›› 2015, Vol. 58 ›› Issue (4): 419-426.

• 研究论文 • 上一篇    下一篇


竺乐庆1,*, 张大兴2, 张真3   

  1. (1. 浙江工商大学计算机与信息工程学院, 杭州310018; 2. 杭州电子科技大学图形图像研究所, 杭州 310012;3. 中国林业科学研究院森林生态与保护研究所, 国家林业局森林保护学重点实验室, 北京 100091)
  • 出版日期:2015-04-20 发布日期:2015-04-20
  • 作者简介:竺乐庆, 女, 1972年生, 博士, 副教授, 主要研究方向为图像处理、模式识别与嵌入式系统, E-mail: zhuleqing@ zjgsu.edu.cn

Feature description of lepidopteran insect wing images based on WLD and HoC and its application in species recognition

ZHU Le-Qing1,*, ZHANG Da-Xing2, ZHANG Zhen3   

  1. (1. School of Computer Science and Information Engineering, Zhejiang Gongshang University, Hangzhou 310018, China; 2. Institute of Graphics and Image, Hangzhou Dianzi University, Hangzhou 310012, China; 3. Key Laboratory of Forest Protection of State Forestry Administration, Research Institute of Forest Ecology, Environment and Protection, Chinese Academy of Forestry, Beijing 100091, China)
  • Online:2015-04-20 Published:2015-04-20

摘要: 【目的】本研究旨在探索使用计算机视觉技术实现对昆虫图像的自动分类方法。【方法】首先通过预处理对采集的昆虫标本图像去除背景,分割出双翅,并对翅图像的位置进行校正。然后把校正后的翅面沿翅伸展的纵向和切向分成4个区域,对每个区域提取WLD(Weber Local Descriptor)和HoC(Histogram of Color)特征并归一化。WLD在灰度图像上提取,反映了翅图像的局部纹理特征。HoC则在HSI(Hue, Saturation, Intensity)颜色空间的彩色图像上提取,反映了每个区域的颜色分布信息。将双翅的各个区域的WLD和HoC按序连接后,得到该昆虫图像的特征向量。使用昆虫图像样本训练集提取到的特征向量训练SVM(Support Vector Machine)分类器,最后使用这些训练得到的分类器即可实现对鳞翅目昆虫的分类识别。【结果】该方法在包含10种576个样本的昆虫图像数据库中进行了测试,取得了100%的独立预测精度,并有理想的时间性能、鲁棒性及稳定性。【结论】实验结果证明了WLD结合HoC是一种有效的鳞翅目昆虫图像特征描述方式。

关键词: 昆虫, 鳞翅目, 图像识别, 颜色直方图, WLD, 支持向量机

Abstract: 【Aim】 This study aims to explore the method to realize the automatic insect image recognition based on computer vision technology. 【Methods】 The captured insect image was first preprocessed to remove the background, and were segmented into two pairs of wings, and the position of the wings was calibrated. Then the calibrated wings were divided into several regions along radial and angular directions. WLD (Weber Local Descriptor) and HoC (Histogram of Color) features were extracted and normalized in each region. The WLD features are extracted on grayscale image, reflecting the local texture feature of wing images. HoC features were extracted on HSI (Hue, Saturation, Intensity) color space, reflecting the color distribution information of the region. The WLD features and HoC features from all the regions of two pairs of wings were concatenated into a feature vector of the insect image. The feature vectors extracted from the insect image samples in training set were used to train the SVMs (Support Vector Machines) which were finally used to classify lepidopteran insects. 【Results】 The proposed method was tested in a database with 576 images and the standalone prediction accuracy was as high as 100%, and the system also demonstrated ideal time performance, good robustness and high stability. 【Conclusion】 The experimental results prove that the combination of WLD and HoC is an effective method for insect image feature description.

Key words: Insect, Lepidoptera, image recognition, Histogram of Color (HoC), Weber Local Descriptor (WLD), Support Vector Machine (SVM)