Abstract
This study compares multiple deep learning architectures for the automated, image-based classification of five locally cultivated black table olive varieties in Turkey: Gemlik, Ayvalik, Uslu, Erkence, and Celebi. Using a dataset of 2500 images, ten architectures - MobileNetV2, EfficientNetB0, EfficientNetV2-S, ResNet50, ResNet101, DenseNet121, InceptionV3, ConvNeXt-Tiny, ViT-B16, and Swin-T - were trained using transfer learning. Model performance was evaluated using accuracy, precision, recall, F1-score, Matthews Correlation Coefficient (MCC), Cohen's Kappa, ROC-AUC, number of parameters, FLOPs, inference time, and generalization gap. EfficientNetV2-S achieved the highest classification accuracy (95.8%), while EfficientNetB0 provided the best trade-off between accuracy and computational complexity. Overall, the results indicate that under limited data conditions, parametric efficiency plays a more critical role than model depth alone.
Abstract (translated)
这项研究比较了多种深度学习架构，用于在土耳其自动分类五种当地栽培的黑橄榄品种：Gemlik、Ayvalik、Uslu、Erkence和Celebi。使用包含2500张图像的数据集，十种架构——MobileNetV2、EfficientNetB0、EfficientNetV2-S、ResNet50、ResNet101、DenseNet121、InceptionV3、ConvNeXt-Tiny、ViT-B16和Swin-T，在迁移学习的背景下进行了训练。模型性能通过准确性、精确性、召回率、F1分数、马修斯相关系数（MCC）、卡帕值、ROC-AUC指标、参数数量、浮点运算数（FLOPs）、推理时间和泛化差距进行评估。EfficientNetV2-S实现了最高的分类准确率（95.8%），而EfficientNetB0在准确性与计算复杂度之间提供了最佳的平衡。总体而言，研究结果表明，在数据受限的情况下，参数效率比模型深度本身扮演着更重要的角色。
URL
https://arxiv.org/abs/2602.18530
