Multispectral detection of invasive alien plants from very high resolution 8-band satellite imagery using probabilistic graphical models

This paper describes the use of probabilistic graphical models, in particular Bayesian networks (BN), for the detection of two problem invasive alien plants, C. odorata and L. camara, from the recently-launched 8-band Worldview-2 satellite imagery acquired on 26 September 2010 over central Swaziland. The main objective of this work was to apply and evaluate the efficacy of the very high resolution multispectral satellite imagery for invasive alien plant detection using BN models. The posterior probabilities (parameters) were estimated using the expectationmaximization (EM) algorithms. A comparison of the results obtained from the models indicates excellent classification accuracies above 95%, and Kappa coefficient values above 0.90. The highest classification accuracies of 99% were obtained when using the newer bands. The main result obtained in this study is that the new Worldview-2 bands are effective for alien invasive species detection. The BN approach considered here provide relatively similar and accurate solutions for the classification of the multispectral image although the L. camara model was marginally competitive relative to the C. odorata model when measured in terms of the Brier score and the logarithmic loss. The results point to the potential usefulness and applicability of the additional bands from the Worldview-2 satellite for very high resolution plant species detection and monitoring.