A high-altitude image target detection method based on RT-DETR

By improving the high-altitude image target detection method based on the RT-DETR architecture and utilizing MAFNet, GC3Block, and FGAD modules, the technical problems in the existing technology are solved, the technical challenges of target detection in high-altitude environments are overcome, and the detection performance is improved, especially the detection accuracy and adaptability.

CN122416264APending Publication Date: 2026-07-17CHINA UNIV OF PETROLEUM (EAST CHINA)
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-05-06
Publication Date
2026-07-17

AI Technical Summary

Technical Problem

High-altitude image target detection suffers from problems such as insufficient multi-scale feature modeling, weak feature fusion targeting, and easy loss of structural information during upsampling, resulting in limited detection accuracy and poor model adaptability.

Method used

An improved RT-DETR architecture is adopted, including the backbone feature extraction network MAFNet, the feature fusion enhancement module GC3Block, and the upsampling enhancement module FGAD. Through multi-scale adaptive fusion, selective fusion, and dynamic upsampling, the feature extraction and fusion capabilities are improved, and the model's adaptability to high-altitude scenes is enhanced.

Benefits of technology

It improves the accuracy and adaptability of high-altitude image target detection, and enhances detection performance, especially the accuracy of target recognition and localization in complex scenes.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122416264A_ABST
    Figure CN122416264A_ABST
Patent Text Reader

Abstract

本发明属于计算机视觉领域,公开一种基于改进RT‑DETR架构的高空图像目标检测方法。首先获取并预处理航拍图像数据集,构建AIR‑DETR模型,将图像输入后依次完成特征提取、融合、上采样增强与目标预测,输出检测结果,并在训练环境中通过训练集与验证集完成模型训练与验证。该模型由骨干网络MAFNet、特征融合模块GC3Block、上采样模块FGAD及混合解码器构成。MAFNet通过通道压缩、多分支深度可分离卷积、自适应融合与残差调制,实现多尺度特征建模;GC3Block基于门控机制实现多层特征选择性融合,减少冗余;FGAD通过焦点引导自适应上采样,实现低分辨率特征的精细恢复。该方法有效提升复杂高空场景中小目标、遮挡及尺度变化目标的检测精度与适应性。
Need to check novelty before this filing date? Find Prior Art