APPLICATION OF ROUGH-SET THEORY AND NEURAL NETWORK AT SUPERFAMILY LEVEL IN INSECT TAXONOMY
Zi Liang · Acta Zootaxonomica Sinica · 2007
Using rough-set theory and neural network analyses of 11 math-morphological features(MMFs)(such as area,perimeter,etc.)from the images of 23 species of insects of the Lepidoptera and Coleoptera families,Noctuoidea,Bombycoidea,Papilionaidea,Scarabaeoidea and Chrysomeloidea,the results are compared with those of ZHAO Han-Qing made by his statistical analysis and indicates that the ranked reliability of MMFS in the identification of insect superfamilies is:from high to low:area,hot-holenumbeperimeter,X-length,form parameter,circularity,roundness-likelihood,eccentricityY-length,lobation,sphericity,roundness-likelihood,eccentricityY-length,lobationform parameter,hot-holenumbe.The results are not completely identical with those of ZHAO Han-Qing made by his statistical analysis,but the most importance of characteristic are identical.The results of pattern recognition by neutral network are completely identical with those of traditional classifications.Accordingly,the conclusion is that this theory applied in insect taxonomy is more idealistic compared with statistical analysis method,and it has great significance at superfamily level when used with rough-set neutral network.