基于分组特征赋权的动态视角图像特征融合  

Feature fusion of dynamic visual angle images based on grouping feature weight

作　　者：张佳琦 张金艺[1] 楼亮亮[2] Zhang Jiaqi;Zhang Jinyi;Lou Liangliang(Key Laboratory of Specialty Fiber Optics and Optical Access Networks,Joint International Research Laboratory of Specialty Fiber Optics and Advanced Communication,Shanghai University,Shanghai 200444,China;Key Laboratory of Wireless Sensor Network&Communication,Shanghai Institute of Microsystem and Information Technology,Chinese Academy of Sciences,Shanghai 200050,China)

机构地区：[1]上海大学特种光纤与光接入网重点实验室特种光纤与先进通信国际合作联合实验室,上海200444 [2]中国科学院上海微系统与信息技术研究所无线传感网与通信重点实验室,上海200050

出　　处：《电子测量技术》2021年第4期144-148,共5页Electronic Measurement Technology

基　　金：高等学校学科创新引智计划(111)资助(D20031);上海市教委重点学科资助项目(J50104)资助。

摘　　要：为了解决移动机器人在目标识别过程中捕获的图像存在多目标干扰和单一视角下目标特征有限,进而导致识别准确率低的问题,提出一种基于分组特征赋权的动态视角图像特征融合方法,该方法通过递进式K均值聚类,对多目标特征进行赋权分组,并且利用LSTM网络实现动态视角下连续图像特征的融合,从而达到提高目标识别的准确率的目的。验证结果表明,在Market-1501数据集上的首位识别率达到了93.80%,平均准确率达到了89.13%,具有较好的实验效果。In order to solve the problem of low recognition accuracy, the image captured by mobile robot in the process of target recognition has multi-target interference and the target feature is limited in a single perspective. In this paper, a method of feature fusion of dynamic visual angle images based on grouping feature weight weights is proposed. In this method, multiple target features are weighted and grouped by progressive K-means clustering, and continuous image features are fused from dynamic perspective by LSTM network, so as to improve the accuracy of target recognition. The verification results show that the first recognition rate on Market-1501 data set reaches 93.80%, and the average accuracy reaches 89.13%, with good experimental results.

关 键 词：分组特征赋权 特征融合 特征提取 目标识别 长短期神经网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]