主办单位:中国气象局沈阳大气环境研究所
国际刊号:ISSN 1673-503X
国内刊号:CN 21-1531/P

气象与环境学报 ›› 2020, Vol. 36 ›› Issue (6): 137-143.doi: 10.3969/j.issn.1673-503X.2020.06.017

• 快报 • 上一篇    下一篇

基于Faster R-CNN的野外环境中蝗虫快速识别

武英洁1(),房世波1,*(),PiotrChudzik2,SimonPearson3,BashirAl-Diri12,冯旭宇4,李云鹏4   

  1. 1. 中国气象科学研究院, 北京 100081
    2. The University of Lincoln, School of Computer Science, Lincoln LN6 7TS, UK
    3. The University of Lincoln, Lincoln Institute for Agri-Food Technology, Lincoln LN6 7TS, UK
    4. 内蒙古自治区生态与农业气象中心, 内蒙古 呼和浩特 021099
  • 收稿日期:2020-07-09 出版日期:2020-12-30 发布日期:2021-01-06
  • 通讯作者: 房世波 E-mail:17863803936@163.com;fangshibo@cma.gov.cn
  • 作者简介:武英洁, 女, 1997年生, 硕士研究生, 主要从事农业遥感、农业气象方面研究, E-mail:17863803936@163.com
  • 基金资助:
    国家自然科学基金项目(61661136005);国家重点研发计划项目(2018YFC1506500);基本科研业务费重点项目(2019Z010)

Rapid identification of locust on fields based on Faster R-CNN

Ying-jie WU1(),Shi-bo FANG1,*(),Chudzik Piotr2,Pearson Simon3,Al-Diri1 Bashir2,Xu-yu FENG4,Yun-peng LI4   

  1. 1. Chinese Academy of Meteorological Sciences, Beijing 100081, China
    2. The University of Lincoln, School of Computer Science, Lincoln LN6 7TS, UK
    3. The University of Lincoln, Lincoln Institute for Agri-Food Technology, Lincoln LN6 7TS, UK
    4. Inner Mongolia Ecology and Agrometeorology Center, Hohhot 021099, China
  • Received:2020-07-09 Online:2020-12-30 Published:2021-01-06
  • Contact: Shi-bo FANG E-mail:17863803936@163.com;fangshibo@cma.gov.cn

摘要:

蝗虫是常见的害虫之一,对农作物和生态系统具有很大的危害,采用常规的方法对蝗虫进行监测存在一定局限性,为了有效应用海量野外影像数据实现对蝗虫实时监测,本文建立了一种基于深度学习网络的蝗虫自动识别模型。利用手机模拟摄像头获取的内蒙古锡林浩特附近草原的280张蝗虫的RGB图像,采用深度学习算法中的Faster R-CNN(Faster Region-based Convolutional Neural Network)网络结构建立了蝗虫识别模型。经验证该模型的精确度为0.756,可以较准确地将蝗虫从野外复杂环境中识别出来,与以往同类研究相比,在识别结果和实用性方面均有较大的进步。该模型是建立蝗虫实时监测系统的基础,可以为蝗虫的防治提供辅助信息,同时该网络结构还可以应用于其他害虫的识别,具有较强的推广性,拓宽了深度学习算法的应用领域。

关键词: 蝗虫, 深度学习, 识别, Faster R-CNN

Abstract:

Locust is the stubborn pest insects which can damage crops and ecosystems.Traditional methods for monitoring locust have many disadvantages.To effectively apply massive field image data to achieve real-time monitoring of locusts, a locust automatic identification model based on a deep learning network was established in this study.Firstly, 280 locust RGB images photographed by the mobile phone camera in a complex field environment from the grasslands of Xilinhot, Inner Mongolia were obtained.Then the Faster R-CNN network structure which performs better in recognition was used.The accuracy of this model is 0.756.The model performs well on locust detection and outperforms the previous methods in the identify results and practicality.The model can accurately identify the locust from the complex environment on fields, which provide auxiliary information for the control of locusts.It is a basis for establishing a real-time monitoring system for monitoring locusts.At the same time, the network structure can also be applied to other pests and diseases' monitor.In addition, the model broadens the application field of deep learning algorithms.

Key words: Locust, Deep learning, Identification, Faster R-CNN

中图分类号: