Optimized Feature Engineering for Dentition based Cattle Age Estimation

Abstract

Accurate age estimation of cattle is crucial for effective herd management, breeding, and health monitoring. In this novel study, a unique methodology for cattle age estimation is introduced using high-resolution images of the teeth and canal, captured at local farms and from cow breeders. This approach involves capturing these images, annotating them to distinguish between teeth and canal, and employing a tailored YOLO v9 deep learning model for detection and segmentation, achieving a mean Average Precision (mAP) of 98% with a confidence threshold of 0.5 to 0.95. The teeth and canal regions are prominent in age computation for experts. After segmenting these Regions of Interest (RoI), conventional feature descriptors were used to extract edge features from the segmented images such as Histogram of Oriented Gradients (HOG). Initial linear regression analysis of these features yielded a Root Mean Square Error (RMSE) close to 52. To enhance predictive performance, personalized feature engineering pipelines incorporating advanced feature engineering and selection techniques were developed. This refinement led to a substantial improvement, reducing RMSE to approximately 0.06 with an R² of 0.99 for HOG features. HOG was selected over Convolutional Neural Networks (CNNs) due to its computational efficiency and suitability for resource-constrained environments. HOG demonstrated strong performance with minimal computational requirements, making it well-suited for real-time applications on mobile devices. While CNNs offer potential for future enhancements, our current approach prioritizes practicality and performance for small-scale applications. Our research significantly advances machine-learning-based cattle age prediction, offering a reliable, scalable solution for agricultural practices and also paving the way for future research in this field.