earticle

영어

In Korea, the proportion of pets as family members is increasing due to the aging population and the increase of single-person households, but the gap in feeding management during the absence of guardians and the problem of feed reading in multi-family households remain unresolved tasks. As a basic study to solve this problem, this paper designed, learned, and evaluated YOLOv5-based face recognition model with the aim of designing an automatic feeding system for companion animal face recognition. The proposed model is based on a scenario in which the face of a companion animal captured by a camera is directly detected in units of bounding boxes to determine whether it is a pre-registered individual, and the feed bin is opened only when it matches the feeding time conditions. In addition, as a result of comparing YOLOv5 and YOLOv8, YOLOv8 recorded higher values with Recall 0.82, Precision 0.83, and YOLOv5 with Recall 0.87, and Precision 0.85. In this study, YOLOv5 was selected as the final face recognition model in consideration of the key requirement to detect individuals without missing them and not misrecognize other individuals in a multi-family environment. The research results are expected to be used as basic data for selecting models and preparing system design standards when implementing an automatic feeding system for companion animals in the future.