AU2020100196A4 - A method of removing rain from single image based on detail supplement - Google Patents
A method of removing rain from single image based on detail supplement Download PDFInfo
- Publication number
- AU2020100196A4 AU2020100196A4 AU2020100196A AU2020100196A AU2020100196A4 AU 2020100196 A4 AU2020100196 A4 AU 2020100196A4 AU 2020100196 A AU2020100196 A AU 2020100196A AU 2020100196 A AU2020100196 A AU 2020100196A AU 2020100196 A4 AU2020100196 A4 AU 2020100196A4
- Authority
- AU
- Australia
- Prior art keywords
- equation
- image
- gru
- diamond
- result
- 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.)
- Ceased
Links
- 238000000034 method Methods 0.000 title claims abstract description 32
- 239000013589 supplement Substances 0.000 title claims abstract description 16
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 claims abstract description 22
- 230000008569 process Effects 0.000 claims abstract description 20
- 229910003460 diamond Inorganic materials 0.000 claims abstract description 18
- 239000010432 diamond Substances 0.000 claims abstract description 18
- 230000000306 recurrent effect Effects 0.000 claims abstract description 15
- 230000007246 mechanism Effects 0.000 claims abstract description 6
- 230000000295 complement effect Effects 0.000 claims abstract description 3
- 238000012549 training Methods 0.000 claims description 9
- 238000000605 extraction Methods 0.000 claims description 4
- 230000006870 function Effects 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 3
- 239000000284 extract Substances 0.000 claims description 2
- 238000005457 optimization Methods 0.000 claims description 2
- 230000000694 effects Effects 0.000 abstract description 4
- 230000006872 improvement Effects 0.000 abstract description 3
- 238000013528 artificial neural network Methods 0.000 abstract description 2
- 238000013135 deep learning Methods 0.000 abstract description 2
- 238000001514 detection method Methods 0.000 abstract description 2
- 238000003709 image segmentation Methods 0.000 abstract description 2
- 230000015572 biosynthetic process Effects 0.000 description 5
- 230000000007 visual effect Effects 0.000 description 5
- 238000013461 design Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 239000000654 additive Substances 0.000 description 1
- 230000000996 additive effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000000739 chaotic effect Effects 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000000704 physical effect Effects 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/20—Image enhancement or restoration by the use of local operators
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
-
- 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
- G06T2207/10016—Video; Image sequence
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
Abstract: Images taken in rainy condition often contain a large number of raindrops, which will affect the image quality. Therefore, images taken in rainy days affect the improvement of application effects in other fields to some extent, such as object detection, image classification, image super-resolution, image segmentation, etc. Hence, we built a algorithm based on deep learning network with three steps to remove raindrops in the images and restore it to a high-quality image. Firstly, We built a generic diamond residual blocks to improve the ability to extract image features of convolution neural network, which can provide rich image context information for subsequent deraining steps. Secondly, Due to the mixability of raindrop shape, size and direction, a recurrent deraining structure is proposed for the network. Through the idea of derain step by step, the mixed raindrop is gradually removed and the image is recovered. Finally, In the process of remove raindrops often accompanied by some loss of image details. Based on this negative effect, a hybrid details complementary mechanism is constructed, which pass the residual information of shallow characteristics subtracted from deeper features of diamond residual block to the next iteration, and combined with the memory of the Gated Recurrent Unit to supplement details again, then details loss is well suppressed.
Description
BACKGROUND AND PURPOSE [0001] Affected by severe weather, especially heavy rain and rainstorm, it is still challenging to detect and recognize images acquired in rainy days for computer vision system. The visual effects of rain are quite complex and the image quality will be affected to some extent. Therefore, the image taken in rainy days will often have a greatly reduced effect when performing other visual tasks. In order to improve other visual tasks, such as image classification, object detection, image segmentation, video tracking etc., we can often first remove the rain from the image to better complete other visual tasks.
