An innovative voting ensemble learning approach for sorting and classifying date fruit varieties

Dates are among Algeria's most significant agricultural crops due to their considerable health and financial benefits. Moreover, they constitute an essential export commodity beyond the hydrocarbon sector. The current traditional methods for classifying and sorting dates are inefficient, time-consuming, and labor-intensive, resulting in a disparity between limited exports and high production levels. This study proposes an Ensemble Learning (EL) model that employs Transfer Learning (TL) techniques to address impediments and enhance date fruit categorization. We evaluate the performance of four classifiers: MobileNetV2, EfficientNet, DenseNet201, and EL soft voting classifier that uses these TL methods, work on a set of 1,619 images of 20 different varieties of Algerian dates. The dataset ranks among the largest benchmarks for varietal variety. The proposed hybrid model has outstanding performance, with a validation accuracy of 98.67% and a classification accuracy of 99.92%. It sets a novel standard in agricultural technology by surpassing all evaluated models in precision, recall, and F1-score. These findings illustrate the approach's capacity to entirely revolutionize date sorting and significantly enhance agricultural productivity and efficiency.