Artificial Neural Network applied as a methodology of mosquito species identification

Abstract

There are about 200 species of mosquitoes (Culicidae) known to be vectors of pathogens that cause diseases in humans. Correct identification of mosquito species is an essential step in the development of effective control strategies for these diseases; recognizing the vectors of pathogens is integral to understanding transmission. Unfortunately, taxonomic identification of mosquitoes is a laborious task, which requires trained experts, and it is jeopardized by the high variability of morphological and molecular characters found within the Culicidae family. In this context, the development of an automatized species identification method would be a valuable and more accessible resource to non-taxonomist and health professionals. In this work, an artificial neural network (ANN) technique was proposed for the identification and classification of 17 species of the genera Anopheles, Aedes, and Culex, based on wing shape characters. We tested the hypothesis that classification using ANN is better than traditional classification by discriminant analysis (DA). Thirty-two wing shape principal components were used as input to a Multilayer Perceptron Classification ANN. The obtained ANN correctly identified species with accuracy rates ranging from 85.70% to 100%, and classified species more efficiently than did the traditional method of multivariate discriminant analysis. The results highlight the power of ANNs to diagnose mosquito species and to partly automatize taxonomic identification. These findings also support the hypothesis that wing venation patterns are species-specific, and thus should be included in taxonomic keys.

Automated identification of insect vectors of Chagas disease in Brazil and Mexico: the Virtual Vector Lab

“…A major source of optimism is that we have achieved very satisfactory correct identification rates working almost exclusively with information on shape of the bugs, which compares well with results of previous such efforts: 84% among five species of papilionid butterflies by Wang et al (2012a), 93% among few species at the ordinal level by Wang et al (2012b), 85.7–100% among 17 species of mosquitoes based on wing-shape characters (Lorenz, Ferraudo & Suesdek, 2015), and 90–98% among seven species of Neuroptera (Yang et al, 2015). From the two-dimensional images available to us, we can, in theory, take advantage of information on shape, size, and coloration.…”

“…Certainly, two-dimensional imagery is easiest to obtain and manage, such that organisms with relatively flattened body forms will be most tractable (e.g., ticks), although wings of other groups (e.g., mosquitoes and sandflies) may also offer opportunities (Zhou, Ling & Rohlf, 1985; Godoy et al, 2014; Lorenz, Ferraudo & Suesdek, 2015). Small size may present some level of challenge, as we will have to manage the complexities of magnification and associated distortion, but our photo apparatus has been designed to allow addition of accessories to facilitate such challenges in imaging.…”

“…Although some advances have been made toward automated identification of mosquitoes (Culicidae) using wing venation (Zhou, Ling & Rohlf, 1985; Lorenz, Ferraudo & Suesdek, 2015), more effort has centered on DNA barcoding as a solution (Cywinska, Hunter & Hebert, 2006; MBI, 2009). For sand flies (Psychodidae), the only steps towards automating identification have been via mass-spectrophotometry of proteins (Mathis et al, 2015) and DNA barcoding (Pinto et al, 2015).…”

Identification of vehicle suspension shock absorber squeak and rattle noise based on wavelet packet transforms and a genetic algorithm-support vector machine

“…The artificial neural network (ANN), support vector machine (SVM) and multiple linear regression (MLR) approaches have been frequently mentioned in the literature [17,18]. Based on extensive empirical research, the ANN and SVM techniques have proven to be very powerful tools to map the nonlinear characteristics between the input features and the output targets, compared to the traditional MLR method [18].…”

Artificial neural networks and geometric morphometric methods as a means for classification: A case‐study using teeth from Carcharhinus sp. (Carcharhinidae)

“…Despite a growing interest in integrating artificial neural networks with GM data, there has been little work on how well these two groups of methods operate relative to or in contrast to each other (although see MacLeod, 2008;Lorenz et al, 2015, Sonnenschein et al, 2015 for exceptions. Sonnenschein et al (2015) used a variety of classification methods, including MLPs, linear discriminate function analysis, and quadratic discriminate function analysis, to classify Drosophila melanogaster wings to sex and genotype based on 12 landmarks and 36 pseudolandmarks.…”