CN116245765A - Image denoising method and system based on enhanced depth expansion convolutional neural network - Google Patents

Image denoising method and system based on enhanced depth expansion convolutional neural network Download PDF

Info

Publication number
CN116245765A
CN116245765A CN202310241500.3A CN202310241500A CN116245765A CN 116245765 A CN116245765 A CN 116245765A CN 202310241500 A CN202310241500 A CN 202310241500A CN 116245765 A CN116245765 A CN 116245765A
Authority
CN
China
Prior art keywords
layer
image
network
neural network
convolutional neural
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.)
Pending
Application number
CN202310241500.3A
Other languages
Chinese (zh)
Inventor
李天平
李萌
冯凯丽
李冠兴
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shandong Normal University
Original Assignee
Shandong Normal University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shandong Normal University filed Critical Shandong Normal University
Priority to CN202310241500.3A priority Critical patent/CN116245765A/en
Publication of CN116245765A publication Critical patent/CN116245765A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/70—Denoising; Smoothing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • 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
    • G06T2207/20081—Training; Learning
    • 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]
    • 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

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

Abstract

The invention provides an image denoising method based on an enhanced depth expansion convolutional neural network, and relates to the field of image denoising. The method comprises the steps of obtaining an original image to be denoised; building an enhanced deep expansion convolutional neural network model with an upper network layer, a lower network layer and an expansion convolutional layer which are in a three-layer parallel structure; inputting an original image into an enhanced deep expansion convolutional neural network model, respectively extracting features of the original image through an upper network layer, a lower network layer and an expansion convolutional layer, fusing the extracted features, and then distributing attention to the fused features by using an attention layer to obtain a residual image; and overlapping the residual image on the original image to obtain a denoising image corresponding to the original image. According to the invention, by reducing the depth of the network and increasing the width of the network, gradient disappearance and gradient explosion are prevented, the training speed of network training can be improved, the image denoising precision is improved, and the calculation cost is reduced, so that the image denoising performance is better improved.

