Construction method of super-resolution convolutional neural network model

A convolutional neural network and convolutional neural network technology, applied in the field of super-resolution convolutional neural network model construction, can solve the problems of large amount of Tensor data, blurring, etc., and achieve reduced memory consumption, no obvious decline, and high speed Improved effect

Pending Publication Date: 2021-02-19
HEILONGJIANG UNIV
View PDF5 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0002] The input of the traditional super-resolution convolutional neural network is usually a pseudo high-resolution image output by the bicubic interpolation algorithm. The resolution of the image has been improved, but it is still very blurry, which needs to be further improved by the convolutional neural network. Quality, such as the SRCNN network, the disadvantage of this method is that the amount of Tensor data flowing in the neural network is large, which is caused by the input of pseudo high-resolution images whose resolution has been increased

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Construction method of super-resolution convolutional neural network model
  • Construction method of super-resolution convolutional neural network model
  • Construction method of super-resolution convolutional neural network model

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0024] according to Figure 1-2 Shown, a kind of construction method of super-resolution convolutional neural network model, described construction method comprises the following steps,

[0025] Step 1: Build a convolutional neural network model;

[0026] Step 2: training the convolutional neural network model of step 1;

[0027] Step 3: Deploy the convolutional neural network model trained in step 2 to realize the image super-resolution function.

[0028] Further, the step 1 specifically includes constructing a convolutional neural network model using two methods of five residual blocks and sub-pixel convolution.

[0029] Further, the construction method of the sub-pixel convolution is specifically to convert the low-resolution feature map (640x360x27) with small length and width dimensions and deep channel depth into a large length and width dimension through pixel shuffling. , high-resolution feature maps (1920x1080x3) with shallow channel depths.

[0030] Further, the ...

Embodiment 2

[0036] Step 1: PSNR (Peak Signal-to-Noise Ratio) and SSIM (Structural Similarity) are evaluation indicators for evaluating the quality of neural network output pictures, the higher the better, and the fewer parameters and training time, the better.

[0037] Due to the use of sub-pixel convolution technology, the images flowing in the network are low-resolution images. The final effect is that the speed of the network inference stage is greatly improved; the memory consumption of the network inference stage is reduced; the final effect of the network is improved.

[0038] If the task is 4 times super-resolution, the size of the low-resolution image is 100*100, and the size of the original high-resolution image is 400*400. If sub-pixel convolution is not used, the network input is a picture pre-enlarged by the bicubic interpolation algorithm , that is, the size of 400*400, such as using sub-pixel convolution, the input does not need to be pre-amplified, that is, the input is 100*...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

PUM

No PUM Login to view more

Abstract

The invention discloses a construction method of a super-resolution convolutional neural network model based on a portable system. The method comprises the steps of 1, constructing a convolutional neural network model; 2, training the convolutional neural network model in the step 1; and 3, deploying the convolutional neural network model trained in the step 2 to realize a super-resolution function of the image. According to the method, an up-sampling algorithm of the sub-pixel convolution layer of the ESPCN is adopted, a network structure before the sub-pixel convolution layer is modified, and a residual block is introduced to construct a network.

Description

technical field [0001] The technical field that the present invention belongs to; Be specifically related to a kind of construction method of super-resolution convolutional neural network model. Background technique [0002] The input of the traditional super-resolution convolutional neural network is usually a pseudo high-resolution image output by the bicubic interpolation algorithm. The resolution of the image has been improved, but it is still very blurry, which needs to be further improved by the convolutional neural network. Quality, such as the SRCNN network, the disadvantage of this method is that the amount of Tensor data flowing in the neural network is large, which is caused by the input of pseudo high-resolution images whose resolution has been increased. In recent years, many network structures for upsampling through the network have been proposed. For example, FSRCNN achieves upsampling through deconvolution layers, and ESPCN achieves upsampling through sub-pix...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

Application Information

Patent Timeline
no application Login to view more
IPC IPC(8): G06T3/40G06N3/04G06N3/08
CPCG06T3/4053G06T3/4046G06N3/08G06N3/045
Inventor 刘明亮王晓航
Owner HEILONGJIANG UNIV
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap