Remote sensing object detection via local parameter-free attention and combined loss

目的遥感目标检测技术在遥感测绘、智慧城市、乡村振兴和国防军事等领域具有广泛应用,但遥感影像存在背景信息复杂、待检目标小且多等特点,导致目标特征随着网络加深淹没在背景信息中,不利于后续检测任务。针对此问题,基于YOLOv5s(you only look once)网络,提出一种局部无参注意力和联合损失的遥感目标检测方法。方法首先提出一种局部无参注意力机制,能够根据当前特征提高局部区域内的目标关注度,而不引入训练参数,以此构建无参注意力残差块,并将其嵌入到骨干网络的不同阶段,实现多尺度目标特征增强和背景信息抑制。在此基础上,利用最优传输距离度量边界框的相似性,构造W-CIoU(Wassertein-complete intersection of union)联合度量方法和相应损失函数,以缓解锚框与真实框差异过大导致的标签误分配问题,降低小目标漏检率。结果在RSOD(remote sensing object detection)和RSSOD(remote sensing super-resolution object detection)数据集上的实验结果表明,该方法在保证模型参数量和复杂度基本不变的前提下,其平均精度均值(mean average precision, mAP)分别达到98.2%及87.4%,相较两种主要对比方法DAB-DETR(dynamic anchor boxes for detection Transformer)和YOLOv8s,在RSOD数据集上分别提升0.8%和0.3%,在RSSOD数据集上分别提升2.9%和2.6%。结论本文方法能够抑制背景信息并增强目标特征,提高检测性能,相较对比方法能更好地满足复杂场景下小目标检测需要。

Authors

Publication Details

Journal
Journal of Image and Graphics
Published
2024-09-12
DOI
https://doi.org/10.11834/jig.240316
Primary Topic
Remote-Sensing Image Classification
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
article

Remote sensing object detection via local parameter-free attention and combined loss

博 山下, Xinyao Zhou, Hong Huang, Weitao Xue
Journal of Image and Graphics
Remote-Sensing Image Classification
article

Remote sensing object detection via local parameter-free attention and combined loss

博 山下, Xinyao Zhou, Hong Huang, Weitao Xue
article en

Abstract

目的遥感目标检测技术在遥感测绘、智慧城市、乡村振兴和国防军事等领域具有广泛应用,但遥感影像存在背景信息复杂、待检目标小且多等特点,导致目标特征随着网络加深淹没在背景信息中,不利于后续检测任务。针对此问题,基于YOLOv5s(you only look once)网络,提出一种局部无参注意力和联合损失的遥感目标检测方法。方法首先提出一种局部无参注意力机制,能够根据当前特征提高局部区域内的目标关注度,而不引入训练参数,以此构建无参注意力残差块,并将其嵌入到骨干网络的不同阶段,实现多尺度目标特征增强和背景信息抑制。在此基础上,利用最优传输距离度量边界框的相似性,构造W-CIoU(Wassertein-complete intersection of union)联合度量方法和相应损失函数,以缓解锚框与真实框差异过大导致的标签误分配问题,降低小目标漏检率。结果在RSOD(remote sensing object detection)和RSSOD(remote sensing super-resolution object detection)数据集上的实验结果表明,该方法在保证模型参数量和复杂度基本不变的前提下,其平均精度均值(mean average precision, mAP)分别达到98.2%及87.4%,相较两种主要对比方法DAB-DETR(dynamic anchor boxes for detection Transformer)和YOLOv8s,在RSOD数据集上分别提升0.8%和0.3%,在RSSOD数据集上分别提升2.9%和2.6%。结论本文方法能够抑制背景信息并增强目标特征,提高检测性能,相较对比方法能更好地满足复杂场景下小目标检测需要。

Journal of Image and GraphicsVol. 30(3)
National Natural Science Foundation of China
Openalex Percentile: Top 75%
Remote-Sensing Image Classification
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Remote sensing object detection via local parameter-free attention and combined loss — 博 山下, Xinyao Zhou, et al. · Journal of Image and Graphics (2024) | TGRS Research Map | TGRS