基于光谱降维和特征融合的高光谱目标跟踪  

Hyperspectral Target Tracking Based on Spectral Dimensionality Reduction and Feature Fusion

作　　者：武丽 汪梦元 黄鲲鹏 田昊翔 仲伟翔 蒲征 王青 WU Li;WANG Mengyuan;HUANG Kunpeng;TIAN Haoxiang;ZHONG Weixiang;PU Zheng;WANG Qing(School of Electronics and Information Engineering Nanjing University of Information Science and Technology,Nanjing 210000 China;School of Electronics and Information Engineering Wuxi University,Wuxi 214000 China;School of Electronics and Information Engineering Harbin Institute of Technology,Harbin 150000 China;School of Physics Xi'an University of Electronic Science and Technology,Xi'an 710000 China)

机构地区：[1]南京信息工程大学电子与信息工程学院,南京210000 [2]无锡学院电子信息工程学院,江苏无锡214000 [3]哈尔滨工业大学电子与信息工程学院,哈尔滨150000 [4]西安电子科技大学物理学院,西安710000

出　　处：《电光与控制》2025年第2期7-12,共6页Electronics Optics & Control

基　　金：国家自然科学基金(62001443,62105258);江苏省自然科学基金(BK20210064);无锡市创新创业资金“太湖之光”科技攻关计划(基础研究)项目(K20221046);无锡学院人才启动基金(2021r007)。

摘　　要：针对现有高光谱视频跟踪算法在目标尺度变化时表现不佳的问题,提出一种基于光谱降维和特征融合的高光谱视频目标跟踪算法。首先,计算目标局部光谱曲线的差值并结合特征值排序和阈值设定获取目标光谱曲线;随后,利用目标光谱曲线与高光谱图像进行光谱角距离计算来实现降维;之后,利用改进的多尺度胶囊网络提取多尺度特征,为利用不同尺度的信息,将降维生成的掩模进行多尺度特征融合;最后,将融合的多尺度特征输入分类和回归胶囊,利用模版更新机制增强跟踪的稳定性和鲁棒性,使得所提算法能够更好地应对尺度变化带来的挑战。实验结果表明,所提算法在应对尺度变化挑战时具有优越性。Aiming at the problem that existing hyperspectral video tracking algorithms perform poorly when target scale variation a hyperspectral video target tracking algorithm based on spectral dimensionality reduction and feature fusion is proposed.Firstly the difference of the local spectral curve of the target is calculated and the spectral curve of the target is obtained by combining eigenvalue sorting and threshold setting.Subsequently the target spectral curve and the hyperspectral image are used to calculate the spectral angular distance to achieve dimensionality reduction.Then the improved multi-scale capsule network is used to extract multi-scale features.In order to use the information of different scales the mask generated by dimensionality reduction is fused with multi-scale features.Finally the fused multi-scale features are input into the classification and regression capsule and the template updating mechanism is used to enhance the stability and robustness of tracking so that the algorithm can better cope with the challenges brought by scale variation.Experimental results indicate the superiority of the proposed algorithm in dealing with the challenge of scale variation.

关 键 词：目标跟踪 光谱降维 胶囊网络 高光谱视频 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]