Rice Grain Classification Using Attention-Based Ensemble Learning Model with Explainable AI

Rice plays a crucial role in global food security, serving as a primary food source for over 50% of the world's population. Accurate classification of rice grain varieties is essential for optimizing agricultural output and quality. Traditional methods rely on visual examination, which can be subjective and time- consuming. This study introduces an ensemble deep learning model that combines attention parallel filters and plain attention CNN models to enhance rice grain classification. The dataset is sourced from the Kaggle repository which includes 75,000 images across five rice varieties. The ensemble model integrates squeeze-and-excitation attention layers for improved feature extraction and classification accuracy. The proposed ensemble model outperformed individual models, achieving an accuracy of 99.81 %, recall of 99.82%, and 99.85% for both the precision and f1-score. The implementation of Grad-CAM visualization confirmed the model's interpretability and focus on relative image regions. The proposed framework demonstrates significant improvements in rice grain classification, offering a reliable solution for agricultural applications.