparatus and device for remote sensing image, and storage medium |
| CN115982418A (en) * | 2023-03-17 | 2023-04-18 | 亿铸科技(杭州)有限责任公司 | Method for improving super-division operation performance of AI (Artificial Intelligence) computing chip |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111080531A (en) * | 2020-01-10 | 2020-04-28 | 北京农业信息技术研究中心 | Super-resolution reconstruction method, system and device for underwater fish image |
| CN111127316A (en) * | 2019-10-29 | 2020-05-08 | 山东大学 | Single face image super-resolution method and system based on SNGAN network |
| CN111626237A (en) * | 2020-05-29 | 2020-09-04 | 中国民航大学 | Crowd counting method and system based on enhanced multi-scale perception network |
| US20200387698A1 (en) * | 2018-07-10 | 2020-12-10 | Tencent Technology (Shenzhen) Company Limited | Hand key point recognition model training method, hand key point recognition method and device |
| CN112288632A (en) * | 2020-10-29 | 2021-01-29 | 福州大学 | Single image super-resolution method and system based on simplified ESRGAN |
| CN112396554A (en) * | 2019-08-14 | 2021-02-23 | 天津大学青岛海洋技术研究院 | Image super-resolution algorithm based on generation countermeasure network |
| CN112435164A (en) * | 2020-11-23 | 2021-03-02 | 浙江工业大学 | Method for simultaneously super-resolution and denoising of low-dose CT lung image based on multi-scale generation countermeasure network |
-
2021
- 2021-03-03 CN CN202110236027.0A patent/CN113034361B/en not_active Expired - Fee Related
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20200387698A1 (en) * | 2018-07-10 | 2020-12-10 | Tencent Technology (Shenzhen) Company Limited | Hand key point recognition model training method, hand key point recognition method and device |
| CN112396554A (en) * | 2019-08-14 | 2021-02-23 | 天津大学青岛海洋技术研究院 | Image super-resolution algorithm based on generation countermeasure network |
| CN111127316A (en) * | 2019-10-29 | 2020-05-08 | 山东大学 | Single face image super-resolution method and system based on SNGAN network |
| CN111080531A (en) * | 2020-01-10 | 2020-04-28 | 北京农业信息技术研究中心 | Super-resolution reconstruction method, system and device for underwater fish image |
| CN111626237A (en) * | 2020-05-29 | 2020-09-04 | 中国民航大学 | Crowd counting method and system based on enhanced multi-scale perception network |
| CN112288632A (en) * | 2020-10-29 | 2021-01-29 | 福州大学 | Single image super-resolution method and system based on simplified ESRGAN |
| CN112435164A (en) * | 2020-11-23 | 2021-03-02 | 浙江工业大学 | Method for simultaneously super-resolution and denoising of low-dose CT lung image based on multi-scale generation countermeasure network |
Non-Patent Citations (3)
| Title |
|---|
| ISOLA P ET AL: "Image-to-image translation with conditional adversarial networks", 《IN PROCEEDINGS OF THE PROCEEDINGS OF THE IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION》 * |
| N. C. RAKOTONIRINA ET AL: "ESRGAN+ : Further Improving Enhanced Super-Resolution Generative Adversarial Network", 《ICASSP 2020 - 2020 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS》 * |
| XU B ET AL: "Empirical evaluation of rectified activations in convolutional network", 《ARXIV:1505.00853》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113516591A (en) * | 2021-07-20 | 2021-10-19 | 海南长光卫星信息技术有限公司 | Remote sensing image super-resolution reconstruction method, device, equipment and storage medium |
| WO2023000158A1 (en) * | 2021-07-20 | 2023-01-26 | 海南长光卫星信息技术有限公司 | Super-resolution reconstruction method, apparatus and device for remote sensing image, and storage medium |
| CN115982418A (en) * | 2023-03-17 | 2023-04-18 | 亿铸科技(杭州)有限责任公司 | Method for improving super-division operation performance of AI (Artificial Intelligence) computing chip |
| CN115982418B (en) * | 2023-03-17 | 2023-05-30 | 亿铸科技(杭州)有限责任公司 | Method for improving super-division operation performance of AI (advanced technology attachment) computing chip |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113034361B (en) | 2022-10-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113034361B (en) | Remote sensing image super-resolution reconstruction method based on improved ESRGAN | |
| CN110033410B (en) | Image reconstruction model training method, image super-resolution reconstruction method and device | |
| Engin et al. | Cycle-dehaze: Enhanced cyclegan for single image dehazing | |
| CN111598778B (en) | Super-resolution reconstruction method for insulator image | |
| CN108921783B (en) | Satellite image super-resolution reconstruction method based on mixed loss function constraint | |
| Xie et al. | Deep coordinate attention network for single image super‐resolution | |
| CN110363068B (en) | High-resolution pedestrian image generation method based on multiscale circulation generation type countermeasure network | |
| CN112699844B (en) | Image super-resolution method based on multi-scale residual hierarchy close-coupled network | |
| CN113096017A (en) | Image super-resolution reconstruction method based on depth coordinate attention network model | |
| CN111640060A (en) | Single image super-resolution reconstruction method based on deep learning and multi-scale residual dense module | |
| Guan et al. | Srdgan: learning the noise prior for super resolution with dual generative adversarial networks | |
| CN110634103A (en) | Image demosaicing method based on generation of countermeasure network | |
| CN112949636A (en) | License plate super-resolution identification method and system and computer readable medium | |
| Fang et al. | High-resolution optical flow and frame-recurrent network for video super-resolution and deblurring | |
| Yu et al. | Split-attention multiframe alignment network for image restoration | |
| CN110751271B (en) | Image traceability feature characterization method based on deep neural network | |
| Zhang et al. | An effective decomposition-enhancement method to restore light field images captured in the dark | |
| CN113379606B (en) | Face super-resolution method based on pre-training generation model | |
| Zheng et al. | Double-branch dehazing network based on self-calibrated attentional convolution | |
| CN111161156A (en) | Deep learning-based underwater pier disease image resolution enhancement method | |
| Wang et al. | Underwater image super-resolution using multi-stage information distillation networks | |
| Chen et al. | Guided dual networks for single image super-resolution | |
| Nie et al. | Context and detail interaction network for stereo rain streak and raindrop removal | |
| CN116823611A (en) | Multi-focus image-based referenced super-resolution method | |
| CN113674154B (en) | Single image super-resolution reconstruction method and system based on generation countermeasure network |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20221014 |