欢迎访问《中国农学通报》,

中国农学通报 ›› 2021, Vol. 37 ›› Issue (5): 88-95.doi: 10.11924/j.issn.1000-6850.casb2020-0051

所属专题: 玉米 烟草种植与生产

• 植物保护·农药 • 上一篇    下一篇

基于U-Net的玉米叶部病斑分割算法

刘永波(), 胡亮, 曹艳, 唐江云, 雷波()   

  1. 四川省农业科学院农业信息与农村经济研究所,成都 610011
  • 收稿日期:2020-04-28 修回日期:2020-10-16 出版日期:2021-02-15 发布日期:2021-02-25
  • 通讯作者: 雷波
  • 作者简介:刘永波,男,1988年出生,四川甘洛人,助理研究员,硕士,主要从事计算机视觉与农业信息技术研究。通信地址:610011 四川省成都市锦江区净居寺路20号附101号,E-mail: dylyb618@163.com
  • 基金资助:
    四川省科技计划项目“基于深度卷积神经网络的玉米病害智能识别与分级鉴定研究”(2018JY0631);四川省科技支撑计划“十三五畜禽育种战略研究与云服务平台建设”(2016NYZ0054);四川省软科学研究计划“旅游业对乡村农户生计韧性的扰动机理及防范对策研究:以四川省为例”(2020JDR0324)

Image Segmentation for Maize Leaf Disease Based on U-Net

Liu Yongbo(), Hu Liang, Cao Yan, Tang Jiangyun, Lei Bo()   

  1. Institute of Agricultural Information and Rural Economy, Sichuan Academy of Agricultural Sciences, Chengdu 610011
  • Received:2020-04-28 Revised:2020-10-16 Online:2021-02-15 Published:2021-02-25
  • Contact: Lei Bo

摘要:

本文旨在提出一种基于U-Net算法模型的玉米病程分级方法,实现对玉米常见4类叶部病害程度的快速、准确、客观分级。该方法以两组U-Net模型并行运算实现对玉米叶部病斑图像的语义分割任务。经测试图像分割试验中病斑分割MIoU值达到93.63%,叶片分割MIoU值达到96.33%,且运算速度均在1秒内完成。试验结果表明,该研究以手机拍照等方式采集数据源,不依赖专业仪器设备即可实现玉米病害快速分级,可取代以往以人工目测进行的病害识别方式,提高了病害分级的准确性和客观性。该模型与物联网设备结合运用,可实现玉米病害预警、降低病害影响、增产增收科技惠农的目标。

关键词: 玉米病害, 图像处理, 全卷积, U-Net, 病程分级

Abstract:

In this paper, a grading method of maize disease course based on U-Net algorithm model is proposed, which can quickly, accurately and objectively classify the degree of four kinds of common maize leaf diseases. In this method, the semantic segmentation of corn leaf disease spot image is realized by parallel operation of two groups of U-Net models. In the test image segmentation experiment, the MIoU value of disease spot segmentation is 93.63%, and the MIoU value of leaf segmentation is 96.33%, and the operation speed is completed in 1 second. The experimental results show that this study collects data sources by means of mobile phone photos, can achieve rapid grading of maize diseases without relying on professional instruments and equipment, and replace the previous method of disease identification by manual visual inspection. The accuracy and objectivity of disease grading are improved. Combined with the Internet of Things equipment, the model can achieve the goal of early warning of maize diseases, reducing the impact of diseases, increasing both production and income by science and technology benefiting farmers.

Key words: maize diseases, image processing, FCN, U-Net, course grading

中图分类号: