基于特征融合的内河船舶尺度自适应跟踪  

Scale self-adaption tracking for inland ship based on feature fusion

作　　者：邹绵璐 秦芮 ZOU Mianlu;QIN Rui(Department of Big Data and Artificial Intelligence,XinyangUniversity,Xinyang,Henan 464000,China)

机构地区：[1]信阳学院大数据与人工智能学院,河南信阳464000

出　　处：《计算机应用文摘》2023年第16期23-25,共3页Chinese Journal of Computer Application

基　　金：基于相关滤波的跟踪算法研究(2022-XJLYB-018);三维向列型液晶流的正则性准则(2022-XJLYB-004)。

摘　　要：文章对以船舶为载体的跟踪系统进行研究,在解决电子巡航系统中船舶跟踪尺度变化干扰问题。为此,基于相关滤波,提出了一种特征融合策略,将卷积神经网络(Convolutional Neural Network,CNN)特征和Hog特征相结合。通过使用CNN获取船舶目标的空间结构、形状等相对抽象的语义信息,再将VGG网络提取的CNN特征与Hog特征进行融合,实现对内河船舶的跟踪。经过在大量船舶数据集上的验证和分析,发现特征融合算法显著提高了跟踪精确度和成功率。这一特征融合算法为船舶跟踪中尺度干扰问题提供了一条高精确度的可行路径。The article focuses on the research of ship based tracking systems,aiming to solve the interference problem of ship tracking scale changes in electronic cruise control systems.Therefore,based on correlation filtering,a feature fusion strategy is proposed,which combines Convolutional Neural Network(CNN)features and Hog features.By using CNN to obtain relatively abstract semantic information such as the spatial structure and shape of ship targets,and then fusing the CNN features extracted by VGG network with Hog features,the tracking of inland ships is achieved.After verification and analysis on a large number of ship datasets,it was found that the feature fusion algorithm significantly improved tracking accuracy and success rate.This feature fusion algorithm provides a high-precision feasible path for the problem of scale interference in ship tracking.

关 键 词：相关滤波 VGG 特征融合 船舶跟踪 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]