Leaf Disease Detection using Transfer Learning with Logistic Regression

Arlea Bless L. Sabarre, Alexandra Nicole S. Navidad, Darwin S. Torbela, and Jetron J. Adtoon*

Computer Engineering Program, College of Engineering Education, University of Mindanao, Philippines

*Corresponding author: jetron.adtoon22@gmail.com 

Abstract

Many plant recognition systems are being cast off to address the serious threat of plant diseases and pathogens concerning the agricultural industry. Computer vision is a one-way method to advance the conventional way of dealing with plant disease detection. However, the result's efficiency using this method entirely depends on the leaf characteristics. Due to the popularity and successful implementation of deep learning, research directed from traditional feature-based methods to deep learning. However, training a new deep learning model from the beginning requires much data and is both costly and time-laborious. In this study, transfer learning, one efficient deep learning method, is applied to recognize and learn useful characteristics directly from the inputted data. Then, a logistic regression classifier for leaf disease classification was used. Four health conditions are addressed in this study, 25 per classification, with a total of 100 samples tested to evaluate the system's accuracy. The result showed 94% accuracy in detecting algal spots, Cercospora, leaf discoloration, and healthy leaves. 

Keywords

leaf, leaf disease, transfer learning, logistic regression, detection.

How to cite?