基于L-FPN的无人机上小目标识别模型轻量化方法
A Lightweight Method for Small Object Detection Models on Unmanned Aerial Vehicles Based on L-FPN
魏昊坤 1刘敬一 1陈金勇 1楚博策 1孙裕鑫 2朱进1
作者信息
- 1. 中国电子科技集团公司第五十四研究所 航天信息应用技术重点实验室,石家庄 050081
- 2. 光电信息控制和安全技术重点实验室,天津 300308
- 折叠
摘要
由于遥感图像拍摄的高度和设备不同导致每张图像的地面采样间隔(GSD)也不同,许多小目标往往易被忽略,遥感图像中旋转框目标检测成为当下研究热点.现有的旋转框检测算法主要面向通用场景下的多尺度目标检测,特征金字塔中特征融合计算操作复杂且耗时,部署到无人机上的边缘端设备时面临很大的挑战.因此本文针对该场景下的小目标检测提出基于L-FPN的无人机上小目标识别模型轻量化方法,首先依据图像的GSD信息进行尺度归一化,然后去除特征金字塔中冗余的高层特征图,最后针对小目标检测调整锚框的尺寸.本方法在DOTA数据集上进行训练验证,结果表明本文提出的基于L-FPN的无人机上小目标识别模型轻量化方法在识别精度与传统模型一致的情况下,模型参数量较原模型减少2.7%,模型大小减少28%,推理速度提升13.24%.
Abstract
Oriented object detection in remote sensing images is a current research hotspot.Due to the var-ying heights and equipment used in capturing remote sensing images,the ground sampling distance(GSD)of each image also varies,causing many small objects to be easily overlooked.Existing rotated object detection al-gorithms are mainly aimed at multi-scale object detection in general scenarios.The feature pyramid network(FPN)has complex and time-consuming fusion computations,which still faces great challenges when deployed on edge devices like UAVs.Therefore,this paper proposes a lightweight method for small object detection in UAVs based on L-FPN.First,normalize the scale according to the GSD information of the image.Second,re-move redundant high-level feature maps in the FPN.Finally,adjust the anchor box sizes for small object detec-tion.The method is trained and validated on the DOTA dataset.Results show that compared to the traditional models,the proposed L-FPN-based lightweight method for small object detection in UAVs achieves consistent recognition accuracy,with 2.7%fewer model parameters,28%smaller model size,and 13.24%faster infer-ence speed.
关键词
目标检测/特征金字塔/模型轻量化/遥感图像/无人机Key words
object detection/feature pyramid/model lightweight/remote sensing images/UAV引用本文复制引用
基金项目
中国博士后科学基金(2021M703021)
河北省重点研发计划(22340301D)
河北省博士后基金(B2021003031)
出版年
2024
国家科技期刊平台
NSTL
万方数据