Impatiens glandulifera, an invasive plant species spreading rapidly throughout Europe, especially along riparian habitats pose a risk to native ecosystems. Hence, it is essential to establish cost-effective methods that enable easy monitoring of the invasive plants. We are developing a methodology for automated, close-range remote sensing using Unmanned Aerial Vehicle (UAV) for the acquisition of high-resolution spectral data, to investigate the population dynamics and effects of I. glandulifera on ecosystem services. We measure the impact of I. glandulifera on soil erosion in river banks within the study area, as the invasive plant was shown to potentially enhance erosion.
The study areas are located within ~25kms from Bayreuth, in riparian areas with known existence of I. glandulifera. For image acquisition, a SOLO (multicopter) from 3DR best-suited UAV for the terrain we were flying over combined with a Parrot Sequoia multiband sensor onboard was used. The post-processing was carried out with Agisoft, ArcMap and QGIS software.
At first, a pixel-based approach was carried out using different classification methods: Normalized Difference Red Edge, Unsupervised/Supervised-classification, Green/Normalised Difference Vegetation Index (GNDVI/NDVI) with GNDVI accuracy being highest among these. Since the pixel-based approach did not yield needed accuracy for distinguishing the invasive plant from the native vegetation we are currently developing rule sets for object-based approach, experimenting with scale, shape and compactness of a multi-resolution segmentation algorithm to produce image objects at the best possible scale in eCognition. Based on its morphology I. glandulifera is easily distinguishable from the native vegetation in its mature stage as it grows taller than the rest of the plants and has very conspicuous pink flowers.
We anticipate this would be the best workflow for very high-resolution images acquired from UAV.