UAVs and Machine Learning Revolutionising Invasive Grass and Vegetation Surveys in Remote Arid Lands

Abstract

The monitoring of invasive grasses and vegetation in remote areas is challenging, costly, and on the ground sometimes dangerous. Satellite and manned aircraft surveys can assist but their use may be limited due to the ground sampling resolution or cloud cover. Straightforward and accurate surveillance methods are needed to quantify rates of grass invasion, offer appropriate vegetation tracking reports, and apply optimal control methods. This paper presents a pipeline process to detect and generate a pixel-wise segmentation of invasive grasses, using buffel grass (Cenchrus ciliaris) and spinifex (Triodia sp.) as examples. The process integrates unmanned aerial vehicles (UAVs) also commonly known as drones, high-resolution red, green, blue colour model (RGB) cameras, and a data processing approach based on machine learning algorithms. The methods are illustrated with data acquired in Cape Range National Park, Western Australia (WA), Australia, orthorectified in Agisoft Photoscan Pro, and processed in Python programming language, scikit-learn, and eXtreme Gradient Boosting (XGBoost) libraries. In total, 342,626 samples were extracted from the obtained data set and labelled into six classes. Segmentation results provided an individual detection rate of 97% for buffel grass and 96% for spinifex, with a global multiclass pixel-wise detection rate of 97%. Obtained results were robust against illumination changes, object rotation, occlusion, background cluttering, and floral density variation.

Prediction Model with High-Performance Constitutive Androstane Receptor (CAR) Using DeepSnap-Deep Learning Approach from the Tox21 10K Compound Library

“…For the RF and XGB, molecular descriptors were calculated using a Python package called Mordred () [64]. Classification experiments were conducted in the Python programming language using specific classifier implementations, RF, and XGB provided by the scikit-learn and rdkit Python packages [76,77,78,79].…”

Technology innovation: advancing capacities for the early detection of and rapid response to invasive species

“…Examples of invasive species detected through the use of drones in combination with image processing include: yellow flag iris (Iris pseudacorus; Baron et al 2018), invasive grasses (Cenchrus ciliaris and Triodia spp. ; Sandino et al 2018), Burmese pythons (Python bivittatus; Gomes 2017), and silk oak trees (Grevillea robusta; Strohecker 2017).…”

“…For example, researchers at the University of Alberta used machine learning to analyze the factors that led to success or failure in 143 attempts to eradicate aquatic invasive species and developed a decision tree for responders in the field to guide their choice of response strategy (Xiao et al 2018). Campbell et al (2015) predict that drones, in combination with infrared cameras, preprogrammed night flights, bait delivery mechanisms, and AI data processing, will be widely adopted for invasive rodent eradication programs within the next five years (see also Sandino et al 2018). Below we provide a brief overview of response tools not previously discussed in this paper.…”

A Sense of Scale

“…We found that the use of the non-destructive SfM method accurately estimated AGB in plots of mixed grasses, including annual and perennial, and forbs in low elevation sagebrush steppe ecosystems. Often a spectral classification is used to discriminate between PFT's or vegetation species before analysis (Akar et al, 2017;Gaston et al, 2018;Jing et al, 2017;Lu & He, 2017). Spectral data from images in this study were not used because they were not radiometrically corrected.…”

“…If available, more training data for sparse classes will improve the accuracy in the classification. The advent of unmanned aerial vehicles (UAVs) allows for field data collection of a large area in less time compared to a field crew (Gaston et al, 2018). UAV flights could be instrumental for collecting test and training data for landscape scale observations and provide imagery to develop a more accurate percent cover of vegetation for individual pixels (Breckenridge et al, 2012).…”