内蒙古农业大学学报(自然科学版)2024,Vol.45Issue(2):51-61,11.DOI:10.16853/j.cnki.1009-3575.2024.02.007
基于LoFTR特征匹配算法的羊只身份识别
Sheep Identification Based on LoFTR Feature Matching Algorithm
摘要
Abstract
In modern agricultural and animal husbandry production,automated livestock identification is a crucial factor in enhanc-ing breeding selection and behavioral research efficiency.To address the issues of time consumption,high costs,and insuffi-cient accuracy associated with traditional sheep identification methods,scholars have recently begun to use deep learning technolo-gy to learn the sheep facial biological features,to further realize the identity recognition.However,identification methods based on the sheep facial images has some problems,such as the instability and fluctuation in recognition performance.Therefore,this study developed a sheep identification method based on matching the feature points.Specifically,the LoFTR algorithm was em-ployed to learn and match the facial feature points of sheep.By learning the biological features of sheep faces,this model achieved the technology of identifying individual identities through matching the feature points.In view of the limitations of single sheep face im-age acquisition angle and unstable recognition effect in previous studies,this study collected multi-angle facial images of 89 small-tailed Han sheep to extract higher-quality sheep facial features and further constructed a sheep facial dataset for model training and recognition.Comparative performance experiments with classical feature point matching models indicated that the LoFTR sheep fa-cial feature point matching model showed higher recognition accuracy and efficiency.This model achieved an average recognition accuracy of 93.53%on sheep facial image datasets,respectively,and the average matching rates also reached 2.39 s.The applica-tion of this method contributed to improve the management efficiency of sheep farms and promote the modernization of sheep farm management.By improving the accuracy and efficiency of sheep facial identification,it effectively enhanced the precision and speed of breeding selection and behavioral research,providing better technical supports for agricultural and animal husbandry produc-tion,and promoting the sustainable development of the industry.关键词
羊脸识别/LoFTR/深度学习/特征点匹配Key words
Sheep face recognition/LoFTR/Deep learning/Feature point matching分类
信息技术与安全科学引用本文复制引用
崔家赫,宣传忠,宋耀邦,王光普..基于LoFTR特征匹配算法的羊只身份识别[J].内蒙古农业大学学报(自然科学版),2024,45(2):51-61,11.基金项目
内蒙古直属高校基本科研业务费项目(BR221032) (BR221032)
内蒙古自治区科技计划项目(2021GG0111) (2021GG0111)