欢迎访问《图学学报》 分享到:

图学学报 ›› 2022, Vol. 43 ›› Issue (4): 667-676.DOI: 10.11996/JG.j.2095-302X.2022040667

• 计算机图形学与虚拟现实 • 上一篇    下一篇

基于VD-MobileNet 网络的 WebAR生活垃圾分类信息可视化方法

  

  1. 1. 西南科技大学计算机科学与技术学院,四川 绵阳 621010;
    2. 四川轻化工大学计算机科学与工程学院,四川 自贡 643002
  • 出版日期:2022-08-31 发布日期:2022-08-15
  • 通讯作者: 吴亚东(1979),男,教授,博士。主要研究方向为可视化与可视分析、人机交互
  • 基金资助:
    四川省科技厅杰青项目(19JCQN0108);四川省重点研发计划项目(2020YFS0360,2020YFG0031)

WebAR garbage classification information visualization method based on VD-MobileNet network

  1. 1. School of Computer Science & Technology, Southwest University of Science and Technology, Mianyang Sichuan 621010, China;
    2. School of Computer Science & Engineering, Sichuan University of Science and Engineering, Zigong Sichuan 643002, China
  • Online:2022-08-31 Published:2022-08-15
  • Contact: WU Ya-dong (1979), professor, Ph.D. His main research interests cover visualization and visual analysis and human-computer interaction
  • Supported by:
    Sichuan Science and Technology Department Jieqing Project (19JCQN0108); Key Research and Development Project of Sichuan
    Province (2020YFS0360, 2020YFG0031)

摘要:

随着我国垃圾分类制度的加速推行,基于虚拟/增强现实技术的垃圾分类应用大量涌现。受识别设备平台及居民应用习惯等方面的影响,针对目前该类应用在便捷性、实用性上存在较大不足,提出了一种基于轻量化神经网络并融合移动增强现实及可视化技术的垃圾分类应用方案。首先,提出了基于深度学习的垃圾分类可变扩张卷积 VD-MobileNet 模型方法能够解决移动设备中存在的计算能力有限、网络庞大等问题,通过在 MobileNet 模型中引入空洞卷积增加感受野、扩大垃圾的特征信息以提升分类精度,引入 LeakyReLU 激活函数优化网络的表达能力;其次,将该模型与 WebAR 技术结合,设计了一款面向移动设备的轻量级垃圾分类信息可视化系统,该系统具备跨平台特性,实现了对分类信息的多元化可视呈现,提供了灵活的交互方式。实验及评估表明,该 VD-MobileNet 模型在垃圾分类数据集中分类效果良好,能够在参数量不变的前提下有效减少计算量,此外结合该模型所设计的 WebAR 应用系统可为用户的垃圾处理事务提供合理有效地协助。

关键词: 垃圾分类, 移动增强现实, MobileNet 模型, 可视化技术, 空洞卷积, WebAR

Abstract:

With the accelerated implementation of the garbage classification regulation in China, many applications for garbage classification based on virtual/augmented reality technologies have sprung up. Under the influence of the identification equipment platform and residents’ habits of using applications, there remain a number of shortcomings in convenience and practicability for this kind of application. A waste classification application scheme was proposed based on a lightweight neural network combined with mobile augmented reality and visualization technology. Firstly, the variable expansion convolution VD-MobileNet model method was proposed for garbage classification based on deep learning, which can solve the problems of limited computing capacity and a huge network of mobile devices. The receptive field was increased by introducing dilated convolution in the MobileNet model. The characteristic information of garbage could be expanded to enhance classification accuracy, and LeakyReLU activation function was
introduced to optimize the expression ability of the network. Secondly, the model was equipped with the WebAR technology, and a lightweight garbage classification information visualization system was designed for mobile devices. This system could operate cross different platforms, realize the diversified visual presentation of classified information, and enable flexible interactions. Experiments and evaluations show that the VD-MobileNet model could achieve excellent classification in the garbage classification data set and can effectively reduce the amount of calculation with constant parameters. In addition, the WebAR application system designed based on the model can provide users with reasonable and effective assistance in garbage disposal.

Key words:

中图分类号: