Decision Support System designed to detect yellow mosaic in Pigeon pea using Computer Vision

Abstract

A real-time identification system for early detection of yellow mosaic disease outbreaks in Pigeon pea has been developed in this study. The prototype is currently for Pigeon pea (Cajanus cajan) species but it is extensible to other species.  The drawback associated with the detection of plant diseases is that it needs an expert from the field who is visually able to distinguish characteristic features and give results. The color feature extraction can increase the results significantly but once a color change is visible to the naked eye, it is too late to apply any remedy. Hence, we are using morphological features to identify diseased leaves. To be able to automate this process we need to identify the key features which can be extracted visually from a leaf and then use them to model the leaf of an infected plant. The selection of appropriate features and the algorithm used for learning are the critical points in this study. We have evaluated the possible features for the identification of diseased leaves and then used them to classify leaves as either infected or healthy using the Classification and Regression Trees decision tree approach. CART is most suited to the data we generate using leaf features as these features are numeric and since they are fourteen in number, it makes them unsuitable for regression.  Several basic morphological, geometric, textural features can be extracted from a plant leaf, but the selection of meaningful features gets us promising results.