基于分层融合机制的超分辨率遥感图像目标检测方法  

Object Detection in Remote Sensing Images Based on Super-Resolution Hierarchical Fusion Mechanism

作　　者：殷凌锋 童旭东 倪欢 YIN Lingfeng;TONG Xudong;NI Huan(School of Remote Sensing and Geomatics Engineering,Nanjing University of Information Technology,Nanjing 210044,China)

机构地区：[1]南京信息工程大学遥感与测绘工程学院,南京210044

出　　处：《航天返回与遥感》2025年第1期123-134,共12页Spacecraft Recovery & Remote Sensing

基　　金：国家高分遥感卫星数据与产品共享交换服务平台项目(2023h277)。

摘　　要：受制于图像光谱信息承载能力限制,基于可见光或红外的单模态目标的检测方法通常难以有效应对遥感图像复杂场景。针对这一问题,文章提出一种基于分层融合机制的超分辨率(超分)遥感图像目标检测方法,有效融合可见光和红外数据信息。首先,基于残差融合模块和单分支增强模块构建分层融合机制,其中残差融合模块整合可见光与红外图像的潜在互补信息,单分支模块进一步强化双模态数据融合特征表达。其次,构建超分辅助分支,增强目标细节特征生成能力,进一步提高检测精度。实验结果表明:文章方法在VEDAI与Drone Vehicle数据集上的检测精度优于传统主流目标检测方法,分别达到了79.45%与81.29%,有效提高了遥感图像目标检测的准确性。Limited by the carrying capacity of image spectral information,single-mode target detection methods based on visible light or infrared are usually difficult to effectively deal with complex scenes of remote sensing images.Aiming at this problem,a super-resolution remote sensing image target detection method based on hierarchical fusion mechanism is proposed,which effectively fuses visible light and infrared data information.Firstly,the residual fusion module and the single-branch module are used to construct a hierarchical fusion mechanism.The residual fusion module combines the potential complementary information of visible and infrared images,and the single-branch module enhances the single-modal features and assists in enhancing the dual-modal data fusion feature expression.Secondly,in order to solve the problem of missing target details in lowresolution images,a super-resolution auxiliary branch is introduced to enhance the ability of target detail feature generation and further improve the detection accuracy.The experimental results show that the detection accuracy(mAP_(50))of the proposed method on the VEDAI and Drone Vehicle datasets is better than that of the existing target detection methods,reaching 79.45%and 81.29%,which effectively improves the accuracy and robustness of remote sensing image target detection in complex environments.

关 键 词：卫星遥感图像 目标检测 超分辨率 特征融合 

分 类 号：TP75[自动化与计算机技术—检测技术与自动化装置] V19[自动化与计算机技术—控制科学与工程]