Description

Image denoising method and system based on enhanced depth expansion convolutional neural network
Technical Field
The invention belongs to the technical field of image denoising, and particularly relates to an image denoising method and system based on an enhanced deep expansion convolutional neural network.
Background
The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art.
Image denoising refers to the process of reducing noise in a digital image, which is a process of restoring a noisy image to a clean image. There are many methods for denoising images, and different denoising methods are used for different noises, and common image denoising modes are divided into three main types: filter-based methods, model-based methods, learning-based methods. Learning-based methods focus on learning the potential mapping of noisy images to clean images and can be categorized into traditional learning-based methods and deep network-based learning methods.
In recent years, depth network-based methods have become the dominant approach because they have achieved more promising denoising results than filtering-based, model-based, and traditional learning-based methods. The inventor finds that the traditional denoising method based on the attention mechanism has more defects, such as complex algorithm, low running speed, poor effect precision, high calculation cost and the like.
Disclosure of Invention
In order to overcome the defects in the prior art, the invention provides the image denoising method and the system based on the enhanced depth expansion convolutional neural network, which not only prevent gradient disappearance and gradient explosion, but also improve the training speed of network training, improve the image denoising precision, reduce the calculation cost and further improve the image denoising performance by reducing the depth of the network and increasing the width of the network.
To achieve the above object, one or more embodiments of the present invention provide the following technical solutions:
the first aspect of the invention provides an image denoising method based on an enhanced deep expansion convolutional neural network.
An image denoising method based on an enhanced depth expansion convolutional neural network comprises the following steps:
acquiring an original image to be denoised;
building an enhanced deep expansion convolutional neural network model with an upper network layer, a lower network layer and an expansion convolutional layer which are in a three-layer parallel structure;
inputting an original image into an enhanced deep expansion convolutional neural network model, respectively extracting features of the original image through an upper network layer, a lower network layer and an expansion convolutional layer, fusing the extracted features, and then distributing attention to the fused features by using an attention layer to obtain a residual image;
and overlapping the residual image on the original image to obtain a denoising image corresponding to the original image.
The second aspect of the invention provides an image denoising system based on an enhanced depth expansion convolutional neural network.
An image denoising system based on an enhanced depth expansion convolutional neural network, comprising:
an original image acquisition module configured to: acquiring an original image to be denoised;
a model building module configured to: building an enhanced deep expansion convolutional neural network model with an upper network layer, a lower network layer and an expansion convolutional layer which are in a three-layer parallel structure;
a feature extraction module configured to: inputting an original image into an enhanced deep expansion convolutional neural network model, respectively extracting features of the original image through an upper network layer, a lower network layer and an expansion convolutional layer, fusing the extracted features, and then distributing attention to the fused features by using an attention layer to obtain a residual image;
a superposition module configured to: and overlapping the residual image on the original image to obtain a denoising image corresponding to the original image.
A third aspect of the present invention provides a computer readable storage medium having stored thereon a program which when executed by a processor performs the steps in the image denoising method based on an enhanced deep dilation convolutional neural network according to the first aspect of the present invention.
A fourth aspect of the invention provides an electronic device comprising a memory, a processor and a program stored on the memory and executable on the processor, the processor implementing the steps in the method for denoising images based on an enhanced deep dilation convolutional neural network according to the first aspect of the invention when the program is executed.
The one or more of the above technical solutions have the following beneficial effects:
the invention provides an image denoising method and system based on an enhanced deep expansion convolutional neural network, and provides an enhanced network model ED-DCNNet, wherein the model reduces the depth of the network, increases the width of the network, enlarges the receptive field, and can extract more effective features from the environment, thereby improving the denoising performance of the network.
The image denoising method and system based on the enhanced depth expansion convolutional neural network, provided by the invention, not only prevent gradient disappearance and gradient explosion, but also improve the training speed of network training, improve the image denoising precision, reduce the calculation cost, and further better improve the image denoising performance.
Additional aspects of the invention will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the invention.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the invention.
Fig. 1 is a flow chart of a method of a first embodiment.
FIG. 2 is a diagram of the overall architecture of an enhanced deep-expanded convolutional neural network.
Fig. 3 is a diagram of the upper and lower network layers.
Fig. 4 is a diagram of an expanded convolution layer structure.
