Using pretrained models in ensemble learning for date fruits multiclass classification.

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Accurate classification of date fruits is vital for quality control, automated sorting, and commercial applications due to their diverse characteristics.
• Deep learning techniques have demonstrated significant advancements in image classification tasks.

Purpose of the Study:

• To classify nine distinct date fruit varieties using four prominent convolutional neural networks (CNNs) and an ensemble learning approach.
• To evaluate the performance of a proposed Dirichlet Ensemble method against individual CNN models.

Main Methods:

• Utilized DenseNet121, MobileNetV2, ResNet18, and VGG16 for date fruit image classification.
• Implemented a Dirichlet Ensemble method that aggregates predictions from individual CNN models.
• Assessed model performance using accuracy, precision, recall, and F1-score metrics.

Main Results:

• The Dirichlet Ensemble achieved superior performance with an accuracy of 98.61%, precision of 98.71%, recall of 98.61%, and an F1-score of 98.62%.
• DenseNet121 (96.92% accuracy) and MobileNetV2 (95.83% accuracy) showed strong performance as standalone models, suitable for systems with limited computing power.
• ResNet18 achieved 92.35% accuracy, outperforming VGG16 (73.24% accuracy), which struggled with complex classification.

Conclusions:

• Ensemble learning, specifically the Dirichlet Ensemble, effectively enhances the accuracy and robustness of date fruit classification.
• The study highlights the potential of deep learning and ensemble methods for agricultural product classification.
• Future research should explore advanced ensemble strategies and fine-tuning for improved food classification models.