Using Boruta-Selected Spectroscopic Wavebands for the Asymptomatic Detection of Fusarium Circinatum Stress

Abstract

High spectral resolution multitemporal data were used to model asymptomatic stress caused by Fusarium circinatum in 3-month old Pinus radiata seedlings. The objectives of the study were: 1) to identify an optimal subset of wavebands that could model asymptomatic stress in P. radiata seedlings and 2) to develop a robust classification model for discriminating healthy and stressed seedlings. To achieve these objectives, spectral data were collected for healthy, infected, and damaged seedlings using a hand-held field spectroradiometer. The data were analyzed, first for combined classes and then for class pairs using the Boruta algorithm. Results indicated that the best discrimination was possible at week three for all classes, with a KHAT value of 0.79 and an out of bag error of 14.00% ( ), using a subset of 107 wavebands. A closer examination of the class pairs, namely healthy-infected (H-I) and infected-damaged (I-D), showed improved discrimination with KHAT values of 0.82 and 0.84, respectively. The H-I class pair was classified using a subset of just 38 wavebands, whereas the I-D class pair was classified using a subset of just 40 wavebands. Overall, this study demonstrated that it is more difficult to discriminate asymptomatic stress when additional stress related classes are present. Nonetheless, the methodology developed in this study has the potential to be operationalized within a nursery environment for the early detection of F. circinatum-induced stress in P. radiata seedlings.

Modelling Water Stress in a Shiraz Vineyard Using Hyperspectral Imaging and Machine Learning

“…MDA measures the changes in OOB error, which results from comparing the OOB error of the original dataset to that of a dataset created through random permutations of variable values. In this study, MDA was utilised to compute VI following the recommendations of [22,54,55]. The MDA VI for a waveband X j is defined by [56]:…”

“…Dimensionality reduction of hyperspectral data using machine learning has been extensively researched (for example see [20,22,23]). The results of our study indicate the VI ranking provided by RF and XGBoost can successfully be used to select a subset of wavebands for classification.…”

Investigating the Utility of Oblique Tree-Based Ensembles for the Classification of Hyperspectral Data

“…For the analysis of plant stress, the high spectral resolution allows for the detection and quantification of a plant’s physiological response to stress [2]. This physiological response is exhibited as subtle variations in a plant’s spectral response, providing the basis for developing stress detection models [3,4]. Hyperspectral data subsequently provides the opportunity to readily monitor pest and disease stress in agricultural crops and forestry, as demonstrated by [3,4,5,6] and others.…”

Agro-Ecological Class Stability Decreases in Response to Climate Change Projections for the Pacific Northwest, USA

“…Similarly, with our dataset, using Random Forest to predict current AECs resulted in comparable overall accuracy and kappa with cross validation on different sets of selected variables using two different variable selection methods. Elsewhere, the Random Forest approach has been successfully applied in various fields of research ranging from micro-array data analysis (Díaz-Uriarte and Alvarez de Andrés, 2006), hyper-spectral data analysis (Adam et al, 2012;Poona and Ismail, 2014), current species distribution studies, as well as in predicting distributions with changing future climate (Schrag et al, 2008;Watling et al, 2012;Langdon and Lawler, 2015). Random Forest belongs to the algorithmic modeling culture (Breiman, 2001) in statistical modeling where the data mechanisms are complex and not known.…”