FIG. 5 (a) is a graph showing the effect of denoising using the ED-DCNNet model of the present invention.
Fig. 5 (b) is a graph of the effect of denoising using the ffdnat model.
Fig. 5 (c) is a noise image at a PSNR value of 18.35 db.
Fig. 5 (d) shows the original image at a PSNR value of 35 db.
Fig. 6 is a system configuration diagram of the second embodiment.
Detailed Description
It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the invention. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of exemplary embodiments according to the present invention.
Embodiments of the invention and features of the embodiments may be combined with each other without conflict.
The invention provides a general idea:
many denoising models for image denoising training have more layers, more complex structures, longer time spent in model training, slower training speed and lower output image precision. In order to increase the training speed and better improve the accuracy of the output image, the invention provides an enhanced network model ED-DCNNet, which reduces the network depth, widens the network width and increases the attention mechanism on the basis of BRDNet. Therefore, the system increases the width of the network by adding a layer of sub-network, enlarges the receptive field, and enables the receptive field to extract more effective characteristics from the environment, thereby improving the denoising performance of the network.
Example 1
The embodiment discloses an image denoising method based on an enhanced depth expansion convolutional neural network.
As shown in fig. 1, the image denoising method based on the enhanced depth expansion convolutional neural network comprises the following steps:
acquiring an original image to be denoised;
building an enhanced deep expansion convolutional neural network model with an upper network layer, a lower network layer and an expansion convolutional layer which are in a three-layer parallel structure;
inputting an original image into an enhanced deep expansion convolutional neural network model, respectively extracting features of the original image through an upper network layer, a lower network layer and an expansion convolutional layer, fusing the extracted features, and then distributing attention to the fused features by using an attention layer to obtain a residual image;
and overlapping the residual image on the original image to obtain a denoising image corresponding to the original image.
Further, the building process of the enhanced deep expansion convolutional neural network model specifically comprises the following steps:
on the basis of a BRDNet network, the depth of an upper network and a lower network is reduced from 17 layers to 10 layers, and a network structure with expansion convolution is added at the same time, so that a three-layer parallel structure comprising the upper network layer, the lower network layer and the expansion convolution layer is formed, and characteristics are extracted;
a fusion layer is arranged behind the three-layer parallel structure, and features extracted from the three-layer parallel structure are fused;
connecting a CA attention layer after the fusion layer;
and connecting a convolution layer after the CA attention layer to obtain the enhanced deep expansion convolution neural network model.
The method comprises the steps of respectively extracting the characteristics of an original image through an upper network layer, a lower network layer and an expansion convolution layer, and specifically comprises the following steps: the upper network layer includes BRN and residual learning, which obtains a potentially clean image by predicting additive gaussian white noise, i.e., the upper network layer can be used to predict noise v, and then use the resulting noise v to generate a clean image x. The lower network layer captures more context information by expanding the acceptance field to extract features. Expanding the convolutional layer extracts more features by increasing the width of the network. The fusion layer is used for connecting three sub-networks, and if the output scale of the three sub-networks is 64x3x3xc, the output scale after the three sub-networks are combined by cascading operation is 64x3x3x3c. The feature layer is input, the weight of each channel is input, and the features can be all concentrated together by using a CA attention mechanism.
Fusing the extracted features, and then distributing attention to the fused features by using an attention layer to obtain residual images, wherein the method specifically comprises the following steps of: the network divides the CNN into several stages according to depth and predicts using feature maps of each stage. These predicted branches are then input into a carefully designed attention module that learns the weights of the predicted branches and aggregates them into the final output.
The CA attention mechanism weights and fuses the features learned in the shallow layer and the deep layer, so that the feature information of each stage is reasonably and effectively processed, more distinguishing features can be learned by the middle layer, and the representation capability of the model is enhanced.
The invention provides an enhanced deep expansion convolutional neural network model ED-DCNNet, which reduces the depth of an upper network and a lower network from 17 layers to 10 layers on the basis of the original BRDNet network, simultaneously increases a network structure with expansion convolutional and adds a CA attention mechanism at the convolutional positions of three networks.
The brdnaet network is a network comprising two layers of subnetworks, with a depth of 17 layers. Wherein, 1-16 layers of the upper network are Conv+BRN+ReLU, and 17 layers are Conv; the 1, 9 and 16 layers of the lower network are Conv+BRN+ReLU, the 2-8 and 10-15 layers are expansion convolution, the last layer is Conv, and the structure proposed by the invention is modified on the BRDNet network.
The overall architecture of the ED-DCNNet network is shown in FIG. 2. The network consists of two parts, an original network and an expanded convolution network. The original network adopts partial upper network and lower network of BRDNet network, in which the upper layer network is formed from RL and BRN, at the same time, in order to reduce receiving domain and reduce calculation cost of network, the lower layer network is added with expansion convolution, and is formed from BRN, RL and expansion convolution.
