Species-identification of wasps using principal component associative memories

This paper presents a novel approach to image-based insect specimen identif ication. exploiting the ability of principal component auto associative mem ories to form trainable classifiers, which may be used to identify unknown images. The system utilises the differences between a pair of reconstructed images produced when the unknown image is included in, and then excluded f rom the training set encoded by the auto associative memory. A non-parametr ic statistical correlation metric, Kendall's t. was used to correlate the r econstructed images. The approach has been applied to the species-identific ation of closely related parasitic wasps based upon their wing venation and pigmentation patterns. (C) 1999 Elsevier Science B.V. All rights reserved.