[0002] For the rapid development of deep learning in recent years, many visual tasks have achieved unexpected improvements. Many studies on derain have also been carried out around the convolutional neural network, and the effect has been gradually improved. According to the results published by all kinds of deraining methods, most of them can achieve the purpose of rain removal, but there is still some room for improvement in the restored image quality. How to ensure that the image after rain removal still has more detailed information still needs continuous research to improve.
[0003] The research history of rain removal task has been more than decade years. Most of the previous works focused on video rain removal. Some solutions often take advantage of the physical properties of rain tracks in an image, such as photometric models, the shape and motion of raindrops, to study raindrops. Due to the significant differences between the dynamically changing raindrops, rain lines and background layers, some schemes propose to more accurately methods to reflect the essential characteristics of rain tracks from the frequency domain. However, rain-removing methods based on frequency domain usually achieve rain-removing by detecting characteristics of raindrops, so it is often difficult to detect shallow rain-trace information or detect by mistake, leading to excessive rain-removing or rain-trace residue. With the development of sparse representation, most researches turn to the rain removal technology of sparse domain. According to the spatial distribution characteristics and optical characteristics of raindrops, the influence of raindrops on images can be regarded as an additive noise, so the problem of rain removal can be transformed into a noise reduction problem to restore image. The core of rain-removing technology based on sparse domain is to separate rain layer and background layer with sparse constraints, and then use the prior knowledge of raindrops and rain lines to conduct dictionary classification and image reconstruction of rain and rain-free. Under the framework of sparse domain, it is beneficial to better highlight the characteristics of raindrops and rain lines then make the background layer and rain layer more linearly separable. However, sparse
2020100196 22 Feb 2020 constraints and prior knowledge are often difficult to be accurately defined, which results in the process of rain removal is still not so accurate and effective. Futhermore, due to the temporal sequence relationship among continuous frames in video, it can exploit more information to achieve the purpose of rain removal. However, for a single image, with only the pre-image information, how to use the current information to better complete the task of rain removal is extremely challenging, which has attracted the attention of many researchers.
[0004] We aim to remove the rain tracks from the rain images and add more detailed information so as to obtain a higher quality restoration image. Due to the complexity of the rain layer, the rainless background layer cannot be directly obtained at one time. By taking advantages of the recurrent steps, the rain can be removed step by step and detailed information can be added step by step. In the end, the gradual method is adopted as the basis of the algorithm. In order to obtain more basic feature information, we built diamond residual blocks with small number of channels at both ends and large number of channels in the middle, which were used as the core recursive training of the recurrent network. In the part of details supplement, because of the diversity of detail, we constructed three kinds of path to add the details of the process: (I) With the depth of the neural network to deepen, there may be a part of loss of image details, we use the differences between original feature maps and feature maps of each iteration, and then the differences pass to the process of next iteration. (II) The results of each iteration contain rich background information. We use skip connection to pass the features of the previous output containing rich picture information to the next iteration, so as to inject and supplement the background information. The loss of some background information is well avoided. (Ill) Finally, we use GRU layer to select more features passing to the next iteration. To avoid passive influences of rain layer, we put GRU layer at back of the residual blocks, which can choose more useful information to pass, and without the rain interference, makes the image quality improved.
FEATURE EXTRACTION BLOCK DESCRIPTION [0001] As we all known, more information can be extracted by increasing the number of feature channels in the convolution layer. However, the increase of feature channels will also lead to the increase of network parameters and computation. Therefore, we designed a diamond residual block, which utilizes the feature channels with smaller dimensions at both ends to reduce network parameters, and extracts more abundant rain layer information by increasing the dimensions of the feature channels in the middle, as shown in the diamond residual block in Figure 1. We use cuboids with different colors
2020100196 22 Feb 2020 to represent feature channels with different dimensions (that is, different number of feature maps).
[0002]The diamond residual block is a structure with 6 convolution layers{1’23,4,5,6}), one f fl
Relu layer 'relu and one GRU module' 8ru. ‘np is defined for input rainy image, x is the feature map of intermediate process and the final output resuIt. The overall feature extraction process can be described as
Equation 1
Equation 2
Equation 3
Equation 4
Equation 5
Equation 6
ECURRENT DERAINING PROCESS DESCRIPTION
001] The distribution of rain is often chaotic, and removing rain trails in an end-to-end network by a one-time method is often difficult or requires complex network design. In order to solve these problems, some previous rain removal methods adopted multi-stage iterative rain removal, and each stage did not need complex design, but only designed shallow residual blocks.
