基于分段组合特征降维的交叉视角目标定位研究  

Research on cross⁃view target location based on segmental combination feature dimension reduction

作　　者：刘琳 赵化启[1] Liu Lin;Zhao Huaqi(School of Information and Electronics Technology,Jiamusi University,Jiamusi 154007,China)

机构地区：[1]佳木斯大学信息电子技术学院,佳木斯154007

出　　处：《现代计算机》2023年第13期39-44,共6页Modern Computer

基　　金：佳木斯大学优秀学科团队项目(JDXKTD⁃2019008);佳木斯大学教育教学改革研究项目(2021JY1⁃49);佳木斯大学国家基金培育项目(JMSUGPZR2022⁃016)。

摘　　要：为提高交叉视角目标定位的精度,提出了一种基于分段组合特征降维的交叉视角目标定位方法。首先使用ResNet⁃50作为主干网络,并选取实例损失函数,提高了目标定位的性能。其次,为去除所提取特征的冗余信息,提出了一种分段组合降维方案对图像全局特征进行降维,保留了特征的主要信息并降低了特征维度,从而提高了目标定位的效率。在University⁃1652数据集上进行验证,实验表明所提方法与降维之前特征匹配相比,AP和Recall@1分别提升了1.08倍和1.1倍,能有效提高定位精度。In order to improve the accuracy of cross‑perspective target positioning,a cross-perspective target localization method based on segmented feature reduction based on segmented combination feature reduction is proposed.Firstly,ResNet-50 is used as the backbone network,and the instance loss function is selected,which improves the performance of target positioning.Secondly,in order to remove the redundant information of the extracted features,a segmented combined dimensionality reduction scheme is proposed to reduce the dimensionality of the global features of the image,retain the main information of the features and reduce the feature dimension,thereby improving the efficiency of target positioning.Verified on the University‑1652 dataset,experiments show that compared with the feature matching before dimensionality reduction,the AP and Recall@1 are improved by 1.08 times and 1.1 times,respectively,which can effectively improve the positioning accuracy.

关 键 词：目标定位 特征提取 特征降维 交叉视角 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]