Artifcial neural networks in online semiautomated pest discriminability: an applied case with 2 Trips species

Abstract

Being faced with practical problems in pest identifcation, we present a methodical paper based on artifcial neural networksto discriminate morphologically very similar species, Trips sambuci Heeger, 1854 and Trips fuscipennis Haliday, 1836 (Tysanoptera:Tripinae), as an applied case for more general use. Te artifcially intelligent system may be successfully applied as a credible, online,semiautomated identifcation tool that extracts hidden information from noisy data, even when the standard characters have muchoverlap and the common morphological keys hint at the practical problem of high morphological plasticity. Statistical analysis of 17characters, measured or determined for each Trips fuscipennis and T. sambuci specimen (reared from larvae in our laboratories),including 15 quantitative morphometric variables, was performed to elucidate morphological plasticity, detect eventual outliers, andvisualize diferences between the studied taxa. Te computational strategy applied in this study includes a set of statistical tools (factoranalysis, correlation analysis, principal component analysis, and linear discriminant analysis) followed by the application of a multilayerperceptron artifcial neural network system, which models functions of almost arbitrary complexity. Tis complex approach has proventhe existence of 2 separate species: T. fuscipennis and T. sambuci. All the specimens could be clearly distinguished with 2 distinctsubgroups for each species, determined by sex. In conclusion, the use of an optimal 3-layer ANN architecture (17, 4, 1) enables fast andreliable 100% classifcation as proven during the extensive verifcation process.