Are wing contours good classifiers for automatic identification in Odonata? A view from the Targeted Odonata Wing Digitization (TOWD) project

Are wing contours good classifiers for automatic identification in Odonata? A view from the Targeted Odonata Wing Digitization (TOWD) project 00

Mayra A. Sáenz Oviedo^a , William R. Kuhn^b , Martin A. Rondon Sepulveda^a , John Abbott^c , Jessica L. Ware^d , Melissa Sanchez-Herrera^d,e ✉️✉️

Department of Epidemiology and Biostatistics. Pontificia Universidad Javeriana, Ak. 7 # 40-62, 10231, Bogotá, Colombia
Discover life in America, Gatlinburg, TN, USA
Department of Museum Research and Collections. University of Alabama Museums. Tuscaloosa, AL 35487, USA
Division of Invertebrate Zoology. American Museum of Natural History. 200 Central Park West at West 79th Street, New York, NY 10024, USA
Faculty of Natural Sciences, Biology Department, Universidad del Rosario, Sede Quinta Mutis. Ak. 26 # 63C-48, 11122, Bogotá, Colombia

International Journal of Odonatology, Volume 25, Pages 96-106, 2022

https://doi.org/10.48156/1388.2022.1917184

Published: 6 December 2022 (Received: 11 June 2022, Accepted: 7 November 2022)

Abstract

In recent decades, a lack of available knowledge about the magnitude, identity and distribution of biodiversity has given way to a taxonomic impediment where species are not being described as fast as the rate of extinction. Using Machine Learning methods based on seven different algorithms (LR, CART, KNN, GNB, LDA, SVM and RFC) we have created an automatic identification approach for odonate genera, through images of wing contours. The training population is composed of the collected specimens that have been digitized in the framework of the NSF funded Odomatic and TOWD projects. Each contour was pre-processed, and 80 coefficients were extracted for each specimen. These form a database with 4656 rows and 80 columns, which was divided into 70% for training and 30% for testing the classifiers. The classifier with the best performance was a Linear Discriminant Analysis (LDA), which discriminated the highest number of classes (100) with an accuracy value of 0.7337, precision of 0.75, recall of 0.73 and a F1 score of 0.73. Additionally, two main confusion groups are reported, among genera within the suborders of Anisoptera and Zygoptera. These confusion groups suggest a need to include other morphological characters that complement the wing information used for the classification of these groups thereby improving accuracy of classification. Likewise, the findings of this work open the door to the application of machine learning methods for the identification of species in Odonata and in insects more broadly which would potentially reduce the impact of the taxonomic impediment.

Keywords: Classification, Machine Learning, supervised, wings

Issue section: Original Article