《工程科学学报》录用稿，htps:/doi.org/10.13374/i,issn2095-9389.2021.01.11.005©北京科技大学2020 工程科学学报DO: 基于改进YOLACT实例分割网络的人耳关键生 理曲线提取 袁立，夏桐，张晓爽 北京科技大学自动化学院，北京100083 ☒通信作者，E-mail:yuan@ustb.edu.cn 版稿 摘要在人耳形状聚类、3D人耳建模、个人定制耳机等相关工作中，获取耳的一些关键生理曲线和关键点的准确 位置非常重要。传统的边缘提取方法对光照和姿势变化非常敏感。本文提出了一种基于ResNeSt和筛选模板策略的改 进YOLACT实例分割网络，分别从定位和分割两方面对原始YOLACT 算法进行改进，通过标注人耳数据集，训练 改进的YOLACT模型，并在预测阶段使用改进的筛选模板策略 可以准确地分割人耳的不同区域并提取关键的生 理曲线。相较于其他方法，本文方法在测试图像集上显示出更 份割精度，且对人耳姿态变化时具有一定的鲁棒 性。 关键词人耳：生理曲线提取：实例分割：改进YO esNeSt 分类号TP391.41 Physiological Curves Extraction of Human Ear Based on Improved YOLACT YUAN Li,XIA Tong ZHANG Xiaoshuang School of Automation.University of Science and Technology Beijing,Beijing 100083,China Corresponding author,Email:lyuan@ustb.edu.cn ABSTRACT In related work such as human ear shape clustering.3D human ear modeling,and personal customized headphones,it is very important to obtain some key physiological curves of the human ear and the accurate positions of key points.Moreover,as an important biological feature,the morphological analysis and classification of the human ear is also of great value for medical work related to the human ear.However,due to the complex morphological structure of the human ear,it is difficult to generate a general standard for the morphological structure of the ear.This paper divides the morphological structure of the human ear into three regions:helix,antihelix and concha,for instance segmentation and key physiological curve extraction.Traditional edge extraction methods are very sensitive to illumination and posture variations. 收稿日期： 基金项目：国家自然科学基金资助项目(61472031)工程科学学报 DOI: 基于改进 YOLACT 实例分割网络的人耳关键生 理曲线提取 袁 立，夏 桐，张晓爽 北京科技大学自动化学院，北京 100083  通信作者，E-mail: lyuan@ustb.edu.cn 摘 要 在人耳形状聚类、3D 人耳建模、个人定制耳机等相关工作中，获取人耳的一些关键生理曲线和关键点的准确 位置非常重要。传统的边缘提取方法对光照和姿势变化非常敏感。本文提出了一种基于 ResNeSt 和筛选模板策略的改 进 YOLACT 实例分割网络，分别从定位和分割两方面对原始 YOLACT 算法进行改进，通过标注人耳数据集，训练 改进的 YOLACT 模型，并在预测阶段使用改进的筛选模板策略，可以准确地分割人耳的不同区域并提取关键的生 理曲线。相较于其他方法，本文方法在测试图像集上显示出更好的分割精度，且对人耳姿态变化时具有一定的鲁棒 性。 关键词 人耳；生理曲线提取；实例分割；改进 YOLACT；ResNeSt 分类号 TP391.41 Physiological Curves Extraction of Human Ear Based on Improved YOLACT YUAN Li , XIA Tong, ZHANG Xiaoshuang School of Automation, University of Science and Technology Beijing, Beijing 100083, China  Corresponding author, E-mail: lyuan@ustb.edu.cn ABSTRACT In related work such as human ear shape clustering, 3D human ear modeling, and personal customized headphones, it is very important to obtain some key physiological curves of the human ear and the accurate positions of key points. Moreover, as an important biological feature, the morphological analysis and classification of the human ear is also of great value for medical work related to the human ear. However, due to the complex morphological structure of the human ear, it is difficult to generate a general standard for the morphological structure of the ear. This paper divides the morphological structure of the human ear into three regions: helix, antihelix and concha, for instance segmentation and key physiological curve extraction. Traditional edge extraction methods are very sensitive to illumination and posture variations. 收稿日期: 基金项目:国家自然科学基金资助项目(61472031) 《工程科学学报》录用稿，https://doi.org/10.13374/j.issn2095-9389.2021.01.11.005 ©北京科技大学 2020 录用稿件，非最终出版稿