| dc.description.abstract | Today, Deep Learning (DL) has transformed many major industries. Agriculture is one
field where DL scientists and researchers are working with farmers to help them utilize
the shrinking resources due to urbanization. However, plant disease, especially crop
plants, is a major threat to global food security. Many types of diseases directly affect
the quality of the fruits, grains, etc …, leading to a decrease in agricultural productivity.
The conventional method of identifying plant disease is by direct observation by naked
eyes. This process is unreliable and subjected to human errors. Several works on deep
learning techniques for leaf disease have been proposed. Most of them built their models
based on limited resolution images on convolutional neural networks (CNNs). In this
article, we want to focus on early disease recognition on plant leaves with small disease
blobs which can only be detected with higher resolution images. We rescale, re-align to
standardize all of our images. We further apply a contrast enhancement method to
improve visualization quality. Our dataset consists of real-life mango leaves collected in
Dong Thap Province, Vietnam. We trained deep learning models to classify three
common diseases such as Anthracnose, Gall Midge, and Powdery Mildew. To improve
our models, we also apply transfer learning for the pre-trained models for the well
known PlantVillage dataset. We also proposed a simpler framework that makes use of
traditional artificial neural networks (ANN) and feature selection (FS). This manuscipt
was accepted and under publishing phase for the book entitled “Handbook of Deep
Learning in Biomedical Engineering”, Elsevier. | en_US |
