浙江农业学报 ›› 2022, Vol. 34 ›› Issue (10): 2230-2239.DOI: 10.3969/j.issn.1004-1524.2022.10.17

• 植物保护 • 上一篇    下一篇

基于颜色矩的土豆、玉米、苹果叶片病害异常检测

张梓婷1(), 韩金玉1, 张东辉1, 李晗1, 李铭源1, 邓志平2, 孙晓勇1,*()   

  1. 1.山东农业大学 信息科学与工程学院,山东 泰安 271000
    2.浙江省农业科学院 病毒学与生物技术研究所,浙江 杭州310021
  • 收稿日期:2021-08-07 出版日期:2022-10-25 发布日期:2022-10-26
  • 通讯作者: 孙晓勇
  • 作者简介:*孙晓勇,E-mail: johnsunx1@126.com
    张梓婷(1998—),女,安徽宿州人,硕士研究生,主要从事大数据与人工智能研究。E-mail: tingzhzi@163.com
  • 基金资助:
    国家自然科学基金(32070684)

Anomaly detection of potato, maize and apple leaf diseases based on color moments

ZHANG Ziting1(), HAN Jinyu1, ZHANG Donghui1, LI Han1, LI Mingyuan1, DENG Zhiping2, SUN Xiaoyong1,*()   

  1. 1. College of Information Science and Engineering, Shandong Agricultural University, Tai’an 271000, Shandong, China
    2. Institute of Virology and Biotechnology, Zhejiang Academy of Agricultural Sciences, Hangzhou 310021, China
  • Received:2021-08-07 Online:2022-10-25 Published:2022-10-26
  • Contact: SUN Xiaoyong

摘要:

农作物病害是影响粮食产量的重要因素之一。目前,大部分研究以已知病害作为数据来源,使用传统机器学习和深度学习方法进行病害识别与分类,这种模型构建方法需要大量的病害数据,而当新发病害出现时,很可能因为检测不到而错过最佳预警时间。为解决该问题,本文拟提出一种仅使用正常农作物叶片数据集作为训练数据便可检测出叶片病害异常的方法。具体地,本研究提出一种基于k-means++聚类与图像分块的农作物叶片病害异常检测方法,通过图像去噪、图像分割、图像截取等预处理操作后,提取图像的颜色矩特征,对训练集进行k-means++聚类,构建比对模型并设置阈值,从而确定测试集异常与否。试验使用的土豆、玉米与苹果数据集均下载于Kaggle网站。通过调整聚类数与分块数,在土豆、玉米和苹果数据集上,识别准确率分别达到了89%、95%、95%以上,并且在玉米和苹果两种数据集上的漏警率为0。

关键词: 叶片病害, 异常检测, 图像分块

Abstract:

Crop diseases are one of the important factors on crop yield. At present, most researches use traditional machine learning and deep learning methods to identify and classify known diseases. However, this kind of model requires a lot of disease data, and it is not suitable for detection of new diseases. To overcome this problem, this work aims to detect the abnormal leaf diseases by using only the healthy leaf data sets as the training data. Specifically, a method of anomaly detection of crop leaf diseases based on k-means++clustering and image blocking is proposed. After image denoising, segmentation, interception and other preprocessing operations, the color moment features of the image are extracted, k-means++clustering is performed on the training set, and the comparison models are constructed and the threshold is set to determine whether the test set is abnormal or not. The datasets of potato, maize and apple leaves used in this experiment are downloaded from Kaggle website. By adjusting the cluster number and block number, the accuracy rate of potato, maize and apple leaf on test set is higher than 89%, 95% and 95%, respectively, and the missing alarm rates on maize and apple are 0.

Key words: leaf diseases, anomaly detection, image blocking

中图分类号: