昆虫图像几何形状特征的提取技术研究①.pdf

( ) 中国农业大学学报 　2003 ,8 3 : 47～50 Journal of China Agricultural University 昆虫图像几何形状特征的提取技术研究① 于新文 　沈佐锐 　高灵旺 　李志红 ( 中国农业大学 农学与生物技术学院,北京 100094) 摘 　要 　为进行昆虫图象的识别和分类 ,对昆虫图象几何形状特征的提取及测量进行了研究。提取了区域面积、 边界周长、孔洞数、偏心率、形状参数、圆形性、似圆度、球状性和叶状性 9 个直观易测的特征。采用判别分析方法 对这些特征进行筛选得到 6 个具有判别意义的特征 ,分别为区域面积、偏心率、形状参数、周长、似圆度、叶状性 ;剔 除了与似圆度具有相关性的圆形性和球状性等特征。利用这 6 个特征对 3 种昆虫进行了识别 ,准确率均达到 100 % ,表明本研究所提取的 6 个昆虫图象特征用于昆虫图象识别是有效的。 关键词 　昆虫图象 ; 特征提取 ; 特征测量 ; 图象识别 ( ) 中图分类号　S 126 ; Q 964 　　　　文章编号 2003 　　　　文献标识码 　A Feature mea suring and extraction for digital image of insects Yu Xinwen , Shen Zuorui , Gao Lingwang , Li Zhihong ( College of Agrono my and Biotechnology , China Agricultural University , Beijing 100094 , China) Abstract 　Basic methods of feature measuring and extraction of 9 shape features such as region area , boundary perimeter , hole number , eccentricity , form factor , roundness , circularity , spherecity and lobation were introduced based on insect binary image. These features were measured on 3 insect species , i. e. , Helicoverpa armigera Hübner ·· with 50 observations , Ostrinia furnacalis Guenee with 40 observations , and Ectomocoris atrox Stal) with 55 observa tions. 6 determinant features , including area , eccentricity , formfactor , perimeter , roundness and lobation , were ex tracted using discriminant analysis method to produce discriminant functions. The recognition result using the discrimi nant functions to the 3 insect species showed that the accuracy is as hi