Detection of Distress and Disease in Deciduous Trees Utilizing Remote Sensing
Presentation Time: Mon, 05/02/2022 - 13:00
Keywords: distressed detection, disease detection, machine learning, multispectral imaging
Locating and identification of plant stress and diseases plays a major role in plant conservation and human safety concerns relating to falling hazards and reduction in fire blocks between structures in medium sized population centers. Overall flora heath can be indicated by visual observations of the chlorophyll and other pigments in the leaves. As outside interference with the plants ability to naturally produce the required nutrients, such as environmental and pathological interference, the visible pigmentation change. In this study, pigment variation is evaluated and analyzed by machine learning methods including image classification for the evaluation of health in deciduous trees. By utilizing multispectral imagery this study compares wavelength values for identified affected individuals showing visual symptoms to located other affected individuals both showing symptomatic and non-symptomatic individuals. Data analysis was conducted utilizing a trained supervised classification, support vector machines and K nearest neighbor method to determine which method was most precise in identifying affected pixels for fast-tracking management evaluations for resource managers. The overall classification accuracy of targeted, healthy, fields, and urban was relatively good, with kappa values ranging from 0.66 to 0.75 and overall accuracy ranging from 70% to 83%. Support vector machines resulted in an accuracy of 82.13% with a kappa coefficient of 0.74 at a 750 point accuracy assessment making it the best method of the two for detections of symptomatic and asymptomatic individuals.
