Automatic Image Labeling Framework Based on Multiple Vision-Language Models for Automation Applications

Agricultural automation and robotics are increasingly transforming modern livestock farming, with computer vision and artificial intelligence algorithms playing a critical role. However, vision-based robotic systems rely heavily on large, well-annotated datasets to train reliable AI models, while manual labeling remains a labor-intensive and time-consuming process. To address this challenge, we propose an automatic image labeling framework based on multiple vision-language models. The framework integrates target point detection, mask generation and fusion, and a mask filtering and voting mechanism to produce high-quality instance masks across diverse livestock species and imaging conditions. Experimental results demonstrate that our method outperforms existing approaches such as Grounded-SAM and SEEM. By enabling efficient dataset creation, the proposed framework supports the development of vision-based perception systems essential for agricultural robotics and intelligent automation.