EARLY DETECTION OF BASAL STEM ROT IN OIL PALMUSING A LIGHTWEIGHT VISION TRANSFORMER BASED ONKNOWLEDGE DISTILLATION FOR EDGE DEVICE IMPLEMENTATION
DOI:
https://doi.org/10.54314/k1j77s61Keywords:
basal stem rot, ganoderma, knowledge distillation, vision transformer, edge computingAbstract
Abstract: Basal stem rot (BSR) caused by Ganoderma boninense is the most destructive disease of oil palm in Southeast Asia, and its early symptoms are subtle enough to be missed during routine visual census. Deep learning models, particularly Vision Transformers (ViT), achieve high accuracy in plant disease recognition, but their size and computational cost prevent deployment on the low-cost devices available to field workers. This study proposes a lightweight transformer for three-class BSR detection (healthy, early BSR, advanced BSR) obtained by distilling knowledge from a ViT-B/16 teacher into a DeiT-Tiny student, followed by INT8 post-training quantization and deployment on a Raspberry Pi 4 and an Android smartphone. A field dataset of 3,600 smartphone images from 1,200 palms was collected and labelled by agronomists, and data were split at the tree level to prevent leakage. The distilled student achieved an accuracy of 94.63%, a macro-F1 of 94.66%, and an early-stage recall of 90.48%, improving on the same student trained without distillation by 3.70 percentage points in accuracy and 6.55 points in early-stage recall while using about 17 times fewer parameters than the teacher. After quantization, the model size fell to approximately 5.8 MB with an inference latency of 54 ms per image on the Raspberry Pi 4. These results show that knowledge distillation is an effective route for bringing transformer-level accuracy to on-site BSR screening in plantations.
Keyword: basal stem rot; edge computing; ganoderma; knowledge distillation; vision transformer.
Abstrak: Busuk pangkal batang (BPB) yang disebabkan oleh Ganoderma boninense merupakan penyakit paling merusak pada kelapa sawit di Asia Tenggara, dan gejala awalnya sangat samar sehingga sering terlewat dalam sensus visual rutin. Model deep learning, khususnya Vision Transformer (ViT), mencapai akurasi tinggi dalam pengenalan penyakit tanaman, tetapi ukuran dan beban komputasinya menghambat penerapan pada perangkat murah yang tersedia bagi pekerja lapangan. Penelitian ini mengusulkan transformer ringan untuk deteksi BPB tiga kelas (sehat, BPB awal, BPB lanjut) yang diperoleh melalui distilasi pengetahuan dari model guru ViT-B/16 ke model murid DeiT-Tiny, dilanjutkan kuantisasi INT8 dan implementasi pada Raspberry Pi 4 serta ponsel Android. Dataset lapangan berisi 3.600 citra ponsel dari 1.200 pohon dikumpulkan dan diberi label oleh ahli agronomi, dengan pembagian data pada tingkat pohon untuk mencegah kebocoran data. Model murid hasil distilasi mencapai akurasi 94,63%, macro-F1 94,66%, dan recall tahap awal 90,48%, lebih baik daripada model murid yang sama tanpa distilasi sebesar 3,70 poin pada akurasi dan 6,55 poin pada recall tahap awal, dengan jumlah parameter sekitar 17 kali lebih sedikit daripada model guru. Setelah kuantisasi, ukuran model turun menjadi sekitar 5,8 MB dengan latensi inferensi 54 ms per citra pada Raspberry Pi 4. Hasil ini menunjukkan bahwa distilasi pengetahuan efektif untuk menghadirkan akurasi setara transformer pada penapisan BPB langsung di kebun.
Kata kunci: busuk pangkal batang; ganoderma; distilasi pengetahuan; komputasi tepi; vision transformer.
References
Corley, R. H. V., & Tinker, P. B. (2023). The oil palm (5th ed.). Wiley Blackwell.
Dosovitskiy, A., Beyer, L., Kolesnikov, A., Weissenborn, D., Zhai, X., Unterthiner, T., Dehghani, M., Minderer, M., Heigold, G., Gelly, S., Uszkoreit, J., & Houlsby, N. (2025). An image is worth 16x16 words: Transformers for image recognition at scale. International Conference on Learning Representations (ICLR).
Ferentinos, K. P. (2024). Deep learning models for plant disease detection and diagnosis. Computers and Electronics in Agriculture, 145, 311–318.
Gou, J., Yu, B., Maybank, S. J., & Tao, D. (2025). Knowledge distillation: A survey. International Journal of Computer Vision, 129(6), 1789–1819.
Hinton, G., Vinyals, O., & Dean, J. (2022). Distilling the knowledge in a neural network. arXiv preprint arXiv:1503.02531.
Howard, A., Sandler, M., Chu, G., Chen, L.-C., Chen, B., Tan, M., Wang, W., Zhu, Y., Pang, R., Vasudevan, V., Le, Q. V., & Adam, H. (2023). Searching for MobileNetV3. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 1314–1324.
Jacob, B., Kligys, S., Chen, B., Zhu, M., Tang, M., Howard, A., Adam, H., & Kalenichenko, D. (2024). Quantization and training of neural networks for efficient integer-arithmetic-only inference. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2704–2713.
Kamilaris, A., & Prenafeta-Boldú, F. X. (2024). Deep learning in agriculture: A survey. Computers and Electronics in Agriculture, 147, 70–90.
Loshchilov, I., & Hutter, F. (2023). Decoupled weight decay regularization. International Conference on Learning Representations (ICLR).
McNemar, Q. (1947). Note on the sampling error of the difference between correlated proportions or percentages. Psychometrika, 12(2), 153–157.
Mehta, S., & Rastegari, M. (2022). MobileViT: Light-weight, general-purpose, and mobile-friendly vision transformer. International Conference on Learning Representations (ICLR).
Mohanty, S. P., Hughes, D. P., & Salathé, M. (2016). Using deep learning for image-based plant disease detection. Frontiers in Plant Science, 7, 1419.
Selvaraju, R. R., Cogswell, M., Das, A., Vedantam, R., Parikh, D., & Batra, D. (2022). Grad-CAM: Visual explanations from deep networks via gradient-based localization. Proceedings of the IEEE International Conference on Computer Vision (ICCV), 618–626.
Shorten, C., & Khoshgoftaar, T. M. (2023). A survey on image data augmentation for deep learning. Journal of Big Data, 6(1), 60.
Tan, M., & Le, Q. (2023). EfficientNet: Rethinking model scaling for convolutional neural networks. Proceedings of the 36th International Conference on Machine Learning (ICML), PMLR 97, 6105–6114.
Thakur, P. S., Khanna, P., Sheorey, T., & Ojha, A. (2022). Trends in vision-based machine learning techniques for plant disease identification: A systematic review. Expert Systems with Applications, 208, 118117.
Touvron, H., Cord, M., Douze, M., Massa, F., Sablayrolles, A., & Jégou, H. (2025). Training data-efficient image transformers & distillation through attention. Proceedings of the 38th International Conference on Machine Learning (ICML), PMLR 139, 10347–10357.
Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, Ł., & Polosukhin, I. (2022). Attention is all you need. Advances in Neural Information Processing Systems, 30.
Downloads
Published
Issue
Section
License
Copyright (c) 2026 Tika Christy, Muhammad Ardiansyah Sembiring, Mustika Fitri Larasati Sibuea

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.










