Eye fundus image detection method and system based on dynamic weighted attention mechanism

A fundus image, dynamic weighting technology, applied in the field of image processing, can solve the problems of increased algorithm complexity, multiple GPU resources, occupation, etc., to ensure the recognition accuracy and reduce the complexity.

Active Publication Date: 2022-07-05
BEIJING ZHENHEALTH TECH CO LTD
View PDF2 Cites 2 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0003] However, in order to extract richer features, the existing segmentation detection methods using deep learning often set the network mo

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
  • Eye fundus image detection method and system based on dynamic weighted attention mechanism
  • Eye fundus image detection method and system based on dynamic weighted attention mechanism
  • Eye fundus image detection method and system based on dynamic weighted attention mechanism

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0079] like figure 2 As shown, this embodiment provides a fundus image detection method based on a dynamic weighted attention mechanism, and the method includes:

[0080] S1. Acquire a fundus image to be used.

[0081] During the shooting process of neonates, due to the limited degree of cooperation, the fundus images captured in most cases are not conducive to the identification of lesions. Therefore, it is necessary to crop the invalid area and image enhancement of the neonatal fundus images.

[0082] Among them, the formula for image enhancement is:

[0083] ;

[0084] in, represents the enhanced image, represents the image after cropping the invalid area, means the standard deviation is Gaussian filter.

[0085] S2. Use the fundus image segmentation model to detect the lesion information of the fundus image; the fundus image segmentation model includes n continuous downsampling layers and n continuous upsampling layers, the nth downsampling layer and the firs...

Embodiment 2

[0111] like Figure 4 As shown, this embodiment provides a fundus image detection system based on a dynamic weighted attention mechanism, and the system includes:

[0112] The data acquisition unit M1 is used to acquire the fundus image to be used;

[0113] The lesion detection unit M2 is used for detecting the lesion information of the fundus image by using the fundus image segmentation model; the fundus image segmentation model includes n continuous downsampling layers and n continuous upsampling layers, and the nth downsampling layer Connect with the first upsampling layer:

[0114] The lesion detection unit M2 specifically includes:

[0115] A downsampling module M21, configured to perform continuous n-layer downsampling on the fundus image by using the fundus image segmentation model to obtain n-layer downsampling output features;

[0116] The attention mechanism weighting module M22 is used to fuse the i-th layer downsampling output feature and the adjacent layer down...

Embodiment 3

[0128] This embodiment provides a fundus image detection system based on a dynamic weighted attention mechanism, including an image acquisition device, a preprocessing module, a segmentation network module, a model training module, a forward reasoning module, a postprocessing module and an output module.

[0129] As the data collection end, the image collection device can collect image data by directly connecting with collection equipment such as neonatal fundus camera, or collect existing fundus data stored on other equipment in advance. For the source of data, such as from equipment , data from the network, and from local storage are not limited.

[0130] The main function of the preprocessing module is to standardize the input image. The standardization process includes two parts, one part is to eliminate the invalid area of ​​the neonatal fundus image and reduce its impact on the network model. During the shooting process of newborns, due to the limited degree of cooperati...

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 relates to an eye fundus image detection method and system based on a dynamic weighted attention mechanism, and relates to the technical field of image processing. And detecting lesion information of the eye fundus image of the premature infant by using the eye fundus image segmentation model. Firstly, continuous down-sampling is carried out on an eye fundus image, dynamic weighted attention fusion is carried out on obtained down-sampling features and down-sampling features obtained in an adjacent layer, then the features after weighted fusion are fused with output features of corresponding up-sampling layers, and finally classification convolution operation is carried out on output of the n up-sampling layer. And obtaining the lesion probability of each pixel. According to the method, hierarchical feature fusion and a dynamic weighted attention mechanism are carried out on the shallow network model, and the purposes of reducing the complexity of algorithm design, reducing the operation time of the algorithm and reducing excessive occupation of GPU resources can be achieved while the recognition accuracy is ensured.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a fundus image detection method and system based on a dynamic weighted attention mechanism. Background technique [0002] Image detection Segmentation detection technology is a classic problem in computer vision research and has become a hot spot in the field of image understanding. Traditional segmentation detection methods include segmenting an image into several disjoint regions based on grayscale, color, spatial texture, geometric shape, etc., so that these features show consistency or similarity in the same region. In recent years, with the continuous development of deep learning technology, image segmentation technology has also made rapid progress, and this technology has been widely used in areas such as unmanned driving, augmented reality, and security monitoring. [0003] However, in order to extract richer features, the existing segmentation detection methods...

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): G06T7/00G06T7/11G06T1/60G06N3/04G06N3/08
CPCG06T7/0012G06T7/11G06T1/60G06N3/08G06T2207/10004G06T2207/20021G06T2207/20081G06T2207/20084G06T2207/30041G06N3/045
Inventor 张冬冬
Owner BEIJING ZHENHEALTH TECH CO LTD
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