Paddy Disease Classification Study: A Deep Convolutional Neural Network Approach

Abstract

One of the most prominent research topics in agriculture relies on the accurate detection of plant diseases for the early prevention of productivity loss. However, most of the strategies that prevent plant diseases lie in the use of chemical substances, which can affect the plant population and be harmful to humans. Under such circumstances, artificial intelligence techniques can provide a powerful tool for early diagnosis without considering the secondary effects of chemical substances. Deep Neural Networks (DNN) have been extensively used in agricultural engineering to provide accurate identification models for preventing plant diseases without considering harmful effects. In this study, paddy disease identification models have been presented considering Convolutional Neural Networks (CNNs) methodologies. Five different classical CNN models, namely Inception-V3, VGG-16, Alex Net, MobileNet V2, and ResNet-18, have been employed over a dataset of 7096 paddy leaves images to compare their performance. The dataset considered in the study consists of five classes of leaves: (a) Healthy leaves, (b) Bacterial Leaf Blight affected leaves, (c) Brown Spot affected leaves, (d) Leaf Blast affected leaves, and (e) Leaf Smut affected leaves. The experimental study indicates that Inception-V3 obtains better results over other tested CNN models in terms of accuracy, which is 96.23%.