<![CDATA[Rhizomes are widely used in Indonesia as cooking spices and herbal ingredients, yet their visual similarity often causes misidentification when recognition relies on manual inspection, leading to inconsistent product quality and economic losses. This study presents an automatic rhizome image classification system based on the MobileNetV3-Large architecture to distinguish eight Indonesian rhizome types (bangle, ginger, kencur, kunci, turmeric, galangal, temu ireng, and temulawak) from RGB images. The dataset is organised by class and processed with a pipeline that includes resizing to 224×224 pixels, image flipping and rotation, brightness adjustment, zoom, and normalisation, before being split into training, validation, and testing subsets with an 80:10:10 ratio. MobileNetV3-Large pretrained on ImageNet is used as a feature extractor with a three layer dense classification head and dropout regularisation, trained using the Adam optimiser with a learning rate of 0.0001 and a checkpoint mechanism to store the best validation model. The proposed model achieves 97.50% accuracy, 97.65% precision, 97.50% recall, and 97.51% f1-score on the test set, indicating stable and balanced performance across all rhizome classes despite their similarity. Compared with earlier rhizome classification approaches based on handcrafted features, which typically report lower accuracies on fewer classes, and with heavier VGG or ResNet backbones, this work provides, to the best of the authors’ knowledge, the first evaluation of MobileNetV3-Large for multi class rhizome classification and shows that it offers a practical and computationally efficient baseline for image based rhizome identification on resource constrained mobile or embedded devices.]]>
Copyrights © 2026