[0002] Each phase in order to reduce the parameters of the network, can use the same residual circulation block, this is equivalent to the overall network structure is divided into N the same subnet, which can use as the feature extraction function, then characteristics will be passed effectively and avoids the additional parameters storage overhead. Figure 2 shows the schematic of recurrent process. The number of subnet can also according to the aspiration to change, make the network more convenient and effective to work.
2020100196 22 Feb 2020 [0003]Step l:Confirm the iteration times T of recurrent progress.
f B
004] Given an Rainny image 7? , and a residual block Jres. The deraining result 1 of first iteration can : described as
A ~ fres (R) Equation 7 ad e {2,3,4, · ·, Γ} , fopowfog recurrent rain removal result can be uniformly expressed as ~ fres (3_i) Equation 8
B radually increase T for training, evaluate the deraining results 'from subjective vision and objective dicators, and confirm the optimal recurrent times T.
005] Step 2: Ensure network easily optimization and prompt feature richness.
006] In the general recurrent processes show in Figure 1), the rainny image R is directly added to ‘ participate in the training of each iteration. In order to avoid the redundancy of rain, the results of evious iterations are only added to the next training process. The cycle process is improved as ~ fres (A-2 + Bt-\) Equation 9
ETAIL SUPPLEMENT METHOD DESCRIPTION [1)001 ] From previous study, we learned that with the deepening of the network, a lot of information will be lost when the extracted information passes through convolution operation, activation function or pooling operation step by step. Especially, the proposed network employs the progressive way to remove rain, which will lead to more and more information loss with the increase of iteration times.
[0002] In order to solve this problem, we put forward a kind of information supplement mechanism in two aspects. The first kind of supplementary information comes from the feature maps extracted from rainy image for the input of each iteration of the diamond residual block. First, the feature maps are obtained from the rainy image through convolution operation and regarded as the original feature information. Because multiple iterations of the diamond residual block are utilized to learn the rain layer information in the
2020100196 22 Feb 2020 network, and each iteration of the diamond residual block will cause certain loss of feature information, we insider to supplement the lost information after each iteration. This lost information can be approximately itaincd by subtracting the information learned by each iteration of the diamond residual block from the iginal feature information and is transferred to the input of the next iteration. In this way, the problem of formation loss can be solved to a certain extent.
003] Considering the difference between the output of each iteration of diamond residual block, the cond kind of information supplement is designed to utilize the output of the previous iteration to provide formation supplement for the output of the next iteration. Therefore, we design a GRU module in the amond recurrent block, which memorizes the feature information learned in each iterative and transmits it the output of the next iteration. As we all know, GRU can solve the long dependency problem in the :twork, including the update gate and reset gate. The update gate can be used to control how much formation from the previous state is brought into the current state, and the reset gate can control how much formation from the previous state is written to the current candidate set. According to these characteristics ' GRU, the proposed network with GRU module can receive the results of the previous iteration more lectively, provide supplementary information for the current output results, and avoid redundant formation.
004] Step 1: Supplement with residual detailXres.
X f
005] The original feature ori are obtained by convolution operation of the original rainny lage R , and can be described as xoH ~ fconV-on (R) Equation 10
The detail residual is defined as the difference between the original feature ori and the convolution result of the fifth layer fc°n*-5 of the diamond residual block, and then the detail residual is taken as the first detail supplement mechanism by concatenate J concat with last iteration result resu“ for each iteration input X feature map result. The process is showed as follows xres-x Ori fconv-sU^ Equation 1 1 % result fconcat res 9 result)
2020100196 22 Feb 2020
006]Step 2: Supplement with selected detailXgru.
