Evaluation of piglet suckling competition index based on YOLOv10 and optical flow direction distribution features

Abstract

With the rapid development of intelligent farming technologies, effectively evaluating piglet competition behaviour during the suckling period has become a key research focus for enhancing livestock management. This paper presents a method for evaluating the piglet suckling competition index, which integrates the YOLOv10 object detection algorithm and optical flow direction distribution features. First, the YOLOv10 model is employed to detect the sow's posture and the positions of the piglets, classifying the sow's posture into lateral recumbency and other postures. Subsequently, precise localisation of the lactation period is achieved by calculating the mask ratio of the piglets within the sow's region and the changes in group activity. Finally, the Farneback optical flow algorithm is utilised to analyse the direction distribution of the optical flow within the piglet region, and the variation coefficient of information entropy is employed to quantify the intensity of piglet suckling competition. Experimental results demonstrate that the proposed method performs well in both object detection and behaviour localisation, achieving a precision of 91.51 % and a recall of 96.04 % for lactation period localisation. Additionally, the method successfully validated the evaluation of piglet suckling competition in different test pens. This study provides technical support for intelligent farming technologies, helping to optimise piglet nutrition management and enhance farming efficiency.