Using DAISY (Digital Automated Identification System) for automated identification of moths of the superfamily Bombycoidea of Borneo / Leong Yuen Munn

Leong, Yuen Munn (2013) Using DAISY (Digital Automated Identification System) for automated identification of moths of the superfamily Bombycoidea of Borneo / Leong Yuen Munn. Masters thesis, University of Malaya.

Preview

PDF (Full Text) Download (1340Kb) | Preview

Abstract

The health of our environment can be measured by monitoring moths. Moths are especially useful as biological indicators given that they are sensitive to environmental changes, widespread and can be found in various habitats. Therefore, by monitoring their ranges and numbers, we can obtain essential clues about our changing environment such as climate change, air pollution and the effects of new farming practices. In this study, I tested DAISY as a tool for automated identification of moths. Images of 210 species of the superfamily Bombycoidea from the book “The Moths of Borneo: Part 3: Lasiocampidae, Eupterotidae, Bombycidae, Brahmaeidae, Saturniidae, Sphingidae” were used as a training data-set for DAISY. The images were pre-processed to (600X400 pixels), mirroring of least torn wing was performed, and the file format was converted from JPEG to TIF. Training and testing of the system were performed using images of the right forewings of moths. Additionally, the hindwings of two species; Actias maenas and A. Selene were included as they have significant shape (very elongated hindwings compared to other species). Three test datasets were then used to evaluate the performance of DAISY: (i) distorted versions of the training images (ii) images from internet resources of same species to the ones in training dataset (Superfamily Bombycoidea), (iii) images from other volumes of The Moths of Borneo of species which were not included in the training dataset, from families; Notodontidae, Lymantriidae, Arctiidae, Drepaninae, Callidulidae, Geometridae, Notuidae and Noctuidae) and (iv) DAISY’s default sample images of species not in the training dataset (Belize sphingids). I classified the results of DAISY identifications into four categories for analysis; (i) test species in training set and correct species is given (true positive, TP), (ii) test species may or may not be in training set but incorrect species name is given (false positive, FP), (iii) test species not in training set and no species name is given (true No-ID, TNI) and (iv) test species in training set but no species name is given (false No-ID, FNI). Based on these criteria I measured the precision as TP / (TP+FP) and overall accurary as (TP+TNI) / (TP+TNI+FP+FNI). Overall, the precision of DAISY across all four test data-sets was 51% while overall accuracy was 50%. I discuss the potential reasons for the low obersved accuracy. Finally, I make recommendations for features to be considered when designing an automated tool for moth identification.

Actions (For repository staff only : Login required)