An image recognition method based on high-order spatial information interaction

CN115272671BActive Publication Date: 2026-07-03TSINGHUA UNIVERSITY
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
TSINGHUA UNIVERSITY
Filing Date
2022-07-22
Publication Date
2026-07-03

AI Technical Summary

Technical Problem

Existing visual models lack explicit high-order spatial interactions in image recognition tasks, resulting in insufficient representational power and difficulty in meeting the needs of complex visual tasks.

Method used

An image recognition method based on high-order spatial information interaction is adopted. By recursively gated convolution and high-order interactive network (HorNet), the spatial interaction capability of the visual model is enhanced. A new basic unit for visual modeling is constructed, including high-order interactive feature pyramid network (HorFPN) to improve the representation capability of the visual model.

Benefits of technology

It significantly improves the performance of visual models in tasks such as image recognition, semantic segmentation, and object detection, and provides stronger image detection capabilities.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115272671B_ABST
    Figure CN115272671B_ABST
Patent Text Reader

Abstract

This invention discloses an image recognition method based on high-order spatial information interaction. The method includes: acquiring initial visual features of an input image; inputting the initial visual features into a high-order interactive network model to obtain first visual features; segmenting the image visual features to obtain multi-dimensional sub-feature maps; performing feature interaction on the multi-dimensional sub-feature maps and outputting a mapping to obtain second visual features; obtaining third visual features based on a deep feature representation of the second visual features; and transforming the third visual features to the required output dimension through the output mapping to obtain the final visual features of the image. This invention can achieve accurate recognition output for various types of images.
Need to check novelty before this filing date? Find Prior Art