Fewer is more:efficient object detection in large aerial images
作者机构：School of AutomationNorthwestern Polytechnical University Zhengzhou Institute of Surveying and Mapping
出 版 物：《Science China(Information Sciences)》 (中国科学:信息科学(英文版))
年 卷 期：2024年第67卷第1期
页 面：105-123页
核心收录：
学科分类：08[工学] 080203[工学-机械设计及理论] 0802[工学-机械工程]
基 金：supported in part by National Natural Science Foundation of China (Grant Nos.62136007,62376223) Natural Science Basic Research Program of Shaanxi (Grant Nos.2021JC-16,2023-JC-ZD-36) Fundamental Research Funds for the Central Universities Doctorate Foundation of Northwestern Polytechnical University (Grant No.CX2021082)
主 题：efficient object detection large aerial images objectness activation network
摘 要：Current mainstream object detection methods for large aerial images usually divide large images into patches and then exhaustively detect the objects of interest on all patches,no matter whether there exist objects or *** paradigm,although effective,is inefficient because the detectors have to go through all patches,severely hindering the inference *** paper presents an objectness activation network(OAN) to help detectors focus on fewer patches but achieve more efficient inference and more accurate results,enabling a simple and effective solution to object detection in large *** brief,OAN is a light fully-convolutional network for judging whether each patch contains objects or not,which can be easily integrated into many object detectors and jointly trained with them *** extensively evaluate our OAN with five advanced *** OAN,all five detectors acquire more than 30.0% speed-up on three large-scale aerial image datasets,meanwhile with consistent accuracy *** extremely large Gaofen-2 images(29200 × 27620 pixels),our OAN improves the detection speed by 70.5%.Moreover,we extend our OAN to driving-scene object detection and 4K video object detection,boosting the detection speed by 112.1% and 75.0%,respectively,without sacrificing the accuracy.
同方期刊数据库