欢迎访问《草业学报》官方网站,今天是 分享到:

草业学报 ›› 2023, Vol. 32 ›› Issue (12): 104-114.DOI: 10.11686/cyxb2023060

• 研究论文 • 上一篇    

基于AlexNet的栽培苜蓿病害识别

李云昊1(), 李仲贤2, 伏帅1, 张忠雪1, 茆士琴1, 冯琦胜1, 梁天刚1(), 李彦忠1   

  1. 1.兰州大学草地农业科技学院,草地农业生态系统国家重点实验室,兰州大学农业农村部牧草创新重点实验室,兰州大学草地农业教育工程研究中心,甘肃 兰州 730020
    2.兰州大学网络安全与信息化办公室,甘肃 兰州 730000
  • 收稿日期:2023-03-02 修回日期:2023-05-26 出版日期:2023-12-20 发布日期:2023-10-18
  • 通讯作者: 梁天刚
  • 作者简介:Corresponding author. E-mail: tgliang@lzu.edu.cn
    李云昊(1996-),男,藏族,甘肃张掖人,在读硕士。E-mail: liyh21@lzu.edu.cn
  • 基金资助:
    中国工程院战略研究与咨询项目(2022-HZ-09);财政部和农业农村部:国家现代农业产业技术体系(CARS-34);甘肃省林业和草原局科技创新项目(kjcx2022010);兰州大学中央高校基本科研业务费专项资金(lzujbky-2022-sp13)

Identification of cultivated alfalfa diseases based on AlexNet

Yun-hao LI1(), Zhong-xian LI2, Shuai FU1, Zhong-xue ZHANG1, Shi-qin MAO1, Qi-sheng FENG1, Tian-gang LIANG1(), Yan-zhong LI1   

  1. 1.College of Pastoral Agriculture Science and Technology,Lanzhou University,State Key Laboratory of Grassland Agro-ecosystem,Key Laboratory of Grassland Livestock Industry Innovation,Ministry of Agriculture and Rural Affairs,Engineering Research Center of Grassland Industry,Ministry of Education,Lanzhou 730020,China
    2.Network Security and Informatization Office,Lanzhou University,Lanzhou 730000,China
  • Received:2023-03-02 Revised:2023-05-26 Online:2023-12-20 Published:2023-10-18
  • Contact: Tian-gang LIANG

摘要:

苜蓿病害的准确快速识别是栽培苜蓿草地病害防治的关键。苜蓿病害鉴别对专业知识和识别工具及检测环境要求较高,传统的苜蓿病害识别往往需要采用显微观察等手段对叶片病害部位进行镜检,存在时效性差、成本高,难以实现大范围多点位的快速识别等弊端。近年来在图像识别领域的计算机视觉和深度学习得到快速发展,为苜蓿病害智能化识别提供了新途径。本研究利用13种常见苜蓿病害图像数据集,基于改进的AlexNet深度学习卷积神经网络,经过300次迭代训练,构建了苜蓿病害识别模型,并对比分析了不同图像输入分辨率的苜蓿病害识别精度。结果表明:13种苜蓿病害最优模型识别总体精度达到72%,最优图像输入尺寸为512像素×512像素;剔除识别精度过低的苜蓿病害样本图片后,褐斑病、霜霉病、炭疽病、黑茎叶斑病和小光壳叶斑病5类苜蓿病害的识别总体精度提高到92%,最优输入尺寸为1200像素×1200像素。这2种模型均能够实现对苜蓿主要病害的快速识别,研究结果可以为苜蓿病害智能检测系统的研发提供图像识别方面的技术支持。

关键词: 苜蓿病害, AlexNet, 目标检测, 深度学习

Abstract:

Accurate and rapid identification of alfalfa diseases is the key to disease prevention and control in alfalfa grassland. The identification of alfalfa diseases requires a high degree of professional knowledge, identification tools, and a suitable detection environment.Traditional methods for identifying alfalfa diseases include microscopic observations and other means to inspect the diseased parts of the leaves to detect pathogen strains. This disadvantages of those methods are their poor timeliness, high cost, and inability to identify diseases rapidly at multiple locations on a large scale. In recent years, computer-aided methods and deep learning in the field of image recognition have developed rapidly, providing new methods for the intelligent identification of alfalfa diseases. In this study, an alfalfa disease identification model was constructed using image datasets of 13 common alfalfa diseases, the improved AlexNet deep learning convolutional neural network, and 300 iterations of training. The recognition accuracy of alfalfa diseases under different image input resolutions was compared and analyzed. The optimal model for identifying 13 types of alfalfa diseases achieved an overall accuracy of 72%, and the optimal size of the image input was 512 pixels×512 pixels. After removing the images of diseased alfalfa samples with low recognition accuracy, the overall recognition accuracy of five alfalfa diseases, namely brown spot disease, downy mildew disease, anthracnose, black stem and leaf spot disease, and little light lenticel spot disease was increased to 92%, and the optimal input image size was 1200 pixels×1200 pixels. These two models are suitable for the rapid identification of major alfalfa diseases. These results provide technical support for the development of intelligent detection systems for alfalfa diseases based on image recognition.

Key words: alfalfa disease, AlexNet, target detection, deep learning