High-Resolution Land Cover Classification Using UAV-Based Hyperspectral Imaging from a Fixed-Wing–VTOL Platform

The advent of unmanned aerial vehicles (UAVs) has significantly advanced Earth observation by providing high-spatial-resolution remote sensing datasets. Among different types of UAVs, fixed-wing drones fly faster and cover longer distances in a single flight. When equipped with hyperspectral sensors, these drones can provide high-resolution data in both the spectral and spatial domains over very large areas in a single flight, with potential applications such as large-scale mapping of real-world objects. Here, a pilot study was carried out by integrating the Headwall Nano-Hyperspec hyperspectral sensor mounted on a fixed-wing–vertical take-off and landing (VTOL) platform. The study was conducted over a 105-acre area with mixed land cover in Tamil Nadu, Southern India. At a flying height of 100m, it generated high-resolution imagery consisting of 269 bands in the spectral range of 400-1000nm, with a spatial resolution of 15 cm. The pre-processed imagery was subjected to dimensionality reduction using the Minimum Noise Fraction (MNF) Transform, followed by supervised classification using machine learning (ML) algorithms, namely artificial neural networks (ANNs), Gaussian process (GP), k-nearest neighbor (KNN), support vector machine (SVM), and random forest (RF). We used the first 10 MNF components, which hold 80.72% of the cumulative variance, to evaluate the ML models, with 300 ground points per class, split 70:30 for training and validation. The GPC outperformed the other models, with an overall accuracy of 85%, followed by RF (84.22%), ANN (82.44%), KNN (78.22%), and finally SVM (77.56%). The best-performing GPC was used to generate the land use/land cover map of the present study area, showing five classes: soil, waterbody, mixed vegetation/trees, built-up, and agricultural land. The resulting high-resolution land cover map demonstrates the potential of fixed-wing UAVs equipped with hyperspectral sensors for accurate, detailed mapping of land surface features across larger areas.