Leveraging Ensemble Deep Learning Model for Fruit Image Recognition with Explainable AI

Precise classification of fruit images is crucial for a range of agricultural and retail purposes. Although there have been improvements, current methods have limits when it comes to dealing with complex feature extraction and ensuring model resilience. This paper introduces a new framework for categorizing fruits using a combination of advanced deep learning models, namely AlexNet, ResNet50, and EfficientNet. The collection of 43,912 photos across 44 fruit classifications underwent thorough pre-processing and data augmentation procedures. Grad-CAM, SHAP, and LIME were employed to achieve model interpretability, offering valuable insights into the decision-making process of the ensemble model. The experimental results indicate that the ensemble model surpasses the performance of individual models, obtaining $96.9 \%$ accuracy, $\mathbf{9 6. 2 \%}$ precision, $94.5 \%$ recall, and $95.4 \%$ F1-score. The proposed ensemble model provides a highly efficient solution for automated fruit classification, with substantial implications for agricultural technology and retail automation.