AI-Driven Date Fruit Classification via Transfer Learning in Smart Agriculture

Abstract

Digital technologies, including the Internet of Things (IoT) and deep learning, are increasingly propelling smart agriculture. A vital aspect is the automated classification of objects for crop assessment and quality control. This study addresses the practical challenge of limited labeled data by investigating the efficacy of Transfer Learning (TL) for multi-grade classification of date fruit varieties. We conduct a rigorous com- parative analysis using three popular pre-trained Convolutional Neural Network (CNN) architectures— VGG16, ResNet50, and Inception V3—benchmarked against traditional CNN methodologies. The exper- imental setup utilizes the TU-DG dataset, which comprises 3,383 images and is partitioned into a 70% training, 20% validation, and 10% testing split. Models are fine-tuned using the Adam optimizer and eval- uated based on accuracy, precision, F1-score, and recall. Our analysis demonstrates that TL significantly surpasses traditional methodologies, which achieved an accuracy of 98%. Specifically, the VGG16 and Inception V3 models achieved a test accuracy, precision, recall, and F1-score of 100%, while ResNet50 achieved 99.85% across all metrics. These results validate the TL approach’s ability to achieve robust re- sults and establish a new benchmark for automated date fruit grading.