ConvNeXt-Driven Multi-Class Animal Prediction Using Deep Feature Learning

Abstract

Automated identification of animal species from visual data is essential for biodiversity conservation, ecological monitoring, and intelligent wildlife management. This paper presents a deep learning framework for large scale multi-class animal species classification using transfer learning with the ConvNeXt-Tiny architecture. A pretrained ConvNeXt model, initialized with ImageNet weights, is fine-tuned on a publicly available dataset comprising 5,400 images across 90 distinct animal classes. The proposed system integrates modern convolutional design principles inspired by vision transformers with the efficiency and inductive biases of conventional CNNs, enabling robust hierarchical feature learning. Data augmentation and dropout regularization are employed to improve generalization and mitigate overfitting. Experimental results demonstrate strong classification performance, achieving an overall accuracy of 93%, with balanced macro precision, recall, and F1-score of 93%. Both quantitative and qualitative evaluations confirm the model’s effectiveness in handling visual variability, background complexity, and inter-class similarity. The proposed framework offers a scalable and reliable solution for automated animal species recognition in real-world ecological and conservation applications.

Introduction

The study focuses on automated multi-class animal species identification using deep learning to support biodiversity conservation, ecological monitoring, and wildlife management. With the proliferation of cameras, drones, and remote sensors, massive volumes of animal imagery are being generated, making manual analysis infeasible. Traditional computer vision methods based on handcrafted features often fail in real-world environments due to variations in lighting, pose, occlusion, and species diversity.

Key Points:

• Advancements in Deep Learning:

  • Convolutional Neural Networks (CNNs) like VGG, ResNet, and EfficientNet have greatly improved image classification accuracy.

  • Transfer learning allows models pretrained on large datasets (e.g., ImageNet) to adapt efficiently to ecological datasets with limited labeled data, improving generalization and reducing training time.

  • Modern architectures like ConvNeXt combine CNN efficiency with transformer-inspired design principles, enhancing feature learning and robustness for complex, fine-grained visual tasks.

• Proposed Framework:

  • The study employs ConvNeXt-Tiny with transfer learning for multi-class animal classification.

  • The network is fine-tuned on a dataset of 5,400 images across 90 animal species, covering a diverse range of morphologies, poses, and habitats.

  • A custom classification head with Global Average Pooling, fully connected layers, dropout, and softmax output ensures discriminative feature learning and mitigates overfitting.

• Dataset and Preprocessing:

  • Images are resized to 224×224 pixels, normalized, and augmented with rotations, translations, shearing, zooming, and flips.

  • Data is split 80:20 into training and testing sets for model evaluation.

• Training:

  • The model is trained for 20 epochs using the Adam optimizer, a learning rate of 1×10??, and sparse categorical cross-entropy loss.

  • High-performance computing resources (AMD Ryzen 7 CPU, 64 GB RAM, NVIDIA RTX 4090 GPU) enable efficient large-scale training.

Conclusion

This study presented a deep learning framework for automated multi-class animal species classification using the ConvNeXt-Tiny architecture with transfer learning. By fine-tuning a model pre-trained on ImageNet and appending a task specific classification head, the proposed system achieved an overall accuracy of 93%, with consistently balanced macro precision, recall, and F1-score across all 90 animal categories. The results confirm that ConvNeXt\'s integration of transformer inspired design principles with conventional CNN inductive biases yields powerful discriminative representations, particularly in the presence of visual diversity, background clutter, and inter class similarity. Data augmentation and dropout regularization further contributed to robust generalization on unseen data. These findings demonstrate the practical viability of the proposed framework for real-world wildlife monitoring, biodiversity conservation, and ecological research applications. Future work may explore hierarchical classification strategies, larger and more imbalanced datasets, and lightweight model variants to support deployment in resource constrained field environments.

