Spectral-Based Classification of Plant Species Groups and Functional Plant Parts in Managed Permanent Grassland

Grassland vegetation typically comprises the species groups grasses, herbs, and legumes. These provide different functional traits feed values. Therefore, knowledge of botanical composition grasslands can enable improved site-specific management livestock feeding. A systematic approach was developed to analyze managed permanent grassland using hyperspectral imaging in a laboratory setting. In first step, images typical plants were recorded, annotated, classified according group plant parts, that is, flowers, leaves, stems. second three machine learning model types—multilayer perceptron (MLP), random forest (RF), partial least squares discriminant analysis (PLS-DA)—were trained with pixel-wise spectral information discriminate parts individual models. The influence radiometric data calibration specific preprocessing steps on overall performance also investigated. While proper negligible our setting, variants, including smoothening derivation spectrum, found be beneficial for classification accuracy. Compared extensively preprocessed data, raw yielded no statistically decreased most cases. Overall, MLP models outperformed PLS-DA RF reached cross-validation accuracies 96.8% 88.6% part classification. obtained insights an essential basis future acquisition vegetation.