Equation 12 x h h h
007] 1 and represent the current input and history state respectively, nt and ' represent the z t* ndidate state and current state respectively, and · and * represent update gate and reset gate spectively. In the process of GRU, two activation functions σ and tan^ are mainly used and then
WWW f «tain three weight coefficients z’ r’ through training.GRU model can be expressed as
| Χ=σ(^· ΙΛ , xt. | Equation 13 |
| rt =a(W; , xt_ | Equation 14 |
| ht = tanh( IF· \rt * ht_x,xt]) | Equation 15 |
| ht = ^-ztyht-x+zt*ht | Equation 16 |
| Xgru = ht | Equation 17 |
X nally, the filtered detail gru are injected into the GRU process of next iteration through concatenate gru .This complementary mechanism can be represented as ht
Equation 18
Claims (3)
1. The procedure of diamond residual block are as follows:
[0001] Construct diamond residual block as feature extractors.
[0002] The diamond residual block is designed the feature channels with smaller dimensions at both ends to reduce network parameters, and extracts more abundant rain layer information by increasing the dimensions of the feature channels in the middle, as shown in the diamond residual block in Figure 1. The diamond residual block is a structure with 6 convolution layers f:r„n {1,2,3,4,5,6)), one Relu layer frelu and one GRU module f. Iin is defined for input rainy image, x is the feature map of intermediate process and the final output xresult. The overall feature extraction process can be described as conv— 1 (4)
Equation 1 conv-3 ^fconv-2
Equation 2 * = freluW
Equation 3 conf -5 conv-A
Equation 4 * = fgruW
Equation 5
X result
(x)
Equation 6
2. The procedures of recurrent network are as follows:
[0001] Step LConfirm the iteration times T of recurrent progress.
[0002] Given an rainny image R , and a residual block ^res. The deraining result of first iteration can be described as
2020100196 22 Feb 2020
Βλ = frARl Equation 7
And /e {2,3,4,---,77} , the following recurrent rain removal result Bt can be uniformly expressed as Bt = fres Equations
Gradually increase T for training, evaluate the deraining results BT from subjective vision and objective indicators, and confirm the optimal recurrent times T.
[0003] Step 2: Ensure network easily optimization and prompt feature richness.
[0004]In the general recurrent processes show in Figure 1), the rainny image R is directly added to ‘ to participate in the training of each iteration. In order to avoid the redundancy of rain, the results of previous iterations are only added to the next training process. The cycle process is improved as
Bt = fres (Bt-2 + ) Equation 9
3. The procedures of detail supplement method are as follows:
[0001] Step 1: Supplement with residual detail xres.
[0002] The original feature xorj are obtained by convolution operation fconv_ori of the original rainny image R , and can be described as = feonv-ori UO Equation 10 [0003] The detail residual is defined as the difference between the original feature xort and the convolution result of the fifth layer fconv_5 of the diamond residual block, and then the detail residual is taken as the first detail supplement mechanism by concatenate fconcat with last iteration result xresult for each iteration input feature map x'resuIt. The process is showed as follows \es = *on - fcOnv-5 (*) Equation 11
X — f (X X result J concat + res 9 result
Equation 12 [0004]Step 2: Supplement with selected detail xgru.
