M2V-SnakeNet: Multi-View Learning with Frozen DINOv2 for Snake Species and Venom Classification
DOI:
https://doi.org/10.62411/faith.3048-3719-431Keywords:
Calibration, Computer vision, Deep learning, Metric-based meta-learning, Multi-view learning, Snake species classification, Venom-status classification, Vision foundation modelsAbstract
Accurate identification of snake species, particularly venom status, is important for emergency medical response and biodiversity monitoring. In practice, this task is challenging because visually similar species may differ in venom status, photographs may capture only part of the animal, and curated datasets are often limited in size. Existing image-based approaches commonly rely on fine-tuning deep networks on small regional datasets, while partial observation, probability calibration, and the relationship between species identity and venom status remain insufficiently addressed. This study proposes M2V-SnakeNet, which couples a frozen self-supervised vision foundation model, DINOv2, with presence-aware multi-view decomposition, cross-view attention fusion, and a prototypical metric-based head, together with a hierarchical venom head linked to the species distribution through a consistency constraint. The framework is designed to represent partial observations without fabricating absent views, while restricting task-specific training to 9.5 of 95.2 million parameters. On a balanced and leakage-audited dataset of seven Indian snake species comprising 1,776 images from a single geographic region, M2V-SnakeNet achieves 93.4% ± 1.5% species accuracy and 96.9% ± 1.4% venom accuracy under five-fold cross-validation. On the fixed held-out split, a five-seed ensemble reaches 97.4% species accuracy and 98.5% venom accuracy, with 99.3% recall for the venomous class. Ablation results identify the frozen DINOv2 encoder as the largest performance contributor, improving species accuracy by 17.4 percentage points over an equal-sized ImageNet-supervised backbone, while learned fusion and multi-view decomposition provide additional gains. External meta-pretraining on a large snake corpus does not improve performance in this setting. These findings indicate that frozen foundation features combined with lightweight task-specific adaptation can provide a competitive low-data approach to joint snake species and venom-status classification, while the limited species and geographic coverage remains an important constraint on generalization.
Downloads
References
J. M. Gutiérrez, J. J. Calvete, A. G. Habib, R. A. Harrison, D. J. Williams, and D. A. Warrell, “Snakebite envenoming,” Nat. Rev. Dis. Prim., vol. 3, no. 1, p. 17063, Sep. 2017, doi: 10.1038/nrdp.2017.63.
J. Aquino, I. L. E. Lima, F. R. Costa, and I. Nunes, “Trends in diagnostic characters of Snakes species (Squamata) in the first two decades of the 21st century,” Zootaxa, vol. 5689, no. 3, pp. 439–450, Sep. 2025, doi: 10.11646/zootaxa.5689.3.2.
A. Patel, L. Cheung, N. Khatod, I. Matijosaitiene, A. Arteaga, and J. W. Gilkey, “Revealing the Unknown: Real-Time Recognition of Galápagos Snake Species Using Deep Learning,” Animals, vol. 10, no. 5, p. 806, May 2020, doi: 10.3390/ani10050806.
M. Vasmatkar, I. Zare, P. Kumbla, S. Pimpalkar, and A. Sharma, “Snake Species Identification and Recognition,” in 2020 IEEE Bombay Section Signature Conference (IBSSC), Dec. 2020, pp. 1–5. doi: 10.1109/IBSSC51096.2020.9332218.
M. Rajabizadeh and M. Rezghi, “A comparative study on image-based snake identification using machine learning,” Sci. Rep., vol. 11, no. 1, p. 19142, Sep. 2021, doi: 10.1038/s41598-021-96031-1.
N. I. Progga, N. Rezoana, M. S. Hossain, R. U. Islam, and K. Andersson, “A CNN Based Model for Venomous and Non-venomous Snake Classification,” in Communications in Computer and Information Science, 2021, pp. 216–231. doi: 10.1007/978-3-030-82269-9_17.
P. Joshi, S. Sati, T. Choudhury, T. Bajaj, K. Kotecha, and A. Sar, “Classification of Venomous and Non-venomous Snakes Using Transfer Learning with MobileNetV2,” in Lecture Notes in Networks and Systems, 2024, pp. 427–438. doi: 10.1007/978-981-97-3810-6_35.
K. Ahmed, M. A. Gad, and A. E. Aboutabl, “Snake species classification using deep learning techniques,” Multimed. Tools Appl., vol. 83, no. 12, pp. 35117–35158, Sep. 2023, doi: 10.1007/s11042-023-16773-0.
M. Iguernane, M. Ouzziki, Y. Es-Saady, M. El Hajji, A. Lansari, and A. Bouazza, “Deep Learning-Based Snake Species Identification for Enhanced Snakebite Management,” AI, vol. 6, no. 2, p. 21, Jan. 2025, doi: 10.3390/ai6020021.
L. Picek, M. Hrúz, A. M. Durso, and I. Bolon, “Overview of SnakeCLEF 2022: Automated Snake Species Identification on a Global Scale,” in CLEF 2022: Conference and Labs of the Evaluation Forum, 2022. [Online]. Available: https://ceur-ws.org/Vol-3180/paper-156.pdf
L. Bloch and C. M. Friedrich, “EfficientNets and Vision Transformers for Snake Species Identification Using Image and Location Information,” in CLEF 2021– Conference and Labs of the Evaluation Forum, 2021. [Online]. Available: https://ceur-ws.org/Vol-2936/paper-126.pdf
R. Chamidullin, M. Šulc, J. Matas, and L. Picek, “A Deep Learning Method for Visual Recognition of Snake Species,” in CLEF 2021– Conference and Labs of the Evaluation Forum, 2021. [Online]. Available: https://ceur-ws.org/Vol-2936/paper-128.pdf
M. Palaniappan, K. Desingu, H. Bharathi, E. A. Chodisetty, and A. Bhaskar, “Deep Learning and Gradient Boosting Ensembles for Classification of Snake Species,” in CLEF 2022: Conference and Labs of the Evaluation Forum, 2022. [Online]. Available: https://ceur-ws.org/Vol-3180/paper-173.pdf
F. Hu et al., “Watch out Venomous Snake Species: A Solution to SnakeCLEF2023,” in Conference and Labs of the Evaluation Forum, Jul. 2023. [Online]. Available: http://arxiv.org/abs/2307.09748
K. Desingu, M. Palaniappan, and J. Kumar, “Snake Species Classification using Transfer Learning Technique,” in CLEF 2021 – Conference and Labs of the Evaluation Forum, 2021. [Online]. Available: https://ceur-ws.org/Vol-2936/paper-133.pdf
L. Kalinathan, P. Balasundaram, P. Ganesh, S. S. Bathala, and R. K. Mukesh, “Automatic Snake Classification using Deep Learning Algorithm,” in CLEF 2021 – Conference and Labs of the Evaluation Forum, 2021. [Online]. Available: https://ceur-ws.org/Vol-2936/paper-135.pdf
H. Wang et al., “Snake-DETR: a lightweight and efficient model for fine-grained snake detection in complex natural environments,” Sci. Rep., vol. 15, no. 1, p. 1282, Jan. 2025, doi: 10.1038/s41598-024-84328-w.
L. Li, R. Kang, W. Huang, and W. Feng, “A Study on Small-Scale Snake Image Classification Based on Improved SimCLR,” Appl. Sci., vol. 15, no. 11, p. 6290, Jun. 2025, doi: 10.3390/app15116290.
A. Miyaguchi, M. Gustineli, A. Fischer, and R. Lundqvist, “Transfer Learning with Self-Supervised Vision Transformers for Snake Identification,” in CLEF 2024: Conference and Labs of the Evaluation Forum, 2024. [Online]. Available: https://ceur-ws.org/Vol-3740/paper-203.pdf
A. M. Durso, G. K. Moorthy, S. P. Mohanty, I. Bolon, M. Salathé, and R. Ruiz de Castañeda, “Supervised Learning Computer Vision Benchmark for Snake Species Identification From Photographs: Implications for Herpetology and Global Health,” Front. Artif. Intell., vol. 4, Apr. 2021, doi: 10.3389/frai.2021.582110.
