Artificial neural networks in online semiautomated pest discriminability: an applied case with 2 Thrips species

摘要: Being faced with practical problems in pest identification, we present a methodical paper based on artificial neural networks to discriminate morphologically very similar species, Thrips sambuci Heeger, 1854 and fuscipennis Haliday, 1836 (Thysanoptera: Thripinae), as an applied case for more general use. The artificially intelligent system may be successfully credible, online, semiautomated identification tool that extracts hidden information from noisy data, even when the standard characters have much overlap common morphological keys hint at problem of high plasticity. Statistical analysis 17 characters, measured or determined each T. specimen (reared larvae our laboratories), including 15 quantitative morphometric variables, was performed elucidate plasticity, detect eventual outliers, visualize differences between studied taxa. computational strategy this study includes set statistical tools (factor analysis, correlation principal component linear discriminant analysis) followed by application multilayer perceptron network system, which models functions almost arbitrary complexity. This complex approach has proven existence 2 separate species: sambuci. All specimens could clearly distinguished distinct subgroups sex. In conclusion, use optimal 3-layer ANN architecture (17, 4, 1) enables fast reliable 100% classification during extensive verification process.