生物多样性 ›› 2020, Vol. 28 ›› Issue (4): 524-533.DOI: 10.17520/biods.2019272

• 技术与方法 • 上一篇    下一篇

国内8款常用植物识别软件的识别能力评价

许展慧1,刘诗尧1,赵莹1,涂文琴1,常诏峰1,张恩涛1,郭靖1,郑迪1,耿鋆1,顾高营1,郭淳鹏1,郭璐璐1,王静1,徐春阳1,彭钏1,杨腾1,崔梦琪1,孙伟成1,张剑坛1,刘皓天1,巴超群1,王鹤琪1,贾竞超1,武金洲1,肖翠2,马克平2,*()   

  1. 1 中国科学院大学, 北京 100049
    2 中国科学院植物研究所植被与环境变化国家重点实验室, 北京 100093
  • 收稿日期:2019-09-01 接受日期:2020-01-21 出版日期:2020-04-20 发布日期:2020-06-15
  • 通讯作者: 马克平
  • 基金资助:
    中国科学院战略性先导科技专项(XDA19050404)

Evaluation of the identification ability of eight commonly used plant identification application softwares in China

Zhanhui Xu1,Shiyao Liu1,Ying Zhao1,Wenqin Tu1,Zhaofeng Chang1,Entao Zhang1,Jing Guo1,Di Zheng1,Jun Geng1,Gaoying Gu1,Chunpeng Guo1,Lulu Guo1,Jing Wang1,Chunyang Xu1,Chuan Peng1,Teng Yang1,Mengqi Cui1,Weicheng Sun1,Jiantan Zhang1,Haotian Liu1,Chaoqun Ba1,Heqi Wang1,Jingchao Jia1,Jinzhou Wu1,Cui Xiao2,Keping Ma2,*()   

  1. 1 University of Chinese Academy of Sciences, Beijing 100049
    2 State Key Laboratory of Vegetation and Environmental Change, Institute of Botany, Chinese Academy of Sciences, Beijing 100093
  • Received:2019-09-01 Accepted:2020-01-21 Online:2020-04-20 Published:2020-06-15
  • Contact: Keping Ma

摘要:

随着智能手机和人工智能技术的发展, 以手机app为载体的植物识别软件慢慢走进公众生活、科普活动和科研活动的各个方面。植物识别app的识别正确率是决定其使用价值和用户体验的关键因素。目前, 国内应用市场上有许多植物识别app, 它们的开发目的和应用范围各异, 软件本身的关注点、数据库来源、算法、硬件要求也存在很大差异。对于不同人群, 植物识别app有不同的意义, 如对于科研人员来说, 识别能力强的app是提高效率的一大工具; 对植物爱好者来说, 具一定准确率的识别app可以作为入门的工具。因此, 对各app的识别能力进行分析与评价显得尤为重要。本文选取了8款常用的app, 分别对400张已准确鉴定的植物图片进行识别, 其中干旱半干旱区、温带、热带和亚热带4个区各选取100张。这些图片共计122科164属340种, 涵盖了乔木、灌木、草本、草质藤本和木质藤本5种生长型, 包含23种国家级保护植物。种、属、科准确识别正确分别计4分、2分、1分, 以此标准对软件识别能力按总得分进行排序, 正确率得分由高到低依次为花帮主、百度识图、花伴侣、形色、花卉识别、植物识别、发现识花、微软识花。

关键词: 植物识别软件, 花帮主, 百度识图, 花伴侣, 形色, 花卉识别, 植物识别, 发现识花, 微软识花

Abstract

Smart phone and artificial intelligence technology development has led to various plant recognition softwares on mobile applications. These applications have gradually entered all aspects of public life, popular science activities, and scientific research activities. Presently, there are many plant recognition apps in China, which have varying development purposes and application scopes. Among these differences include variation in software concerns, database sources, algorithms, and hardware which could implicate large discrepancies between apps, making it important to analyze and evaluate the accuracy, scope of application and potential use of each software. In this paper, eight apps were selected to identify 400 accurately identified plant photos, 100 photos being chosen from arid and semi-arid zones, temperate zones, tropical zones, and subtropical zones, respectively. In total, these photos belong to 122 families, 164 genera and 340 species, covering five growth forms of trees, shrubs, herbs, herbaceous vines and woody vines, as well as 23 national protected plant species. Accurate identification of species, genera and families was scored 4, 2 and 1 points, respectively. The software recognition ability was sorted according to total scores, and the results are as follows: HuaBangZhu, Baidu-Shitu, HuaBanLv, XingSe, Huahui-Shibie, Zhiwu-Shibie, Faxian-Shihua, Flower Recognition.

Key words: Plant recognition software, HuaBangZhu, Baidu-Shitu, HuaBanLv, XingSe, Huahui-Shibie, Zhiwu-Shibie, Faxian-Shihua, Flower recognition