Performance analysis of various deep learning models for detecting rice diseases

Abstract

<p>A major portion of the world’s population relies on rice as a staple diet, hence rice is essential to maintaining food security worldwide. Unfortunately, rice crops are susceptible to a number of illnesses that, if detected and treated promptly, can result in significant output losses. Expert visual inspection is a time-consuming and arbitrary part of the conventional procedures for diagnosing diseases in rice. An effective method for automated illness diagnosis in agriculture has evolved in recent years: deep learning, a branch of artificial intelligence. The objective of this research is to compare AlexNet, DCNN, MobileNet, GoogleNet, VGG16, ResNet50 and Xception, these are various deep learning models in order to choose the one that would produce the highest levels of accuracy, precision recall, specificity, and F1-score for detecting rice diseases. In this study we train the model for nine different types of rice diseases named as Rice Blast (Pyricularia oryzae), Rice Sheath Blight (Rhizoctonia Solani), Bacterial Leaf Blight (Xanthomonas oryzae pv. oryzae), Tungro Disease, Rice Grassy Stunt Virus (RGSV), Rice Yellow Mottle Virus (RYMV), Bakanae Disease (Fusarium moniliforme), Brown Spot (Cochliobolus miyabeanus) and Rice Tungro Bacilliform Virus (RTBV) with 30,000 images. For this we used the secondary dataset for analyzing the performance of models. We trained the model for both normalized and non-normalized dataset. After comparing the various models we get the better result from ResNet50 model with accuracy of 97.50%.</p>