Evaluating DINOv2 Architecture for Radish Plant Disease Severity Classification

Öz

Early detection of leaf diseases is crucial to prevent yield loss and support sustainable agriculture. However, traditional visual inspections are time-consuming, subjective, and difficult to implement across large agricultural fields. To address these limitations, this study proposes a deep learning-based methodology for the automatic classification of leaf disease severity in radish plants utilizing high-resolution Unmanned Aerial Vehicle (UAV) imagery. Initially, the K-means clustering algorithm was applied to segment the radish plants from complex field backgrounds, specifically isolating them from bare ground and mulching film. The extracted plant regions were subsequently categorized into three severity levels: healthy, mild disease, and severe disease. For the classification task, DINOv2 (Distillation with No Labels v2), a state-of-the-art self-supervised Vision Transformer (ViT) architecture recognized for its robust global feature extraction capabilities, was employed. The pre-trained DINOv2 model was fine-tuned specifically on the augmented radish dataset. The experimental results demonstrate the high efficacy of the proposed approach. The model achieved an overall accuracy of 92.7%, alongside a macro-averaged precision of 92.78%, a recall of 92.70%, and an F1-score of 92.72% on the unseen test dataset. Based on the confusion matrix analysis, the model successfully distinguished between 'healthy' and 'severe disease' instances with high precision. Although the transitional 'mild disease' class presented visual similarities, the model maintained stable predictive performance. These findings indicate that the integration of UAV imagery with advanced ViT models such as DINOv2 provides a practical and highly accurate solution for monitoring agricultural leaf diseases. Ultimately, this approach can facilitate rapid decision-making and optimize resource management in precision agriculture applications.

Anahtar Kelimeler

• Precision Agriculture
• Unmanned Aerial Vehicle (UAV)
• Deep Learning
• DINOv2
• Plant Disease Detection

Turp Bitkilerinde Yaprak Hastalığı Şiddetinin Sınıflandırılması İçin DINOv2 Mimarisinin Değerlendirilmesi

Öz

Yaprak hastalıklarının erken tespit edilmesi, verim kaybını önlemek ve sürdürülebilir tarımı desteklemek için oldukça önemlidir. Ancak, gözleme dayalı geleneksel yöntemler zaman alıcı, öznel ve büyük tarım arazilerinde uygulanması zor bir yapıdadır. Bu kısıtlamalara bir çözüm olarak bu çalışmada, insansız hava aracı ile alınan yüksek çözünürlüklü görüntüler kullanılarak turp bitkilerindeki yaprak hastalığı şiddetini otomatik olarak sınıflandıran derin öğrenme tabanlı bir metodoloji önerilmektedir. İlk olarak, turp bitkilerini çıplak toprak ve malç filmi gibi karmaşık tarla arka planından ayırmak için K-means kümeleme algoritması uygulanmıştır. Ardından, elde edilen bitki bölgeleri üç farklı şiddet seviyesine ayrılmıştır: sağlıklı, hafif hastalıklı ve şiddetli hastalıklı. Sınıflandırma görevi için, güçlü küresel öznitelik çıkarma yeteneğiyle bilinen, güncel bir öz-denetimli görme transformatörü olan DINOv2 mimarisi kullanılmıştır. Önceden eğitilmiş DINOv2 modeli, veri artırımı uygulanmış turp veri seti üzerinde özel olarak ince ayarla yeniden eğitilmiştir. Deneysel sonuçlar, önerilen yaklaşımın yüksek düzeyde etkili olduğunu göstermektedir. Model, daha önce görülmemiş test verisi üzerinde %92,7 oranında bir genel doğruluğun yanı sıra %92,78 kesinlik, %92,70 duyarlılık ve %92,72 F1-skoru elde etmiştir. Karışıklık matrisi analizine göre model, 'sağlıklı' ve 'şiddetli hastalıklı' vakaları yüksek bir hassasiyetle birbirinden ayırmayı başarmıştır. Geçiş aşaması olan 'hafif hastalıklı' sınıf görsel benzerlikler sunsa da model kararlı bir tahmin performansı sergilemeye devam etmiştir. Bu bulgular, İHA görüntülerinin DINOv2 gibi gelişmiş ViT modelleriyle entegrasyonunun, tarımsal yaprak hastalıklarının izlenmesinde pratik ve son derece doğru bir çözüm sunduğunu ortaya koymaktadır. Sonuç olarak bu yaklaşım, hassas tarım uygulamalarında karar alma süreçlerini hızlandırabilir ve kaynak yönetimini optimize edebilir.

