Peer-reviewed
Detection of <b>Silybum marianum</b> infection with <b>Microbotryum silybum</b> using VNIR field spectroscopy
• Identification of a smut fungus with no visual signs on leaves using non-destructive spectroscopy. • The method was applied in-situ on live plants using low cost tools. • Three innovative classifiers were tested to evaluate their performance. • An independent dataset was used for the validation of the results.
Microbotryum silybum is a smut fungus infecting Silybum marianum (milk thistle) weed and is currently investigated as a means for its biological control. Although the fungus' detection is important for the evaluation of biological control effectiveness and decision making, in-situ diagnosis is not always possible. The presented approach describes the identification of systemically infected S. marianum plants by using field spectroscopy and hierarchical self-organizing maps. An experimental field that contained both healthy and artificially inoculated S. marianum plants was used to acquire leaf spectra using a handheld visible and near-infrared spectrometer (310-1100 nm). Three supervised hierarchical self-organizing models, including Supervised Kohonen Network (SKN), Counter propagation Artificial Neural Network (CP-ANN) and XY-Fusion network (XY-F) were utilized for the identification of the systemically infected S. marianum plants. As input features to the classifiers, the pre-processed spectral signatures were used. The pre-processing of the spectra included normalisation, second derivative and principal component extraction. The systemically infected S. marianum identification rates using SKN and CP-ANN reached high overall accuracy (up to 90%) and even higher using the XY-F (95.16%). The results demonstrate the potential for a high accuracy identification of systemically infected S. marianum plants during vegetative growth, with the assistance of hierarchical self-organizing maps
Microbotryum silybum is a smut fungus infecting Silybum marianum (milk thistle) weed and is currently investigated as a means for its biological control. Although the fungus' detection is important for the evaluation of biological control effectiveness and decision making, in-situ diagnosis is not always possible. The presented approach describes the identification of systemically infected S. marianum plants by using field spectroscopy and hierarchical self-organizing maps. An experimental field that contained both healthy and artificially inoculated S. marianum plants was used to acquire leaf spectra using a handheld visible and near-infrared spectrometer (310-1100 nm). Three supervised hierarchical self-organizing models, including Supervised Kohonen Network (SKN), Counter propagation Artificial Neural Network (CP-ANN) and XY-Fusion network (XY-F) were utilized for the identification of the systemically infected S. marianum plants. As input features to the classifiers, the pre-processed spectral signatures were used. The pre-processing of the spectra included normalisation, second derivative and principal component extraction. The systemically infected S. marianum identification rates using SKN and CP-ANN reached high overall accuracy (up to 90%) and even higher using the XY-F (95.16%). The results demonstrate the potential for a high accuracy identification of systemically infected S. marianum plants during vegetative growth, with the assistance of hierarchical self-organizing maps
Article, 2017