Rice Plant Disease Detection System Using Transfer Learning with MobilenetV3Large

Abstract

In this study, we address that foliar diseases of rice (Oryza sativa L.) pose a serious threat to agricultural productivity and propose an effective method for disease detection using Convolutional Neural Network (CNN). We use transfer learning on the MobilenetV3Large model to improve the model's performance. Our study involves a curated dataset containing images of infected rice leaves, followed by a careful preprocessing step. This dataset is then used to train a CNN model. The results show a commendable accuracy rate of over 90% and almost reaching 95% when the model is trained over 200 epochs. The model performance graph shows a consistent upward trend in accuracy coupled with decreasing loss during the training process. Furthermore, the classification results highlight the ability of the model to discriminate between different types of diseases affecting rice leaves. This study demonstrates the effectiveness of our proposed method and positions it as a valuable tool for leaf disease detection in rice. By providing faster and more accurate control measures, our approach has the potential to significantly improve agricultural productivity. The successful application of the CNN model using MobilenetV3Large highlights its adaptability and robust performance in addressing the pressing problem of rice leaf diseases and provides a promising path for future advances in precision agriculture.