BEIJING -- Chinese scientists have successfully developed a lightweight model for beef cattle behavior recognition from quadruped robot video in grassland pastures, improving the efficiency of herd feeding and management.

The lightweight model MASM-YOLO was proposed by the Agricultural Information Institute of the Chinese Academy of Agricultural Sciences, and the relevant research was published in Computers and Electronics in Agriculture.

Accurate and rapid identification of typical cattle behaviors is fundamental to disease diagnosis, estrus monitoring, calving prediction, and health assessment.

MASM-YOLO enables precise multi-behavior detection under complex conditions, suitable for real-time execution on board a mobile robot.

By integrating the Multi-Scale Focus and Extraction Network, the Adaptive Decomposition and Alignment Head, and other technologies, MASM-YOLO addresses key challenges, including lighting variations, motion blur, and occlusions within cattle groups.

MASM-YOLO achieves rapid recognition of six typical behaviors of beef cattle, including feeding, resting, locomotion and licking. It strikes an optimal balance between recognition accuracy and computational efficiency.

This model provides key technical support for the full-scale development of grazing robots.