大理大学学报 ›› 2022, Vol. 7 ›› Issue (12): 8-14.

• 数学与计算机科学 • 上一篇    下一篇

基于迁移学习的垃圾图像分类模型的比较研究

  

  1. 1.大理大学数学与计算机学院,云南大理 6710032.大理大学工程实训中心,云南大理 671003

    3.大理大学农学与生物科学学院,云南大理 671003

  • 收稿日期:2022-04-21 出版日期:2022-12-15 发布日期:2022-12-15
  • 通讯作者: 杨邓奇,教授,博士,E-mail:dengqiyang@163.com。
  • 作者简介:牛镱潼,硕士研究生,主要从事图像识别研究。
  • 基金资助:
    国家自然科学基金项目(31960119)

Comparative Study of Garbage Image Classification Models Based on Transfer Learning

  1. 1. College of Mathematics and Computer Dali University Dali Yunnan 671003 China 2. Engineering Training Center Dali University Dali Yunnan 671003 China3.College of Agriculture and Biology Science Dali University Dali Yunnan 671003 China

     

  • Received:2022-04-21 Online:2022-12-15 Published:2022-12-15

摘要: 诸多深度神经网络模型被应用于垃圾图像进行自动识别与分类,并且取得了很好的效果。当前主流的深度神经网络包括基于注意力机制的神经网络和基于卷积的神经网络。现有关于垃圾分类的研究多以基于卷积的神经网络模型为主,而基于注意力机制的神经网络模型在垃圾分类方面尚未有尝试。这2个类型的深度神经网络在小规模的垃圾分类数据集上的表现哪个更好,是值得探索的问题。对5种具有代表性的模型进行系统的比较研究,实验表明,与基于卷积的神经网络模型相比,在小规模垃圾分类数据集上,纯注意力机制的深度神经网络模型表现出更好的性能,为垃圾分类模型的选择提供了参考。

关键词: 垃圾分类, 深度卷积神经网络, 迁移学习, 注意力机制

Abstract: Many deep neural network models have been applied to automatic identification and classification of garbage images and have achieved good results. Current mainstream deep neural networks include attention mechanism-based neural networks and convolution-based neural networks. The existing research on garbage classification is mainly based on convolution-based neural network model, while the attention mechanism-based neural network model has not been tried in garbage classification. Which of these two types of deep neural networks performs better on small-scale garbage classification data sets is worth exploring. A systematic comparative study of several representative models shows that compared with convolution-based neural network model, the deep neural network model with pure attention mechanism on small-scale garbage classification data sets shows better performance, which provides a reference for garbage classification model selection. 

Key words: garbage classification, deep convolution neural network, transfer learning, attention mechanism

中图分类号: