Severity Net: A Comparative Analysis of Deep Learning Approaches for Rhizome Rot Disease Severity Recognition in Turmeric Plant

Abstract

In turmeric cultivation, the timely identification and precise recognition of disease severity play an essential role in maintaining the economic sustainability of this crop. This study addresses the critical need for accurate recognition of rhizome rot disease severity in turmeric cultivation, which is a pivotal factor in sustaining economic viability. The efficiency of deep learning algorithms, including Convolutional Neural Network (CNN), VGG19, and a hybrid CNN-SVM, is explored for disease severity recognition. A comprehensive dataset of 4591 images covering a range of disease severity from healthy plants to advanced rhizome rot stages (10% to 100%) was meticulously prepared for this study. This dataset was utilized to train the deep learning models to predict the disease severity accurately, and the performance of each model was assessed using evaluation metrics. Among the algorithms, CNN showed the most promising performance with an accuracy of 91.78%, while CNN-SVM also demonstrated competitive results. The VGG19 model, however, struggled with overfitting issues, resulting in lower accuracy. The results of this study demonstrate the potential of using deep learning approaches for reliable disease severity recognition, which can be further extended for other diseases and crops to develop an effective tool for agricultural disease management.