EfficientNetB0-based Rice Disease Classification: A Strategic Fine-Tuning Approach

Rice, a major stape food, is crucial for both sustenance and the living of millions of farmers. Detecting rice leaf diseases is vital to prevent widespread crop losses, safeguarding farmers’ incomes and global food supply. While current detection systems face accuracy and robustness challenges, deep learning models, particularly the EfficientNetB0, show promise. Despite their sensitivity to noise and variations in leaf appearance, ongoing research addresses these issues. This study introduces the use of EfficientNetB0 for accurate classification, incorporating a fine-tuning strategy in a brief five-epoch training period. This approach optimally balances model adaptation and prevents overfitting, leveraging the deep neural network’s ability to capture complex hierarchical features. The limited fine-tuning duration addresses potential risks while maximizing the model’s robust features. This unique amalgamation of advanced architecture and tailored training strategy aims to achieve efficient and accurate disease classification. The research signifies a step toward revolutionizing rice disease detection, emphasizing the importance of striking a balance between cutting-edge methodologies and practical considerations in the pursuit of agricultural sustainability. Before Fine Tuning the Training Accuracy was $\mathbf{9 8. 4 6 \%}$, Validation Accuracy was $\mathbf{9 9. 1 5 \%}$ and Test Accuracy was $\mathbf{9 8. 4 4 \%}$.After Fine Tuning there was a significant increase in the accuracy. The Training Accuracy came to be $\mathbf{9 9. 7 7 \%}$, Validation Accuracy is $\mathbf{9 9. 4 3 \%}$ and Test Accuracy: 99.12%