A Comparison of Analytical Approaches for the Spectral Discrimination and Characterisation of Mite Infestations on Banana Plants

This research investigates the capability of field-based spectroscopy (350–2500 nm) for discriminating banana plants (Cavendish subgroup Williams) infested with spider mites from those unaffected. Spider are considered a major threat to agricultural production, as they occur on over 1000 plant species, including varieties. Plants were grown under controlled glasshouse environment remove any influence other than imposed treatment (presence or absence mites). The spectroradiometer measurements undertaken leaf clip three infestation events. From resultant spectral data, various classification models evaluated partial least squares discriminant analysis (PLSDA), K-nearest neighbour, support vector machines and back propagation neural network. Wavelengths found have significant response presence extracted using competitive adaptive reweighted sampling (CARS), sub-window permutation (SPA) random frog (RF) benchmarked models. CARS SPA provided high detection success (86% prediction accuracy), wavelengths be corresponding red edge near-infrared portions spectrum. As there is limited access operational commercial hyperspectral imaging additional complexity, multispectral camera (Sequoia) was assessed detecting mite impacts plants. Simulated bands able provide level accuracy (prediction 82%) based PLSDA model, band being most important, followed by edge, green bands. Multispectral vegetation indices trialled simple threshold-based method normalised difference index (GNDVI), which achieved 82% accuracy. investigation determined that remote sensing approaches can an accurate infestations, sensors having potential more commercially accessible means outbreaks.