Click on “Download PDF” for the PDF version or on the title for the HTML version.
If you are not an ASABE member or if your employer has not arranged for access to the full-text, Click here for options.
Identifying Tomato Diseases from Leave Images Using Deep Convolutional Neural Networks
Published by the American Society of Agricultural and Biological Engineers, St. Joseph, Michigan www.asabe.org
Citation: 2021 ASABE Annual International Virtual Meeting 2100291.(doi:10.13031/aim.202100291)
Authors: Hsueh-Hung Cheng, Yu-Lun Dai, Chu-Ping Lin, Jin-Hsing Huang, Shih-Fang Chen, Yan-Fu Kuo
Keywords: Deep learning, disease identification, tomato disease
Abstract. Plant diseases could cause severe yield loss of tomatoes. Identifying the diseases is the first step to control them. Conventionally, experienced farmers or pathologists observe the lesions on tomato leaves to identify the diseases. However, tomato industry in Taiwan faces labor shortage. Experienced farmers or pathologists might not always be available. This study proposed to develop a service to identify eight categories of tomato diseases. The service comprised of two deep learning models. The first deep learning model was an anomaly detection model built to ensure that the input images submitted by users were legitimate. It consisted of a feature extractor – VGG-11 – and an isolation forest algorithm. The second deep learning model was a YOLOv4 model trained to identify eight categories of disease symptoms of infected tomato leaves. The eight categories of disease symptoms were: (1) specks – including bacterial spot (caused by Xanthomonas euvesicatoria, Xanthomonas vesicatoria, or Xanthomonas perforans) at early stage, early blight (caused by Alternaria solani) at early stage, target spot (caused by Corynespora cassiicola) at early stage, and gray leaf spot (caused by Stemphylium solani) (2) scorches & spots with yellow halos – including bacterial spot (caused by Xanthomonas euvesicatoria, Xanthomonas vesicatoria, or Xanthomonas perforans) at late stage (3) powdery mildew – including powdery mildew I (caused by Erysiphe orontii, Erysiphe cichoracearum, or Oidium sp.) (4) rings spots – including early blight (caused by Alternaria solani) at late stage and target spot (caused by Corynespora cassiicola) at late stage (5) yellow spots – including leaf mold (caused by Fulvia fulva, or Pseudocercospora atromarginalis) and powdery mildew II (caused by Leveillula sp.) (6) water-soaked lesions – including late blight (caused by Phytophthora infestans) (7) serpentine mines – including leaf miner (caused by Liriomyza bryoniae) (8) systemic disorders – including diseases caused by tomato yellow leaf curl virus or tomato chlorosis virus. The developed anomaly detection model achieved an accuracy of 98.0% in anomalous image classification. The developed YOLOv4 model achieved a mean average precision of 77.27% in detecting the eight categories of tomato disease symptoms.
(Download PDF) (Export to EndNotes)