计算机与现代化 ›› 2022, Vol. 0 ›› Issue (04): 38-44.

• 图像处理 • 上一篇    下一篇

基于全局自注意力的小麦图像识别

  

  1. (四川大学电子信息学院,四川成都610065)
  • 出版日期:2022-05-07 发布日期:2022-05-07
  • 作者简介:何晨曦(1996—),男,河南信阳人,硕士研究生,研究方向:计算机视觉,E-mail: 1824413744@qq.com; 通信作者:王正勇(1969—),女,四川成都人,副教授,硕士生导师,博士,研究方向:图像处理与模式识别,通信与信息处理,计算机视觉,E-mail: wangzheny@scu.edu.cn; 卿粼波(1982—),男,四川成都人,副教授,博士生导师,博士,研究方向:多媒体通信与信息系统,人工智能与计算机视觉,嵌入式系统,E-mail: qing_lb@scu.edu.cn; 何小海(1964—),男,四川绵阳人,教授,博士生导师,博士,研究方向:图像处理,模式识别,图像通信,E-mail: nic5602@scu.edu.cn; 吴小强(1971—),男,四川成都人,高级工程师,硕士,研究方向:计算机应用与模式识别,E-mail: 2396480@qq.com。
  • 基金资助:
    国家自然科学基金资助项目(61871278)

Wheat Image Recognition Based on Global Self-attention

  1. (College of Electronics and Information Engineering, Sichuan University, Chengdu 610065, China) 
  • Online:2022-05-07 Published:2022-05-07

摘要: 在实际应用场景下,通过图像识别的方式来识别小麦的病虫害具有极大的挑战性。与以往纯粹基于卷积神经网络(Convolutional Neural Network, CNN)的方法相比,将小麦图像转换成一系列视觉语言,并从全局视角进行小麦识别的方法是更可行和实用的。运用Convolutional Visual Transformers(CVT)来解决小麦识别分为2个环节。首先,利用2分支CNN生成的2种特征图来实现注意选择性融合(Attentional Selective Fusion, ASF)。ASF通过融合多个特征和全局-局部注意力来获取有区别的信息,并投射成一系列的视觉语言。其次,受Transformers在自然语言处理方面的成功启发,用全局自注意力来建模这些视觉语言之间的关系。将CVT与经典分类网络LeNet-5、ResNet-18、VGG-16、EfficientNet对比,识别率有所提升,同时该方法具有良好的泛化能力。

关键词: 小麦识别, 全局-局部注意, Transformer, 全局自注意力

Abstract: In the actual application scenario, it is very challenging to identify wheat diseases and pests by image recognition. Compared with the previous methods based solely on convolutional neural network (CNN), the method of converting wheat images into a series of visual languages and recognizing wheat from a global perspective is more feasible and practical. The use of convolutional visual Transformers (CVT) to solve wheat recognition is divided into two links. First, two feature maps generated by two-branch CNN are used to realize attentional selective fusion (ASF). ASF obtains different information by fusing multiple features and global-local attention, and projects it into a series of visual languages. Secondly, inspired by the success of Transformers in natural language processing, global self-attention is used to model the relationship between these visual languages. Compared with classical classification networks LeNet-5, ResNet-18, VGG-16 and EfficientNet, CVT improves the recognition rate, and this method has good generalization ability.

Key words: wheat recognition, global-local attention, Transformer, global self-attention