CN117876692A - Feature weighted connection guided single-image remote sensing image denoising method - Google Patents
Feature weighted connection guided single-image remote sensing image denoising method Download PDFInfo
- Publication number
- CN117876692A CN117876692A CN202410270382.3A CN202410270382A CN117876692A CN 117876692 A CN117876692 A CN 117876692A CN 202410270382 A CN202410270382 A CN 202410270382A CN 117876692 A CN117876692 A CN 117876692A
- Authority
- CN
- China
- Prior art keywords
- denoising
- image
- feature
- remote sensing
- neurons
- 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 61
- 238000012549 training Methods 0.000 claims abstract description 15
- 238000005070 sampling Methods 0.000 claims abstract description 10
- 210000002569 neuron Anatomy 0.000 claims description 37
- 239000011159 matrix material Substances 0.000 claims description 10
- 238000010586 diagram Methods 0.000 claims description 7
- 238000000605 extraction Methods 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 3
- 239000013589 supplement Substances 0.000 claims description 3
- 230000001502 supplementing effect Effects 0.000 claims description 3
- 238000013528 artificial neural network Methods 0.000 abstract description 3
- 230000003321 amplification Effects 0.000 abstract 1
- 238000003199 nucleic acid amplification method Methods 0.000 abstract 1
- 230000006870 function Effects 0.000 description 4
- 238000013527 convolutional neural network Methods 0.000 description 3
- 238000001914 filtration Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000004438 eyesight Effects 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000011158 quantitative evaluation Methods 0.000 description 1
- 230000008054 signal transmission Effects 0.000 description 1
- 230000016776 visual perception Effects 0.000 description 1
Classifications
-
- 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
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a feature weighted connection guided single-image remote sensing image denoising method, which belongs to the technical field of image denoising and is used for image denoising, and comprises the steps of carrying out data amplification on an input single image by utilizing random Bernoulli sampling, training a neural network, fixing trained multiple groups of neural network parameters and obtaining multiple denoising sub-models; and (3) carrying out random Dropout discarding part of parameters on all the denoising submodels, inputting the target image into all the submodels, and taking the average value of all the submodels as a result. The invention comprehensively considers the feature weighting and the self-supervision constraint to improve the effectiveness and the robustness of the model. The method is a blind denoising method which only needs single-image input, can be used for solving the problems that supervised learning training data are limited and denoising cannot be effectively performed, and can also solve the problem that detail textures of images are blurred after self-supervised learning denoising, so that the denoising performance of remote sensing images is greatly improved.
Description
Technical Field
The invention discloses a single-image remote sensing image denoising method guided by feature weighting connection, and belongs to the technical field of image denoising.
Background
Due to the interference of the device and the long-distance transmission of signals, the remote sensing image is usually affected by noise, so that the edge texture details are blurred, and the image quality is reduced. The existence of noise not only interferes with the visual perception of the remote sensing image, but also reduces the precision of subsequent image processing, and brings much trouble to target detection and image segmentation. Therefore, denoising an optical remote sensing image with large noise is an indispensable task in order to obtain a clear and high-quality remote sensing image. Changing the hardware device may generally eliminate periodic noise in the remote sensing image. However, due to the presence of shot noise, many random noises are still widely present in the remote sensing image. Many researchers have attempted to eliminate noise signals by image processing methods.
Disclosure of Invention
The invention aims to provide a single-image remote sensing image denoising method guided by feature weighting connection, which aims to solve the problem of low image denoising precision in the prior art.
A feature weighted connection guided single-image remote sensing image denoising method comprises the following steps:
s1, using random Bernoulli sampling to input images to obtain image pairsRepeating the sampling for a plurality of times to obtain an image pair set +.>；
S2, designing a dynamic characteristic weighting branch module and a neuron branch module;
the dynamic characteristic weighting branch module learns channel information and space information from the characteristic diagram through a convolution-based extractor, and constructs a dynamic matrix by using the channel information and the space information, and the dynamic matrix supplements correlation information of neurons;
the neuron branching module calculates the activation degree of each neuron according to the correlation among the neurons;
s3, designing a feature weight connection module based on the structure of the U-net, and connecting a feature weighting module of the same layer in the encoder and the decoder;
s4, after the modules in the S2 and the S3 are designed, an untrained denoising model is formed, wherein the denoising model comprises a dynamic characteristic weighting branch module, a neuron branch module and a characteristic weight connecting module;
model training is carried out, and the denoising device is optimized by minimizing a loss function;
s5, model speculation, and a trained denoising modelA series of new denoising submodels { are obtained after Dropout is carried outNew random bernoulli samples, different from the training phase, are performed on the noise image and an instance of the bernoulli samples is fed to each newly formed denoiser.
S1 comprises, randomly bernoulli sampling:
；/>；
where M represents the number of image pairs, M represents the mth image pair,representing the matrix by multiplying correspondingly element by element, < >>Representing a two-dimensional vector sample in a bernoulli sample.
S2 comprises an energy function of S2.1. Degree of activationThe method comprises the following steps:
；/>；/>；；
；/>；/>；；
in the method, in the process of the invention,、/>、/>、/>、/>and->Is an intermediate parameter->Is corresponding to->Weights of individual neurons, +.>For the t-th target neuron, +.>Refers to other neurons in the same channel.
S2 comprises, S2.2 solving for the minimum energy of the target neuronThe method comprises the following steps:
；
usingThe degree of importance of the neurons is reflected.
Corresponding weight matrix->Is self-adaptive, automatically generated by a weight generator; applying a convolution-based feature extractor, the extractor pair size being +.>Is operated on to obtain a characteristic map of the size +.>Is defined as the characteristic vector of the target neuron->Is to be used for the general purpose of (a)Track dimension weights; after traversing all neurons, a size of +.>Weight vector +.>Arranged according to the spatial position of the corresponding neurons to obtain a size of +.>Weight matrix->Weight matrix->The method is used for supplementing the inter-channel correlation information and improving the capability of the attenuation module for extracting similar features of different dimensions in the remote sensing image.
S3, providing the feature weights learned from the encoder to corresponding feature weighting modules in the decoder, wherein the weighting modules in the decoder learn the feature weights from the current features and learn the weight thermodynamic diagrams from the corresponding weighting modules in the encoder, so that the feature weights can be transferred between the encoder and the decoder, and the consistency of the feature extraction stage and the generation stage in the overall denoising task is ensured.
S4 comprises the following steps:
；
in the method, in the process of the invention,for the parameter +.>Is>Is->Is included in the sample component of (a).
Compared with the prior art, the invention has the following beneficial effects: the invention provides a single-image remote sensing image denoising method guided by feature weighting connection, which comprehensively considers feature weighting and self-supervision constraint to improve the effectiveness and robustness of a model. The method is a blind denoising method which only needs single-image input, can be used for solving the problems that supervised learning training data are limited and denoising cannot be effectively performed, and can also solve the problem that detail textures of images are blurred after self-supervised learning denoising, so that the denoising performance of remote sensing images is greatly improved.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the present invention more apparent, the technical solutions in the present invention will be clearly and completely described below, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
A feature weighted connection guided single-image remote sensing image denoising method comprises the following steps:
s1, using random Bernoulli sampling to input images to obtain image pairsRepeating the sampling for a plurality of times to obtain an image pair set +.>；
S2, designing a dynamic characteristic weighting branch module and a neuron branch module;
the dynamic characteristic weighting branch module learns channel information and space information from the characteristic diagram through a convolution-based extractor, and constructs a dynamic matrix by using the channel information and the space information, and the dynamic matrix supplements correlation information of neurons;
the neuron branching module calculates the activation degree of each neuron according to the correlation among the neurons;
s3, designing a feature weight connection module based on the structure of the U-net, and connecting a feature weighting module of the same layer in the encoder and the decoder;
s4, after the modules in the S2 and the S3 are designed, an untrained denoising model is formed, wherein the denoising model comprises a dynamic characteristic weighting branch module, a neuron branch module and a characteristic weight connecting module;
model training is carried out, and the denoising device is optimized by minimizing a loss function;
s5, model speculation, and a trained denoising modelA series of new denoising submodels { are obtained after Dropout is carried outNew random bernoulli samples, different from the training phase, are performed on the noise image and an instance of the bernoulli samples is fed to each newly formed denoiser.
