Browsing by Author "Poona, Nitesh Keshavelal"

Now showing 1 - 3 of 3
  • Thumbnail Image
    Item
    Investigating the utility of oblique tree-based ensembles for the classification of hyperspectral data
    (MDPI, 2016-11) Poona, Nitesh Keshavelal; Van Niekerk, Adriaan; Ismail, Riyad
    Ensemble classifiers are being widely used for the classification of spectroscopic data. In this regard, the random forest (RF) ensemble has been successfully applied in an array of applications, and has proven to be robust in handling high dimensional data. More recently, several variants of the traditional RF algorithm including rotation forest (rotF) and oblique random forest (oRF) have been applied to classifying high dimensional data. In this study we compare the traditional RF, rotF, and oRF (using three different splitting rules, i.e., ridge regression, partial least squares, and support vector machine) for the classification of healthy and infected Pinus radiata seedlings using high dimensional spectroscopic data. We further test the robustness of these five ensemble classifiers to reduced spectral resolution by spectral resampling (binning) of the original spectral bands. The results showed that the three oblique random forest ensembles outperformed both the traditional RF and rotF ensembles. Additionally, the rotF ensemble proved to be the least robust of the five ensembles tested. Spectral resampling of the original bands provided mixed results. Nevertheless, the results demonstrate that using spectral resampled bands is a promising approach to classifying asymptomatic stress in Pinus radiata seedlings.
  • Thumbnail Image
    Item
    Modelling water stress in a Shiraz Vineyard using hyperspectral imaging and machine learning
    (MDPI, 2018) Loggenberg, Kyle; Strever, Albert; Greyling, Berno; Poona, Nitesh Keshavelal
    The detection of water stress in vineyards plays an integral role in the sustainability of high-quality grapes and prevention of devastating crop loses. Hyperspectral remote sensing technologies combined with machine learning provides a practical means for modelling vineyard water stress. In this study, we applied two ensemble learners, i.e., random forest (RF) and extreme gradient boosting (XGBoost), for discriminating stressed and non-stressed Shiraz vines using terrestrial hyperspectral imaging. Additionally, we evaluated the utility of a spectral subset of wavebands, derived using RF mean decrease accuracy (MDA) and XGBoost gain. Our results show that both ensemble learners can effectively analyse the hyperspectral data. When using all wavebands (p = 176), RF produced a test accuracy of 83.3% (KHAT (kappa analysis) = 0.67), and XGBoost a test accuracy of 80.0% (KHAT = 0.6). Using the subset of wavebands (p = 18) produced slight increases in accuracy ranging from 1.7% to 5.5% for both RF and XGBoost. We further investigated the effect of smoothing the spectral data using the Savitzky-Golay filter. The results indicated that the Savitzky-Golay filter reduced model accuracies (ranging from 0.7% to 3.3%). The results demonstrate the feasibility of terrestrial hyperspectral imagery and machine learning to create a semi-automated framework for vineyard water stress modelling.
  • Thumbnail Image
    Item
    A remote sensing-machine learning framework for modelling forest health
    (Stellenbosch : Stellenbosch University, 2020-12) Poona, Nitesh Keshavelal; Van Niekerk, Adriaan; Ismail, Riyad; Stellenbosch University. Faculty of Arts and Social Sciences. Dept. of Geography and Environmental Studies.
    ENGLISH ABSTRACT: The utility of remote sensing data, in particular high dimensional spectroscopy data, is now widely used for the detection and monitoring of pest and disease in agriculture and forestry. Coupled with advanced data analytics, spectroscopic data can provide a wealth of information regarding vegetation health, and successfully demonstrates the utility of spectroscopic data and advanced machine learning (ML) algorithms, i.e. tree-based ensemble learners, by developing a remote sensing-machine learning framework for forest health assessment and monitoring. Specifically, the research investigates the use of spectroscopic data for modelling Fusarium circinatum stress in Pinus radiata and Pinus patula. The research first investigated the utility of novel wrapper feature selection algorithms embedded with the random forest (RF) learner to develop classification models for discriminating healthy, infected, and damaged P. radiata and P. patula seedlings within a nursery environment. Results showed that reducing data dimensionality results in improved model accuracies. More importantly, the results showed that the RF-Boruta framework yielded the best results. Two RF variants were subsequently explored, namely oblique random forest (oRF), and rotation forest (rotF). The performances of oRF and rotF were benchmarked against those of traditional RF. All models were evaluated in terms of their ability to discriminate healthy and stressed Pinus seedlings. Spectral resampling was employed to reduce data dimensionality. The oRF model yielded the best results, with oRF svm (oRF employing support vector machine as splitting model) proving to be the most robust. To extend the utility of model building, the research developed normalised difference two-band spectral indices for real-time F. circinatum stress detection. The Boruta algorithm was employed to identify relevant bands, which were used to derive two-band indices. The indices were compared with an extensive list of currently available indices, identified from the literature, to assess the value thereof. Indices were evaluated within univariate and multivariate paradigms, with the latter proving more adept at classifying healthy, damaged, and infected seedlings.The use of high spatial resolution satellite remote sensing imagery for modelling pitch canker in P. radiata trees in a commercial plantation was also evaluated. This exploration served to complement the remote sensing-machine learning framework developed for the nursery environment. In this component of the research, an artificial neural network model was used (whereas tree-based ensemble models were used in the former elements of the research). Results highlight the potential of using high spatial resolution satellite remote sensing for mapping and monitoring of pitch canker infected trees. Overall, the research successfully demonstrated that high spectral and high spatial resolution remotely sensed data, coupled with advanced data analytics, i.e. tree-based ensemble learners and wrapper algorithms, provides a potentially operational and economically viable framework for F. circinatum management within a nursery and plantation environment.