Classification of Water Stress in Sunagoke Moss Using Color Texture and Neural Networks

Abstract

The general appearance of a plant is the most obvious indicator of its physiological well- being. Color Co-occurrence Matrix (CCM) texture features, extracted from a set of 1095 images were used to classify water stress in Sunagoke moss (Rhacomitrium canescens) using Neural Networks (NN). An Excess Green Water Stress Index (EGWSI) was developed and used to quantify water stress in the sample. The CCM texture features were extracted from: red-green-blue (RGB); hue-saturation-intensity (HSI) and CIE's (Comission Internationale de LEclairage) LAB and XYZ color spaces. The HSI texture features achieved 99.45% water stress classification efficiency. They were followed by RGB, XYZ and LAB texture features with classification efficiencies of 99.07%, 98.83 and 96.3% in that order respectfully. The HSI textures features displayed a higher ability and reliability to classify water stress in Sunagoke moss and can be used for stress detection under varying light intensities. A significant accomplishment of this study was the detection of both flood and draught water stress in a plant that exhibits a high level of desiccation tolerance. This provides an opportunity for the possibility of allowing plants to control their own bioproduction environments.