Image processing-based automatic tooth segmentation and age estimation in sheep using deep learning

Accurate determination of sheep age is essential for optimizing meat quality, reproductive efficiency, feeding strategies, and market value. Conventional age estimation methods rely on subjective visual inspection of dental structures and are prone to inconsistency and human error. Therefore, automated solutions capable of providing objective and reproducible assessments are needed.

This study proposes a fully automated deep learning and image processing framework for sheep age estimation using dental images. Class imbalance in the dataset was addressed through systematic data augmentation. The images were then segmented using the You Only Look Once version 8 (YOLOv8) algorithm to isolate anatomically relevant tooth regions. Several Convolutional Neural Network (CNN) architectures were evaluated and compared, including VGG16, ResNet50, EfficientNetB0, MobileNetV2, and Xception, fine-tuned via transfer learning, as well as a custom BasicCNN model. A graphical user interface (GUI) was also developed and deployed as a publicly accessible, containerized application to provide a practical and user-friendly implementation of the prediction system.

Among the evaluated models, EfficientNetB0 achieved the highest performance, attaining an overall accuracy of 95 %, with 97 % for the 3-12-month and 2-3-year groups, 92 % for the 1-1.5-year group, and 93 % for the 1.5-2-year group. These results demonstrate that combining automatic segmentation with transfer learning substantially improves model generalization and classification accuracy.

The proposed framework offers a robust, automated, and scalable solution for sheep age estimation. By eliminating manual assessment, the system contributes to precision livestock farming and supports informed decision-making in agricultural practice. The integration of deep learning, automatic segmentation, and a user-friendly interface highlights its potential for broader adoption in real-world field applications.