[0005] x, and ht_x represent the current input and history state respectively, ht and ht represent the candidate state and current state respectively, and zt and rt represent update gate and reset gate respectively. In the process of GRU, two activation functions σ and tanh are mainly used and then obtain three weight coefficients Wz, Wr, W through training.GRU model f can be expressed as
[0006]Finally, the filtered detail xgru are injected into the GRU process of next iteration through concatenate f. This complementary mechanism can be represented as
Vl fconcat (Xgru ’ ^/-1
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| AU2020100196A AU2020100196A4 (en) | 2020-02-08 | 2020-02-08 | A method of removing rain from single image based on detail supplement |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| AU2020100196A AU2020100196A4 (en) | 2020-02-08 | 2020-02-08 | A method of removing rain from single image based on detail supplement |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| AU2020100196A4 true AU2020100196A4 (en) | 2020-03-19 |
Family
ID=69805022
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| AU2020100196A Ceased AU2020100196A4 (en) | 2020-02-08 | 2020-02-08 | A method of removing rain from single image based on detail supplement |
Country Status (1)
| Country | Link |
|---|---|
| AU (1) | AU2020100196A4 (en) |
Cited By (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111462013A (en) * | 2020-04-03 | 2020-07-28 | 西安交通大学 | Single-image rain removing method based on structured residual learning |
| CN111738932A (en) * | 2020-05-13 | 2020-10-02 | 合肥师范学院 | Automatic rain removing method for photographed image of vehicle-mounted camera |
| CN112085680A (en) * | 2020-09-09 | 2020-12-15 | 腾讯科技（深圳）有限公司 | Image processing method and device, electronic equipment and storage medium |
| CN112184573A (en) * | 2020-09-15 | 2021-01-05 | 西安理工大学 | Context aggregation residual single image rain removing method based on convolutional neural network |
| CN112507327A (en) * | 2020-08-31 | 2021-03-16 | 华南理工大学 | Weather detection and processing method based on machine learning |
| CN112541880A (en) * | 2020-12-07 | 2021-03-23 | 南昌航空大学 | Deep learning branch image based rain removing system and control method |
| CN112651898A (en) * | 2021-01-12 | 2021-04-13 | 北京大学 | Video rain removing method and device based on memory enhancement |
| CN112734675A (en) * | 2021-01-19 | 2021-04-30 | 西安理工大学 | Image rain removing method based on pyramid model and non-local enhanced dense block |
| CN112907479A (en) * | 2021-03-05 | 2021-06-04 | 西安电子科技大学 | Residual single image rain removing method based on attention mechanism |
| CN113033687A (en) * | 2021-04-02 | 2021-06-25 | 西北工业大学 | Target detection and identification method under rain and snow weather condition |
| CN113129226A (en) * | 2021-03-24 | 2021-07-16 | 西安理工大学 | ConvGRU-U-Net-based computed ghost imaging reconstruction algorithm |
| CN113256538A (en) * | 2021-06-23 | 2021-08-13 | 浙江师范大学 | Unsupervised rain removal method based on deep learning |
| CN113379643A (en) * | 2021-06-29 | 2021-09-10 | 西安理工大学 | Image denoising method based on NSST domain and Res2Net network |
| CN113378704A (en) * | 2021-06-09 | 2021-09-10 | 武汉理工大学 | Multi-target detection method, equipment and storage medium |
| CN113393385A (en) * | 2021-05-12 | 2021-09-14 | 广州工程技术职业学院 | Unsupervised rain removal method, system, device and medium based on multi-scale fusion |
| CN113487622A (en) * | 2021-05-25 | 2021-10-08 | 中国科学院自动化研究所 | Head and neck organ image segmentation method and device, electronic equipment and storage medium |
| CN113870129A (en) * | 2021-09-15 | 2021-12-31 | 大连理工大学 | Video rain removing method based on spatial perception and time difference learning |
| CN114627005A (en) * | 2022-02-16 | 2022-06-14 | 武汉大学 | Rain density classification guided dual-stage single image rain removing method |
| CN114638768A (en) * | 2022-05-19 | 2022-06-17 | 武汉大学 | Image rain removing method, system and equipment based on dynamic association learning network |
| CN114820379A (en) * | 2022-05-12 | 2022-07-29 | 中南大学 | Image rain layer removing method for generating countermeasure network based on attention dual residual error |
