浙江农业学报 ›› 2017, Vol. 29 ›› Issue (4): 668-675.DOI: 10.3969/j.issn.1004-1524.2017.04.22

• 生物系统工程 • 上一篇    下一篇

基于多特征融合的植物叶片识别研究

高良1, 闫民2, 赵方1, *   

  1. 1.北京林业大学 信息学院,北京 100083;
    2.北京林业大学 工学院,北京 100083
  • 收稿日期:2016-11-30 出版日期:2017-04-20 发布日期:2017-04-27
  • 通讯作者: 赵方,E-mail:fangzhao@bjfu.edu.cn
  • 作者简介:高良(1989—),男,山东潍坊人,硕士研究生,主要研究方向为图像处理与模式识别。E-mail:gaoliang_2008@163.com
  • 基金资助:
    国家自然科学基金项目(11272061)

Plant leaf recognition based on fusion of multiple features

GAO Liang1, YAN Min2, ZHAO Fang1, *   

  1. 1. School of Information Science and Technology, Beijing Forestry University, Beijing 100083, China;
    2. School of Technology, Beijing Forestry University, Beijing 100083, China
  • Received:2016-11-30 Online:2017-04-20 Published:2017-04-27

摘要: 植物叶片识别作为植物自动分类识别的重要分支,有着很高的实际应用价值。针对当前叶片特征描述存在的局限和叶片识别准确率较低的实际,以叶片图像为研究对象,首先对图像进行预处理,在提取叶片几何特征和纹理特征的基础上,设计描述叶片轮廓的距离矩阵和角点矩阵,通过计算基于几何特征、纹理特征和角点距离矩阵的综合相似度对叶片进行精确识别。对Flavia数据集中的32类共计960幅叶片图像进行训练和测试,结果表明,基于叶片图像多特征融合的识别方法对叶片特征描述能力更强,识别准确率更高,对Flavia数据集的识别率可达97.50%,具有较好的识别效果。

关键词: 叶片识别, 几何特征, 纹理特征, 角点距离矩阵, 综合相似度

Abstract: As an important branch of plant automatic classification and recognition, plant leaf recognition is of great value in practical application. In view of the limitation of description methods for leaf features and the problem of low accuracy of plant leaf recognition, leaf images were used as recognition objects in this paper. An image preprocessing algorithm was proposed to ensure getting the features of leaf images accurately. In addition to the geometric features and texture features, the leaf profile was described by distance matrix and corner matrix, and the leaf could be identified more precisely by calculating the comprehensive similarity of geometric features, texture features and corner distance matrix. Experiments were performed on Flavia dataset of 960 images divided into 32 classes. Compared with other recognition methods, the method proposed in this paper achieved better recognition effect. The experimental results showed that the recognition accuracy reached 97.50% with high practicability.

Key words: leaf recognition, geometric features, texture features, corner distance matrix, comprehensive similarity

中图分类号: