Unlock instant, AI-driven research and patent intelligence for your innovation.

Lightweight super-resolution image reconstruction method based on deep learning

An image reconstruction and super-resolution technology, applied in image data processing, graphics and image conversion, instruments, etc., can solve problems such as difficult deployment of mobile devices, deep neural network, etc.

Pending Publication Date: 2021-09-14
HANGZHOU DIANZI UNIV
View PDF0 Cites 1 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0004] The present invention solves the defect that the current neural network depth is too large and it is difficult to deploy on mobile devices

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
  • Lightweight super-resolution image reconstruction method based on deep learning
  • Lightweight super-resolution image reconstruction method based on deep learning
  • Lightweight super-resolution image reconstruction method based on deep learning

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0025] The present invention will be further described below in conjunction with the accompanying drawings and specific implementation.

[0026] Such as figure 1 Shown is the difference between the Tiny-FSRCNN network model and FSRCNN. The Tiny-FSRCNN network model is mainly divided into feature extraction, compression, mapping and reconstruction. The main functions of each part are introduced in detail below:

[0027] A) Feature extraction

[0028] Compared with the 5×5 receptive field in the first layer of FSRCNN, Conv(5,d,1) requires a 5×5 convolution kernel for convolution operations. Here, a smaller 3×3 receptive field and a larger The 64-channel convolution kernel reduces the complexity of the design in the hardware implementation stage of the algorithm while the feature extraction effect is not greatly reduced. The size of all convolution kernels is 3×3, so that the hardware design will be more efficient. It is simple, and uses a smaller 3×3 convolution design, which...

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 lightweight super-resolution image reconstruction method based on deep learning, and the method achieves the optimization of a classical FSRCNN network model from the hardware implementation through the reference of the method advantages of a shallow convolutional network, proposes a lightweight super-resolution shallow convolutional neural network Tiny-FSRCNN, and designs a lightweight super-resolution image reconstruction method based on deep learning according to the method; through calculation of a hardware circuit acceleration method, the image reconstruction speed of a small terminal is increased, and the quality of a reconstructed image is guaranteed.

Description

technical field [0001] The invention belongs to the field of artificial intelligence and relates to a lightweight super-resolution image reconstruction method based on deep learning. Background technique [0002] In recent years, image super-resolution algorithms based on deep learning methods have achieved remarkable results. Since 2012, the emergence of the AlexNet network has made the convolutional neural network deeper and wider, which has greatly improved the performance of the convolutional neural network and promoted the development of the convolutional neural network field. In order to improve the performance of the network, many scholars have started deeper network attempts, such as VGG and GoogLeNet, and deeper network structures have brought better training results. [0003] However, most of the research focuses on the improvement of the objective accuracy index of the algorithm on various benchmark datasets, thus adopting a network with a high dimensionality in ...

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
Patent Type & Authority Applications(China)
IPC IPC(8): G06T3/40
CPCG06T3/4053
Inventor 孙浩高恒洋岳晨曦
Owner HANGZHOU DIANZI UNIV