Detecting and Mapping Invasive Species Across Riparian Corridors via Object Detection Approaches in UAV Imagery: An Example of Impatiens glandulifera

TL;DR

A new tool uses drone images and computer vision to detect invasive Impatiens glandulifera flowers in hard-to-reach river areas.

Contribution

A semi-automatic thresholding tool (SATT) is introduced for identifying invasive plant flowers in UAV imagery with high precision.

Findings

The SATT achieved 79%-96% precision and 73%-86% mean average precision in identifying Impatiens glandulifera flowers.

The tool converts detection results into GIS-compatible vector formats for mapping invasive species hotspots.

The SATT requires minimal training data and is accessible to nonexperts via a graphical user interface.

Abstract

Riparian zones in the United Kingdom have high species diversity but are prone to anthropogenic changes and alien plant invasions, like
Impatiens glandulifera
. However, identification can be challenging due to poor accessibility or visibility via tree canopies. UAVs provide a means to access previously inaccessible areas and capture imagery of the area. In this study, a method is introduced to identify the flowers of invasive species (
Impatiens glandulifera
) and map their locations using a computer vision framework and oblique image capture methods. The process includes thresholding images, image masking, blurring, ellipsoid shape search, noise reduction, and contour extraction. Locations are determined using camera parameters, EXIF data, and the average flower size, then converted into vector format for GIS software. This method is wrapped into a single executable program named the semi‐automatic thresholding tool (SATT). A validation set of 312 UAV images from the River Elwy, North Wales, showed high precision (79%–96%) and mean average precision (mAP) scores of 73%–86%. This demonstrates that the SATT consistently and correctly identifies
Impatiens glandulifera
flowers from UAV imagery, making it effective for identifying hotspots and targeting management techniques along riparian corridors. The tool has been wrapped into a single‐file executable program with a graphical user interface, enabling nonexperts to use the tool without the need of any software installation. Overall, the tool obtains consistent detection levels of abundance/or flower density across the study site. The tool also does not require an extensive amount of training data, and the intuitive design of the software enables nonexperts to utilize the tool and modify parameter values to adapt it to their needs.

Riparian zones in the United Kingdom are diverse but prone to anthropogenic changes and invasive species like
impatiens glandulifera
. This study introduces a method to identify and map these flowers using a computer vision framework and UAV imagery, wrapped into an executable program called the semi‐automatic thresholding tool (SATT). Validation showed high precision and consistent identification, making the SATT effective for targeting management techniques along riparian corridors and accessible for nonexperts.

Figures