AI-based automated weight prediction in cattle for herd health surveillance.

Abstract

Early and accurate monitoring of livestock health is critical for effective disease prevention, welfare assurance, and sustainable farm management. Labor-intensive and stressful livestock weighing methods remain a major bottleneck for effective herd health surveillance in large-scale operations. This study presents a data-driven Walk-Over Weighing System (WoWS) enhanced with Fast Fourier Transform (FFT) and machine learning (ML) algorithms to provide a non-invasive, automated solution for real-time weight estimation in cattle. Dynamic weight signals from 86 dairy cows were collected twice daily during routine milking using a walk-over-weighing (WoWS) platform at the Burdur MAKU farm. Raw force-time signals were pre-processed and transformed using FFT to reduce noise and extract spectral-domain features relevant for weight estimation. Six ML models, including Support Vector Regression (SVR), were evaluated for prediction performance. The SVR model yielded the highest accuracy (MAE: 2.3 kg, R²: 0.999). The system's functionality was further extended through integration with Internet of Things (IoT) frameworks for real-time data collection and anomaly detection. Heatmaps and time-aligned weight distributions validated the system's robustness under dynamic field conditions. This FFT- and AI-enhanced WoWS offers a scalable and effective tool for herd-level health surveillance by enabling continuous monitoring, early detection of abnormal weight trends (e.g., weight loss due to disease onset or inadequate feeding), and remote decision-making. The proposed system supports One Health principles by reducing manual handling, minimizing animal stress, improving welfare, and lowering labor demands, thereby contributing to more sustainable and efficient livestock-farming practices. Future directions include expanding multi-sensor integration and epidemiological modeling for more comprehensive livestock health management.