S1 comprises, randomly bernoulli sampling:
；/>；
where M represents the number of image pairs, M represents the mth image pair,representing the matrix by multiplying correspondingly element by element, < >>Representing a two-dimensional vector sample in a bernoulli sample.
S2 comprises an energy function of S2.1. Degree of activationThe method comprises the following steps:
；/>；/>；；
；/>；/>；；
in the method, in the process of the invention,、/>、/>、/>、/>and->Is an intermediate parameter->Is corresponding to->Weights of individual neurons, +.>For the t-th target neuron, +.>Refers to other neurons in the same channel.
S2 comprises, S2.2 solving for the minimum energy of the target neuronThe method comprises the following steps:
；
usingThe degree of importance of the neurons is reflected.
Corresponding weight matrix->Is self-adaptive, automatically generated by a weight generator; applying a convolution-based feature extractor, the extractor pair size being +.>Is operated on to obtain a characteristic map of the size +.>Is defined as the characteristic vector of the target neuron->Channel dimension weights of (a); after traversing all neurons, a size of +.>Weight vector +.>Arranged according to the spatial positions of the corresponding neurons to obtainThe size is +.>Weight matrix of (2)Weight matrix->The method is used for supplementing the inter-channel correlation information and improving the capability of the attenuation module for extracting similar features of different dimensions in the remote sensing image.
S3, providing the feature weights learned from the encoder to corresponding feature weighting modules in the decoder, wherein the weighting modules in the decoder learn the feature weights from the current features and learn the weight thermodynamic diagrams from the corresponding weighting modules in the encoder, so that the feature weights can be transferred between the encoder and the decoder, and the consistency of the feature extraction stage and the generation stage in the overall denoising task is ensured.
S4 comprises the following steps:
；
in the method, in the process of the invention,for the parameter +.>Is>Is->Is included in the sample component of (a).
In the prior art, the main remote sensing image denoising methods are as follows.
1. One of the conventional remote sensing image denoising methods is a denoising algorithm based on filtering, and the main idea is to smooth an image to preserve information and calculate the relation between a target pixel and an adjacent pixel to remove noise. Early, filter-based image denoising methods utilized local information of an image. Another conventional method is statistical learning. It learns patterns and knowledge from the data to learn statistical properties of the image and noise signals to determine parameters of the method, such as scale thresholds and kernel sizes. Although both statistical learning methods and filtering-based methods achieve satisfactory performance in remote sensing image denoising tasks, there are still some problems: (1) These methods can only be used for specific images where the noise strength is known; (2) Prior to denoising, the user needs to manually adjust the super parameters to achieve optimal performance.
2. The denoising method based on supervised learning is mostly based on deep learning, and paired clean images and noisy images are needed as training data. Early researchers used primarily multi-layer perceptrons (MLPs) or feedforward neural networks to extract image features, but had unstable denoising problems. With the advent of AlexNet and its widespread use in the field of computer vision, convolutional Neural Networks (CNNs) have become one of the most important basic methods for image denoising. The supervised denoising method has a key disadvantage. In the absence of a large amount of effective training data, particularly in the remote sensing image denoising task, it is difficult to obtain a clean remote sensing image and perform supervised learning and training on a noisy remote sensing image pair, which results in poor performance of the supervised image denoising method.
3. The self-supervision learning-based denoising method effectively solves the problem that the supervision-based method needs high-quality and large-scale training data. When training a convolutional neural network using only noise images as input data, the model will be over-fitted. In an image denoising task, this typically indicates that the output image is consistent with the input image. In fact, the reason for the overfitting is that the model only learns the identity mapping, rather than implementing a de-noised mapping. The self-supervised image denoising method can realize blind denoising without a clean image, but generally causes image blurring and loss of some details.
The invention provides a single-image remote sensing image denoising method guided by feature weighting connection, which comprehensively considers feature weighting and self-supervision constraint to improve the effectiveness and robustness of a model. The method is a blind denoising method which only needs single-image input, can be used for solving the problems that supervised learning training data are limited and denoising cannot be effectively performed, and can also solve the problem that detail textures of images are blurred after self-supervised learning denoising, so that the denoising performance of remote sensing images is greatly improved.
