The KLASIFIKASI PENYAKIT PADA TANAMAN DAUN SINGKONG MENGGUNAKAN VISION TRANSFORMER: KLASIFIKASI PENYAKIT PADA TANAMAN DAUN SINGKONG MENGGUNAKAN VISION TRANSFORMER

Rakan Nabila, Luthfi
Dwi Putro Wicaksono, Aditya

POSITIF
Vol 11 No 2 (2025): Positif : Jurnal Sistem dan Teknologi Informasi

This research develops a cassava leaf disease classification model using Vision Transformer (ViT) to identify four types of diseases and healthy leaves. With a dataset from Kaggle (3,000 images/class), the TinyViT model was tested through parameter variations to achieve optimal performance. Results showed that the combination of SGD, 50 epochs, and batch size 32 gave the highest validation accuracy (83.16%), outperforming Adam/AdamW. Despite overfitting (100% training accuracy), the model showed good generalization with 81% precision and recall. These findings confirm the potential of ViT in plant disease detection, while highlighting the need to address overfitting through further regularization. Future research can explore dataset expansion and fine-tuning for accuracy improvement.