RL is fully known as ReLu, a rectifying linear unit, capable of preventing gradient extinction or explosion; BRN is batch reorganization to solve the small batch and internal covariate displacement problem.
In order to improve the denoising performance, the invention increases the width of the sub-network, introduces an expanded convolution network, wherein the basic layer of the expanded convolution network consists of a Dilated Conv, BN and RL, the Dilated Conv is the expanded convolution with a certain factor, and the network can achieve better denoising effect under the condition of not increasing the depth of the network through expansion operation, thereby obtaining higher calculation efficiency.
The related Conv is expansion convolution, so that the calculation cost can be saved, the width of a network can be increased, and more context information can be extracted; BN is BRN and has the same meaning as above.
(1) EBRD module composed of upper network and lower network
As shown in fig. 3, in this module, the depth of the upper layer network (first network) is 10, which is composed of two different types of layers of conv+brn+relu and Conv, wherein layers 1 to 10 are conv+brn+relu, and layer 11 is Conv. The first layer has a size of c×3×3×64, the 2 nd to 10 th layers have a size of 64×3×3×64, and the 11 th layer has a size of 64×3×3×c.
Conv+BRN+ReLU means that convolution, batch reforming and rectification are sequentially realized by linear units, and the aim is to improve the performance; conv is convolution, which is the meaning of superimposing the extracted feature information, and belongs to a function of operation.
The lower network (second network) is a lower network with depth of 10, the layers 2-8 are expansion convolution, the layers 2-8 can receive more information from a wider field with the help of expansion factors, and the receptive field can acquire more information from the environment. In addition, the EBRD module has only 10 layers, the number of layers is shallower, gradient disappearance or explosion cannot be caused, and the denoising calculation cost is reduced.
In order to obtain the optimal parameters of the network, the mean square error is selected for calculation. Let x and y be a clean image and a noisy image, respectively, when given a training dataset { xj, yj } nj=1, the EBRD module uses the minimum rectification linear unit to obtain the corresponding model to predict the residual image f (y), and then converts the noisy image into a clean image by x=y-f (y). One such equation can be derived by training samples: f (y) =y-x.
Thus, the following optimal parameters can be obtained by the loss function of the module:
Figure BDA0004124325750000081
wherein y is a noise image; x is the clean image; f (y) is the residual image; n represents the number of noise image patches and θ represents the parameters of the model.
It follows that feature variables in the image can be extracted more effectively by expanding the convolution kernel.
(2) DC module including dilated convolution layer
The module is a depth residual network with extended convolution for image denoising, which aims to automatically learn a training pair with residual noise mapped onto a clean image by designing an efficient architecture with fewer layers. The expansion convolution with a certain expansion factor can effectively increase the receptive field of the convolution layer, and the expansion of the size of the filter can effectively expand the receptive field, so that the module achieves a better denoising effect under the condition of not increasing the depth of the convolution layer.
As shown in FIG. 4, the convolution layer of the module is 10 layers, the first layer is composed of Conv+ReLU+BN, the 2 nd to 10 th layers are composed of related Conv+BN+ReLU, and the expansion factor is certain, so that the exponential expansion of the receptive field can be better supported on the premise of not sacrificing the resolution and the coverage rate, the receptive field is expanded through expansion convolution, and more external information can be received by CNN, so that the image denoising efficiency is effectively improved. Most important among these is the development of a residual model to learn the noise mapping of the noise image to the clean image using the combination of dilation convolution.
To increase the receptive field, we choose to linearly expand the receive domain by stacking several convolutional layers, using a 3x3 filter in the CNN, so that the size of the receive field per layer is doubled. The use of such a small-sized filter not only can reduce the number of model parameters to a greater extent to expand the reception domain, but also can filter natural images more effectively.
In a convolution algorithm, the size of the receiving domain may be calculated according to a given setting:
Figure BDA0004124325750000091
where l represents the input size, k represents the size of the convolution filter, p represents zero plus, s represents the step size.
To expand equation (2), we introduce an expansion factor d, the output dimensions are shown below:
Figure BDA0004124325750000092
let rl, sl (where sl=1), dl be the receive field size, the step size, and the spreading factor of the convolutional layer l, respectively, the filter size k is a predetermined value for all convolutional layers, so the receptive field for layer l can be expressed as:
r l ＝r l-1 +(k-1)*d l (4)
to demonstrate the effectiveness of this technique in detail, we used images in the sliding iron exploration database to train the model, while training eight advanced models of BM3D, IRCNN, dudeNet, EPLL, CSF, dnCNN, FFDNet, BRDNet for the same number of iterations, and tested on three reference datasets of CBSD68, kodak24, and McMaster, respectively, we used identifiable peak signal-to-noise ratio (PSNR) and visual effect to verify the denoising effect, and if the PSNR value of the denoising method on the test dataset is large, it was demonstrated that the denoising method has good performance.
