Citrus Gummosis disease severity classification using participatory sensing, remote sensing and weather data

Phythophthora disease, Gummosis, in Citrus crop results in huge yield losses every year. The purpose of this study is to detect the Gummosis disease infested fields using remote sensing and meteorological data. We present the use of participatory sensing framework to collect the reliable ground truth data about the disease incidences. Various vegetation, soil and water indices have been derived from Landsat-8 data and morphometric parameters of watershed from ASTER DEM. These parameters along with meteorological data are used as a large feature set which is then reduced using the LASSO and Elastic Net regularization techniques. We evaluate SVM, Naive Bayes and ANN based classifiers for three class classification between healthy, low and high disease infested fields. Results depicts that a combination of LASSO regularization with ANN classifier outscores the other approaches considered in this paper.