Preliminary Development of an Artificial Intelligence-Based Measurement System for Automated Feed Monitoring and Feed Weight Estimation

This paper presents the preliminary development and experimental validation of an artificial intelligence–based measurement system for automated feed monitoring and direct feed weight estimation in the context of Precision Livestock Farming. The proposed approach relies exclusively on RGB imaging and supervised deep learning techniques to estimate feed mass, deliberately avoiding explicit three-dimensional geometric reconstruction or density-based calibration procedures typically adopted in existing solutions. A single-stage object detection and classification framework is employed to jointly localize the feed heap and infer its weight class in a single forward pass. The model is trained on a large, carefully annotated dataset covering a wide range of feed weights, with controlled quantization and extensive variability in illumination conditions and feed arrangement, in order to enhance robustness and generalization capability. Experimental results show excellent detection performance and a strong correlation between predicted and reference weights, with most estimation errors confined to adjacent quantized classes. Beyond conventional computer vision metrics, the proposed system is explicitly treated as a measurement instrument. A metrological characterization is therefore performed, including calibration for gain and offset correction and a quantitative evaluation of measurement uncertainty following ISO GUM principles [1]. The results demonstrate a near-unitary calibrated response and an expanded uncertainty compatible with the requirements of feed intake monitoring applications. Compared to camera-based 3D systems, the proposed RGBonly solution significantly reduces hardware complexity, computational burden, and deployment constraints, while maintaining high accuracy and repeatability. These features make the system particularly suitable for large-scale farm deployment and for integration within multimodal sensing platforms aimed at improving feed efficiency assessment and sustainability in livestock production systems.