兽类学报 ›› 2019, Vol. 39 ›› Issue (1): 43-51.DOI: 10.16829/j.slxb.150211

• • 上一篇    下一篇

基于发育网络识别模型的大熊猫面部识别

侯金 郑伯川 李玉杰 白文科 齐桂兰 董峻妃 杨泽静 张晋东   

  1. 西南野生动植物资源保护教育部重点实验室,西华师范大学生命科学学院珍稀动植物研究所
  • 出版日期:2019-01-30 发布日期:2019-01-18
  • 通讯作者: 张晋东 E-mail:zhangjd224@163.com
  • 基金资助:
    国家自然科学基金面上项目(41571517;31572293);西华师范大学英才基金(17YC358)

Facial recognition of giant pandas based on developmental network recognition mode

Houjin, Zheng Bochuan, LI Yujie, Bai Wenke, QI Guilan, Dong Junfei, Yang Zejing, Zhang Jindong   

  1. Key Laboratory of Southwest China Wildlife Resources Conservation,Institute of Rare Animals and Plants of School of Life Sciences, China West Normal University
  • Online:2019-01-30 Published:2019-01-18

摘要: 个体识别是动物行为学与生态学研究工作的基础,也是制定珍稀野生动物保护政策的重要依据。为了丰富大熊猫个体识别和种群数量调查的方法,我们于2017年7月分别在四川省雅安市碧峰峡大熊猫基地和四川省汶川县耿达镇的中华大熊猫苑共计拍摄18只大熊猫个体,每只大熊猫拍摄6~13张高质量面部照片(共计131张),利用发育网络(Developmental Network)建立大熊猫面部识别模型。利用此模型对存在部分背景的大熊猫面部照片进行识别检测,得到的个体识别率为79.41%,对完全去除背景的大熊猫面部照片进行识别检测,得到的个体识别率为58.82%。研究表明,发育网络具有足够的大熊猫个体识别能力,不同背景比例的照片对大熊猫个体识别的实际结果具有较大的影响。随着发育网络识别模型的发展,我们建议更多的野生动物保护研究者结合这一技术深入地开展珍稀野生动物(如大熊猫)个体识别研究,逐步提高识别准确度,并应用到关键区域大规模的动物调查中。

关键词: 个体识别, 发育网络, 大熊猫

Abstract: Individual identification is basis of animal behavior and ecology research, and is also important for the development of wildlife protection policies. In order to investigate new methods of giant panda (Ailuropoda melanoleuca) individual identification and population surveying, we photographed 18 giant pandas from Bifengxia giant panda base in Ya'an and the Chinese Giant Panda Garden in Gengda, Wenchuan County, Sichuan province in July 2017. We took high-quality facial photos of each giant panda 6 ~ 13 times for a total of 131 photographs. Using a developmental Network, we formulated a model for the facial recognition of the giant panda. We used the model to test photos which had some background features, with a successful individual recognition rate of 79.41%. We also tested photos in which the background was completely removed, which resulted in an individual recognition rate of 58.82%. This study shows that the developmental Network has adequate discriminatory ability for giant panda individual identification, and that the proportion of background features has severe influence on the actual results of the model. With the progress of the developmental network recognition model, we suggest that more wildlife researchers pursue this application of novel technology to conduct individual identification research of rare wildlife species such as the giant panda, gradually improve identification accuracies, and apply the model to large-scale wildlife investigations in key regions.

Key words: Individual identification, Developmental network, Giant panda