Check patentability & draft patents in minutes with Patsnap Eureka AI!

Double-image recognition and classification method based on attention mechanism and multi-size information extraction

A technology of information extraction, recognition and classification, applied in image analysis, details involving image stitching, image enhancement, etc., can solve the problems of low image classification accuracy and incomplete features, and achieve the goal of improving classification accuracy and enhancing expression ability Effect

Active Publication Date: 2022-04-29
UNIV OF SCI & TECH BEIJING
View PDF17 Cites 2 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0004] In order to solve the above-mentioned technical problems, the object of the present invention is to provide a double-image recognition and classification method based on attention mechanism and multi-size information extraction, so as to solve the problem that the existing image classification method only inputs one image of the surface of the object and ignores the other images of the object. The characteristics of the situation lead to the problem of incomplete features and low image classification accuracy

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
  • Double-image recognition and classification method based on attention mechanism and multi-size information extraction
  • Double-image recognition and classification method based on attention mechanism and multi-size information extraction
  • Double-image recognition and classification method based on attention mechanism and multi-size information extraction

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0060] Embodiment 1: On the one hand, the present invention provides a kind of dual image recognition and classification method based on attention mechanism and multi-size information extraction, comprising:

[0061] Acquiring two images of the object to be classified; wherein, the two images are images taken under different conditions at the same position;

[0062] A parallel multi-scale feature extraction network fuses information of different sizes to extract features of two images;

[0063] Using the dual image spatial attention module to fuse the features of the two images from a spatial perspective to obtain spatial fusion feature information;

[0064] Using the dual image channel attention module to fuse the features of the two images from the perspective of the channel to obtain channel fusion feature information;

[0065] The features extracted by the various methods are fused and interacted, and the formed fused information is input to the classification network to ...

Embodiment 2

[0095] Embodiment 2: as figure 1 As shown, the embodiment of the present invention provides a double image recognition and classification method based on attention mechanism and multi-size information extraction, the method includes:

[0096] S1, acquiring two images of the object to be classified; wherein, the two images are images taken at the same position under different conditions;

[0097] It should be noted that, since the surface of the same object captured by the camera is different in different situations, different images exhibit different characteristics. Therefore, in this example, in order to solve the problem that a single image does not have obvious characteristics of an object, two images of the same object in different situations are used to predict the category of the object. Image features in different situations can complement each other to improve the recognition accuracy of the classification model.

[0098] S2, the parallel multi-scale feature extract...

Embodiment 1

[0108] In this embodiment, roughness samples of different sandpaper grinding and polishing types are used to verify the effect of the dual image recognition and classification model based on attention mechanism and multi-scale information extraction. Set different shooting angles to collect roughness images of the surface of the sample polished by sandpaper. The data set divides the roughness into four grades according to different sandpaper types. Table 1 shows the roughness ranges corresponding to different roughness levels and the number of images per angle. The roughness category is 320 in 320-60s, which means the mesh number of sandpaper, and 60s means sanding for 60 seconds. The number of pictures taken at each angle is 160, and the shooting angles are 0 degree, 15 degree, 30 degree and 45 degree. The cross-entropy loss function and the Adam optimizer are used to iteratively update the parameters of the model, the training is iterated 20 times, the size of each batch i...

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 double-image recognition and classification method based on an attention mechanism and multi-size information extraction. The method comprises the following steps: acquiring two images of an object to be classified; wherein the two images are the images shot at the same position under different conditions; the parallel multi-size feature extraction network fuses information of different sizes so as to extract features of different sizes of the two images at the same time; fusing the features of the two images from the angle of space by using a double-image space attention module to obtain space fusion feature information; fusing the features of the two images from the angle of a channel by using a double-image channel attention module to obtain channel fusion feature information; and the features extracted by various methods are mutually interactively fused, and formed fusion information is input into a classification network to obtain the category of the object to be classified. According to the method, two images of the same object under different conditions are combined, the problem that feature information of a single image is not comprehensive is solved, and the classification accuracy is improved.

Description

technical field [0001] The invention relates to the technical fields of image classification and artificial intelligence, in particular to a dual image recognition and classification method based on attention mechanism and multi-size information extraction. Background technique [0002] During the process of the camera taking images of the surface of the object, different shooting angles, light sources of different wavelengths and positions form different images. Affected by the shooting angle and light source, the image taken in a single case cannot fully reflect the real condition of the object. If two objects are taken at the same angle or light source, the images are similar, but the images taken at different angles or light sources are different. Neural networks take similar images as input and can easily predict the categories of different objects as the same category. [0003] Due to the rapid development of deep learning, image classification methods based on convo...

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): G06K9/62G06T3/40G06T5/50G06V10/40G06V10/774G06V10/771
CPCG06T5/50G06T3/4038G06T2207/20221G06T2200/32G06F18/213G06F18/214
Inventor 张桃红郭徐徐范素丽
Owner UNIV OF SCI & TECH BEIJING
Features
  • R&D
  • Intellectual Property
  • Life Sciences
  • Materials
  • Tech Scout
Why Patsnap Eureka
  • Unparalleled Data Quality
  • Higher Quality Content
  • 60% Fewer Hallucinations
Social media
Patsnap Eureka Blog
Learn More

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

© 2025 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com