I. Bolon, L. Picek, A. M. Durso, G. Alcoba, F. Chappuis, and R. Ruiz de Castañeda, “An artificial intelligence model to identify snakes from across the world: Opportunities and challenges for global health and herpetology,” PLoS Negl. Trop. Dis., vol. 16, no. 8, p. e0010647, Aug. 2022, doi: 10.1371/journal.pntd.0010647.
L. Picek, R. Chamidullin, M. Hrúz, and A. M. Durso, “Overview of SnakeCLEF 2023: Snake Identification in Medically Important Scenarios,” in CLEF 2023: Conference and Labs of the Evaluation Forum, 2023. [Online]. Available: https://ceur-ws.org/Vol-3497/paper-168.pdf
F. M. Firnando, D. R. I. M. Setiadi, A. R. Muslikh, and S. W. Iriananda, “Analyzing InceptionV3 and InceptionResNetV2 with Data Augmentation for Rice Leaf Disease Classification,” J. Futur. Artif. Intell. Technol., vol. 1, no. 1, pp. 1–11, May 2024, doi: 10.62411/faith.2024-4.
iNaturalist, “Snakes (suborder Serpentes),” iNaturalist. https://www.inaturalist.org/taxa/85553-Serpentes (accessed Jan. 02, 2026).
K. Muhamada et al., “Exploring Machine Learning and Deep Learning Techniques for Occluded Face Recognition: A Comprehensive Survey and Comparative Analysis,” J. Futur. Artif. Intell. Technol., vol. 1, no. 2, pp. 160–173, Sep. 2024, doi: 10.62411/faith.2024-30.
M. Oquab et al., “DINOv2: Learning Robust Visual Features without Supervision,” Trans. Mach. Learn. Res., Feb. 2024, doi: 10.48550/arxiv.2304.07193.
A. Vaswani et al., “Attention Is All You Need,” in 31st Conference on Neural Information Processing Systems (NIPS 2017), 2017. [Online]. Available: http://arxiv.org/abs/1706.03762
J. Snell, K. Swersky, and R. S. Zemel, “Prototypical Networks for Few-shot Learning,” in NIPS’17: Proceedings of the 31st International Conference on Neural Information Processing Systems, Jun. 2017, pp. 4080–4090. [Online]. Available: http://arxiv.org/abs/1703.05175
I. Loshchilov and F. Hutter, “Decoupled Weight Decay Regularization,” ArXiv. Jan. 04, 2019. [Online]. Available: http://arxiv.org/abs/1711.05101
R. Wightman et al., “PyTorch Image Models (timm),” Zenodo. https://github.com/huggingface/pytorch-image-models
C. Guo, G. Pleiss, Y. Sun, and K. Q. Weinberger, “On Calibration of Modern Neural Networks,” in Proceedings of the 34th International Conference on Machine Learning, Aug. 2017, pp. 1321–1330. [Online]. Available: http://arxiv.org/abs/1706.04599
L. van der Maaten and G. Hinton, “Visualizing Data using t-SNE,” J. Mach. Learn. Res., vol. 9, no. 86, pp. 2579−2605, 2008, [Online]. Available: https://www.jmlr.org/papers/v9/vandermaaten08a.html
A. Dosovitskiy et al., “An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale,” in ICLR 2021, Jun. 2021. [Online]. Available: http://arxiv.org/abs/2010.11929
J. Deng, W. Dong, R. Socher, L.-J. Li, Kai Li, and Li Fei-Fei, “ImageNet: A large-scale hierarchical image database,” in 2009 IEEE Conference on Computer Vision and Pattern Recognition, Jun. 2009, pp. 248–255. doi: 10.1109/CVPR.2009.5206848.
Downloads
Published
How to Cite
Issue
Section
License
Copyright (c) 2026 Hamza Maqsood Khan, Wakeel Ahmad, Syed M. Adnan
This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.
