Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

Medical image super-resolution reconstruction method based on multi-attention residual feature fusion

A technology of super-resolution reconstruction and feature fusion, applied in neural learning methods, image data processing, graphics and image conversion, etc., can solve problems such as insufficient network representation ability and insufficient learning ability, so as to improve image super-resolution reconstruction performance, solve Effects of Insufficient Network Learning Ability

Pending Publication Date: 2021-08-24
ZHEJIANG UNIV OF TECH
View PDF0 Cites 6 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0004] In order to overcome the existing deficiencies, the present invention improves the above two problems, and proposes a medical image super-resolution reconstruction method based on multi-attention residual feature fusion, which combines channel attention mechanism and spatial attention The two modules of the mechanism allow the network to weight the features of the input image, so as to solve the problem of insufficient network representation ability caused by the equal processing of the input features; the residual attention feature fusion module (RAFF) is proposed, which combines the neural network Local feature fusion, coupled with global feature fusion, can make full use of the local residual features in the entire network, thereby solving the problem that the deep neural network is too deep and the network learning ability is insufficient

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
  • Medical image super-resolution reconstruction method based on multi-attention residual feature fusion
  • Medical image super-resolution reconstruction method based on multi-attention residual feature fusion
  • Medical image super-resolution reconstruction method based on multi-attention residual feature fusion

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0039] The present invention will be further described below in conjunction with the accompanying drawings.

[0040] refer to figure 1 with figure 2 , a medical image super-resolution reconstruction method based on multi-attention residual feature fusion, combining channel attention and spatial attention modules, weighting the features of the input image from multiple aspects, plus local and global residuals The features are fused and a novel super-resolution reconstruction method is constructed, including the following steps:

[0041] 1) Processing of training samples:

[0042] In order to make the proposed super-resolution network model have corresponding high-resolution images (i.e. HR) and low-resolution images (i.e. LR) in the retraining process, the present invention first performs a high-definition medical image original data set with a resolution of 512x512. 2x and 4x bicubic downsampling, resulting in low-resolution datasets at 256x256 and 128x128 resolutions.

...

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

A medical image super-resolution reconstruction method based on multi-attention residual feature fusion combines two modules of a channel attention mechanism and a space attention mechanism, and enables a network to perform weighting processing on features of an input image, thereby solving the problem of insufficient network characterization capability caused by equal processing of the input features. A residual attention feature fusion module (RAFF) is provided, local features in the neural network are fused, and global features are fused, so that local residual features in the whole network can be more fully utilized, and the problem that the deep neural network is too deep and the learning ability of the network is insufficient is solved. According to the method, super-resolution reconstruction is effectively carried out on the medical image, and the image super-resolution reconstruction performance of a deep super-resolution network can be improved.

Description

technical field [0001] The invention relates to a medical image super-resolution reconstruction method based on multi-attention residual feature fusion. Background technique [0002] In recent years, deep convolutional neural networks have achieved very good results in super-resolution on a single image. A common method for image super-resolution is the instance-based method, which uses high-resolution images and low-resolution The high-resolution image information is used to generate a super-resolution version that approximates the original high-resolution image. There are also many methods of using super-resolution on medical images. The super-resolution reconstruction of medical images helps to improve the accuracy and objectivity of computer-aided clinical medical diagnosis and quantitative analysis of diseases. Therefore, the medical images that need It is necessary to perform super-resolution reconstruction. [0003] The defects of the existing technology are: most o...

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): G06T3/40G06K9/62G06N3/04G06N3/08
CPCG06T3/4053G06N3/08G06N3/045G06F18/253
Inventor 徐涵杰管秋陆正威韦子晗陈奕州
Owner ZHEJIANG UNIV OF TECH
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products