Lumpy Skin Disease (LSD) is a rapidly spreading transboundary disease in cattle, primarily caused by the Lumpy Skin Disease Virus (LSDV), which belongs to the Capripoxvirus genus of the Poxviridae family. The disease is characterized by fever, nodules on the skin, mucous membrane damage, enlarged lymph nodes, and significant decline in productivity including milk yield and fertility. Traditional diagnosis methods such as physical inspection and laboratory testing (e.g., PCR or ELISA) often prove insufficient for early detection and are not feasible in rural or resource-constrained areas due to cost, infrastructure, and time limitations.
This project proposes a deep learning-based solution using Convolutional Neural Networks (CNNs) to detect LSD from digital images of cattle skin. A carefully curated image dataset containing Normal, Mild, and Severe cases of LSD-affected cattle was used for training the model. All images were labeled and preprocessed through resizing, normalization, and augmentation techniques to improve model generalization.
The CNN model architecture was designed using TensorFlow and Keras, consisting of multiple convolutional, pooling, and dense layers optimized through experimentation. The training process included techniques such as dropout, data augmentation, and early stopping to avoid overfitting and improve robustness.
After training, the model showed promising accuracy in distinguishing between the severity levels of the disease. To simplify usability for real-world applications, the classification output was mapped to a binary
Introduction
Lumpy Skin Disease (LSD) is a contagious viral illness affecting cattle, leading to economic losses due to symptoms like skin nodules, fever, reduced milk production, abortions, and death. Traditional diagnostic methods like PCR and ELISA are accurate but impractical for rural settings due to cost and equipment requirements.
To address this, the study proposes a Convolutional Neural Network (CNN)-based image classification system that can detect LSD from cattle skin images. The model categorizes images into Normal, Mild, and Severe but outputs a simplified binary decision — Positive (diseased) or Negative (healthy) — for practical use by farmers and veterinarians, especially in low-resource areas via a command-line interface (CLI) without internet dependence.
Research Objectives:
Develop a CNN model for classifying LSD severity levels.
Preprocess and augment a dataset of ~4000 real cattle images.
Implement binary classification logic from multiclass CNN outputs.
Evaluate performance using accuracy, precision, recall, F1-score, and confusion matrix.
Deploy the model using a Python CLI for image-based prediction.
Promote scalable AI use in rural veterinary practices.
Research Hypotheses:
CNNs can achieve >85% accuracy in detecting LSD.
Binary output enhances usability with minimal accuracy loss.
Image preprocessing improves model accuracy.
The system can run offline on standard devices, making it viable for remote use.
Literature Review Highlights:
Traditional LSD diagnosis methods are effective but not scalable.
Deep learning, particularly CNNs, excels in image-based diagnosis.
Few existing models focus on LSD image classification under varied real-world conditions.
There is a lack of public LSD image datasets and real-time deployment tools.
Preprocessing: Resizing to 128×128 pixels, normalization, and augmentation.
Model Architecture:
2 convolution + pooling layers
Dense layer with dropout
Softmax output for 3 classes
Training: 15–25 epochs using the Adam optimizer with early stopping.
Deployment: Saved as .h5 model; Python script provides CLI-based predictions grouped into "LSD Positive" or "LSD Negative."
Performance Analysis:
CNN Accuracy: 93.2%
Precision/Recall/F1-Score: All ~0.93 (higher than Random Forest and Decision Tree models).
Validation Accuracy: Improved from 68% to 95% over 20 epochs.
Loss Curves: Training and validation loss steadily decreased, showing good convergence and no overfitting.
Conclusion
This project focused on developing a robust and accurate deep learning model using Convolutional Neural Networks (CNNs) for the detection of Lumpy Skin Disease (LSD) in cattle based on symptomatic input data. Given the economic and agricultural impact of LSD, especially in rural farming communities, this work aimed to deliver a fast, reliable, and scalable diagnostic tool.
The CNN model was trained and validated over 20 epochs, achieving a peak training accuracy of 98% and validation accuracy of 95%. The consistently improving learning curves indicate that the model was successful in capturing the underlying patterns of the disease symptoms and demonstrated strong generalization capabilities on unseen data. The absence of overfitting further emphasizes the model’s robustness.
The system\'s successful performance reinforces the potential of deep learning in veterinary medicine and livestock disease monitoring. Compared to manual diagnosis—which may be slow, subjective, and dependent on expert availability—this AI-driven approach ensures faster detection, higher precision, and greater accessibility, especially in underserved or remote regions. The implementation of such a system can significantly aid in early disease identification, timely intervention, and reduced economic loss in the dairy and livestock sectors.
References
[1] Abutarboush, H. F., et al. (2021). Deep learning applications in veterinary medicine: A review. Veterinary Sciences, 8(10), 236. https://doi.org/10.3390/vetsci8100236
[2] Gari, G., Bonnet, P., Roger, F., Waret-Szkuta, A., Grosbois, V., & Kampa, J. (2012). Epidemiological aspects and financial impact of Lumpy Skin Disease in Ethiopia. Preventive Veterinary Medicine, 102(4), 274–283. https://doi.org/10.1016/j.prevetmed.2011.08.004
[3] Litjens, G., Kooi, T., Bejnordi, B. E., Setio, A. A., Ciompi, F., Ghafoorian, M., ... & Sánchez, C. I. (2017). A survey on deep learning in medical image analysis. Medical Image Analysis, 42, 60–88. https://doi.org/10.1016/j.media.2017.07.005
[4] Abrahams, R., et al. (2019). Lumpy skin disease. In: OIE Technical Disease Card. World Organisation for Animal Health. Available: https://www.woah.org/
[5] LeCun, Y., Bengio, Y., & Hinton, G. (2015). Deep learning. Nature, 521(7553), 436–444. https://doi.org/10.1038/nature14539
[6] Selvaraju, R. R., et al. (2017). Grad-CAM: Visual explanations from deep networks via gradient-based localization. Proceedings of the IEEE International Conference on Computer Vision (ICCV), 618–626. https://doi.org/10.1109/ICCV.2017.74
[7] Khan, M. K., et al. (2020). Artificial Intelligence for Livestock Disease Prediction: A Review of Current Applications and Future Directions. Animals, 10(12), 2282. https://doi.org/10.3390/ani10122282