Ensemble Ensemble of Deep Convolutional Neural Networks for Predicting Blast Disease Severity in Paddy

Ensemble of Deep Convolutional Neural Networks for Predicting Blast Disease Severity in Paddy

Abstract

Background Blast disease poses a significant threat to paddy crops worldwide, causing substantial yield losses under favorable environmental conditions. This study aims to develop an efficient method for predicting blast disease severity in paddy plants using convolutional neural network (CNN) models. Three popular CNN architectures, VGG16, ResNet50, and InceptionV3, are employed in this work. Initially, the ImageNet dataset is used to pre-train these models for a identification task and then fine-tuned on a dataset for blast disease severity classification based on the Percentage Disease Index (PDI) score. The final model is constructed as an ensemble of the three CNN networks, combining their outputs using a weighted averaging method.

Results The efficiency of the suggested ensemble model is demonstrated by experimental data, achieving impressive training accuracy, validation accuracy, and testing accuracy of 96.09%, 94.44%, and 88%, respectively, using a dataset of 18,865 labeled images.

Conclusions The findings highlight the ability of deep learning