We trained and tested the various models as they were run with Intel Core i7-6700 CPU, 16GBRAM and Nvidia GOn the PC of the eForce GTX 1080Ti Graphic Processing Unit (GPU), nvidia CUDA 8.0 and cuDNN 7.5 are selected to accelerate the computing power of the GPU. The development environment of the software is Windows10.0+Python 3.7+Tensorflow1.3.0+PyCharm2021.2.31, and the learning rate is 1×10 -3 Each subnetwork has a depth of 10 and the ed-DCNNet has a network depth of 11.
The PSNR can be calculated by the following formula:
Figure BDA0004124325750000101
where MAX denotes the maximum pixel value of each image, MSE is the error between the true clean image and the predicted clean image,
Figure BDA0004124325750000102
wherein->
Figure BDA0004124325750000103
And->
Figure BDA0004124325750000104
Representing the pixels from points (i, j) of a given clean image and a restored clean image, respectively.
For images in the dataset to be tested, we will test them at different noise levels and at different datasets. Table one shows the results of tests on three different data sets of DBSD68, kodak24 and McMaster at noise levels of 15, 25, 35, 50 and 70, and compares with the eight advanced methods.
It shows the average PSNR values of each data set at different noise levels, and it can be seen from table one that the PSNR values of ED-DCNNet are higher than those of other networks at different data sets and different noise levels, so that the denoising effect of ED-DCNNet is better.
Fig. 5 shows the visualization of one image of our chosen BSD68 test set with ffdnat and noisy images at noise level δ=35. The four graphs in fig. 5 are comparative experiments, verifying which method has the greatest PSNR value at the same noise level and is most effective. It can be seen that the frame proposed herein works best.
Table 1 comparison with other advanced methods on different data sets
Figure BDA0004124325750000111
At a noise level of 35 in fig. 5, the BSD68 data set has one image denoised:
fig. 5 (a) shows the denoising effect of ED-DCNNet according to the present invention: PSNR value 34.55dB;
fig. 5 (b) is a denoising effect using the FFDNet model: PSNR value is 32.45dB;
FIG. 5 (c) noise/18.35 dB, which is a picture at a noise picture PSNR value of 18.35 dB;
fig. 5 (d) Original/35dB, i.e. Original image with PSNR value of 35 dB.
The invention compares the training time in addition to the image denoising effect and mode. Under the same experimental environment, the model provided by the invention has improved training speed and performance, and the time spent for training the Waterloo Exploration data set for 50 times is reduced by about 19 hours compared with BRDNet.
Example two
The embodiment discloses an image denoising system based on an enhanced depth expansion convolutional neural network.
As shown in fig. 6, an image denoising system based on an enhanced depth-expanded convolutional neural network, comprising:
an original image acquisition module configured to: acquiring an original image to be denoised;
a model building module configured to: building an enhanced deep expansion convolutional neural network model with an upper network layer, a lower network layer and an expansion convolutional layer which are in a three-layer parallel structure;
a feature extraction module configured to: inputting an original image into an enhanced deep expansion convolutional neural network model, respectively extracting features of the original image through an upper network layer, a lower network layer and an expansion convolutional layer, fusing the extracted features, and then distributing attention to the fused features by using an attention layer to obtain a residual image;
a superposition module configured to: and overlapping the residual image on the original image to obtain a denoising image corresponding to the original image.
Example III
An object of the present embodiment is to provide a computer-readable storage medium.
A computer readable storage medium having stored thereon a computer program which when executed by a processor performs the steps in an enhanced deep dilation convolutional neural network based image denoising method as described in embodiment 1 of the present disclosure.
Example IV
An object of the present embodiment is to provide an electronic apparatus.
An electronic device comprising a memory, a processor, and a program stored on the memory and executable on the processor, which when executed, implements the steps in an image denoising method based on an enhanced deep dilation convolutional neural network as described in embodiment 1 of the present disclosure.
The steps involved in the devices of the second, third and fourth embodiments correspond to those of the first embodiment of the method, and the detailed description of the embodiments can be found in the related description section of the first embodiment. The term "computer-readable storage medium" should be taken to include a single medium or multiple media including one or more sets of instructions; it should also be understood to include any medium capable of storing, encoding or carrying a set of instructions for execution by a processor and that cause the processor to perform any one of the methods of the present invention.
It will be appreciated by those skilled in the art that the modules or steps of the invention described above may be implemented by general-purpose computer means, alternatively they may be implemented by program code executable by computing means, whereby they may be stored in storage means for execution by computing means, or they may be made into individual integrated circuit modules separately, or a plurality of modules or steps in them may be made into a single integrated circuit module. The present invention is not limited to any specific combination of hardware and software.
While the foregoing description of the embodiments of the present invention has been presented in conjunction with the drawings, it should be understood that it is not intended to limit the scope of the invention, but rather, it is intended to cover all modifications or variations within the scope of the invention as defined by the claims of the present invention.

