网络与信息安全学报 ›› 2021, Vol. 7 ›› Issue (3): 141-148.doi: 10.11959/j.issn.2096-109x.2021061

• 学术论文 • 上一篇    下一篇

基于多视图半监督集成学习的人体动作识别算法

陈圣楠1,2, 范新民1,2   

  1. 1 福建师范大学,福建 福州 350117
    2 网络安全与教育信息化福建省高校工程研究中心,福建 福州 350117
  • 修回日期:2021-04-18 出版日期:2021-06-15 发布日期:2021-06-01
  • 作者简介:陈圣楠(1991- ),女,福建福州人,福建师范大学助理研究员,主要研究方向为数据挖掘、网络安全应用
    范新民(1971- ),男,福建建瓯人,福建师范大学研究员,主要研究方向为网络安全应用、教育信息化、教育大数据
  • 基金资助:
    国家自然科学基金(71701019);福建省中青年教师教育科研项目(JAT200071)

Human action recognition method based on multi-view semi-supervised ensemble learning

Shengnan CHEN1,2, Xinmin FAN1,2   

  1. 1 Fujian Normal University, Fuzhou 350117, China
    2 Fujian Engineering Research Center, Network Security and Educational Informatization, Fuzhou 350117, China
  • Revised:2021-04-18 Online:2021-06-15 Published:2021-06-01
  • Supported by:
    The National Natural Science Foundation of China(71701019);Educational Research Projects of Young and Middle-aged Teachers in Fujian Province(JAT200071)

摘要:

移动设备上难以获取大量标签样本,而训练不足导致分类模型在人体动作识别上表现欠佳。针对这一问题,提出一种基于多视图半监督集成学习的人体动作识别算法。首先,利用两种内置传感器收集的数据构建两个特征视图,将两个视图和两种基分类器进行组合构建协同学习框架;然后,根据多分类任务重新定义置信度,结合主动学习思想在迭代过程中控制预测伪标签结果;使用LightGBM对扩充后的训练集进行学习。实验结果表明,算法的精确率、召回率和F1值较高,能稳定、准确地识别多种人体动作。

关键词: 人体动作识别, 半监督学习, 小样本, LightGBM

Abstract:

Mass labeled data are hard to get in mobile devices.Inadequate training leads to bad performance of classifiers in human action recognition.To tackle this problem, a multi-view semi-supervised ensemble learning method was proposed.First, data of two different inertial sensors was used to construct two feature views.Two feature views and two base classifiers were combined to construct co-training framework.Then, the confidence degree was redefined in multi-class task and was combined with active learning method to control predict pseudo-label result in each iteration.Finally, extended training data was used as input to train LightGBM.Experiments show that the method has good performance in precision rate, recall rate and F1 value, which can effectively detect different human action.

Key words: human action recognition, semi-supervised learning, few-shot, LightGBM

中图分类号: 

No Suggested Reading articles found!