Unmanned Aerial Vehicle (UAV)-Based Mapping of Iris Pseudacorus L. Invasion in Laguna del Sauce (Uruguay) Coast

TL;DR

The paper addresses the challenge of mapping an invasive aquatic plant, Iris pseudacorus L. (Yellow Flag Iris), in Laguna del Sauce, Uruguay, to support management actions. It introduces a UAV-based multispectral workflow using a small quadcopter and a four-band camera, coupled with a semi-supervised detection pipeline that leverages spectral indices $NDVI = \frac{NIR-RED}{NIR+RED}$, $CI\text{-}GREEN = \frac{NIR}{GREEN} - 1$, and $CI\text{-}EDGE = \frac{NIR}{RED EDGE} - 1$ to generate hypothesis regions, subsequently refined with Top-Hat filtering and morphological operations. Starting from 10,517 hypothesis regions, the method reduces to 524 connected regions likely containing YFI, validated on a separate image set, demonstrating robustness to timing differences. The approach offers a low-cost, scalable solution for high-resolution IAS monitoring that can inform management decisions, buffer-zone planning, and integration with citizen-science observations like iNaturalist.

Abstract

Biological invasions pose a significant threat to the sustainability of water sources. Efforts are increasingly being made to prevent invasions, eradicate established invaders, or control them. Remote sensing (RS) has long been recognized as a potential tool to aid in this effort, for example, by mapping the distribution of invasive species or identifying areas at risk of invasion. This paper provides a detailed explanation of a process for mapping the actual distribution of invasive species. This article presents a case studie on the detection of invasive Iris Pseudacorus L. using multispectral data captured by small Unmanned Aerial Vehicles (UAVs). The process involved spectral feature mapping followed by semi-supervised classification, which produced accurate maps of these invasive.

Figures (10)

• Figure 1: On the left of the figure, the distribution map of the Yellow Flag Iris in Uruguay is shown. The study area in Laguna del Sauce (Maldonado/Uruguay) is indicated on the right. Source: Developed by the authors based on the webpage NaturalistaUYNaturalistaUY and the reserach conducted by G. Minuti minuti2022
• Figure 2: This figure shows the original image captured on the left, and the red rectangle defines the manual labeled YFI regions sector. The right image shows in detail the sets of connected-pixels in the mask $\mathbf{K}$, with a different color for each disjoint set of the list $[t^{(i)}]_{i=1,...,L}$.
• Figure 3: The figure depicts how OPEN and CLOSE morphological operations perform on 1D signal (not binary). Source: image modified from gonzalez:2018.
• Figure 4: The figure shows the five channels employed to detect the first hypotheses with YFI plants. The second and third rows depict the result of the OPEN and CLOSE operations, respectively, using the same kernel.
• Figure 5: The figure shows binarization on the five channels using the thresholds found by the OPEN and CLOSE channels maps on the overall image.