Anahtar Kelimeler

• Hassas Tarım
• İnsansız Hava Aracı (İHA)
• Derin Öğrenme
• DINOv2
• Bitki Hastalığı Tespiti

Kaynakça

1. Savary, S., Willocquet, L., Pethybridge, S. J., Esker, P., McRoberts, N., & Nelson, A. (2019). The global burden of pathogens and pests on major food crops. Nature Ecology & Evolution, 3(3), 430-439.
2. Oerke, E. C. (2006). Crop losses to pests. The Journal of Agricultural Science, 144(1), 31-43.
3. Mace, M. E. (2012). Fungal wilt diseases of plants. New York: Elsevier Academic Press.
4. Dang, L. M., Hassan, S. I., Im, S., Sangaiah, A. K., Mehmood, I., Rho, S., Seo, S., & Moon, H. (2017). UAV based wilt detection system via convolutional neural networks. Sustainable Computing: Informatics and Systems, 16, 33–41.
5. Zhang, C., & Kovacs, J. M. (2012). The application of small unmanned aerial systems for precision agriculture: A review. Precision Agriculture, 13(6), 693-712.
6. Radoglou-Grammatikis, P., Sarigiannidis, P., Lagkas, T., & Moscholios, I. (2020). A compilation of UAV applications for precision agriculture. Computer Networks, 172, 107148.
7. Tsouros, D. C., Bibi, S., & Macris, P. G. (2019). A review on UAV-based applications for precision agriculture. Information, 10(11), 349.
8. Maes, W. H., & Steppe, K. (2019). Perspectives for remote sensing with unmanned aerial vehicles in precision agriculture. Trends in Plant Science, 24(2), 152-164.

9. Hamuda, E., Glavin, M., & Jones, E. (2016). A survey of image processing techniques for plant extraction and segmentation in the field. Computers and Electronics in Agriculture, 125, 184-199.
10. Barbedo, J. G. A. (2019). A review on the use of unmanned aerial vehicles and imaging sensors for monitoring and assessing plant stresses. Drones, 3(2), 40.
11. Wang, Z., Wang, K., Pan, S., & Han, Y. (2018). Segmentation of crop disease images with an improved K-means clustering algorithm. Applied Engineering in Agriculture, 34(2), 277-289.
12. Kerkech, M., Hafiane, A., & Canals, R. (2018). Deep leaning approach with colorimetric spaces and vegetation indices for vine diseases detection in UAV images. Computers and Electronics in Agriculture, 155, 237-243.
13. Tetila, E. C., Machado, B. B., Astolfi, G., Belete, N. A. d. S., Amorim, W. P., Roel, A. R., & Pistori, H. (2020). Detection and classification of soybean pests using deep learning with UAV images. Computers and Electronics in Agriculture, 179, 105836.
14. LeCun, Y., Bengio, Y., & Hinton, G. (2015). Deep learning. Nature, 521(7553), 436-444.
15. Kamilaris, A., & Prenafeta-Boldú, F. X. (2018). Deep learning in agriculture: A survey. Computers and Electronics in Agriculture, 147, 70-90.
16. Caron, M., Touvron, H., Misra, I., Jégou, H., Mairal, J., Bojanowski, P., & Joulin, A. (2021). Emerging properties in self-supervised vision transformers. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), (pp. 9650-9660), 11-17 October 2021, Montreal, QC, Canada.
17. Reedha, R., Dericquebourg, E., Canals, R., & Hafiane, A. (2022). Transformer neural network for weed and crop classification of high resolution UAV images. Remote Sensing, 14(3), 592.
18. Oquab, M., Darcet, T., Moutakanni, T., Vo, H., Szafraniec, M., Khalidov, V., … Bojanowski, P. (2023). DINOv2: Learning robust visual features without supervision. arXiv preprint.
19. Espejo-Garcia, B., Güldenring, R., Nalpantidis, L., & Fountas, S. (2025). Foundation vision models in agriculture: DINOv2, LoRA and knowledge distillation for disease and weed identification. Computers and Electronics in Agriculture, 239, 110900. https://doi.org/10.1016/j.compag.2025.110900
20. Bai, C., Zhang, L., Gao, L., Peng, L., Li, P., & Yang, L. (2024). DINOV2-FCS: a model for fruit leaf disease classification and severity prediction. Frontiers in Plant Science, 15, 1475282. https://doi.org/10.3389/fpls.2024.1475282
21. Bhagat, S., Kokare, M., Haswani, V., Hambarde, P., Taori, T., Ghante, P. H., & Patil, D. K. (2024). Advancing real-time plant disease detection: A lightweight deep learning approach and novel dataset for pigeon pea crop. Smart Agricultural Technology, 7, 100408.
22. Resource-Efficient ViT Team. (2025). Resource-efficient few-shot plant disease classification via quantized low-rank adapters in Vision Transformers. ResearchGate Preprints. https://doi.org/10.21203/rs.3.rs-9014055/v1
23. Doherty, K., Gurinas, M., Samsoe, E., Casper, C., Larkin, B., Ramsey, P., ... & Salakhutdinov, R. (2025). Ground-truthed and high-resolution drone images of the leafy spurge weed plant (Euphorbia esula). Scientific Data, 12(1), 746.
24. Barbedo, J. G. A. (2018). Impact of dataset size and variety on the effectiveness of deep learning and transfer learning for plant disease classification. Computers and Electronics in Agriculture, 153, 46-53.
25. Shorten, C., & Khoshgoftaar, T. M. (2019). A survey on image data augmentation for deep learning. Journal of Big Data, 6(60), 1–48.