In the present invention, the image pairsThe images in (a) are different from each other, but the collection of the images contains almost complete pixel information, so that the information of the original image is almost free from loss after a plurality of random Bernoulli sampling. Energy->Target neuron->The greater the distinction from other neurons, the more attention should be paid in the denoising task. A weighting module in the decoder learns feature weights from the current features and learns weight thermodynamic diagrams from corresponding weighting modules in the encoder. It enables feature weights to be passed between the encoder and decoder, ensuring consistency in the overall denoising task for the feature extraction stage and the generation stage. Only those pixels that are masked can measure each pair +.>Is a loss of (2). Due to->Is randomly generated whenWhen large enough, the difference across all image pixels can be measured by the total loss of all pairs.
The method of the present invention was quantitatively compared with the images after denoising by other self-monitoring methods as shown in table 1.
TABLE 1 quantitative comparison of denoised images
。
As can be seen from table 1, the image denoising effect of the method on the UC-Merced data set and the OPTIMAL-31 data set is superior to that of other self-supervision image denoising methods on quantitative evaluation standards (peak signal to noise ratio PSNR and structural similarity SSIM).
The above embodiments are only for illustrating the technical aspects of the present invention, not for limiting the same, and although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those skilled in the art that: the technical solutions described in the foregoing embodiments may be modified or some or all of the technical features may be replaced with other technical solutions, which do not depart from the scope of the technical solutions of the embodiments of the present invention.
Claims (7)
1. A feature weighted connection guided single-image remote sensing image denoising method is characterized by comprising the following steps:
s1, using random Bernoulli sampling to input images to obtain image pairsRepeating the sampling for a plurality of times to obtain an image pair set +.>；
S2, designing a dynamic characteristic weighting branch module and a neuron branch module;
the dynamic characteristic weighting branch module learns channel information and space information from the characteristic diagram through a convolution-based extractor, and constructs a dynamic matrix by using the channel information and the space information, and the dynamic matrix supplements correlation information of neurons;
the neuron branching module calculates the activation degree of each neuron according to the correlation among the neurons;
s3, designing a feature weight connection module based on the structure of the U-net, and connecting a feature weighting module of the same layer in the encoder and the decoder;
s4, after the modules in the S2 and the S3 are designed, an untrained denoising model is formed, wherein the denoising model comprises a dynamic characteristic weighting branch module, a neuron branch module and a characteristic weight connecting module;
model training is carried out, and the denoising device is optimized by minimizing a loss function;
s5, model speculation, and a trained denoising modelA series of new denoising submodels { are obtained after Dropout is carried outNew random bernoulli samples, different from the training phase, are performed on the noise image and an instance of the bernoulli samples is fed to each newly formed denoiser.
2. The feature weighted connection guided single-image remote sensing image denoising method according to claim 1, wherein S1 comprises the following steps:
；/>；
where M represents the number of image pairs, M represents the mth image pair,representing the matrix by multiplying correspondingly element by element, < >>Representing a two-dimensional vector sample in a bernoulli sample.
3. The feature weighted connection guided single image remote sensing image denoising method as claimed in claim 2, wherein S2 comprises an energy function of S2.1. Degree of activationThe method comprises the following steps:
；/>；/>；/>；
；/>；/>；；
in the method, in the process of the invention,、/>、/>、/>、/>and->Is an intermediate parameter->Is corresponding to->Weights of individual neurons, +.>For the t-th target neuron, +.>Refers to other neurons in the same channel.
4. A feature weighted connection guided single image remote sensing image denoising method according to claim 3, wherein S2 comprises, S2.2 solving for the minimum energy of the target neuronThe method comprises the following steps:
；
usingThe degree of importance of the neurons is reflected.
5. The method for denoising a feature weighted connection guided single-view remote sensing image of claim 4,corresponding weight matrix->Is self-adaptive, automatically generated by a weight generator; applying a convolution-based feature extractor, the extractor pair size being +.>Is operated on to obtain a characteristic map of the size +.>Is defined as the characteristic vector of the target neuron->Channel dimension weights of (a); after traversing all neurons, a size of +.>Weight vector +.>Arranged according to the spatial position of the corresponding neurons to obtain a size of +.>Weight matrix->Weight matrix->The method is used for supplementing the inter-channel correlation information and improving the capability of the attenuation module for extracting similar features of different dimensions in the remote sensing image.