References

[1] M. A. Rahman, T. M. Berhe, L. Borgianni, M. S. Ahmed, C. Bua, D. Adami, and S. Giordano, “A scalable framework for deploying AI-powered wildlife monitoring in resource-limited field environments,” IEEE Access, vol. 13, pp. 145023–145041, 2025, doi: 10.1109/ACCESS.2025.3598927. [2] H. Nguyen, S. J. Maclagan, T. D. Nguyen, T. Nguyen, P. Flemons, K. Andrews, E. G. Ritchie, and D. Phung, “Animal recognition and identification with deep convolutional neural networks for automated wildlife monitoring,” in Proc. 2017 IEEE Int. Conf. Data Science and Advanced Analytics (DSAA), 2017, pp. 40–49, doi: 10.1109/DSAA.2017.31. [3] W. H. S. Antônio, M. Da Silva, R. S. Miani, and J. R. Souza, “A proposal of an animal detection system using machine learning,” Applied Artificial Intelligence, vol. 33, no. 13, pp. 1093–1106, 2019, doi: 10.1080/08839514.2019.1673993. [4] A. Shalika and L. Seneviratne, “Animal classification system based on image processing and support vector machine,” Journal of Computer and Communications, vol. 4, pp. 12–21, 2016, doi: 10.4236/jcc.2016.41002. [5] W. H. E. Mg, T. T. Zin, P. Tin, M. Aikawa, K. Honkawa, and Y. Horii, “Automated cattle monitoring system for calving time prediction using trajectory data embedded time series analysis,” IEEE Open Journal of the Industrial Electronics Society, vol. 6, pp. 216–234, 2025, doi: 10.1109/OJIES.2025.3533663. [6] Z. U. Rehman, A. Syed, A. Tayab, G. G. Tejani, D. S. Khafaga, and E.-S. M. El-Kenawy, “Improving real-time animal detection using group sparsity in YOLOv8: A solution for animal-toy differentiation,” Computers, Materials & Continua, vol. 86, no. 2, pp. 1–25, 2026, doi: 10.32604/cmc.2025.070310. [7] M. B. Oliveira, H. S. Bernardino, A. B. Vieira, and D. A. Augusto, “Classification of animal species via deep neural networks and species distribution modeling: A systematic review,” Artificial Intelligence Review, vol. 58, no. 8, p. 230, 2025, doi: 10.1007/s10462-024-11074-w. [8] W. Feng, W. Ju, A. Li, W. Bao, and J. Zhang, “High-efficiency progressive transmission and automatic recognition of wildlife monitoring images with WISNs,” IEEE Access, vol. 7, pp. 161412–161423, 2019, doi: 10.1109/ACCESS.2019.2951596. [9] M. Willi, R. T. Pitman, A. W. Cardoso, C. Locke, A. Swanson, A. Boyer, M. Veldthuis, and L. Fortson, “Identifying animal species in camera trap images using deep learning and citizen science,” Methods in Ecology and Evolution, 2018, doi: 10.1111/2041-210X.13099. [10] S. S. Khan, N. V. Doohan, M. Gupta, S. Jaffari, A. Chourasia, K. Joshi, and B. Panchal, “Hybrid deep learning approach for enhanced animal breed classification and prediction,” Traitement du Signal, vol. 40, no. 5, pp. 2087–2099, 2023, doi: 10.18280/ts.400526. [11] A. He, X. Li, X. Wu, C. Su, J. Chen, S. Xu, and X. Guo, “ALSS-YOLO: An adaptive lightweight channel split and shuffling network for TIR wildlife detection in UAV imagery,” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 17, pp. 17308–17326, 2024, doi: 10.1109/JSTARS.2024.3461172. [12] R. A. Rajagukguk, S.-Y. Lee, J.-Y. Park, K. F. Daniel, C.-R. Lee, Z. Chen, D. Liu, T. Norton, J. Park, and S.-W. Hong, “Deep learning for visual animal monitoring (detection, tracking, pose estimation, and behavior classification): A comprehensive review,” Smart Agricultural Technology, vol. 12, p. 101539, 2025, doi: 10.1016/j.atech.2025.101539. [13] Y. Wu, X. Liu, J. Hao, Z. Xing, H. Xi, W. Sun, and Q. Yao, “MBAF-MSCNet: Multi-branch fusion for multi-scale context in infrared wildlife detection,” IEEE Access, vol. 13, pp. 165830–165843, 2025, doi: 10.1109/ACCESS.2025.3605738. [14] H. S. de A. Martins, G. S. Ferrante, and R. I. Meneguette, “Descriptor: A Brazilian roadkill animals for object detection GAN-augmented dataset (BRAGAN),” IEEE Data Descriptions, pp. 1–10, 2026, doi: 10.1109/IEEEDATA.2026.3664029. [15] A. Negreiro, R. E. P. Ferreira, T. Bresolin, G. L. Menezes, E. Casella, D. Döpfer, L. L. Hernandez, G. J. M. Rosa, and J. R. R. Dórea, “Artificial intelligence for animal identification: Development of a computer vision system for reliable dairy cattle traceability across developmental phases,” Journal of Dairy Science, 2026, doi: 10.3168/jds.2025-27653. [16] Q. Zhang, K. Ahmed, N. Sharda, and H. Wang, “A comprehensive survey of animal identification: Exploring data sources, AI advances, classification obstacles and the role of taxonomy,” International Journal of Intelligent Systems, vol. 2024, no. 1, p. 7033535, 2024, doi: 10.1155/2024/7033535. [17] A. B. Mughal, R. U. Khan, A. U. Rehman, and A. Bermak, “Deep learning for dynamic wildlife monitoring: A real-time approach,” IEEE Access, vol. 13, pp. 147422–147448, 2025, doi: 10.1109/ACCESS.2025.3600625. [18] K. Parkavi, A. Ganguly, S. Sharma, A. Banerjee, and K. Kejriwal, “Enhancing road safety: Detection of animals on highways during night,” IEEE Access, vol. 13, pp. 36877–36896, 2025, doi: 10.1109/ACCESS.2025.3545490. [19] T. Battu and D. S. R. Lakshmi, “Animal image identification and classification using deep neural networks techniques,” Measurement: Sensors, vol. 25, p. 100611, 2023, doi: 10.1016/j.measen.2022.100611. [20] M. Mulero-Pázmány, S. Hurtado, C. Barba-González, M. L. Antequera-Gómez, F. Díaz-Ruiz, R. Real, I. Navas-Delgado, and J. F. Aldana-Montes, “Addressing significant challenges for animal detection in camera trap images: A novel deep learning-based approach,” Scientific Reports, vol. 15, no. 1, p. 16191, 2025, doi: 10.1038/s41598-025-90249-z.

Copyright

Copyright © 2026 Ketan Kanjiya, Piyush Sonani, Upendrasinh Zala. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.