Fine-Tuned Convolutional Neural Network Model for Rice Leaf Disease Prediction

Abstract

Rice stands as a crucial sustenance for the majority of the global population, underscoring its cultivation's significance within agriculture. Nonetheless, the vitality of rice crops is persistently imperilled by a multitude of leaf diseases, culminating in substantial declines in yield. Prompt and accurate diagnosis of these diseases holds paramount importance for effective disease management and safeguarding food security. In recent times, convolutional neural networks (CNNs), have been identified as remarkable models in achieving optimal results in image-processing tasks, including the identification of plant diseases and health diseases. The proposed work has been employed with a CNN model for rice leaf disease classification. The open-source data repository “Kaggle” has been used as the source of the rice leaf disease dataset collection. Further, the CNN model has been trained for the automatic recognition and classification of four different kinds of rice leave diseases namely bacterial blight, blast, brown spot, and tungro. To build this CNN model, fine-tuning has been performed by changing the number of epochs and learning rate along with optimizers. The results show that the model has achieved the highest accuracy of 93.33% at epoch 14, whereas, at epoch 75 it has been identified as 75.56%. Furthermore, when the epoch was again increased to 100, the accuracy started to increase and resulted in 81.53% at epoch 100. In future, the model can be trained with the augmented images to achieve optimal results in the increased number of epochs.

Topics & Concepts