航空科学技术2025，Vol.36Issue(3)：111-118,8.DOI:10.19452/j.issn1007-5453.2025.03.014
基于目标图像块激活的航空图像目标检测技术研究
Research on Aerial Image Object Detection Based on Object Image Block Activation
摘要
Abstract
Aiming at the key challenges of large image size and dense distribution of targets,which are commonly found in aerial image target detection,this paper pioneers the target image block activation strategy,which not only solves the inefficiency problem of traditional methods in processing aerial images,but also significantly improves the accuracy performance in complex scenes,which is of great significance in promoting the development of aerial image target detection technology.Existing target detection techniques for aerial images are processed by simply cropping image blocks,which is not only inefficient,but also leads to waste of resources and increase of false detection rate due to the detection of a large number of invalid image blocks.Therefore,this paper proposes a target image block activation module(TIBAM),which enables the detector to focus on valid image blocks containing the target by introducing a convolutional attention mechanism,thus reducing the waste of resources on invalid image blocks and realizing the intelligent recognition and processing of potentially valid image blocks.The introduction of TIBAM brings more than 17%improvement in the inference speed for the detector,and also achieves a stable improvement in detection accuracy,providing an innovative solution for efficient and accurate aerial image target detection.This paper fully demonstrate the practical application value and wide applicability of the TIBAM module by integrating it in the mainstream one-stage detector Retinanet and two-stage detector Faster-RCNN,as well as validating it on the Visdrone aerial image dataset.关键词
高效目标检测/航空图像/注意力机制/卷积神经网络Key words
efficient target detection/aerial images/attention mechanism/convolutional neural network分类
航空航天引用本文复制引用
张佳,冯婕,张骏鹏,朱潇雨..基于目标图像块激活的航空图像目标检测技术研究[J].航空科学技术,2025,36(3):111-118,8.基金项目
航空科学基金(2019ZC081002) Aeronautical Science Foundation of China(2019ZC081002) （2019ZC081002）