6. The feature weighted connection guided single image remote sensing image denoising method of claim 5, wherein S3 comprises providing feature weights learned from the encoder to corresponding feature weighting modules in the decoder, wherein the weighting modules in the decoder learn feature weights from current features and learn weight thermodynamic diagrams from corresponding weighting modules in the encoder, enabling feature weights to be passed between the encoder and decoder, ensuring consistency of feature extraction and generation phases in overall denoising task.
7. The feature weighted connection guided single-image remote sensing image denoising method according to claim 6, wherein S4 comprises:
；
in the method, in the process of the invention,for the parameter +.>Is>Is->Is included in the sample component of (a).
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410270382.3A CN117876692B (en) | 2024-03-11 | 2024-03-11 | Feature weighted connection guided single-image remote sensing image denoising method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410270382.3A CN117876692B (en) | 2024-03-11 | 2024-03-11 | Feature weighted connection guided single-image remote sensing image denoising method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117876692A true CN117876692A (en) | 2024-04-12 |
| CN117876692B CN117876692B (en) | 2024-05-17 |
Family
ID=90594999
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410270382.3A Active CN117876692B (en) | 2024-03-11 | 2024-03-11 | Feature weighted connection guided single-image remote sensing image denoising method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117876692B (en) |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110349103A (en) * | 2019-07-01 | 2019-10-18 | 昆明理工大学 | It is a kind of based on deep neural network and jump connection without clean label image denoising method |
| CN113538281A (en) * | 2021-07-21 | 2021-10-22 | 深圳大学 | Image denoising method and device, computer equipment and storage medium |
| AU2021105153A4 (en) * | 2021-08-09 | 2021-11-11 | Yunshigao Technology Company Limited | An unsupervised learning of point cloud denoising |
| CN113989256A (en) * | 2021-11-08 | 2022-01-28 | 北京市测绘设计研究院 | Detection model optimization method, detection method and detection device for remote sensing image building |
| CN114460648A (en) * | 2022-01-27 | 2022-05-10 | 东北石油大学 | 3D convolutional neural network-based self-supervision 3D seismic data random noise suppression method |
| CN114511473A (en) * | 2022-04-19 | 2022-05-17 | 武汉大学 | Hyperspectral remote sensing image denoising method based on unsupervised adaptive learning |
| CN115146667A (en) * | 2022-04-22 | 2022-10-04 | 中国石油大学(华东) | Multi-scale seismic noise suppression method based on curvelet transform and multi-branch deep self-coding |
| CN115908206A (en) * | 2023-03-13 | 2023-04-04 | 中国石油大学(华东) | Remote sensing image defogging method based on dynamic characteristic attention network |
| WO2023055689A1 (en) * | 2021-09-29 | 2023-04-06 | Subtle Medical, Inc. | Systems and methods for noise-aware self-supervised enhancement of images using deep learning |
| CN116645283A (en) * | 2023-05-10 | 2023-08-25 | 曲阜师范大学 | Low-dose CT image denoising method based on self-supervision perceptual loss multi-scale convolutional neural network |
| CN117422619A (en) * | 2023-09-06 | 2024-01-19 | 中国石油大学(北京) | Training method of image reconstruction model, image reconstruction method, device and equipment |
-
2024
- 2024-03-11 CN CN202410270382.3A patent/CN117876692B/en active Active
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110349103A (en) * | 2019-07-01 | 2019-10-18 | 昆明理工大学 | It is a kind of based on deep neural network and jump connection without clean label image denoising method |
| CN113538281A (en) * | 2021-07-21 | 2021-10-22 | 深圳大学 | Image denoising method and device, computer equipment and storage medium |
| AU2021105153A4 (en) * | 2021-08-09 | 2021-11-11 | Yunshigao Technology Company Limited | An unsupervised learning of point cloud denoising |
| WO2023055689A1 (en) * | 2021-09-29 | 2023-04-06 | Subtle Medical, Inc. | Systems and methods for noise-aware self-supervised enhancement of images using deep learning |
| CN113989256A (en) * | 2021-11-08 | 2022-01-28 | 北京市测绘设计研究院 | Detection model optimization method, detection method and detection device for remote sensing image building |
| CN114460648A (en) * | 2022-01-27 | 2022-05-10 | 东北石油大学 | 3D convolutional neural network-based self-supervision 3D seismic data random noise suppression method |
| CN114511473A (en) * | 2022-04-19 | 2022-05-17 | 武汉大学 | Hyperspectral remote sensing image denoising method based on unsupervised adaptive learning |
| CN115146667A (en) * | 2022-04-22 | 2022-10-04 | 中国石油大学(华东) | Multi-scale seismic noise suppression method based on curvelet transform and multi-branch deep self-coding |
| CN115908206A (en) * | 2023-03-13 | 2023-04-04 | 中国石油大学(华东) | Remote sensing image defogging method based on dynamic characteristic attention network |
| CN116645283A (en) * | 2023-05-10 | 2023-08-25 | 曲阜师范大学 | Low-dose CT image denoising method based on self-supervision perceptual loss multi-scale convolutional neural network |
| CN117422619A (en) * | 2023-09-06 | 2024-01-19 | 中国石油大学(北京) | Training method of image reconstruction model, image reconstruction method, device and equipment |
Non-Patent Citations (3)
| Title |
|---|
| YUHUI QUAN: "Self2Self With Dropout: Learning Self-Supervised Denoising From Single Image", 《PROCEEDINGS OF THE IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR)》, 3 December 2020 (2020-12-03), pages 1890 - 1898 * |
| 张一飞;陈忠;张峰;欧阳超;: "基于栈式去噪自编码器的遥感图像分类", 计算机应用, no. 2, 15 December 2016 (2016-12-15) * |
| 贾晓芬;郭永存;柴华荣;赵佰亭;黄友锐;: "深立井井壁图像的卷积神经网络去噪方法", 西安交通大学学报, no. 06, 21 March 2019 (2019-03-21) * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117876692B (en) | 2024-05-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Tian et al. | Deep learning on image denoising: An overview | |
| CN111209952B (en) | Underwater target detection method based on improved SSD and migration learning | |
| CN112233038B (en) | True image denoising method based on multi-scale fusion and edge enhancement | |
| CN110599409B (en) | Convolutional neural network image denoising method based on multi-scale convolutional groups and parallel | |
| CN108765319B (en) | Image denoising method based on generation countermeasure network | |
| CN109035149B (en) | License plate image motion blur removing method based on deep learning | |
| CN111275643B (en) | Real noise blind denoising network system and method based on channel and space attention | |
| CN112233026A (en) | SAR image denoising method based on multi-scale residual attention network | |
| CN111242862A (en) | Multi-scale fusion parallel dense residual convolution neural network image denoising method | |
| CN111340716B (en) | Image deblurring method for improving double-discrimination countermeasure network model | |
| CN111861894A (en) | Image motion blur removing method based on generating type countermeasure network | |
| Min et al. | Blind deblurring via a novel recursive deep CNN improved by wavelet transform | |
| CN111968047A (en) | Adaptive optical image blind restoration method based on generating type countermeasure network | |
| CN114723630A (en) | Image deblurring method and system based on cavity double-residual multi-scale depth network | |
| CN115272303A (en) | Textile fabric defect degree evaluation method, device and system based on Gaussian blur | |
| Abdulah et al. | Review Study of Image De-Noising on Digital Image Processing and Applications | |
| Le et al. | LD-Net: An efficient lightweight denoising model based on convolutional neural network | |
| CN117876692B (en) | Feature weighted connection guided single-image remote sensing image denoising method | |
| CN116309178A (en) | Visible light image denoising method based on self-adaptive attention mechanism network | |
| CN114419341B (en) | Convolutional neural network image recognition method based on transfer learning improvement | |
| CN115760589A (en) | Image optimization method and device for motion blurred image | |
| Xiao et al. | Underwater image classification based on image enhancement and information quality evaluation | |
| CN114022469A (en) | Cigarette appearance defect image classification method and system | |
| CN113379641A (en) | Single image rain removing method and system based on self-coding convolutional neural network | |
| Biswas | Impulse noise suppression in color images using median filter and deep learning |
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 |