Claims (10)

1. The image denoising method based on the enhanced depth expansion convolutional neural network is characterized by comprising the following steps of:
acquiring an original image to be denoised;
building an enhanced deep expansion convolutional neural network model with an upper network layer, a lower network layer and an expansion convolutional layer which are in a three-layer parallel structure;
inputting an original image into an enhanced deep expansion convolutional neural network model, respectively extracting features of the original image through an upper network layer, a lower network layer and an expansion convolutional layer, fusing the extracted features, and then distributing attention to the fused features by using an attention layer to obtain a residual image;
and overlapping the residual image on the original image to obtain a denoising image corresponding to the original image.
2. The image denoising method based on the enhanced deep expansion convolutional neural network according to claim 1, wherein the building process of the enhanced deep expansion convolutional neural network model specifically comprises the following steps:
on the basis of a BRDNet network, the depth of an upper network and a lower network is reduced from 17 layers to 10 layers, and a network structure with expansion convolution is added at the same time, so that a three-layer parallel structure comprising the upper network layer, the lower network layer and the expansion convolution layer is formed, and characteristics are extracted;
a fusion layer is arranged behind the three-layer parallel structure, and features extracted from the three-layer parallel structure are fused;
connecting a CA attention layer after the fusion layer;
and connecting a convolution layer after the CA attention layer to obtain the enhanced deep expansion convolution neural network model.
3. The image denoising method based on the enhanced deep expansion convolutional neural network as claimed in claim 2, wherein the upper network consists of Conv+BRN+ReLU and Conv layers of two different types, wherein the depth is 10, the 1 st to 10 th layers are Conv+BRN+ReLU, and the 11 th layer is Conv; the first layer has a size of c×3×3×64, the 2 nd to 10 th layers have a size of 64×3×3×64, and the 11 th layer has a size of 64×3×3×c.
4. The image denoising method based on the enhanced deep expansion convolutional neural network according to claim 2, wherein the lower network consists of Conv+BRN+ReLU and expansion convolutional layers of two different types, wherein the depth is 10, the 1 st layer is Conv+BRN+ReLU, the 2 nd to 8 th layers are expansion convolutional layers, and the 11 th layer is Conv.
5. The image denoising method based on the enhanced deep expansion convolutional neural network according to claim 2, wherein the expansion convolutional layer is 10 layers, the first layer is composed of conv+relu+bn, the 2 nd to 10 th layers are composed of dimated conv+bn+relu, and the 11 th layer is Conv.
6. The image denoising method based on an enhanced deep dilation convolutional neural network of claim 5, wherein in the convolution algorithm, the size of the receiving domain is expressed as:
Figure FDA0004124325740000021
where l represents the input size, k represents the size of the filter, p represents zero plus, s represents the step size, and d represents the expansion factor.
7. The image denoising method based on enhanced deep dilation convolutional neural network of claim 5, wherein in the dilation convolutional layer, the receptive field of the first layer can be expressed as:
r l ＝r l-1 +(k-1)*d l
wherein rl and dl are the received field size, the spreading factor of the convolution layer l, and the filter size k, respectively.
8. The image denoising system based on the enhanced depth expansion convolutional neural network is characterized in that: comprising the following steps:
an original image acquisition module configured to: acquiring an original image to be denoised;
a model building module configured to: building an enhanced deep expansion convolutional neural network model with an upper network layer, a lower network layer and an expansion convolutional layer which are in a three-layer parallel structure;
a feature extraction module configured to: inputting an original image into an enhanced deep expansion convolutional neural network model, respectively extracting features of the original image through an upper network layer, a lower network layer and an expansion convolutional layer, fusing the extracted features, and then distributing attention to the fused features by using an attention layer to obtain a residual image;
a superposition module configured to: and overlapping the residual image on the original image to obtain a denoising image corresponding to the original image.
9. A computer readable storage medium having a program stored thereon, which when executed by a processor, implements the steps in the enhanced deep dilation convolutional neural network based image denoising method of any one of claims 1-7.
10. An electronic device comprising a memory, a processor and a program stored on the memory and executable on the processor, wherein the processor performs the steps in the enhanced deep dilation convolutional neural network-based image denoising method as claimed in any one of claims 1 to 7 when the program is executed.
CN202310241500.3A 2023-03-14 2023-03-14 Image denoising method and system based on enhanced depth expansion convolutional neural network Pending CN116245765A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310241500.3A CN116245765A (en) 2023-03-14 2023-03-14 Image denoising method and system based on enhanced depth expansion convolutional neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310241500.3A CN116245765A (en) 2023-03-14 2023-03-14 Image denoising method and system based on enhanced depth expansion convolutional neural network

