湖北农业科学 ›› 2023, Vol. 62 ›› Issue (10): 212-217.doi: 10.14088/j.cnki.issn0439-8114.2023.10.036

• 信息工程 • 上一篇    下一篇

一种应用于云南省外侵物种识别的边缘计算模型

罗玲1a, 宋科1a, 王皓1a, 资彩飞2, 奉伟1a, 杜铭铭1a, 孙仲享1b, 曹志勇1a   

  1. 1a.云南农业大学, 大数据学院,昆明 650500;
    1b.云南农业大学, 植物保护学院,昆明 650500;
    2.云南高创人才服务有限公司,昆明 650221
  • 收稿日期:2023-04-14 发布日期:2023-11-14
  • 通讯作者: 曹志勇(1976-),男,云南曲靖人,教授,博士,主要从事农业环境信息处理、图像采集与处理研究,(电话)13708718797(电子信箱)czy@ynau.edu.cn。
  • 作者简介:罗玲(1999-),女,云南曲靖人,在读硕士研究生,研究方向为图像识别,(电话)15287954671(电子信箱)3139970477@qq.com。
  • 基金资助:
    国家重点研发计划项目(2021YFD1400200); 云南省重大科技专项计划项目(202302AE090020)

An edge computing model applied to the identification of invasive species in Yunnan Province

LUO Ling1a, SONG Ke1a, WANG Hao1a, ZI Cai-fei2, FENG Wei1a, DU Ming-ming1a, SUN Zhong-xiang1b, CAO Zhi-yong1a   

  1. 1a. College of Big Data, Kunming 650500, China;
    1b. College of Plant Protection, Yunnan Agricultural University, Kunming 650500, China;
    2. Yunnan Gaochuang Human Resource Service Co., Ltd., Kunming 650221,China
  • Received:2023-04-14 Published:2023-11-14

摘要: 基于MobileNet模型迁移对云南省4种主要外侵物种(鬼针草、喀西茄、水花生和紫茎泽兰)图像进行识别,将宽度倍率为1.0和1.4的MobileNet-v2模型分别应用在本研究数据集上进行试验,分析了MobileNet-v2网络模型识别不稳定的原因,通过增加通道注意力机制模块、更新激活函数和压缩网络层数对模型进行改进。结果表明,改进后的MobileNet-v2模型识别准确率达96.8%,模型参数量仅为 1 535 093。改进后的MobileNet-v2模型识别准确率高、模型参数量少,适合部署于边缘端,能更好地应用于云南省外侵物种防治领域。

关键词: 外侵物种, 边缘计算模型, MobileNet-v2模型, 云南省

Abstract: Based on the MobileNet model migration, four main invasive species (Bidens bipinnata, Cassia plant, water peanut, and Eupatorium adenophorum) images in Yunnan Province were identified. The MobileNet-v2 models with width ratios of 1.0 and 1.4 were applied to the dataset in this study for experiments. The reasons for the unstable recognition of the MobileNet-v2 network model were analyzed, and the model was improved by adding channel attention mechanism modules, updating activation functions, and compressing network layers. The results showed that the improved MobileNet-v2 model had a recognition accuracy of 96.8%, and the model parameter quantity was only 15 359 093. The improved MobileNet-v2 model had high recognition accuracy and fewer model parameters, making it suitable for deployment at the edge, and could be better applied to the field of invasive species control in Yunnan Province.

Key words: invasive species, edge computing model, MobileNet-v2 model, Yunnan Province

中图分类号: