Multiclass classifier based on deep learning for detection of citrus disease using fluorescence imaging spectroscopy

Author(s)

Neves, Ruan F. O. ; Wetterich, Caio B. ; Sousa, Elaine P. M. ; Marcassa, Luis G.

• Document type: Journal article
• Source: Laser Physics; v. 33, n. 5, p. 9-pg., 2023-05-01.

Abstract

In this work, we have combined fluorescence imaging spectroscopy (FIS) and supervised learning methods to identify and discriminate between citrus canker, Huanglongbing, and other leaf symptoms. Our goal is to differentiate these diseases and nutrient conditions without prior eye assessment of symptoms. Five supervised learning methods were evaluated. Our results show that by combining FIS with a convolutional neural network (AlexNet), it is possible to identify the disease of a sample with up to 95% accuracy. An enormous gain of time and a substantial cost reduction were achieved by this approach compared to polymerase chain reaction-based methods. (AU)

• FAPESP's process: 11/22275-6 - Fluorescence Imaging using variable liquid crystal tunable filter for citrus diseases
• Grantee: Caio Bruno Wetterich
• Support Opportunities: Scholarships in Brazil - Doctorate
• FAPESP's process: 13/11023-1 - Use o fluorescence images for a comparative study of diseases in citrus samples from Florida and São Paulo
• Grantee: Caio Bruno Wetterich
• Support Opportunities: Scholarships abroad - Research Internship - Doctorate
• FAPESP's process: 10/16536-9 - Fluorescence Imaging Applied to Citrus Diseases
• Grantee: Luis Gustavo Marcassa
• Support Opportunities: Regular Research Grants