Hyperspectral Classification Of Invasive Alien Plants (IAPS) From Co-Occurring Mangroves Using Ensemble Learning Model

Though hyperspectral remote sensing has been widely used for vegetation classification, there is a gap in the specific application of these techniques for differentiating invasive alien plants (IAPs) from native mangrove species, as most existing studies have primarily relied on image-based approaches rather than detailed spectral data analysis. This study explores the use of hyperspectral remote sensing and classification algorithms to differentiate between mangrove species and IAPs. Laboratory spectra from 10 species (7 mangroves, 3 IAPs) were analyzed across three spectral modes: reflectance spectra (RS), first derivative spectra (FD), and second derivative spectra (SD). A one-way ANOVA identified significant wavelengths, which were reduced using Principal Component Analysis (PCA) and Recursive Feature Elimination using cross-validation (RFECV). Classification models, namely SVM, KNN, and an SVM+KNN ensemble were used to classify the spectra and evaluated using stratified k-fold cross-validation. Results showed that significant wavelengths were consistently found in the Visible (V) and Near Infrared (N) regions, highlighting their importance for accurate species differentiation. RS and FD modes yielded high classification accuracy, with KNN slightly outperforming other models. The SVM + KNN ensemble improved performance in SD mode $(\mathrm{OA}=0.9)$ as compared to SVM $(\mathrm{OA}=0.85)$ and $\mathrm{KNN}(\mathrm{OA}=0.83)$. The study demonstrates the potential of spectral data and machine learning models for accurate IAP detection, offering a valuable tool for managing invasive species.