Publications (1)

Publication Number Publication Date
CN116245765A true CN116245765A (en) 2023-06-09

Family

ID=86634895

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310241500.3A Pending CN116245765A (en) 2023-03-14 2023-03-14 Image denoising method and system based on enhanced depth expansion convolutional neural network

Country Status (1)

Country Link
CN (1) CN116245765A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116893175A (en) * 2023-07-04 2023-10-17 千顺智能(珠海)有限公司 Multi-station multi-defect unified connector measurement system based on deep learning

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116893175A (en) * 2023-07-04 2023-10-17 千顺智能(珠海)有限公司 Multi-station multi-defect unified connector measurement system based on deep learning

Similar Documents

Publication Publication Date Title
CN111028146B (en) Image super-resolution method for generating countermeasure network based on double discriminators
CN109035149B (en) License plate image motion blur removing method based on deep learning
CN109671026B (en) Gray level image noise reduction method based on void convolution and automatic coding and decoding neural network
CN109523470B (en) Depth image super-resolution reconstruction method and system
CN111709883B (en) Image detection method, device and equipment
CN111476249B (en) Construction method of multi-scale large-receptive-field convolutional neural network
CN110189260B (en) Image noise reduction method based on multi-scale parallel gated neural network
CN111369450B (en) Method and device for removing mole marks
CN110969089B (en) Lightweight face recognition system and recognition method in noise environment
CN111353956B (en) Image restoration method and device, computer equipment and storage medium
CN112634146A (en) Multi-channel CNN medical CT image denoising method based on multiple attention mechanisms
CN116245765A (en) Image denoising method and system based on enhanced depth expansion convolutional neural network
CN108230269B (en) Grid removing method, device and equipment based on depth residual error network and storage medium
CN111784699A (en) Method and device for carrying out target segmentation on three-dimensional point cloud data and terminal equipment
CN110782001A (en) Improved method for using shared convolution kernel based on group convolution neural network
CN116977651B (en) Image denoising method based on double-branch and multi-scale feature extraction
CN116523888B (en) Pavement crack detection method, device, equipment and medium
CN111861940A (en) Image toning enhancement method based on condition continuous adjustment
WO2023206343A1 (en) Image super-resolution method based on image pre-training strategy
CN116091844A (en) Image data processing method and system based on edge calculation
CN113256528B (en) Low-illumination video enhancement method based on multi-scale cascade depth residual error network
CN112734673B (en) Low-illumination image enhancement method and system based on multi-expression fusion
CN114529450A (en) Face image super-resolution method based on improved depth iterative cooperative network
CN112308772B (en) Super-resolution reconstruction method based on deep learning local and non-local information
Shen et al. Itsrn++: Stronger and better implicit transformer network for continuous screen content image super-resolution

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