粗糙集神经网络在昆虫总科阶元分类学上的应用  

Application of a rough-set neural network to superfamily level in insect taxonomy

作　　者：杜瑞卿[1] 褚学英[1] 王庆林[1] 赵秋红[2] 庞发虎[1] 

机构地区：[1]南阳师范学院生命科学系,河南南阳473061 [2]南阳师范学院数学系,河南南阳473061

出　　处：《中国农业大学学报》2007年第1期33-38,共6页Journal of China Agricultural University

基　　金：南阳师范学院青年科学研究资助项目(nytc2004k01)

摘　　要：为研究粗糙集和神经网络相结合方法的实践性,以及昆虫的数学形态特征在总科阶元上作为分类特征的可行性、可靠性和重要性。从总科角度对鳞翅目(Lepidoptera)和鞘翅目(Coleoptera)5个总科23种虫体图像中提取的昆虫面积、周长等11项数学形态特征进行粗糙集神经网络分析。结果表明:在总科阶元上,11项特征的可靠性大小为,面积、亮斑数>周长、横轴长、形状参数、圆形性、似圆度、偏心率>纵轴长、叶状性、球状性,与赵汗青等的统计分析结果中属性特征的重要性大多数一致;神经网络模式识别结果与传统分类结果完全一致。应用粗糙集理论进行昆虫数学形态特征分类结果准确;在昆虫总科阶元分类上粗糙集神经网络较统计学方法具有优势。To study the feasibility, reliability and importance of taxonomy character at superfamily level in insect math- morphological features by a combination approach of the rough-set theory and the neutral network we analyzed It mathmorphological features （MMFs）, such as area, perimeter, and so on, from the images of 23 species of insects, in five superfamilies of Lepidoptera and Coleoptera, Noctuoidea, Bombycoidea, Papilionoidea, Scarabaeoidea and chrysomeloidea, The results were compared with those of the statistical analysis reported by Zhao Hanqing. The rank of reliability MMFS in the identification of insect superfamilies is from high to low; the area and hot-holenumbe are grater than perimeter; the X-length, form parameter, circularity, roundness-likelihood, eccentricity greater then Y-length, lobation, sphericity, The results are not completely identical with those from Zhao Hanqing, but the most characteristics are identical, The results of pattern recognition by neutral network are completely identical with those of traditional classifications. This theory applied to insect taxonomy is better compared with the statistical analysis method and has a great significance at superfamily level when used with rough-set neutral network.

关 键 词：昆虫分类 粗糙集 神经网络 数学形态特征 

分 类 号：Q964[生物学—昆虫学]