| CN114972105A (en) * | 2022-06-10 | 2022-08-30 | 江苏海洋大学 | Single image rain removing method based on multi-scale fusion residual error network |
| CN117392545A (en) * | 2023-10-26 | 2024-01-12 | 南昌航空大学 | SAR image target detection method based on deep learning |
-
2020
- 2020-02-08 AU AU2020100196A patent/AU2020100196A4/en not_active Ceased
Cited By (36)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111462013B (en) * | 2020-04-03 | 2022-03-22 | 西安交通大学 | Single-image rain removing method based on structured residual learning |
| CN111462013A (en) * | 2020-04-03 | 2020-07-28 | 西安交通大学 | Single-image rain removing method based on structured residual learning |
| CN111738932A (en) * | 2020-05-13 | 2020-10-02 | 合肥师范学院 | Automatic rain removing method for photographed image of vehicle-mounted camera |
| CN112507327A (en) * | 2020-08-31 | 2021-03-16 | 华南理工大学 | Weather detection and processing method based on machine learning |
| CN112085680B (en) * | 2020-09-09 | 2023-12-12 | 腾讯科技（深圳）有限公司 | Image processing method and device, electronic equipment and storage medium |
| CN112085680A (en) * | 2020-09-09 | 2020-12-15 | 腾讯科技（深圳）有限公司 | Image processing method and device, electronic equipment and storage medium |
| CN112184573A (en) * | 2020-09-15 | 2021-01-05 | 西安理工大学 | Context aggregation residual single image rain removing method based on convolutional neural network |
| CN112184573B (en) * | 2020-09-15 | 2023-05-16 | 西安理工大学 | Single image rain removing method for context aggregation residual error based on convolutional neural network |
| CN112541880A (en) * | 2020-12-07 | 2021-03-23 | 南昌航空大学 | Deep learning branch image based rain removing system and control method |
| CN112651898A (en) * | 2021-01-12 | 2021-04-13 | 北京大学 | Video rain removing method and device based on memory enhancement |
| CN112651898B (en) * | 2021-01-12 | 2023-04-07 | 北京大学 | Video rain removing method and device based on memory enhancement |
| CN112734675B (en) * | 2021-01-19 | 2024-02-09 | 西安理工大学 | Image rain removing method based on pyramid model and non-local enhanced dense block |
| CN112734675A (en) * | 2021-01-19 | 2021-04-30 | 西安理工大学 | Image rain removing method based on pyramid model and non-local enhanced dense block |
| CN112907479A (en) * | 2021-03-05 | 2021-06-04 | 西安电子科技大学 | Residual single image rain removing method based on attention mechanism |
| CN112907479B (en) * | 2021-03-05 | 2023-04-18 | 西安电子科技大学 | Residual single image rain removing method based on attention mechanism |
| CN113129226A (en) * | 2021-03-24 | 2021-07-16 | 西安理工大学 | ConvGRU-U-Net-based computed ghost imaging reconstruction algorithm |
| CN113129226B (en) * | 2021-03-24 | 2023-06-23 | 西安理工大学 | ConvGRU-U-Net-based computing ghost imaging reconstruction algorithm |
| CN113033687A (en) * | 2021-04-02 | 2021-06-25 | 西北工业大学 | Target detection and identification method under rain and snow weather condition |
| CN113393385B (en) * | 2021-05-12 | 2024-01-02 | 广州工程技术职业学院 | Multi-scale fusion-based unsupervised rain removing method, system, device and medium |
| CN113393385A (en) * | 2021-05-12 | 2021-09-14 | 广州工程技术职业学院 | Unsupervised rain removal method, system, device and medium based on multi-scale fusion |
| CN113487622B (en) * | 2021-05-25 | 2023-10-31 | 中国科学院自动化研究所 | Head-neck organ image segmentation method, device, electronic equipment and storage medium |
| CN113487622A (en) * | 2021-05-25 | 2021-10-08 | 中国科学院自动化研究所 | Head and neck organ image segmentation method and device, electronic equipment and storage medium |
| CN113378704A (en) * | 2021-06-09 | 2021-09-10 | 武汉理工大学 | Multi-target detection method, equipment and storage medium |
| CN113256538A (en) * | 2021-06-23 | 2021-08-13 | 浙江师范大学 | Unsupervised rain removal method based on deep learning |
| CN113379643A (en) * | 2021-06-29 | 2021-09-10 | 西安理工大学 | Image denoising method based on NSST domain and Res2Net network |
| CN113870129A (en) * | 2021-09-15 | 2021-12-31 | 大连理工大学 | Video rain removing method based on spatial perception and time difference learning |
| CN114627005B (en) * | 2022-02-16 | 2024-04-12 | 武汉大学 | Rain density classification guided double-stage single image rain removing method |
