Identifying and localizing the disease spots of late blight on tomato leaves using deep convolutional neural networks
Journal
2019 ASABE Annual International Meeting
Date Issued
2019-01-01
Author(s)
Lin, Chu Ping
Cheng, Hsueh Hung
YAN-FU KUO
Huang, Jin Hsing
SHIH-FANG CHEN
Ke, Yan Ling
Lin, Chu Ping
Cheng, Hsueh Hung
YAN-FU KUO
Huang, Jin Hsing
SHIH-FANG CHEN
Abstract
© 2019 ASABE Annual International Meeting. All rights reserved. Tomato late blight is an infamous disease due to causing severe tomato yield loss. Phytophthora infestans, the causal pathogen of tomato late blight, could disseminate to all the cultivation regions in a suitable weather condition and destroy all the crop in weeks. In order to prevent severe disease spreading, early symptom identification of the disease is important to take actions for disease control. Late blight symptoms include from irregularly shaped water-soaked to brown necrotic lesions on plant leaves and stems. Conventionally, the identification of late blight deeply relies on the experience of tomato farmers. However, the symptoms of late blight might be confused with the atypical symptoms and lesions of some other diseases, confusing not only the well-experienced farmers but also the inexperienced plant pathologists. This study proposed to identify tomato late blight using leaf images and deep learning. A Navigator-teacher-scrutinizer network (NTS-Net) was developed to localize and identify the putative late blight lesions of tomato leaves. The developed NTS-Net model achieved a mean accuracy of 99.76% in diseased and healthy plant identification and also achieved a precision of 50% in lesions localization.
Subjects
Deep learning | Disease identification | Navigator-teacher-scrutinizer network | Tomato late blight
Type
conference paper