Shiddiq Saiful Aziz, Siti Kania Kushadiani, Dini Rahmawati, Alya Dwinanda, Sri Rahayu
Hoya species identification is often challenging due to morphological similarities, particularly in leaf venation patterns and floral structures. This study aims to evaluate the performance of three lightweight convolutional neural network (CNN) architectures - MobileNetV2, EfficientNetB0, and MobileNetV3 - in classifying 29 flower species and 26 leaf species of Hoya. A dataset of 1,116 images was collected and augmented using geometric and photometric transformations. Transfer learning from ImageNet-pretrained weights was applied, and models were evaluated using accuracy, precision, recall, and F1-score metrics, supported by confusion matrix analysis. MobileNetV3 achieved the highest performance, with accuracies of 89.19% for flowers and 87.79% for leaves, followed by EfficientNetB0, while MobileNetV2 yielded the lowest scores. Misclassifications mainly occurred among morphologically similar species. These results demonstrate that MobileNetV3 is well-suited for real-time Hoya identification in resource-constrained environments, and future research may integrate additional feature modalities or ensemble approaches to improve performance. © 2025 IEEE.
Politeknik Negeri Bandung, Department of Electrical Engineering, Bandung Barat, Indonesia; National Research and Innovation Agency, Research Center for Data and Information Sciences, Jakarta, Indonesia; Universitas Pendidikan Indonesia, Department of Electrical Engineering, Bandung, Indonesia; National Research and Innovation Agency, Research Center for Applied Botany, Jakarta, Indonesia
Research at a Glance
Register to unlock
Topics & SDG Alignment
Register to unlock
Collaboration
Register to unlock
Author Profile (Selected)
Register to unlock
References Overview
Register to unlock
Journal & Source
Register to unlock
Metadata & Integrity
Register to unlock