| CN114627005A (en) * | 2022-02-16 | 2022-06-14 | 武汉大学 | Rain density classification guided dual-stage single image rain removing method |
| CN114820379B (en) * | 2022-05-12 | 2024-04-26 | 中南大学 | Image rain-like layer removing method for generating countermeasure network based on attention dual residual error |
| CN114820379A (en) * | 2022-05-12 | 2022-07-29 | 中南大学 | Image rain layer removing method for generating countermeasure network based on attention dual residual error |
| CN114638768A (en) * | 2022-05-19 | 2022-06-17 | 武汉大学 | Image rain removing method, system and equipment based on dynamic association learning network |
| CN114638768B (en) * | 2022-05-19 | 2022-08-09 | 武汉大学 | Image rain removing method, system and equipment based on dynamic association learning network |
| CN114972105B (en) * | 2022-06-10 | 2024-03-29 | 江苏海洋大学 | Single image rain removing method based on multi-scale fusion residual error network |
| CN114972105A (en) * | 2022-06-10 | 2022-08-30 | 江苏海洋大学 | Single image rain removing method based on multi-scale fusion residual error network |
| CN117392545A (en) * | 2023-10-26 | 2024-01-12 | 南昌航空大学 | SAR image target detection method based on deep learning |
| CN117392545B (en) * | 2023-10-26 | 2024-02-09 | 南昌航空大学 | SAR image target detection method based on deep learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| AU2020100196A4 (en) | A method of removing rain from single image based on detail supplement | |
| Yu et al. | Learning a discriminative feature network for semantic segmentation | |
| CN112308860B (en) | Earth observation image semantic segmentation method based on self-supervision learning | |
| CN115049936B (en) | High-resolution remote sensing image-oriented boundary enhanced semantic segmentation method | |
| CN111127354B (en) | Single-image rain removing method based on multi-scale dictionary learning | |
| CN110706269B (en) | Binocular vision SLAM-based dynamic scene dense modeling method | |
| CN110189260B (en) | Image noise reduction method based on multi-scale parallel gated neural network | |
| Yi et al. | Efficient and accurate multi-scale topological network for single image dehazing | |
| Yao et al. | Improving depth gradient continuity in transformers: A comparative study on monocular depth estimation with cnn | |
| Yang et al. | End-to-end rain removal network based on progressive residual detail supplement | |
| CN113066025A (en) | Image defogging method based on incremental learning and feature and attention transfer | |
| CN112651360A (en) | Skeleton action recognition method under small sample | |
| CN115797808A (en) | Unmanned aerial vehicle inspection defect image identification method, system, device and medium | |
| CN104253994B (en) | A kind of night monitoring video real time enhancing method merged based on sparse coding | |
| CN113011438B (en) | Bimodal image significance detection method based on node classification and sparse graph learning | |
| CN113962905A (en) | Single image rain removing method based on multi-stage feature complementary network | |
| CN116645369B (en) | Anomaly detection method based on twin self-encoder and two-way information depth supervision | |
| Yuan et al. | Recurrent structure attention guidance for depth super-resolution | |
| CN107194932B (en) | Adaptive background reconstruction algorithm based on exponential forgetting | |
| CN115131364B (en) | Method for segmenting medical image based on Transformer | |
| CN116385289A (en) | Progressive inscription character image restoration model and restoration method | |
| Jia et al. | Single-image snow removal based on an attention mechanism and a generative adversarial network | |
| CN114463237A (en) | Real-time video rain removing method based on global motion compensation and inter-frame time domain correlation | |
| Li et al. | Feature attention parallel aggregation network for single image haze removal | |
| Tan et al. | DBSwin: Transformer based dual branch network for single image deraining |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| FGI | Letters patent sealed or granted (innovation patent) | ||
| MK22 | Patent ceased section 143a(d), or expired - non payment of renewal fee or expiry |