A statistical approach to identify diseased leaf from healthy

Abstract

The economic advancement of a country depends on the green revolution. Productivity enhancement plays a significant part in improvising green revolution for country like India. Plant nutrition should be monitored to improve the yield in agriculture. This could be done either manually or by automatic procedure. Manual procedure drags the process. So the best way is to identify the mal nutrient of a plant by categorizing the diseased plant from the healthy plant. In this work, the leaf of Pongamia Pinnata species is taken from Mendeley Data. These leaves possess RGB color. They are converted into gray space along with contrast enhancement for further processing. Texture and tone features such as gray-level co-occurrence matric, statistical, histogram and probability measures are extracted from the contrast enhanced images. Statistical-based t test is conducted to find the significant features in categorizing the leaves into diseased and healthy. Among the 29 features, results demonstrate that energy is the most significant ( <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$p=0.0002$</tex> ) feature followed by maximum probability ( <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$p=0.0047$</tex> ) and information measure of correlation ( <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$p=0.0073$</tex> ). A good correlation ( <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$r=0.566$</tex> ) is observed for the feature energy with the out class, namely healthy and diseased. This work thus involves automation in the process of identifying the diseased leaf from healthy.